Skip to contentSkip to conversation

Second InfiniBand merge candidate patch set

PATCH Superseded Roland Dreier 5 participants 65 messages Dec 20, 2004 – Dec 29, 2004

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

v4 cover Second InfiniBand merge candidate patch set

The following series of patches is the latest version of the OpenIB InfiniBand drivers. We believe that this version is suitable for merging when 2.6.11 opens (or into -mm immediately), although of course we are willing to go through as many more iterations as required to fix any remaining issues.

The previous posting last week did not generate many comments. We have fixed all the minor glitches identified, and updated the copyright/license comments in the source files so that no external files or URLs other than the main COPYING file are referenced (the license remains unchanged as dual GPL/BSD).

In addition a fair bit of testing has been done and several bugs have been fixed over the past week. The IP-over-InfiniBand driver now appears to work well on a broad range of hardware.

Merging these patches should be low-risk: the only non-documentation changes outside of the new drivers/infiniband tree are a small set of networking patches to handle devices of type AF_INFINIBAND where appropriate. These patches have been posted to [email protected] multiple times.

Linus/Andrew - is there any further that you want to see done, or do these patches look acceptable to you?

Thanks,
Roland Dreier
OpenIB Alliance
www.openib.org

#1 lore raw

PATCH v4 1/24 Add core InfiniBand support (public headers)

Add public headers for core InfiniBand support. This can be thought of as a midlayer that provides an abstraction between low-level hardware drivers and upper level protocols (such as IP-over-InfiniBand).

drivers/infiniband/include/ib_cache.h new file +53 -0
@@ -0,0 +1,53 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: ib_cache.h 1349 2004-12-16 21:09:43Z roland $
+ */
+
+#ifndef _IB_CACHE_H
+#define _IB_CACHE_H
+
+#include <ib_verbs.h>
+
+int ib_cached_gid_get(struct ib_device    *device,
+		      u8                   port,
+		      int                  index,
+		      union ib_gid        *gid);
+int ib_cached_pkey_get(struct ib_device    *device_handle,
+		       u8                   port,
+		       int                  index,
+		       u16                 *pkey);
+int ib_cached_pkey_find(struct ib_device    *device,
+			u8                   port,
+			u16                  pkey,
+			u16                 *index);
+
+#endif /* _IB_CACHE_H */
drivers/infiniband/include/ib_fmr_pool.h new file +92 -0

92 lines not on this page · raw message

drivers/infiniband/include/ib_pack.h new file +245 -0

245 lines not on this page · raw message

drivers/infiniband/include/ib_verbs.h new file +1236 -0

1,236 lines not on this page · raw message

#2 lore raw

PATCH v4 2/24 Add core InfiniBand support

Add implementation of core InfiniBand support. This can be thought of as a midlayer that provides an abstraction between low-level hardware drivers and upper level protocols (such as IP-over-InfiniBand).

drivers/infiniband/Kconfig new file +10 -0
@@ -0,0 +1,10 @@
+menu "InfiniBand support"
+
+config INFINIBAND
+	tristate "InfiniBand support"
+	---help---
+	  Core support for InfiniBand (IB).  Make sure to also select
+	  any protocols you wish to use as well as drivers for your
+	  InfiniBand hardware.
+
+endmenu
drivers/infiniband/Makefile new file +1 -0
@@ -0,0 +1 @@
+obj-$(CONFIG_INFINIBAND)		+= core/
drivers/infiniband/core/Makefile new file +6 -0
@@ -0,0 +1,6 @@
+EXTRA_CFLAGS += -Idrivers/infiniband/include
+
+obj-$(CONFIG_INFINIBAND) +=	ib_core.o
+
+ib_core-y :=			packer.o ud_header.o verbs.o sysfs.o \
+				device.o fmr_pool.o cache.o
drivers/infiniband/core/cache.c new file +328 -0

328 lines not on this page · raw message

drivers/infiniband/core/core_priv.h new file +52 -0

52 lines not on this page · raw message

drivers/infiniband/core/device.c new file +614 -0

614 lines not on this page · raw message

drivers/infiniband/core/fmr_pool.c new file +507 -0

507 lines not on this page · raw message

drivers/infiniband/core/packer.c new file +201 -0

201 lines not on this page · raw message

drivers/infiniband/core/sysfs.c new file +695 -0

695 lines not on this page · raw message

drivers/infiniband/core/ud_header.c new file +365 -0

365 lines not on this page · raw message

drivers/infiniband/core/verbs.c new file +433 -0

221 lines not on this page · raw message

The rest of this diff is not shown. View the raw message.

Part of the diff is not shown here. View the raw message.

#3 lore raw

PATCH v4 3/24 Hook up drivers/infiniband

Add the appropriate lines to drivers/Kconfig and drivers/Makefile so that the kernel configuration and build systems know about drivers/infiniband.

drivers/Kconfig +2 -0
@@ -54,4 +54,6 @@
 
 source "drivers/usb/Kconfig"
 
+source "drivers/infiniband/Kconfig"
+
 endmenu
drivers/Makefile +1 -0
@@ -59,4 +59,5 @@
 obj-$(CONFIG_EISA)		+= eisa/
 obj-$(CONFIG_CPU_FREQ)		+= cpufreq/
 obj-$(CONFIG_MMC)		+= mmc/
+obj-$(CONFIG_INFINIBAND)	+= infiniband/
 obj-y				+= firmware/
#4 lore raw

PATCH v4 4/24 Add InfiniBand MAD (management datagram) support (public headers)

Add public headers for handling InfiniBand MADs (management datagrams), including sending and receiving MADs as well as passing MADs on to local agents.

drivers/infiniband/include/ib_mad.h new file +404 -0
@@ -0,0 +1,404 @@
+/*
+ * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
+ * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
+ * Copyright (c) 2004 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
+ * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id$
+ */
+
+#if !defined( IB_MAD_H )
+#define IB_MAD_H
+
+#include <ib_verbs.h>
+
+/* Management base version */
+#define IB_MGMT_BASE_VERSION			1
+
+/* Management classes */
+#define IB_MGMT_CLASS_SUBN_LID_ROUTED		0x01
+#define IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE	0x81
+#define IB_MGMT_CLASS_SUBN_ADM			0x03
+#define IB_MGMT_CLASS_PERF_MGMT			0x04
+#define IB_MGMT_CLASS_BM			0x05
+#define IB_MGMT_CLASS_DEVICE_MGMT		0x06
+#define IB_MGMT_CLASS_CM			0x07
+#define IB_MGMT_CLASS_SNMP			0x08
+#define IB_MGMT_CLASS_VENDOR_RANGE2_START	0x30
+#define IB_MGMT_CLASS_VENDOR_RANGE2_END		0x4F
+
+/* Management methods */
+#define IB_MGMT_METHOD_GET			0x01
+#define IB_MGMT_METHOD_SET			0x02
+#define IB_MGMT_METHOD_GET_RESP			0x81
+#define IB_MGMT_METHOD_SEND			0x03
+#define IB_MGMT_METHOD_TRAP			0x05
+#define IB_MGMT_METHOD_REPORT			0x06
+#define IB_MGMT_METHOD_REPORT_RESP		0x86
+#define IB_MGMT_METHOD_TRAP_REPRESS		0x07
+
+#define IB_MGMT_METHOD_RESP			0x80
+
+#define IB_MGMT_MAX_METHODS			128
+
+#define IB_QP0		0
+#define IB_QP1		__constant_htonl(1)
+#define IB_QP1_QKEY	0x80010000
+
+struct ib_grh {
+	u32		version_tclass_flow;
+	u16		paylen;
+	u8		next_hdr;
+	u8		hop_limit;
+	union ib_gid	sgid;
+	union ib_gid	dgid;
+} __attribute__ ((packed));
+
+struct ib_mad_hdr {
+	u8	base_version;
+	u8	mgmt_class;
+	u8	class_version;
+	u8	method;
+	u16	status;
+	u16	class_specific;
+	u64	tid;
+	u16	attr_id;
+	u16	resv;
+	u32	attr_mod;
+} __attribute__ ((packed));
+
+struct ib_rmpp_hdr {
+	u8	rmpp_version;
+	u8	rmpp_type;
+	u8	rmpp_rtime_flags;
+	u8	rmpp_status;
+	u32	seg_num;
+	u32	paylen_newwin;
+} __attribute__ ((packed));
+
+struct ib_mad {
+	struct ib_mad_hdr	mad_hdr;
+	u8			data[232];
+} __attribute__ ((packed));
+
+struct ib_rmpp_mad {
+	struct ib_mad_hdr	mad_hdr;
+	struct ib_rmpp_hdr	rmpp_hdr;
+	u8			data[220];
+} __attribute__ ((packed));
+
+struct ib_vendor_mad {
+	struct ib_mad_hdr	mad_hdr;
+	struct ib_rmpp_hdr	rmpp_hdr;
+	u8			reserved;
+	u8			oui[3];
+	u8			data[216];
+} __attribute__ ((packed));
+
+struct ib_mad_agent;
+struct ib_mad_send_wc;
+struct ib_mad_recv_wc;
+
+/**
+ * ib_mad_send_handler - callback handler for a sent MAD.
+ * @mad_agent: MAD agent that sent the MAD.
+ * @mad_send_wc: Send work completion information on the sent MAD.
+ */
+typedef void (*ib_mad_send_handler)(struct ib_mad_agent *mad_agent,
+				    struct ib_mad_send_wc *mad_send_wc);
+
+/**
+ * ib_mad_snoop_handler - Callback handler for snooping sent MADs.
+ * @mad_agent: MAD agent that snooped the MAD.
+ * @send_wr: Work request information on the sent MAD.
+ * @mad_send_wc: Work completion information on the sent MAD.  Valid
+ *   only for snooping that occurs on a send completion.
+ *
+ * Clients snooping MADs should not modify data referenced by the @send_wr
+ * or @mad_send_wc.
+ */
+typedef void (*ib_mad_snoop_handler)(struct ib_mad_agent *mad_agent,
+				     struct ib_send_wr *send_wr,
+				     struct ib_mad_send_wc *mad_send_wc);
+
+/**
+ * ib_mad_recv_handler - callback handler for a received MAD.
+ * @mad_agent: MAD agent requesting the received MAD.
+ * @mad_recv_wc: Received work completion information on the received MAD.
+ *
+ * MADs received in response to a send request operation will be handed to
+ * the user after the send operation completes.  All data buffers given
+ * to registered agents through this routine are owned by the receiving
+ * client, except for snooping agents.  Clients snooping MADs should not
+ * modify the data referenced by @mad_recv_wc.
+ */
+typedef void (*ib_mad_recv_handler)(struct ib_mad_agent *mad_agent,
+				    struct ib_mad_recv_wc *mad_recv_wc);
+
+/**
+ * ib_mad_agent - Used to track MAD registration with the access layer.
+ * @device: Reference to device registration is on.
+ * @qp: Reference to QP used for sending and receiving MADs.
+ * @recv_handler: Callback handler for a received MAD.
+ * @send_handler: Callback handler for a sent MAD.
+ * @snoop_handler: Callback handler for snooped sent MADs.
+ * @context: User-specified context associated with this registration.
+ * @hi_tid: Access layer assigned transaction ID for this client.
+ *   Unsolicited MADs sent by this client will have the upper 32-bits
+ *   of their TID set to this value.
+ * @port_num: Port number on which QP is registered
+ */
+struct ib_mad_agent {
+	struct ib_device	*device;
+	struct ib_qp		*qp;
+	ib_mad_recv_handler	recv_handler;
+	ib_mad_send_handler	send_handler;
+	ib_mad_snoop_handler	snoop_handler;
+	void			*context;
+	u32			hi_tid;
+	u8			port_num;
+};
+
+/**
+ * ib_mad_send_wc - MAD send completion information.
+ * @wr_id: Work request identifier associated with the send MAD request.
+ * @status: Completion status.
+ * @vendor_err: Optional vendor error information returned with a failed
+ *   request.
+ */
+struct ib_mad_send_wc {
+	u64			wr_id;
+	enum ib_wc_status	status;
+	u32			vendor_err;
+};
+
+/**
+ * ib_mad_recv_buf - received MAD buffer information.
+ * @list: Reference to next data buffer for a received RMPP MAD.
+ * @grh: References a data buffer containing the global route header.
+ *   The data refereced by this buffer is only valid if the GRH is
+ *   valid.
+ * @mad: References the start of the received MAD.
+ */
+struct ib_mad_recv_buf {
+	struct list_head	list;
+	struct ib_grh		*grh;
+	struct ib_mad		*mad;
+};
+
+/**
+ * ib_mad_recv_wc - received MAD information.
+ * @wc: Completion information for the received data.
+ * @recv_buf: Specifies the location of the received data buffer(s).
+ * @mad_len: The length of the received MAD, without duplicated headers.
+ *
+ * For received response, the wr_id field of the wc is set to the wr_id
+ *   for the corresponding send request.
+ */
+struct ib_mad_recv_wc {
+	struct ib_wc		*wc;
+	struct ib_mad_recv_buf	recv_buf;
+	int			mad_len;
+};
+
+/**
+ * ib_mad_reg_req - MAD registration request
+ * @mgmt_class: Indicates which management class of MADs should be receive
+ *   by the caller.  This field is only required if the user wishes to
+ *   receive unsolicited MADs, otherwise it should be 0.
+ * @mgmt_class_version: Indicates which version of MADs for the given
+ *   management class to receive.
+ * @oui: Indicates IEEE OUI when mgmt_class is a vendor class
+ *   in the range from 0x30 to 0x4f. Otherwise not used.
+ * @method_mask: The caller will receive unsolicited MADs for any method
+ *   where @method_mask = 1.
+ */
+struct ib_mad_reg_req {
+	u8	mgmt_class;
+	u8	mgmt_class_version;
+	u8	oui[3];
+	DECLARE_BITMAP(method_mask, IB_MGMT_MAX_METHODS);
+};
+
+/**
+ * ib_register_mad_agent - Register to send/receive MADs.
+ * @device: The device to register with.
+ * @port_num: The port on the specified device to use.
+ * @qp_type: Specifies which QP to access.  Must be either
+ *   IB_QPT_SMI or IB_QPT_GSI.
+ * @mad_reg_req: Specifies which unsolicited MADs should be received
+ *   by the caller.  This parameter may be NULL if the caller only
+ *   wishes to receive solicited responses.
+ * @rmpp_version: If set, indicates that the client will send
+ *   and receive MADs that contain the RMPP header for the given version.
+ *   If set to 0, indicates that RMPP is not used by this client.
+ * @send_handler: The completion callback routine invoked after a send
+ *   request has completed.
+ * @recv_handler: The completion callback routine invoked for a received
+ *   MAD.
+ * @context: User specified context associated with the registration.
+ */
+struct ib_mad_agent *ib_register_mad_agent(struct ib_device *device,
+					   u8 port_num,
+					   enum ib_qp_type qp_type,
+					   struct ib_mad_reg_req *mad_reg_req,
+					   u8 rmpp_version,
+					   ib_mad_send_handler send_handler,
+					   ib_mad_recv_handler recv_handler,
+					   void *context);
+
+enum ib_mad_snoop_flags {
+	/*IB_MAD_SNOOP_POSTED_SENDS	   = 1,*/
+	/*IB_MAD_SNOOP_RMPP_SENDS	   = (1<<1),*/
+	IB_MAD_SNOOP_SEND_COMPLETIONS	   = (1<<2),
+	/*IB_MAD_SNOOP_RMPP_SEND_COMPLETIONS = (1<<3),*/
+	IB_MAD_SNOOP_RECVS		   = (1<<4)
+	/*IB_MAD_SNOOP_RMPP_RECVS	   = (1<<5),*/
+	/*IB_MAD_SNOOP_REDIRECTED_QPS	   = (1<<6)*/
+};
+
+/**
+ * ib_register_mad_snoop - Register to snoop sent and received MADs.
+ * @device: The device to register with.
+ * @port_num: The port on the specified device to use.
+ * @qp_type: Specifies which QP traffic to snoop.  Must be either
+ *   IB_QPT_SMI or IB_QPT_GSI.
+ * @mad_snoop_flags: Specifies information where snooping occurs.
+ * @send_handler: The callback routine invoked for a snooped send.
+ * @recv_handler: The callback routine invoked for a snooped receive.
+ * @context: User specified context associated with the registration.
+ */
+struct ib_mad_agent *ib_register_mad_snoop(struct ib_device *device,
+					   u8 port_num,
+					   enum ib_qp_type qp_type,
+					   int mad_snoop_flags,
+					   ib_mad_snoop_handler snoop_handler,
+					   ib_mad_recv_handler recv_handler,
+					   void *context);
+
+/**
+ * ib_unregister_mad_agent - Unregisters a client from using MAD services.
+ * @mad_agent: Corresponding MAD registration request to deregister.
+ *
+ * After invoking this routine, MAD services are no longer usable by the
+ * client on the associated QP.
+ */
+int ib_unregister_mad_agent(struct ib_mad_agent *mad_agent);
+
+/**
+ * ib_post_send_mad - Posts MAD(s) to the send queue of the QP associated
+ *   with the registered client.
+ * @mad_agent: Specifies the associated registration to post the send to.
+ * @send_wr: Specifies the information needed to send the MAD(s).
+ * @bad_send_wr: Specifies the MAD on which an error was encountered.
+ *
+ * Sent MADs are not guaranteed to complete in the order that they were posted.
+ */
+int ib_post_send_mad(struct ib_mad_agent *mad_agent,
+		     struct ib_send_wr *send_wr,
+		     struct ib_send_wr **bad_send_wr);
+
+/**
+ * ib_coalesce_recv_mad - Coalesces received MAD data into a single buffer.
+ * @mad_recv_wc: Work completion information for a received MAD.
+ * @buf: User-provided data buffer to receive the coalesced buffers.  The
+ *   referenced buffer should be at least the size of the mad_len specified
+ *   by @mad_recv_wc.
+ *
+ * This call copies a chain of received RMPP MADs into a single data buffer,
+ * removing duplicated headers.
+ */
+void ib_coalesce_recv_mad(struct ib_mad_recv_wc *mad_recv_wc,
+			  void *buf);
+
+/**
+ * ib_free_recv_mad - Returns data buffers used to receive a MAD to the
+ *   access layer.
+ * @mad_recv_wc: Work completion information for a received MAD.
+ *
+ * Clients receiving MADs through their ib_mad_recv_handler must call this
+ * routine to return the work completion buffers to the access layer.
+ */
+void ib_free_recv_mad(struct ib_mad_recv_wc *mad_recv_wc);
+
+/**
+ * ib_cancel_mad - Cancels an outstanding send MAD operation.
+ * @mad_agent: Specifies the registration associated with sent MAD.
+ * @wr_id: Indicates the work request identifier of the MAD to cancel.
+ *
+ * MADs will be returned to the user through the corresponding
+ * ib_mad_send_handler.
+ */
+void ib_cancel_mad(struct ib_mad_agent *mad_agent,
+		   u64 wr_id);
+
+/**
+ * ib_redirect_mad_qp - Registers a QP for MAD services.
+ * @qp: Reference to a QP that requires MAD services.
+ * @rmpp_version: If set, indicates that the client will send
+ *   and receive MADs that contain the RMPP header for the given version.
+ *   If set to 0, indicates that RMPP is not used by this client.
+ * @send_handler: The completion callback routine invoked after a send
+ *   request has completed.
+ * @recv_handler: The completion callback routine invoked for a received
+ *   MAD.
+ * @context: User specified context associated with the registration.
+ *
+ * Use of this call allows clients to use MAD services, such as RMPP,
+ * on user-owned QPs.  After calling this routine, users may send
+ * MADs on the specified QP by calling ib_mad_post_send.
+ */
+struct ib_mad_agent *ib_redirect_mad_qp(struct ib_qp *qp,
+					u8 rmpp_version,
+					ib_mad_send_handler send_handler,
+					ib_mad_recv_handler recv_handler,
+					void *context);
+
+/**
+ * ib_process_mad_wc - Processes a work completion associated with a
+ *   MAD sent or received on a redirected QP.
+ * @mad_agent: Specifies the registered MAD service using the redirected QP.
+ * @wc: References a work completion associated with a sent or received
+ *   MAD segment.
+ *
+ * This routine is used to complete or continue processing on a MAD request.
+ * If the work completion is associated with a send operation, calling
+ * this routine is required to continue an RMPP transfer or to wait for a
+ * corresponding response, if it is a request.  If the work completion is
+ * associated with a receive operation, calling this routine is required to
+ * process an inbound or outbound RMPP transfer, or to match a response MAD
+ * with its corresponding request.
+ */
+int ib_process_mad_wc(struct ib_mad_agent *mad_agent,
+		      struct ib_wc *wc);
+
+#endif /* IB_MAD_H */
drivers/infiniband/include/ib_smi.h new file +96 -0

96 lines not on this page · raw message

#5 lore raw

PATCH v4 5/24 Add InfiniBand MAD (management datagram) support

Add support for handling InfiniBand MADs (management datagrams), including sending and receiving MADs as well as passing MADs on to local agents.

This is required for an SM (subnet manager) to discover and configure the host, since the SM's query MADs must be passed to the local SMA (subnet management agent). In addition, this support is used by upper level protocols to send queries to and receive responses from the SM.

drivers/infiniband/core/Makefile +3 -1
@@ -1,6 +1,8 @@
 EXTRA_CFLAGS += -Idrivers/infiniband/include
 
-obj-$(CONFIG_INFINIBAND) +=	ib_core.o
+obj-$(CONFIG_INFINIBAND) +=	ib_core.o ib_mad.o
 
 ib_core-y :=			packer.o ud_header.o verbs.o sysfs.o \
 				device.o fmr_pool.o cache.o
+
+ib_mad-y :=			mad.o smi.o agent.o
drivers/infiniband/core/agent.c new file +399 -0

399 lines not on this page · raw message

drivers/infiniband/core/mad.c new file +2632 -0

2,592 lines not on this page · raw message

The rest of this diff is not shown. View the raw message.

Part of the diff is not shown here. View the raw message.

#6 lore raw

PATCH v4 6/24 Add InfiniBand MAD (management datagram) support (private headers)

Add MAD layer private implementation headers.

drivers/infiniband/core/agent.h new file +55 -0
@@ -0,0 +1,55 @@
+/*
+ * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
+ * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
+ * Copyright (c) 2004 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
+ * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id$
+ */
+
+#ifndef __AGENT_H_
+#define __AGENT_H_
+
+extern spinlock_t ib_agent_port_list_lock;
+
+extern int ib_agent_port_open(struct ib_device *device,
+			      int port_num);
+
+extern int ib_agent_port_close(struct ib_device *device, int port_num);
+
+extern int agent_send(struct ib_mad_private *mad,
+		      struct ib_grh *grh,
+		      struct ib_wc *wc,
+		      struct ib_device *device,
+		      int port_num);
+
+#endif	/* __AGENT_H_ */
drivers/infiniband/core/agent_priv.h new file +64 -0

64 lines not on this page · raw message

drivers/infiniband/core/mad_priv.h new file +194 -0

194 lines not on this page · raw message

#7 lore raw

PATCH v4 7/24 Add InfiniBand MAD SMI support

Add MAD layer SMI (Subnet Management Interface) code.

drivers/infiniband/core/smi.c new file +234 -0
@@ -0,0 +1,234 @@
+/*
+ * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
+ * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
+ * Copyright (c) 2004 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
+ * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id$
+ */
+
+#include <ib_smi.h>
+
+
+/*
+ * Fixup a directed route SMP for sending
+ * Return 0 if the SMP should be discarded
+ */
+int smi_handle_dr_smp_send(struct ib_smp *smp,
+			   u8 node_type,
+			   int port_num)
+{
+	u8 hop_ptr, hop_cnt;
+
+	hop_ptr = smp->hop_ptr;
+	hop_cnt = smp->hop_cnt;
+
+	/* See section 14.2.2.2, Vol 1 IB spec */
+	if (!ib_get_smp_direction(smp)) {
+		/* C14-9:1 */
+		if (hop_cnt && hop_ptr == 0) {
+			smp->hop_ptr++;
+			return (smp->initial_path[smp->hop_ptr] ==
+				port_num);
+		}
+
+		/* C14-9:2 */
+		if (hop_ptr && hop_ptr < hop_cnt) {
+			if (node_type != IB_NODE_SWITCH)
+				return 0;
+
+			/* smp->return_path set when received */
+			smp->hop_ptr++;
+			return (smp->initial_path[smp->hop_ptr] ==
+				port_num);
+		}
+
+		/* C14-9:3 -- We're at the end of the DR segment of path */
+		if (hop_ptr == hop_cnt) {
+			/* smp->return_path set when received */
+			smp->hop_ptr++;
+			return (node_type == IB_NODE_SWITCH ||
+				smp->dr_dlid == IB_LID_PERMISSIVE);
+		}
+
+		/* C14-9:4 -- hop_ptr = hop_cnt + 1 -> give to SMA/SM */
+		/* C14-9:5 -- Fail unreasonable hop pointer */
+		return (hop_ptr == hop_cnt + 1);
+
+	} else {
+		/* C14-13:1 */
+		if (hop_cnt && hop_ptr == hop_cnt + 1) {
+			smp->hop_ptr--;
+			return (smp->return_path[smp->hop_ptr] ==
+				port_num);
+		}
+
+		/* C14-13:2 */
+		if (2 <= hop_ptr && hop_ptr <= hop_cnt) {
+			if (node_type != IB_NODE_SWITCH)
+				return 0;
+
+			smp->hop_ptr--;
+			return (smp->return_path[smp->hop_ptr] ==
+				port_num);
+		}
+
+		/* C14-13:3 -- at the end of the DR segment of path */
+		if (hop_ptr == 1) {
+			smp->hop_ptr--;
+			/* C14-13:3 -- SMPs destined for SM shouldn't be here */
+			return (node_type == IB_NODE_SWITCH ||
+				smp->dr_slid == IB_LID_PERMISSIVE);
+		}
+
+		/* C14-13:4 -- hop_ptr = 0 -> should have gone to SM */
+		if (hop_ptr == 0)
+			return 1;
+
+		/* C14-13:5 -- Check for unreasonable hop pointer */
+		return 0;
+	}
+}
+
+/*
+ * Adjust information for a received SMP
+ * Return 0 if the SMP should be dropped
+ */
+int smi_handle_dr_smp_recv(struct ib_smp *smp,
+			   u8 node_type,
+			   int port_num,
+			   int phys_port_cnt)
+{
+	u8 hop_ptr, hop_cnt;
+
+	hop_ptr = smp->hop_ptr;
+	hop_cnt = smp->hop_cnt;
+
+	/* See section 14.2.2.2, Vol 1 IB spec */
+	if (!ib_get_smp_direction(smp)) {
+		/* C14-9:1 -- sender should have incremented hop_ptr */
+		if (hop_cnt && hop_ptr == 0)
+			return 0;
+
+		/* C14-9:2 -- intermediate hop */
+		if (hop_ptr && hop_ptr < hop_cnt) {
+			if (node_type != IB_NODE_SWITCH)
+				return 0;
+
+			smp->return_path[hop_ptr] = port_num;
+			/* smp->hop_ptr updated when sending */
+			return (smp->initial_path[hop_ptr+1] <= phys_port_cnt);
+		}
+
+		/* C14-9:3 -- We're at the end of the DR segment of path */
+		if (hop_ptr == hop_cnt) {
+			if (hop_cnt)
+				smp->return_path[hop_ptr] = port_num;
+			/* smp->hop_ptr updated when sending */
+
+			return (node_type == IB_NODE_SWITCH ||
+				smp->dr_dlid == IB_LID_PERMISSIVE);
+		}
+
+		/* C14-9:4 -- hop_ptr = hop_cnt + 1 -> give to SMA/SM */
+		/* C14-9:5 -- fail unreasonable hop pointer */
+		return (hop_ptr == hop_cnt + 1);
+
+	} else {
+
+		/* C14-13:1 */
+		if (hop_cnt && hop_ptr == hop_cnt + 1) {
+			smp->hop_ptr--;
+			return (smp->return_path[smp->hop_ptr] ==
+				port_num);
+		}
+
+		/* C14-13:2 */
+		if (2 <= hop_ptr && hop_ptr <= hop_cnt) {
+			if (node_type != IB_NODE_SWITCH)
+				return 0;
+
+			/* smp->hop_ptr updated when sending */
+			return (smp->return_path[hop_ptr-1] <= phys_port_cnt);
+		}
+
+		/* C14-13:3 -- We're at the end of the DR segment of path */
+		if (hop_ptr == 1) {
+			if (smp->dr_slid == IB_LID_PERMISSIVE) {
+				/* giving SMP to SM - update hop_ptr */
+				smp->hop_ptr--;
+				return 1;
+			}
+			/* smp->hop_ptr updated when sending */
+			return (node_type == IB_NODE_SWITCH);
+		}
+
+		/* C14-13:4 -- hop_ptr = 0 -> give to SM */
+		/* C14-13:5 -- Check for unreasonable hop pointer */
+		return (hop_ptr == 0);
+	}
+}
+
+/*
+ * Return 1 if the received DR SMP should be forwarded to the send queue
+ * Return 0 if the SMP should be completed up the stack
+ */
+int smi_check_forward_dr_smp(struct ib_smp *smp)
+{
+	u8 hop_ptr, hop_cnt;
+
+	hop_ptr = smp->hop_ptr;
+	hop_cnt = smp->hop_cnt;
+
+	if (!ib_get_smp_direction(smp)) {
+		/* C14-9:2 -- intermediate hop */
+		if (hop_ptr && hop_ptr < hop_cnt)
+			return 1;
+
+		/* C14-9:3 -- at the end of the DR segment of path */
+		if (hop_ptr == hop_cnt)
+			return (smp->dr_dlid == IB_LID_PERMISSIVE);
+
+		/* C14-9:4 -- hop_ptr = hop_cnt + 1 -> give to SMA/SM */
+		if (hop_ptr == hop_cnt + 1)
+			return 1;
+	} else {
+		/* C14-13:2 */
+		if (2 <= hop_ptr && hop_ptr <= hop_cnt)
+			return 1;
+
+		/* C14-13:3 -- at the end of the DR segment of path */
+		if (hop_ptr == 1)
+			return (smp->dr_slid != IB_LID_PERMISSIVE);
+	}
+	return 0;
+}
drivers/infiniband/core/smi.h new file +67 -0

67 lines not on this page · raw message

#8 lore raw

PATCH v4 8/24 Add InfiniBand SA (Subnet Administration) query support

Add support for sending queries to the SA (Subnet Administration). In particular the PathRecord and MCMember (multicast group member) used by the IP-over-InfiniBand driver are implemented.

drivers/infiniband/core/Makefile +3 -1
@@ -1,8 +1,10 @@
 EXTRA_CFLAGS += -Idrivers/infiniband/include
 
-obj-$(CONFIG_INFINIBAND) +=	ib_core.o ib_mad.o
+obj-$(CONFIG_INFINIBAND) +=	ib_core.o ib_mad.o ib_sa.o
 
 ib_core-y :=			packer.o ud_header.o verbs.o sysfs.o \
 				device.o fmr_pool.o cache.o
 
 ib_mad-y :=			mad.o smi.o agent.o
+
+ib_sa-y :=			sa_query.o
drivers/infiniband/core/sa_query.c new file +866 -0

866 lines not on this page · raw message

drivers/infiniband/include/ib_sa.h new file +280 -0

280 lines not on this page · raw message

#9 lore raw

PATCH v4 9/24 Add Mellanox HCA low-level driver

Add a low-level driver for Mellanox MT23108 and MT25208 HCAs. The MT25208 is only fully supported when in MT23108 compatibility mode; only the very beginnings of support for native MT25208 mode (required for HCAs without local memory) is present.

(As a side note, I believe this driver would be the first in-tree consumer of the PCI MSI/MSI-X API)

drivers/infiniband/Kconfig +2 -0
@@ -7,4 +7,6 @@
 	  any protocols you wish to use as well as drivers for your
 	  InfiniBand hardware.
 
+source "drivers/infiniband/hw/mthca/Kconfig"
+
 endmenu
drivers/infiniband/Makefile +1 -0
@@ -1 +1,2 @@
 obj-$(CONFIG_INFINIBAND)		+= core/
+obj-$(CONFIG_INFINIBAND_MTHCA)		+= hw/mthca/
drivers/infiniband/hw/mthca/Kconfig new file +26 -0
@@ -0,0 +1,26 @@
+config INFINIBAND_MTHCA
+	tristate "Mellanox HCA support"
+	depends on PCI && INFINIBAND
+	---help---
+	  This is a low-level driver for Mellanox InfiniHost host
+	  channel adapters (HCAs), including the MT23108 PCI-X HCA
+	  ("Tavor") and the MT25208 PCI Express HCA ("Arbel").
+
+config INFINIBAND_MTHCA_DEBUG
+	bool "Verbose debugging output"
+	depends on INFINIBAND_MTHCA
+	default n
+	---help---
+	  This option causes the mthca driver produce a bunch of debug
+	  messages.  Select this is you are developing the driver or
+	  trying to diagnose a problem.
+
+config INFINIBAND_MTHCA_SSE_DOORBELL
+	bool "SSE doorbell code"
+	depends on INFINIBAND_MTHCA && X86 && !X86_64
+	default n
+	---help---
+	  This option will have the mthca driver use SSE instructions
+	  to ring hardware doorbell registers.  This may improve
+	  performance for some workloads, but the driver will not run
+	  on processors without SSE instructions.
drivers/infiniband/hw/mthca/Makefile new file +12 -0
@@ -0,0 +1,12 @@
+EXTRA_CFLAGS += -Idrivers/infiniband/include
+
+ifdef CONFIG_INFINIBAND_MTHCA_DEBUG
+EXTRA_CFLAGS += -DDEBUG
+endif
+
+obj-$(CONFIG_INFINIBAND_MTHCA) += ib_mthca.o
+
+ib_mthca-y :=	mthca_main.o mthca_cmd.o mthca_profile.o mthca_reset.o \
+		mthca_allocator.o mthca_eq.o mthca_pd.o mthca_cq.o \
+		mthca_mr.o mthca_qp.o mthca_av.o mthca_mcg.o mthca_mad.o \
+		mthca_provider.o
drivers/infiniband/hw/mthca/mthca_allocator.c new file +179 -0

179 lines not on this page · raw message

drivers/infiniband/hw/mthca/mthca_config_reg.h new file +55 -0

55 lines not on this page · raw message

drivers/infiniband/hw/mthca/mthca_dev.h new file +391 -0

391 lines not on this page · raw message

drivers/infiniband/hw/mthca/mthca_doorbell.h new file +123 -0

123 lines not on this page · raw message

drivers/infiniband/hw/mthca/mthca_main.c new file +933 -0

933 lines not on this page · raw message

#10 lore raw

PATCH v4 10/24 Add Mellanox HCA low-level driver (midlayer interface)

Add midlayer interface code for Mellanox HCA driver.

drivers/infiniband/hw/mthca/mthca_provider.c new file +626 -0
@@ -0,0 +1,626 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: mthca_provider.c 1349 2004-12-16 21:09:43Z roland $
+ */
+
+#include <ib_mad.h>
+#include <ib_smi.h>
+
+#include "mthca_dev.h"
+#include "mthca_cmd.h"
+
+static int mthca_query_device(struct ib_device *ibdev,
+			      struct ib_device_attr *props)
+{
+	struct ib_mad *in_mad  = NULL;
+	struct ib_mad *out_mad = NULL;
+	int err = -ENOMEM;
+	u8 status;
+
+	in_mad  = kmalloc(sizeof *in_mad, GFP_KERNEL);
+	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	props->fw_ver        = to_mdev(ibdev)->fw_ver;
+
+	memset(in_mad, 0, sizeof *in_mad);
+	in_mad->mad_hdr.base_version       = 1;
+	in_mad->mad_hdr.mgmt_class     	   = IB_MGMT_CLASS_SUBN_LID_ROUTED;
+	in_mad->mad_hdr.class_version  	   = 1;
+	in_mad->mad_hdr.method         	   = IB_MGMT_METHOD_GET;
+	in_mad->mad_hdr.attr_id   	   = IB_SMP_ATTR_NODE_INFO;
+
+	err = mthca_MAD_IFC(to_mdev(ibdev), 1,
+			    1, in_mad, out_mad,
+			    &status);
+	if (err)
+		goto out;
+	if (status) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	props->vendor_id      = be32_to_cpup((u32 *) (out_mad->data + 76)) &
+		0xffffff;
+	props->vendor_part_id = be16_to_cpup((u16 *) (out_mad->data + 70));
+	props->hw_ver         = be16_to_cpup((u16 *) (out_mad->data + 72));
+	memcpy(&props->sys_image_guid, out_mad->data + 44, 8);
+	memcpy(&props->node_guid,      out_mad->data + 52, 8);
+
+	err = 0;
+ out:
+	kfree(in_mad);
+	kfree(out_mad);
+	return err;
+}
+
+static int mthca_query_port(struct ib_device *ibdev,
+			    u8 port, struct ib_port_attr *props)
+{
+	struct ib_mad *in_mad  = NULL;
+	struct ib_mad *out_mad = NULL;
+	int err = -ENOMEM;
+	u8 status;
+
+	in_mad  = kmalloc(sizeof *in_mad, GFP_KERNEL);
+	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	memset(in_mad, 0, sizeof *in_mad);
+	in_mad->mad_hdr.base_version       = 1;
+	in_mad->mad_hdr.mgmt_class     	   = IB_MGMT_CLASS_SUBN_LID_ROUTED;
+	in_mad->mad_hdr.class_version  	   = 1;
+	in_mad->mad_hdr.method         	   = IB_MGMT_METHOD_GET;
+	in_mad->mad_hdr.attr_id   	   = IB_SMP_ATTR_PORT_INFO;
+	in_mad->mad_hdr.attr_mod           = cpu_to_be32(port);
+
+	err = mthca_MAD_IFC(to_mdev(ibdev), 1,
+			    port, in_mad, out_mad,
+			    &status);
+	if (err)
+		goto out;
+	if (status) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	props->lid               = be16_to_cpup((u16 *) (out_mad->data + 56));
+	props->lmc               = (*(u8 *) (out_mad->data + 74)) & 0x7;
+	props->sm_lid            = be16_to_cpup((u16 *) (out_mad->data + 58));
+	props->sm_sl             = (*(u8 *) (out_mad->data + 76)) & 0xf;
+	props->state             = (*(u8 *) (out_mad->data + 72)) & 0xf;
+	props->port_cap_flags    = be32_to_cpup((u32 *) (out_mad->data + 60));
+	props->gid_tbl_len       = to_mdev(ibdev)->limits.gid_table_len;
+	props->pkey_tbl_len      = to_mdev(ibdev)->limits.pkey_table_len;
+	props->qkey_viol_cntr    = be16_to_cpup((u16 *) (out_mad->data + 88));
+
+ out:
+	kfree(in_mad);
+	kfree(out_mad);
+	return err;
+}
+
+static int mthca_modify_port(struct ib_device *ibdev,
+			     u8 port, int port_modify_mask,
+			     struct ib_port_modify *props)
+{
+	return 0;
+}
+
+static int mthca_query_pkey(struct ib_device *ibdev,
+			    u8 port, u16 index, u16 *pkey)
+{
+	struct ib_mad *in_mad  = NULL;
+	struct ib_mad *out_mad = NULL;
+	int err = -ENOMEM;
+	u8 status;
+
+	in_mad  = kmalloc(sizeof *in_mad, GFP_KERNEL);
+	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	memset(in_mad, 0, sizeof *in_mad);
+	in_mad->mad_hdr.base_version       = 1;
+	in_mad->mad_hdr.mgmt_class     	   = IB_MGMT_CLASS_SUBN_LID_ROUTED;
+	in_mad->mad_hdr.class_version  	   = 1;
+	in_mad->mad_hdr.method         	   = IB_MGMT_METHOD_GET;
+	in_mad->mad_hdr.attr_id   	   = IB_SMP_ATTR_PKEY_TABLE;
+	in_mad->mad_hdr.attr_mod           = cpu_to_be32(index / 32);
+
+	err = mthca_MAD_IFC(to_mdev(ibdev), 1,
+			    port, in_mad, out_mad,
+			    &status);
+	if (err)
+		goto out;
+	if (status) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	*pkey = be16_to_cpu(((u16 *) (out_mad->data + 40))[index % 32]);
+
+ out:
+	kfree(in_mad);
+	kfree(out_mad);
+	return err;
+}
+
+static int mthca_query_gid(struct ib_device *ibdev, u8 port,
+			   int index, union ib_gid *gid)
+{
+	struct ib_mad *in_mad  = NULL;
+	struct ib_mad *out_mad = NULL;
+	int err = -ENOMEM;
+	u8 status;
+
+	in_mad  = kmalloc(sizeof *in_mad, GFP_KERNEL);
+	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	memset(in_mad, 0, sizeof *in_mad);
+	in_mad->mad_hdr.base_version       = 1;
+	in_mad->mad_hdr.mgmt_class     	   = IB_MGMT_CLASS_SUBN_LID_ROUTED;
+	in_mad->mad_hdr.class_version  	   = 1;
+	in_mad->mad_hdr.method         	   = IB_MGMT_METHOD_GET;
+	in_mad->mad_hdr.attr_id   	   = IB_SMP_ATTR_PORT_INFO;
+	in_mad->mad_hdr.attr_mod           = cpu_to_be32(port);
+
+	err = mthca_MAD_IFC(to_mdev(ibdev), 1,
+			    port, in_mad, out_mad,
+			    &status);
+	if (err)
+		goto out;
+	if (status) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	memcpy(gid->raw, out_mad->data + 48, 8);
+
+	memset(in_mad, 0, sizeof *in_mad);
+	in_mad->mad_hdr.base_version       = 1;
+	in_mad->mad_hdr.mgmt_class     	   = IB_MGMT_CLASS_SUBN_LID_ROUTED;
+	in_mad->mad_hdr.class_version  	   = 1;
+	in_mad->mad_hdr.method         	   = IB_MGMT_METHOD_GET;
+	in_mad->mad_hdr.attr_id   	   = IB_SMP_ATTR_GUID_INFO;
+	in_mad->mad_hdr.attr_mod           = cpu_to_be32(index / 8);
+
+	err = mthca_MAD_IFC(to_mdev(ibdev), 1,
+			    port, in_mad, out_mad,
+			    &status);
+	if (err)
+		goto out;
+	if (status) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	memcpy(gid->raw + 8, out_mad->data + 40 + (index % 8) * 16, 8);
+
+ out:
+	kfree(in_mad);
+	kfree(out_mad);
+	return err;
+}
+
+static struct ib_pd *mthca_alloc_pd(struct ib_device *ibdev)
+{
+	struct mthca_pd *pd;
+	int err;
+
+	pd = kmalloc(sizeof *pd, GFP_KERNEL);
+	if (!pd)
+		return ERR_PTR(-ENOMEM);
+
+	err = mthca_pd_alloc(to_mdev(ibdev), pd);
+	if (err) {
+		kfree(pd);
+		return ERR_PTR(err);
+	}
+
+	return &pd->ibpd;
+}
+
+static int mthca_dealloc_pd(struct ib_pd *pd)
+{
+	mthca_pd_free(to_mdev(pd->device), to_mpd(pd));
+	kfree(pd);
+
+	return 0;
+}
+
+static struct ib_ah *mthca_ah_create(struct ib_pd *pd,
+				     struct ib_ah_attr *ah_attr)
+{
+	int err;
+	struct mthca_ah *ah;
+
+	ah = kmalloc(sizeof *ah, GFP_KERNEL);
+	if (!ah)
+		return ERR_PTR(-ENOMEM);
+
+	err = mthca_create_ah(to_mdev(pd->device), to_mpd(pd), ah_attr, ah);
+	if (err) {
+		kfree(ah);
+		return ERR_PTR(err);
+	}
+
+	return &ah->ibah;
+}
+
+static int mthca_ah_destroy(struct ib_ah *ah)
+{
+	mthca_destroy_ah(to_mdev(ah->device), to_mah(ah));
+	kfree(ah);
+
+	return 0;
+}
+
+static struct ib_qp *mthca_create_qp(struct ib_pd *pd,
+				     struct ib_qp_init_attr *init_attr)
+{
+	struct mthca_qp *qp;
+	int err;
+
+	switch (init_attr->qp_type) {
+	case IB_QPT_RC:
+	case IB_QPT_UC:
+	case IB_QPT_UD:
+	{
+		qp = kmalloc(sizeof *qp, GFP_KERNEL);
+		if (!qp)
+			return ERR_PTR(-ENOMEM);
+
+		qp->sq.max    = init_attr->cap.max_send_wr;
+		qp->rq.max    = init_attr->cap.max_recv_wr;
+		qp->sq.max_gs = init_attr->cap.max_send_sge;
+		qp->rq.max_gs = init_attr->cap.max_recv_sge;
+
+		err = mthca_alloc_qp(to_mdev(pd->device), to_mpd(pd),
+				     to_mcq(init_attr->send_cq),
+				     to_mcq(init_attr->recv_cq),
+				     init_attr->qp_type, init_attr->sq_sig_type,
+				     init_attr->rq_sig_type, qp);
+		qp->ibqp.qp_num = qp->qpn;
+		break;
+	}
+	case IB_QPT_SMI:
+	case IB_QPT_GSI:
+	{
+		qp = kmalloc(sizeof (struct mthca_sqp), GFP_KERNEL);
+		if (!qp)
+			return ERR_PTR(-ENOMEM);
+
+		qp->sq.max    = init_attr->cap.max_send_wr;
+		qp->rq.max    = init_attr->cap.max_recv_wr;
+		qp->sq.max_gs = init_attr->cap.max_send_sge;
+		qp->rq.max_gs = init_attr->cap.max_recv_sge;
+
+		qp->ibqp.qp_num = init_attr->qp_type == IB_QPT_SMI ? 0 : 1;
+
+		err = mthca_alloc_sqp(to_mdev(pd->device), to_mpd(pd),
+				      to_mcq(init_attr->send_cq),
+				      to_mcq(init_attr->recv_cq),
+				      init_attr->sq_sig_type, init_attr->rq_sig_type,
+				      qp->ibqp.qp_num, init_attr->port_num,
+				      to_msqp(qp));
+		break;
+	}
+	default:
+		/* Don't support raw QPs */
+		return ERR_PTR(-ENOSYS);
+	}
+
+	if (err) {
+		kfree(qp);
+		return ERR_PTR(err);
+	}
+
+        init_attr->cap.max_inline_data = 0;
+
+	return &qp->ibqp;
+}
+
+static int mthca_destroy_qp(struct ib_qp *qp)
+{
+	mthca_free_qp(to_mdev(qp->device), to_mqp(qp));
+	kfree(qp);
+	return 0;
+}
+
+static struct ib_cq *mthca_create_cq(struct ib_device *ibdev, int entries)
+{
+	struct mthca_cq *cq;
+	int nent;
+	int err;
+
+	cq = kmalloc(sizeof *cq, GFP_KERNEL);
+	if (!cq)
+		return ERR_PTR(-ENOMEM);
+
+	for (nent = 1; nent <= entries; nent <<= 1)
+		; /* nothing */
+
+	err = mthca_init_cq(to_mdev(ibdev), nent, cq);
+	if (err) {
+		kfree(cq);
+		cq = ERR_PTR(err);
+	} else
+		cq->ibcq.cqe = nent - 1;
+
+	return &cq->ibcq;
+}
+
+static int mthca_destroy_cq(struct ib_cq *cq)
+{
+	mthca_free_cq(to_mdev(cq->device), to_mcq(cq));
+	kfree(cq);
+
+	return 0;
+}
+
+static int mthca_req_notify_cq(struct ib_cq *cq, enum ib_cq_notify notify)
+{
+	mthca_arm_cq(to_mdev(cq->device), to_mcq(cq),
+		     notify == IB_CQ_SOLICITED);
+	return 0;
+}
+
+static inline u32 convert_access(int acc)
+{
+	return (acc & IB_ACCESS_REMOTE_ATOMIC ? MTHCA_MPT_FLAG_ATOMIC       : 0) |
+	       (acc & IB_ACCESS_REMOTE_WRITE  ? MTHCA_MPT_FLAG_REMOTE_WRITE : 0) |
+	       (acc & IB_ACCESS_REMOTE_READ   ? MTHCA_MPT_FLAG_REMOTE_READ  : 0) |
+	       (acc & IB_ACCESS_LOCAL_WRITE   ? MTHCA_MPT_FLAG_LOCAL_WRITE  : 0) |
+	       MTHCA_MPT_FLAG_LOCAL_READ;
+}
+
+static struct ib_mr *mthca_get_dma_mr(struct ib_pd *pd, int acc)
+{
+	struct mthca_mr *mr;
+	int err;
+
+	mr = kmalloc(sizeof *mr, GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(-ENOMEM);
+
+	err = mthca_mr_alloc_notrans(to_mdev(pd->device),
+				     to_mpd(pd)->pd_num,
+				     convert_access(acc), mr);
+
+	if (err) {
+		kfree(mr);
+		return ERR_PTR(err);
+	}
+
+	return &mr->ibmr;
+}
+
+static struct ib_mr *mthca_reg_phys_mr(struct ib_pd       *pd,
+				       struct ib_phys_buf *buffer_list,
+				       int                 num_phys_buf,
+				       int                 acc,
+				       u64                *iova_start)
+{
+	struct mthca_mr *mr;
+	u64 *page_list;
+	u64 total_size;
+	u64 mask;
+	int shift;
+	int npages;
+	int err;
+	int i, j, n;
+
+	/* First check that we have enough alignment */
+	if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK))
+		return ERR_PTR(-EINVAL);
+
+	if (num_phys_buf > 1 &&
+	    ((buffer_list[0].addr + buffer_list[0].size) & ~PAGE_MASK))
+		return ERR_PTR(-EINVAL);
+
+	mask = 0;
+	total_size = 0;
+	for (i = 0; i < num_phys_buf; ++i) {
+		if (buffer_list[i].addr & ~PAGE_MASK)
+			return ERR_PTR(-EINVAL);
+		if (i != 0 && i != num_phys_buf - 1 &&
+		    (buffer_list[i].size & ~PAGE_MASK))
+			return ERR_PTR(-EINVAL);
+
+		total_size += buffer_list[i].size;
+		if (i > 0)
+			mask |= buffer_list[i].addr;
+	}
+
+	/* Find largest page shift we can use to cover buffers */
+	for (shift = PAGE_SHIFT; shift < 31; ++shift)
+		if (num_phys_buf > 1) {
+			if ((1ULL << shift) & mask)
+				break;
+		} else {
+			if (1ULL << shift >=
+			    buffer_list[0].size +
+			    (buffer_list[0].addr & ((1ULL << shift) - 1)))
+				break;
+		}
+
+	buffer_list[0].size += buffer_list[0].addr & ((1ULL << shift) - 1);
+	buffer_list[0].addr &= ~0ull << shift;
+
+	mr = kmalloc(sizeof *mr, GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(-ENOMEM);
+
+	npages = 0;
+	for (i = 0; i < num_phys_buf; ++i)
+		npages += (buffer_list[i].size + (1ULL << shift) - 1) >> shift;
+
+	if (!npages)
+		return &mr->ibmr;
+
+	page_list = kmalloc(npages * sizeof *page_list, GFP_KERNEL);
+	if (!page_list) {
+		kfree(mr);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	n = 0;
+	for (i = 0; i < num_phys_buf; ++i)
+		for (j = 0;
+		     j < (buffer_list[i].size + (1ULL << shift) - 1) >> shift;
+		     ++j)
+			page_list[n++] = buffer_list[i].addr + ((u64) j << shift);
+
+	mthca_dbg(to_mdev(pd->device), "Registering memory at %llx (iova %llx) "
+		  "in PD %x; shift %d, npages %d.\n",
+		  (unsigned long long) buffer_list[0].addr,
+		  (unsigned long long) *iova_start,
+		  to_mpd(pd)->pd_num,
+		  shift, npages);
+
+	err = mthca_mr_alloc_phys(to_mdev(pd->device),
+				  to_mpd(pd)->pd_num,
+				  page_list, shift, npages,
+				  *iova_start, total_size,
+				  convert_access(acc), mr);
+
+	if (err) {
+		kfree(mr);
+		return ERR_PTR(err);
+	}
+
+	kfree(page_list);
+	return &mr->ibmr;
+}
+
+static int mthca_dereg_mr(struct ib_mr *mr)
+{
+	mthca_free_mr(to_mdev(mr->device), to_mmr(mr));
+	kfree(mr);
+	return 0;
+}
+
+static ssize_t show_rev(struct class_device *cdev, char *buf)
+{
+	struct mthca_dev *dev = container_of(cdev, struct mthca_dev, ib_dev.class_dev);
+	return sprintf(buf, "%x\n", dev->rev_id);
+}
+
+static ssize_t show_fw_ver(struct class_device *cdev, char *buf)
+{
+	struct mthca_dev *dev = container_of(cdev, struct mthca_dev, ib_dev.class_dev);
+	return sprintf(buf, "%x.%x.%x\n", (int) (dev->fw_ver >> 32),
+		       (int) (dev->fw_ver >> 16) & 0xffff,
+		       (int) dev->fw_ver & 0xffff);
+}
+
+static ssize_t show_hca(struct class_device *cdev, char *buf)
+{
+	struct mthca_dev *dev = container_of(cdev, struct mthca_dev, ib_dev.class_dev);
+	switch (dev->hca_type) {
+	case TAVOR:        return sprintf(buf, "MT23108\n");
+	case ARBEL_COMPAT: return sprintf(buf, "MT25208 (MT23108 compat mode)\n");
+	case ARBEL_NATIVE: return sprintf(buf, "MT25208\n");
+	default:           return sprintf(buf, "unknown\n");
+	}
+}
+
+static CLASS_DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
+static CLASS_DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
+static CLASS_DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
+
+static struct class_device_attribute *mthca_class_attributes[] = {
+	&class_device_attr_hw_rev,
+	&class_device_attr_fw_ver,
+	&class_device_attr_hca_type
+};
+
+int mthca_register_device(struct mthca_dev *dev)
+{
+	int ret;
+	int i;
+
+	strlcpy(dev->ib_dev.name, "mthca%d", IB_DEVICE_NAME_MAX);
+	dev->ib_dev.node_type            = IB_NODE_CA;
+	dev->ib_dev.phys_port_cnt        = dev->limits.num_ports;
+	dev->ib_dev.dma_device           = &dev->pdev->dev;
+	dev->ib_dev.class_dev.dev        = &dev->pdev->dev;
+	dev->ib_dev.query_device         = mthca_query_device;
+	dev->ib_dev.query_port           = mthca_query_port;
+	dev->ib_dev.modify_port          = mthca_modify_port;
+	dev->ib_dev.query_pkey           = mthca_query_pkey;
+	dev->ib_dev.query_gid            = mthca_query_gid;
+	dev->ib_dev.alloc_pd             = mthca_alloc_pd;
+	dev->ib_dev.dealloc_pd           = mthca_dealloc_pd;
+	dev->ib_dev.create_ah            = mthca_ah_create;
+	dev->ib_dev.destroy_ah           = mthca_ah_destroy;
+	dev->ib_dev.create_qp            = mthca_create_qp;
+	dev->ib_dev.modify_qp            = mthca_modify_qp;
+	dev->ib_dev.destroy_qp           = mthca_destroy_qp;
+	dev->ib_dev.post_send            = mthca_post_send;
+	dev->ib_dev.post_recv            = mthca_post_receive;
+	dev->ib_dev.create_cq            = mthca_create_cq;
+	dev->ib_dev.destroy_cq           = mthca_destroy_cq;
+	dev->ib_dev.poll_cq              = mthca_poll_cq;
+	dev->ib_dev.req_notify_cq        = mthca_req_notify_cq;
+	dev->ib_dev.get_dma_mr           = mthca_get_dma_mr;
+	dev->ib_dev.reg_phys_mr          = mthca_reg_phys_mr;
+	dev->ib_dev.dereg_mr             = mthca_dereg_mr;
+	dev->ib_dev.attach_mcast         = mthca_multicast_attach;
+	dev->ib_dev.detach_mcast         = mthca_multicast_detach;
+	dev->ib_dev.process_mad          = mthca_process_mad;
+
+	ret = ib_register_device(&dev->ib_dev);
+	if (ret)
+		return ret;
+
+	for (i = 0; i < ARRAY_SIZE(mthca_class_attributes); ++i) {
+		ret = class_device_create_file(&dev->ib_dev.class_dev,
+					       mthca_class_attributes[i]);
+		if (ret) {
+			ib_unregister_device(&dev->ib_dev);
+			return ret;
+		}
+	}
+
+	return 0;
+}
+
+void mthca_unregister_device(struct mthca_dev *dev)
+{
+	ib_unregister_device(&dev->ib_dev);
+}
drivers/infiniband/hw/mthca/mthca_provider.h new file +225 -0

225 lines not on this page · raw message

#11 lore raw

PATCH v4 11/24 Add Mellanox HCA low-level driver (FW commands)

Add firmware command processing code for Mellanox HCA driver.

drivers/infiniband/hw/mthca/mthca_cmd.c new file +1573 -0
@@ -0,0 +1,1573 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: mthca_cmd.c 1349 2004-12-16 21:09:43Z roland $
+ */
+
+#include <linux/sched.h>
+#include <linux/pci.h>
+#include <linux/errno.h>
+#include <asm/io.h>
+
+#include "mthca_dev.h"
+#include "mthca_config_reg.h"
+#include "mthca_cmd.h"
+
+#define CMD_POLL_TOKEN 0xffff
+
+enum {
+	HCR_IN_PARAM_OFFSET    = 0x00,
+	HCR_IN_MODIFIER_OFFSET = 0x08,
+	HCR_OUT_PARAM_OFFSET   = 0x0c,
+	HCR_TOKEN_OFFSET       = 0x14,
+	HCR_STATUS_OFFSET      = 0x18,
+
+	HCR_OPMOD_SHIFT        = 12,
+	HCA_E_BIT              = 22,
+	HCR_GO_BIT             = 23
+};
+
+enum {
+	/* initialization and general commands */
+	CMD_SYS_EN          = 0x1,
+	CMD_SYS_DIS         = 0x2,
+	CMD_MAP_FA          = 0xfff,
+	CMD_UNMAP_FA        = 0xffe,
+	CMD_RUN_FW          = 0xff6,
+	CMD_MOD_STAT_CFG    = 0x34,
+	CMD_QUERY_DEV_LIM   = 0x3,
+	CMD_QUERY_FW        = 0x4,
+	CMD_ENABLE_LAM      = 0xff8,
+	CMD_DISABLE_LAM     = 0xff7,
+	CMD_QUERY_DDR       = 0x5,
+	CMD_QUERY_ADAPTER   = 0x6,
+	CMD_INIT_HCA        = 0x7,
+	CMD_CLOSE_HCA       = 0x8,
+	CMD_INIT_IB         = 0x9,
+	CMD_CLOSE_IB        = 0xa,
+	CMD_QUERY_HCA       = 0xb,
+	CMD_SET_IB          = 0xc,
+	CMD_ACCESS_DDR      = 0x2e,
+	CMD_MAP_ICM         = 0xffa,
+	CMD_UNMAP_ICM       = 0xff9,
+	CMD_MAP_ICM_AUX     = 0xffc,
+	CMD_UNMAP_ICM_AUX   = 0xffb,
+	CMD_SET_ICM_SIZE    = 0xffd,
+
+	/* TPT commands */
+	CMD_SW2HW_MPT 	    = 0xd,
+	CMD_QUERY_MPT 	    = 0xe,
+	CMD_HW2SW_MPT 	    = 0xf,
+	CMD_READ_MTT        = 0x10,
+	CMD_WRITE_MTT       = 0x11,
+	CMD_SYNC_TPT        = 0x2f,
+
+	/* EQ commands */
+	CMD_MAP_EQ          = 0x12,
+	CMD_SW2HW_EQ 	    = 0x13,
+	CMD_HW2SW_EQ 	    = 0x14,
+	CMD_QUERY_EQ        = 0x15,
+
+	/* CQ commands */
+	CMD_SW2HW_CQ 	    = 0x16,
+	CMD_HW2SW_CQ 	    = 0x17,
+	CMD_QUERY_CQ 	    = 0x18,
+	CMD_RESIZE_CQ       = 0x2c,
+
+	/* SRQ commands */
+	CMD_SW2HW_SRQ 	    = 0x35,
+	CMD_HW2SW_SRQ 	    = 0x36,
+	CMD_QUERY_SRQ       = 0x37,
+
+	/* QP/EE commands */
+	CMD_RST2INIT_QPEE   = 0x19,
+	CMD_INIT2RTR_QPEE   = 0x1a,
+	CMD_RTR2RTS_QPEE    = 0x1b,
+	CMD_RTS2RTS_QPEE    = 0x1c,
+	CMD_SQERR2RTS_QPEE  = 0x1d,
+	CMD_2ERR_QPEE       = 0x1e,
+	CMD_RTS2SQD_QPEE    = 0x1f,
+	CMD_SQD2SQD_QPEE    = 0x38,
+	CMD_SQD2RTS_QPEE    = 0x20,
+	CMD_ERR2RST_QPEE    = 0x21,
+	CMD_QUERY_QPEE      = 0x22,
+	CMD_INIT2INIT_QPEE  = 0x2d,
+	CMD_SUSPEND_QPEE    = 0x32,
+	CMD_UNSUSPEND_QPEE  = 0x33,
+	/* special QPs and management commands */
+	CMD_CONF_SPECIAL_QP = 0x23,
+	CMD_MAD_IFC         = 0x24,
+
+	/* multicast commands */
+	CMD_READ_MGM        = 0x25,
+	CMD_WRITE_MGM       = 0x26,
+	CMD_MGID_HASH       = 0x27,
+
+	/* miscellaneous commands */
+	CMD_DIAG_RPRT       = 0x30,
+	CMD_NOP             = 0x31,
+
+	/* debug commands */
+	CMD_QUERY_DEBUG_MSG = 0x2a,
+	CMD_SET_DEBUG_MSG   = 0x2b,
+};
+
+/*
+ * According to Mellanox code, FW may be starved and never complete
+ * commands.  So we can't use strict timeouts described in PRM -- we
+ * just arbitrarily select 60 seconds for now.
+ */
+#if 0
+/*
+ * Round up and add 1 to make sure we get the full wait time (since we
+ * will be starting in the middle of a jiffy)
+ */
+enum {
+	CMD_TIME_CLASS_A = (HZ + 999) / 1000 + 1,
+	CMD_TIME_CLASS_B = (HZ +  99) /  100 + 1,
+	CMD_TIME_CLASS_C = (HZ +   9) /   10 + 1
+};
+#else
+enum {
+	CMD_TIME_CLASS_A = 60 * HZ,
+	CMD_TIME_CLASS_B = 60 * HZ,
+	CMD_TIME_CLASS_C = 60 * HZ
+};
+#endif
+
+enum {
+	GO_BIT_TIMEOUT = HZ * 10
+};
+
+struct mthca_cmd_context {
+	struct completion done;
+	struct timer_list timer;
+	int               result;
+	int               next;
+	u64               out_param;
+	u16               token;
+	u8                status;
+};
+
+static inline int go_bit(struct mthca_dev *dev)
+{
+	return readl(dev->hcr + HCR_STATUS_OFFSET) &
+		swab32(1 << HCR_GO_BIT);
+}
+
+static int mthca_cmd_post(struct mthca_dev *dev,
+			  u64 in_param,
+			  u64 out_param,
+			  u32 in_modifier,
+			  u8 op_modifier,
+			  u16 op,
+			  u16 token,
+			  int event)
+{
+	int err = 0;
+
+	if (down_interruptible(&dev->cmd.hcr_sem))
+		return -EINTR;
+
+	if (event) {
+		unsigned long end = jiffies + GO_BIT_TIMEOUT;
+
+		while (go_bit(dev) && time_before(jiffies, end)) {
+			set_current_state(TASK_RUNNING);
+			schedule();
+		}
+	}
+
+	if (go_bit(dev)) {
+		err = -EAGAIN;
+		goto out;
+	}
+
+	/*
+	 * We use writel (instead of something like memcpy_toio)
+	 * because writes of less than 32 bits to the HCR don't work
+	 * (and some architectures such as ia64 implement memcpy_toio
+	 * in terms of writeb).
+	 */
+	__raw_writel(cpu_to_be32(in_param >> 32),           dev->hcr + 0 * 4);
+	__raw_writel(cpu_to_be32(in_param & 0xfffffffful),  dev->hcr + 1 * 4);
+	__raw_writel(cpu_to_be32(in_modifier),              dev->hcr + 2 * 4);
+	__raw_writel(cpu_to_be32(out_param >> 32),          dev->hcr + 3 * 4);
+	__raw_writel(cpu_to_be32(out_param & 0xfffffffful), dev->hcr + 4 * 4);
+	__raw_writel(cpu_to_be32(token << 16),              dev->hcr + 5 * 4);
+
+	/* __raw_writel may not order writes. */
+	wmb();
+
+	__raw_writel(cpu_to_be32((1 << HCR_GO_BIT)                |
+				 (event ? (1 << HCA_E_BIT) : 0)   |
+				 (op_modifier << HCR_OPMOD_SHIFT) |
+				 op),                       dev->hcr + 6 * 4);
+
+out:
+	up(&dev->cmd.hcr_sem);
+	return err;
+}
+
+static int mthca_cmd_poll(struct mthca_dev *dev,
+			  u64 in_param,
+			  u64 *out_param,
+			  int out_is_imm,
+			  u32 in_modifier,
+			  u8 op_modifier,
+			  u16 op,
+			  unsigned long timeout,
+			  u8 *status)
+{
+	int err = 0;
+	unsigned long end;
+
+	if (down_interruptible(&dev->cmd.poll_sem))
+		return -EINTR;
+
+	err = mthca_cmd_post(dev, in_param,
+			     out_param ? *out_param : 0,
+			     in_modifier, op_modifier,
+			     op, CMD_POLL_TOKEN, 0);
+	if (err)
+		goto out;
+
+	end = timeout + jiffies;
+	while (go_bit(dev) && time_before(jiffies, end)) {
+		set_current_state(TASK_RUNNING);
+		schedule();
+	}
+
+	if (go_bit(dev)) {
+		err = -EBUSY;
+		goto out;
+	}
+
+	if (out_is_imm) {
+		memcpy_fromio(out_param, dev->hcr + HCR_OUT_PARAM_OFFSET, sizeof (u64));
+		be64_to_cpus(out_param);
+	}
+
+	*status = be32_to_cpu(__raw_readl(dev->hcr + HCR_STATUS_OFFSET)) >> 24;
+
+out:
+	up(&dev->cmd.poll_sem);
+	return err;
+}
+
+void mthca_cmd_event(struct mthca_dev *dev,
+		     u16 token,
+		     u8  status,
+		     u64 out_param)
+{
+	struct mthca_cmd_context *context =
+		&dev->cmd.context[token & dev->cmd.token_mask];
+
+	/* previously timed out command completing at long last */
+	if (token != context->token)
+		return;
+
+	context->result    = 0;
+	context->status    = status;
+	context->out_param = out_param;
+
+	context->token += dev->cmd.token_mask + 1;
+
+	complete(&context->done);
+}
+
+static void event_timeout(unsigned long context_ptr)
+{
+	struct mthca_cmd_context *context =
+		(struct mthca_cmd_context *) context_ptr;
+
+	context->result = -EBUSY;
+	complete(&context->done);
+}
+
+static int mthca_cmd_wait(struct mthca_dev *dev,
+			  u64 in_param,
+			  u64 *out_param,
+			  int out_is_imm,
+			  u32 in_modifier,
+			  u8 op_modifier,
+			  u16 op,
+			  unsigned long timeout,
+			  u8 *status)
+{
+	int err = 0;
+	struct mthca_cmd_context *context;
+
+	if (down_interruptible(&dev->cmd.event_sem))
+		return -EINTR;
+
+	spin_lock(&dev->cmd.context_lock);
+	BUG_ON(dev->cmd.free_head < 0);
+	context = &dev->cmd.context[dev->cmd.free_head];
+	dev->cmd.free_head = context->next;
+	spin_unlock(&dev->cmd.context_lock);
+
+	init_completion(&context->done);
+
+	err = mthca_cmd_post(dev, in_param,
+			     out_param ? *out_param : 0,
+			     in_modifier, op_modifier,
+			     op, context->token, 1);
+	if (err)
+		goto out;
+
+	context->timer.expires  = jiffies + timeout;
+	add_timer(&context->timer);
+
+	wait_for_completion(&context->done);
+	del_timer_sync(&context->timer);
+
+	err = context->result;
+	if (err)
+		goto out;
+
+	*status = context->status;
+	if (*status)
+		mthca_dbg(dev, "Command %02x completed with status %02x\n",
+			  op, *status);
+
+	if (out_is_imm)
+		*out_param = context->out_param;
+
+out:
+	spin_lock(&dev->cmd.context_lock);
+	context->next = dev->cmd.free_head;
+	dev->cmd.free_head = context - dev->cmd.context;
+	spin_unlock(&dev->cmd.context_lock);
+
+	up(&dev->cmd.event_sem);
+	return err;
+}
+
+/* Invoke a command with an output mailbox */
+static int mthca_cmd_box(struct mthca_dev *dev,
+			 u64 in_param,
+			 u64 out_param,
+			 u32 in_modifier,
+			 u8 op_modifier,
+			 u16 op,
+			 unsigned long timeout,
+			 u8 *status)
+{
+	if (dev->cmd.use_events)
+		return mthca_cmd_wait(dev, in_param, &out_param, 0,
+				      in_modifier, op_modifier, op,
+				      timeout, status);
+	else
+		return mthca_cmd_poll(dev, in_param, &out_param, 0,
+				      in_modifier, op_modifier, op,
+				      timeout, status);
+}
+
+/* Invoke a command with no output parameter */
+static int mthca_cmd(struct mthca_dev *dev,
+		     u64 in_param,
+		     u32 in_modifier,
+		     u8 op_modifier,
+		     u16 op,
+		     unsigned long timeout,
+		     u8 *status)
+{
+	return mthca_cmd_box(dev, in_param, 0, in_modifier,
+			     op_modifier, op, timeout, status);
+}
+
+/*
+ * Invoke a command with an immediate output parameter (and copy the
+ * output into the caller's out_param pointer after the command
+ * executes).
+ */
+static int mthca_cmd_imm(struct mthca_dev *dev,
+			 u64 in_param,
+			 u64 *out_param,
+			 u32 in_modifier,
+			 u8 op_modifier,
+			 u16 op,
+			 unsigned long timeout,
+			 u8 *status)
+{
+	if (dev->cmd.use_events)
+		return mthca_cmd_wait(dev, in_param, out_param, 1,
+				      in_modifier, op_modifier, op,
+				      timeout, status);
+	else
+		return mthca_cmd_poll(dev, in_param, out_param, 1,
+				      in_modifier, op_modifier, op,
+				      timeout, status);
+}
+
+/*
+ * Switch to using events to issue FW commands (should be called after
+ * event queue to command events has been initialized).
+ */
+int mthca_cmd_use_events(struct mthca_dev *dev)
+{
+	int i;
+
+	dev->cmd.context = kmalloc(dev->cmd.max_cmds *
+				   sizeof (struct mthca_cmd_context),
+				   GFP_KERNEL);
+	if (!dev->cmd.context)
+		return -ENOMEM;
+
+	for (i = 0; i < dev->cmd.max_cmds; ++i) {
+		dev->cmd.context[i].token = i;
+		dev->cmd.context[i].next = i + 1;
+		init_timer(&dev->cmd.context[i].timer);
+		dev->cmd.context[i].timer.data     =
+			(unsigned long) &dev->cmd.context[i];
+		dev->cmd.context[i].timer.function = event_timeout;
+	}
+
+	dev->cmd.context[dev->cmd.max_cmds - 1].next = -1;
+	dev->cmd.free_head = 0;
+
+	sema_init(&dev->cmd.event_sem, dev->cmd.max_cmds);
+	spin_lock_init(&dev->cmd.context_lock);
+
+	for (dev->cmd.token_mask = 1;
+	     dev->cmd.token_mask < dev->cmd.max_cmds;
+	     dev->cmd.token_mask <<= 1)
+		; /* nothing */
+	--dev->cmd.token_mask;
+
+	dev->cmd.use_events = 1;
+	down(&dev->cmd.poll_sem);
+
+	return 0;
+}
+
+/*
+ * Switch back to polling (used when shutting down the device)
+ */
+void mthca_cmd_use_polling(struct mthca_dev *dev)
+{
+	int i;
+
+	dev->cmd.use_events = 0;
+
+	for (i = 0; i < dev->cmd.max_cmds; ++i)
+		down(&dev->cmd.event_sem);
+
+	kfree(dev->cmd.context);
+
+	up(&dev->cmd.poll_sem);
+}
+
+int mthca_SYS_EN(struct mthca_dev *dev, u8 *status)
+{
+	u64 out;
+	int ret;
+
+	ret = mthca_cmd_imm(dev, 0, &out, 0, 0, CMD_SYS_EN, HZ, status);
+
+	if (*status == MTHCA_CMD_STAT_DDR_MEM_ERR)
+		mthca_warn(dev, "SYS_EN DDR error: syn=%x, sock=%d, "
+			   "sladdr=%d, SPD source=%s\n",
+			   (int) (out >> 6) & 0xf, (int) (out >> 4) & 3,
+			   (int) (out >> 1) & 7, (int) out & 1 ? "NVMEM" : "DIMM");
+
+	return ret;
+}
+
+int mthca_SYS_DIS(struct mthca_dev *dev, u8 *status)
+{
+	return mthca_cmd(dev, 0, 0, 0, CMD_SYS_DIS, HZ, status);
+}
+
+int mthca_MAP_FA(struct mthca_dev *dev, int count,
+		 struct scatterlist *sglist, u8 *status)
+{
+	u32 *inbox;
+	dma_addr_t indma;
+	int lg;
+	int nent = 0;
+	int i, j;
+	int err = 0;
+	int ts = 0;
+
+	inbox = pci_alloc_consistent(dev->pdev, PAGE_SIZE, &indma);
+	memset(inbox, 0, PAGE_SIZE);
+
+	for (i = 0; i < count; ++i) {
+		/*
+		 * We have to pass pages that are aligned to their
+		 * size, so find the least significant 1 in the
+		 * address or size and use that as our log2 size.
+		 */
+		lg = ffs(sg_dma_address(sglist + i) | sg_dma_len(sglist + i)) - 1;
+		if (lg < 12) {
+			mthca_warn(dev, "Got FW area not aligned to 4K (%llx/%x).\n",
+				   (unsigned long long) sg_dma_address(sglist + i),
+				   sg_dma_len(sglist + i));
+			err = -EINVAL;
+			goto out;
+		}
+		for (j = 0; j < sg_dma_len(sglist + i) / (1 << lg); ++j, ++nent) {
+			*((__be64 *) (inbox + nent * 4 + 2)) =
+				cpu_to_be64((sg_dma_address(sglist + i) +
+					     (j << lg)) |
+					    (lg - 12));
+			ts += 1 << (lg - 10);
+			if (nent == PAGE_SIZE / 16) {
+				err = mthca_cmd(dev, indma, nent, 0, CMD_MAP_FA,
+						CMD_TIME_CLASS_B, status);
+				if (err || *status)
+					goto out;
+				nent = 0;
+			}
+		}
+	}
+
+	if (nent) {
+		err = mthca_cmd(dev, indma, nent, 0, CMD_MAP_FA,
+				CMD_TIME_CLASS_B, status);
+	}
+
+	mthca_dbg(dev, "Mapped %d KB of host memory for FW.\n", ts);
+
+out:
+	pci_free_consistent(dev->pdev, PAGE_SIZE, inbox, indma);
+	return err;
+}
+
+int mthca_UNMAP_FA(struct mthca_dev *dev, u8 *status)
+{
+	return mthca_cmd(dev, 0, 0, 0, CMD_UNMAP_FA, CMD_TIME_CLASS_B, status);
+}
+
+int mthca_RUN_FW(struct mthca_dev *dev, u8 *status)
+{
+	return mthca_cmd(dev, 0, 0, 0, CMD_RUN_FW, CMD_TIME_CLASS_A, status);
+}
+
+int mthca_QUERY_FW(struct mthca_dev *dev, u8 *status)
+{
+	u32 *outbox;
+	dma_addr_t outdma;
+	int err = 0;
+	u8 lg;
+
+#define QUERY_FW_OUT_SIZE             0x100
+#define QUERY_FW_VER_OFFSET            0x00
+#define QUERY_FW_MAX_CMD_OFFSET        0x0f
+#define QUERY_FW_ERR_START_OFFSET      0x30
+#define QUERY_FW_ERR_SIZE_OFFSET       0x38
+
+#define QUERY_FW_START_OFFSET          0x20
+#define QUERY_FW_END_OFFSET            0x28
+
+#define QUERY_FW_SIZE_OFFSET           0x00
+#define QUERY_FW_CLR_INT_BASE_OFFSET   0x20
+#define QUERY_FW_EQ_ARM_BASE_OFFSET    0x40
+#define QUERY_FW_EQ_SET_CI_BASE_OFFSET 0x48
+
+	outbox = pci_alloc_consistent(dev->pdev, QUERY_FW_OUT_SIZE, &outdma);
+	if (!outbox) {
+		return -ENOMEM;
+	}
+
+	err = mthca_cmd_box(dev, 0, outdma, 0, 0, CMD_QUERY_FW,
+			    CMD_TIME_CLASS_A, status);
+
+	if (err)
+		goto out;
+
+	MTHCA_GET(dev->fw_ver,   outbox, QUERY_FW_VER_OFFSET);
+	/*
+	 * FW subminor version is at more signifant bits than minor
+	 * version, so swap here.
+	 */
+	dev->fw_ver = (dev->fw_ver & 0xffff00000000ull) |
+		((dev->fw_ver & 0xffff0000ull) >> 16) |
+		((dev->fw_ver & 0x0000ffffull) << 16);
+
+	MTHCA_GET(lg, outbox, QUERY_FW_MAX_CMD_OFFSET);
+	dev->cmd.max_cmds = 1 << lg;
+
+	mthca_dbg(dev, "FW version %012llx, max commands %d\n",
+		  (unsigned long long) dev->fw_ver, dev->cmd.max_cmds);
+
+	if (dev->hca_type == ARBEL_NATIVE) {
+		MTHCA_GET(dev->fw.arbel.fw_pages,       outbox, QUERY_FW_SIZE_OFFSET);
+		MTHCA_GET(dev->fw.arbel.clr_int_base,   outbox, QUERY_FW_CLR_INT_BASE_OFFSET);
+		MTHCA_GET(dev->fw.arbel.eq_arm_base,    outbox, QUERY_FW_EQ_ARM_BASE_OFFSET);
+		MTHCA_GET(dev->fw.arbel.eq_set_ci_base, outbox, QUERY_FW_EQ_SET_CI_BASE_OFFSET);
+		mthca_dbg(dev, "FW size %d KB\n", dev->fw.arbel.fw_pages << 2);
+
+		/*
+		 * Arbel page size is always 4 KB; round up number of
+		 * system pages needed.
+		 */
+		dev->fw.arbel.fw_pages =
+			(dev->fw.arbel.fw_pages + (1 << (PAGE_SHIFT - 12)) - 1) >>
+			(PAGE_SHIFT - 12);
+
+		mthca_dbg(dev, "Clear int @ %llx, EQ arm @ %llx, EQ set CI @ %llx\n",
+			  (unsigned long long) dev->fw.arbel.clr_int_base,
+			  (unsigned long long) dev->fw.arbel.eq_arm_base,
+			  (unsigned long long) dev->fw.arbel.eq_set_ci_base);
+	} else {
+		MTHCA_GET(dev->fw.tavor.fw_start, outbox, QUERY_FW_START_OFFSET);
+		MTHCA_GET(dev->fw.tavor.fw_end,   outbox, QUERY_FW_END_OFFSET);
+
+		mthca_dbg(dev, "FW size %d KB (start %llx, end %llx)\n",
+			  (int) ((dev->fw.tavor.fw_end - dev->fw.tavor.fw_start) >> 10),
+			  (unsigned long long) dev->fw.tavor.fw_start,
+			  (unsigned long long) dev->fw.tavor.fw_end);
+	}
+
+out:
+	pci_free_consistent(dev->pdev, QUERY_FW_OUT_SIZE, outbox, outdma);
+	return err;
+}
+
+int mthca_ENABLE_LAM(struct mthca_dev *dev, u8 *status)
+{
+	u8 info;
+	u32 *outbox;
+	dma_addr_t outdma;
+	int err = 0;
+
+#define ENABLE_LAM_OUT_SIZE         0x100
+#define ENABLE_LAM_START_OFFSET     0x00
+#define ENABLE_LAM_END_OFFSET       0x08
+#define ENABLE_LAM_INFO_OFFSET      0x13
+
+#define ENABLE_LAM_INFO_HIDDEN_FLAG (1 << 4)
+#define ENABLE_LAM_INFO_ECC_MASK    0x3
+
+	outbox = pci_alloc_consistent(dev->pdev, ENABLE_LAM_OUT_SIZE, &outdma);
+	if (!outbox)
+		return -ENOMEM;
+
+	err = mthca_cmd_box(dev, 0, outdma, 0, 0, CMD_ENABLE_LAM,
+			    CMD_TIME_CLASS_C, status);
+
+	if (err)
+		goto out;
+
+	if (*status == MTHCA_CMD_STAT_LAM_NOT_PRE)
+		goto out;
+
+	MTHCA_GET(dev->ddr_start, outbox, ENABLE_LAM_START_OFFSET);
+	MTHCA_GET(dev->ddr_end,   outbox, ENABLE_LAM_END_OFFSET);
+	MTHCA_GET(info,           outbox, ENABLE_LAM_INFO_OFFSET);
+
+	if (!!(info & ENABLE_LAM_INFO_HIDDEN_FLAG) !=
+	    !!(dev->mthca_flags & MTHCA_FLAG_DDR_HIDDEN)) {
+		mthca_info(dev, "FW reports that HCA-attached memory "
+			   "is %s hidden; does not match PCI config\n",
+			   (info & ENABLE_LAM_INFO_HIDDEN_FLAG) ?
+			   "" : "not");
+	}
+	if (info & ENABLE_LAM_INFO_HIDDEN_FLAG)
+		mthca_dbg(dev, "HCA-attached memory is hidden.\n");
+
+	mthca_dbg(dev, "HCA memory size %d KB (start %llx, end %llx)\n",
+		  (int) ((dev->ddr_end - dev->ddr_start) >> 10),
+		  (unsigned long long) dev->ddr_start,
+		  (unsigned long long) dev->ddr_end);
+
+out:
+	pci_free_consistent(dev->pdev, ENABLE_LAM_OUT_SIZE, outbox, outdma);
+	return err;
+}
+
+int mthca_DISABLE_LAM(struct mthca_dev *dev, u8 *status)
+{
+	return mthca_cmd(dev, 0, 0, 0, CMD_SYS_DIS, CMD_TIME_CLASS_C, status);
+}
+
+int mthca_QUERY_DDR(struct mthca_dev *dev, u8 *status)
+{
+	u8 info;
+	u32 *outbox;
+	dma_addr_t outdma;
+	int err = 0;
+
+#define QUERY_DDR_OUT_SIZE         0x100
+#define QUERY_DDR_START_OFFSET     0x00
+#define QUERY_DDR_END_OFFSET       0x08
+#define QUERY_DDR_INFO_OFFSET      0x13
+
+#define QUERY_DDR_INFO_HIDDEN_FLAG (1 << 4)
+#define QUERY_DDR_INFO_ECC_MASK    0x3
+
+	outbox = pci_alloc_consistent(dev->pdev, QUERY_DDR_OUT_SIZE, &outdma);
+	if (!outbox)
+		return -ENOMEM;
+
+	err = mthca_cmd_box(dev, 0, outdma, 0, 0, CMD_QUERY_DDR,
+			    CMD_TIME_CLASS_A, status);
+
+	if (err)
+		goto out;
+
+	MTHCA_GET(dev->ddr_start, outbox, QUERY_DDR_START_OFFSET);
+	MTHCA_GET(dev->ddr_end,   outbox, QUERY_DDR_END_OFFSET);
+	MTHCA_GET(info,           outbox, QUERY_DDR_INFO_OFFSET);
+
+	if (!!(info & QUERY_DDR_INFO_HIDDEN_FLAG) !=
+	    !!(dev->mthca_flags & MTHCA_FLAG_DDR_HIDDEN)) {
+		mthca_info(dev, "FW reports that HCA-attached memory "
+			   "is %s hidden; does not match PCI config\n",
+			   (info & QUERY_DDR_INFO_HIDDEN_FLAG) ?
+			   "" : "not");
+	}
+	if (info & QUERY_DDR_INFO_HIDDEN_FLAG)
+		mthca_dbg(dev, "HCA-attached memory is hidden.\n");
+
+	mthca_dbg(dev, "HCA memory size %d KB (start %llx, end %llx)\n",
+		  (int) ((dev->ddr_end - dev->ddr_start) >> 10),
+		  (unsigned long long) dev->ddr_start,
+		  (unsigned long long) dev->ddr_end);
+
+out:
+	pci_free_consistent(dev->pdev, QUERY_DDR_OUT_SIZE, outbox, outdma);
+	return err;
+}
+
+int mthca_QUERY_DEV_LIM(struct mthca_dev *dev,
+			struct mthca_dev_lim *dev_lim, u8 *status)
+{
+	u32 *outbox;
+	dma_addr_t outdma;
+	u8 field;
+	u16 size;
+	int err;
+
+#define QUERY_DEV_LIM_OUT_SIZE             0x100
+#define QUERY_DEV_LIM_MAX_SRQ_SZ_OFFSET     0x10
+#define QUERY_DEV_LIM_MAX_QP_SZ_OFFSET      0x11
+#define QUERY_DEV_LIM_RSVD_QP_OFFSET        0x12
+#define QUERY_DEV_LIM_MAX_QP_OFFSET         0x13
+#define QUERY_DEV_LIM_RSVD_SRQ_OFFSET       0x14
+#define QUERY_DEV_LIM_MAX_SRQ_OFFSET        0x15
+#define QUERY_DEV_LIM_RSVD_EEC_OFFSET       0x16
+#define QUERY_DEV_LIM_MAX_EEC_OFFSET        0x17
+#define QUERY_DEV_LIM_MAX_CQ_SZ_OFFSET      0x19
+#define QUERY_DEV_LIM_RSVD_CQ_OFFSET        0x1a
+#define QUERY_DEV_LIM_MAX_CQ_OFFSET         0x1b
+#define QUERY_DEV_LIM_MAX_MPT_OFFSET        0x1d
+#define QUERY_DEV_LIM_RSVD_EQ_OFFSET        0x1e
+#define QUERY_DEV_LIM_MAX_EQ_OFFSET         0x1f
+#define QUERY_DEV_LIM_RSVD_MTT_OFFSET       0x20
+#define QUERY_DEV_LIM_MAX_MRW_SZ_OFFSET     0x21
+#define QUERY_DEV_LIM_RSVD_MRW_OFFSET       0x22
+#define QUERY_DEV_LIM_MAX_MTT_SEG_OFFSET    0x23
+#define QUERY_DEV_LIM_MAX_AV_OFFSET         0x27
+#define QUERY_DEV_LIM_MAX_REQ_QP_OFFSET     0x29
+#define QUERY_DEV_LIM_MAX_RES_QP_OFFSET     0x2b
+#define QUERY_DEV_LIM_MAX_RDMA_OFFSET       0x2f
+#define QUERY_DEV_LIM_RSZ_SRQ_OFFSET        0x33
+#define QUERY_DEV_LIM_ACK_DELAY_OFFSET      0x35
+#define QUERY_DEV_LIM_MTU_WIDTH_OFFSET      0x36
+#define QUERY_DEV_LIM_VL_PORT_OFFSET        0x37
+#define QUERY_DEV_LIM_MAX_GID_OFFSET        0x3b
+#define QUERY_DEV_LIM_MAX_PKEY_OFFSET       0x3f
+#define QUERY_DEV_LIM_FLAGS_OFFSET          0x44
+#define QUERY_DEV_LIM_RSVD_UAR_OFFSET       0x48
+#define QUERY_DEV_LIM_UAR_SZ_OFFSET         0x49
+#define QUERY_DEV_LIM_PAGE_SZ_OFFSET        0x4b
+#define QUERY_DEV_LIM_MAX_SG_OFFSET         0x51
+#define QUERY_DEV_LIM_MAX_DESC_SZ_OFFSET    0x52
+#define QUERY_DEV_LIM_MAX_SG_RQ_OFFSET      0x55
+#define QUERY_DEV_LIM_MAX_DESC_SZ_RQ_OFFSET 0x56
+#define QUERY_DEV_LIM_MAX_QP_MCG_OFFSET     0x61
+#define QUERY_DEV_LIM_RSVD_MCG_OFFSET       0x62
+#define QUERY_DEV_LIM_MAX_MCG_OFFSET        0x63
+#define QUERY_DEV_LIM_RSVD_PD_OFFSET        0x64
+#define QUERY_DEV_LIM_MAX_PD_OFFSET         0x65
+#define QUERY_DEV_LIM_RSVD_RDD_OFFSET       0x66
+#define QUERY_DEV_LIM_MAX_RDD_OFFSET        0x67
+#define QUERY_DEV_LIM_EEC_ENTRY_SZ_OFFSET   0x80
+#define QUERY_DEV_LIM_QPC_ENTRY_SZ_OFFSET   0x82
+#define QUERY_DEV_LIM_EEEC_ENTRY_SZ_OFFSET  0x84
+#define QUERY_DEV_LIM_EQPC_ENTRY_SZ_OFFSET  0x86
+#define QUERY_DEV_LIM_EQC_ENTRY_SZ_OFFSET   0x88
+#define QUERY_DEV_LIM_CQC_ENTRY_SZ_OFFSET   0x8a
+#define QUERY_DEV_LIM_SRQ_ENTRY_SZ_OFFSET   0x8c
+#define QUERY_DEV_LIM_UAR_ENTRY_SZ_OFFSET   0x8e
+#define QUERY_DEV_LIM_MTT_ENTRY_SZ_OFFSET   0x90
+#define QUERY_DEV_LIM_MPT_ENTRY_SZ_OFFSET   0x92
+#define QUERY_DEV_LIM_PBL_SZ_OFFSET         0x96
+#define QUERY_DEV_LIM_BMME_FLAGS_OFFSET     0x97
+#define QUERY_DEV_LIM_RSVD_LKEY_OFFSET      0x98
+#define QUERY_DEV_LIM_LAMR_OFFSET           0x9f
+#define QUERY_DEV_LIM_MAX_ICM_SZ_OFFSET     0xa0
+
+	outbox = pci_alloc_consistent(dev->pdev, QUERY_DEV_LIM_OUT_SIZE, &outdma);
+	if (!outbox)
+		return -ENOMEM;
+
+	err = mthca_cmd_box(dev, 0, outdma, 0, 0, CMD_QUERY_DEV_LIM,
+			    CMD_TIME_CLASS_A, status);
+
+	if (err)
+		goto out;
+
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_SRQ_SZ_OFFSET);
+	dev_lim->max_srq_sz = 1 << field;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_QP_SZ_OFFSET);
+	dev_lim->max_qp_sz = 1 << field;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_QP_OFFSET);
+	dev_lim->reserved_qps = 1 << (field & 0xf);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_QP_OFFSET);
+	dev_lim->max_qps = 1 << (field & 0x1f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_SRQ_OFFSET);
+	dev_lim->reserved_srqs = 1 << (field >> 4);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_SRQ_OFFSET);
+	dev_lim->max_srqs = 1 << (field & 0x1f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_EEC_OFFSET);
+	dev_lim->reserved_eecs = 1 << (field & 0xf);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_EEC_OFFSET);
+	dev_lim->max_eecs = 1 << (field & 0x1f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_CQ_SZ_OFFSET);
+	dev_lim->max_cq_sz = 1 << field;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_CQ_OFFSET);
+	dev_lim->reserved_cqs = 1 << (field & 0xf);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_CQ_OFFSET);
+	dev_lim->max_cqs = 1 << (field & 0x1f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_MPT_OFFSET);
+	dev_lim->max_mpts = 1 << (field & 0x3f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_EQ_OFFSET);
+	dev_lim->reserved_eqs = 1 << (field & 0xf);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_EQ_OFFSET);
+	dev_lim->max_eqs = 1 << (field & 0x7);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_MTT_OFFSET);
+	dev_lim->reserved_mtts = 1 << (field >> 4);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_MRW_SZ_OFFSET);
+	dev_lim->max_mrw_sz = 1 << field;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_MRW_OFFSET);
+	dev_lim->reserved_mrws = 1 << (field & 0xf);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_MTT_SEG_OFFSET);
+	dev_lim->max_mtt_seg = 1 << (field & 0x3f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_REQ_QP_OFFSET);
+	dev_lim->max_requester_per_qp = 1 << (field & 0x3f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_RES_QP_OFFSET);
+	dev_lim->max_responder_per_qp = 1 << (field & 0x3f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_RDMA_OFFSET);
+	dev_lim->max_rdma_global = 1 << (field & 0x3f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_ACK_DELAY_OFFSET);
+	dev_lim->local_ca_ack_delay = field & 0x1f;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MTU_WIDTH_OFFSET);
+	dev_lim->max_mtu        = field >> 4;
+	dev_lim->max_port_width = field & 0xf;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_VL_PORT_OFFSET);
+	dev_lim->max_vl    = field >> 4;
+	dev_lim->num_ports = field & 0xf;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_GID_OFFSET);
+	dev_lim->max_gids = 1 << (field & 0xf);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_PKEY_OFFSET);
+	dev_lim->max_pkeys = 1 << (field & 0xf);
+	MTHCA_GET(dev_lim->flags, outbox, QUERY_DEV_LIM_FLAGS_OFFSET);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_UAR_OFFSET);
+	dev_lim->reserved_uars = field >> 4;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_UAR_SZ_OFFSET);
+	dev_lim->uar_size = 1 << ((field & 0x3f) + 20);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_PAGE_SZ_OFFSET);
+	dev_lim->min_page_sz = 1 << field;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_SG_OFFSET);
+	dev_lim->max_sg = field;
+
+	MTHCA_GET(size, outbox, QUERY_DEV_LIM_MAX_DESC_SZ_OFFSET);
+	dev_lim->max_desc_sz = size;
+
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_QP_MCG_OFFSET);
+	dev_lim->max_qp_per_mcg = 1 << field;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_MCG_OFFSET);
+	dev_lim->reserved_mgms = field & 0xf;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_MCG_OFFSET);
+	dev_lim->max_mcgs = 1 << field;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_PD_OFFSET);
+	dev_lim->reserved_pds = field >> 4;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_PD_OFFSET);
+	dev_lim->max_pds = 1 << (field & 0x3f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_RDD_OFFSET);
+	dev_lim->reserved_rdds = field >> 4;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_RDD_OFFSET);
+	dev_lim->max_rdds = 1 << (field & 0x3f);
+
+	MTHCA_GET(size, outbox, QUERY_DEV_LIM_EEC_ENTRY_SZ_OFFSET);
+	dev_lim->eec_entry_sz = size;
+	MTHCA_GET(size, outbox, QUERY_DEV_LIM_QPC_ENTRY_SZ_OFFSET);
+	dev_lim->qpc_entry_sz = size;
+	MTHCA_GET(size, outbox, QUERY_DEV_LIM_EEEC_ENTRY_SZ_OFFSET);
+	dev_lim->eeec_entry_sz = size;
+	MTHCA_GET(size, outbox, QUERY_DEV_LIM_EQPC_ENTRY_SZ_OFFSET);
+	dev_lim->eqpc_entry_sz = size;
+	MTHCA_GET(size, outbox, QUERY_DEV_LIM_EQC_ENTRY_SZ_OFFSET);
+	dev_lim->eqc_entry_sz = size;
+	MTHCA_GET(size, outbox, QUERY_DEV_LIM_CQC_ENTRY_SZ_OFFSET);
+	dev_lim->cqc_entry_sz = size;
+	MTHCA_GET(size, outbox, QUERY_DEV_LIM_SRQ_ENTRY_SZ_OFFSET);
+	dev_lim->srq_entry_sz = size;
+	MTHCA_GET(size, outbox, QUERY_DEV_LIM_UAR_ENTRY_SZ_OFFSET);
+	dev_lim->uar_scratch_entry_sz = size;
+
+	mthca_dbg(dev, "Max QPs: %d, reserved QPs: %d, entry size: %d\n",
+		  dev_lim->max_qps, dev_lim->reserved_qps, dev_lim->qpc_entry_sz);
+	mthca_dbg(dev, "Max CQs: %d, reserved CQs: %d, entry size: %d\n",
+		  dev_lim->max_cqs, dev_lim->reserved_cqs, dev_lim->cqc_entry_sz);
+	mthca_dbg(dev, "Max EQs: %d, reserved EQs: %d, entry size: %d\n",
+		  dev_lim->max_eqs, dev_lim->reserved_eqs, dev_lim->eqc_entry_sz);
+	mthca_dbg(dev, "reserved MPTs: %d, reserved MTTs: %d\n",
+		  dev_lim->reserved_mrws, dev_lim->reserved_mtts);
+	mthca_dbg(dev, "Max PDs: %d, reserved PDs: %d, reserved UARs: %d\n",
+		  dev_lim->max_pds, dev_lim->reserved_pds, dev_lim->reserved_uars);
+	mthca_dbg(dev, "Max QP/MCG: %d, reserved MGMs: %d\n",
+		  dev_lim->max_pds, dev_lim->reserved_mgms);
+
+	mthca_dbg(dev, "Flags: %08x\n", dev_lim->flags);
+
+	if (dev->hca_type == ARBEL_NATIVE) {
+		MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSZ_SRQ_OFFSET);
+		dev_lim->hca.arbel.resize_srq = field & 1;
+		MTHCA_GET(size, outbox, QUERY_DEV_LIM_MTT_ENTRY_SZ_OFFSET);
+		dev_lim->hca.arbel.mtt_entry_sz = size;
+		MTHCA_GET(size, outbox, QUERY_DEV_LIM_MPT_ENTRY_SZ_OFFSET);
+		dev_lim->hca.arbel.mpt_entry_sz = size;
+		MTHCA_GET(field, outbox, QUERY_DEV_LIM_PBL_SZ_OFFSET);
+		dev_lim->hca.arbel.max_pbl_sz = 1 << (field & 0x3f);
+		MTHCA_GET(dev_lim->hca.arbel.bmme_flags, outbox,
+			  QUERY_DEV_LIM_BMME_FLAGS_OFFSET);
+		MTHCA_GET(dev_lim->hca.arbel.reserved_lkey, outbox,
+			  QUERY_DEV_LIM_RSVD_LKEY_OFFSET);
+		MTHCA_GET(field, outbox, QUERY_DEV_LIM_LAMR_OFFSET);
+		dev_lim->hca.arbel.lam_required = field & 1;
+		MTHCA_GET(dev_lim->hca.arbel.max_icm_sz, outbox,
+			  QUERY_DEV_LIM_MAX_ICM_SZ_OFFSET);
+
+		if (dev_lim->hca.arbel.bmme_flags & 1)
+			mthca_dbg(dev, "Base MM extensions: yes "
+				  "(flags %d, max PBL %d, rsvd L_Key %08x)\n",
+				  dev_lim->hca.arbel.bmme_flags,
+				  dev_lim->hca.arbel.max_pbl_sz,
+				  dev_lim->hca.arbel.reserved_lkey);
+		else
+			mthca_dbg(dev, "Base MM extensions: no\n");
+
+		mthca_dbg(dev, "Max ICM size %lld MB\n",
+			  (unsigned long long) dev_lim->hca.arbel.max_icm_sz >> 20);
+	} else {
+		MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_AV_OFFSET);
+		dev_lim->hca.tavor.max_avs = 1 << (field & 0x3f);
+	}
+
+out:
+	pci_free_consistent(dev->pdev, QUERY_DEV_LIM_OUT_SIZE, outbox, outdma);
+	return err;
+}
+
+int mthca_QUERY_ADAPTER(struct mthca_dev *dev,
+			struct mthca_adapter *adapter, u8 *status)
+{
+	u32 *outbox;
+	dma_addr_t outdma;
+	int err;
+
+#define QUERY_ADAPTER_OUT_SIZE             0x100
+#define QUERY_ADAPTER_VENDOR_ID_OFFSET     0x00
+#define QUERY_ADAPTER_DEVICE_ID_OFFSET     0x04
+#define QUERY_ADAPTER_REVISION_ID_OFFSET   0x08
+#define QUERY_ADAPTER_INTA_PIN_OFFSET      0x10
+
+	outbox = pci_alloc_consistent(dev->pdev, QUERY_ADAPTER_OUT_SIZE, &outdma);
+	if (!outbox)
+		return -ENOMEM;
+
+	err = mthca_cmd_box(dev, 0, outdma, 0, 0, CMD_QUERY_ADAPTER,
+			    CMD_TIME_CLASS_A, status);
+
+	if (err)
+		goto out;
+
+	MTHCA_GET(adapter->vendor_id, outbox, QUERY_ADAPTER_VENDOR_ID_OFFSET);
+	MTHCA_GET(adapter->device_id, outbox, QUERY_ADAPTER_DEVICE_ID_OFFSET);
+	MTHCA_GET(adapter->revision_id, outbox, QUERY_ADAPTER_REVISION_ID_OFFSET);
+	MTHCA_GET(adapter->inta_pin, outbox, QUERY_ADAPTER_INTA_PIN_OFFSET);
+
+out:
+	pci_free_consistent(dev->pdev, QUERY_DEV_LIM_OUT_SIZE, outbox, outdma);
+	return err;
+}
+
+int mthca_INIT_HCA(struct mthca_dev *dev,
+		   struct mthca_init_hca_param *param,
+		   u8 *status)
+{
+	u32 *inbox;
+	dma_addr_t indma;
+	int err;
+
+#define INIT_HCA_IN_SIZE             	 0x200
+#define INIT_HCA_FLAGS_OFFSET        	 0x014
+#define INIT_HCA_QPC_OFFSET          	 0x020
+#define  INIT_HCA_QPC_BASE_OFFSET    	 (INIT_HCA_QPC_OFFSET + 0x10)
+#define  INIT_HCA_LOG_QP_OFFSET      	 (INIT_HCA_QPC_OFFSET + 0x17)
+#define  INIT_HCA_EEC_BASE_OFFSET    	 (INIT_HCA_QPC_OFFSET + 0x20)
+#define  INIT_HCA_LOG_EEC_OFFSET     	 (INIT_HCA_QPC_OFFSET + 0x27)
+#define  INIT_HCA_SRQC_BASE_OFFSET   	 (INIT_HCA_QPC_OFFSET + 0x28)
+#define  INIT_HCA_LOG_SRQ_OFFSET     	 (INIT_HCA_QPC_OFFSET + 0x2f)
+#define  INIT_HCA_CQC_BASE_OFFSET    	 (INIT_HCA_QPC_OFFSET + 0x30)
+#define  INIT_HCA_LOG_CQ_OFFSET      	 (INIT_HCA_QPC_OFFSET + 0x37)
+#define  INIT_HCA_EQPC_BASE_OFFSET   	 (INIT_HCA_QPC_OFFSET + 0x40)
+#define  INIT_HCA_EEEC_BASE_OFFSET   	 (INIT_HCA_QPC_OFFSET + 0x50)
+#define  INIT_HCA_EQC_BASE_OFFSET    	 (INIT_HCA_QPC_OFFSET + 0x60)
+#define  INIT_HCA_LOG_EQ_OFFSET      	 (INIT_HCA_QPC_OFFSET + 0x67)
+#define  INIT_HCA_RDB_BASE_OFFSET    	 (INIT_HCA_QPC_OFFSET + 0x70)
+#define INIT_HCA_UDAV_OFFSET         	 0x0b0
+#define  INIT_HCA_UDAV_LKEY_OFFSET   	 (INIT_HCA_UDAV_OFFSET + 0x0)
+#define  INIT_HCA_UDAV_PD_OFFSET     	 (INIT_HCA_UDAV_OFFSET + 0x4)
+#define INIT_HCA_MCAST_OFFSET        	 0x0c0
+#define  INIT_HCA_MC_BASE_OFFSET         (INIT_HCA_MCAST_OFFSET + 0x00)
+#define  INIT_HCA_LOG_MC_ENTRY_SZ_OFFSET (INIT_HCA_MCAST_OFFSET + 0x12)
+#define  INIT_HCA_MC_HASH_SZ_OFFSET      (INIT_HCA_MCAST_OFFSET + 0x16)
+#define  INIT_HCA_LOG_MC_TABLE_SZ_OFFSET (INIT_HCA_MCAST_OFFSET + 0x1b)
+#define INIT_HCA_TPT_OFFSET              0x0f0
+#define  INIT_HCA_MPT_BASE_OFFSET        (INIT_HCA_TPT_OFFSET + 0x00)
+#define  INIT_HCA_MTT_SEG_SZ_OFFSET      (INIT_HCA_TPT_OFFSET + 0x09)
+#define  INIT_HCA_LOG_MPT_SZ_OFFSET      (INIT_HCA_TPT_OFFSET + 0x0b)
+#define  INIT_HCA_MTT_BASE_OFFSET        (INIT_HCA_TPT_OFFSET + 0x10)
+#define INIT_HCA_UAR_OFFSET              0x120
+#define  INIT_HCA_UAR_BASE_OFFSET        (INIT_HCA_UAR_OFFSET + 0x00)
+#define  INIT_HCA_UAR_PAGE_SZ_OFFSET     (INIT_HCA_UAR_OFFSET + 0x0b)
+#define  INIT_HCA_UAR_SCATCH_BASE_OFFSET (INIT_HCA_UAR_OFFSET + 0x10)
+
+	inbox = pci_alloc_consistent(dev->pdev, INIT_HCA_IN_SIZE, &indma);
+	if (!inbox)
+		return -ENOMEM;
+
+	memset(inbox, 0, INIT_HCA_IN_SIZE);
+
+#if defined(__LITTLE_ENDIAN)
+	*(inbox + INIT_HCA_FLAGS_OFFSET / 4) &= ~cpu_to_be32(1 << 1);
+#elif defined(__BIG_ENDIAN)
+	*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |= cpu_to_be32(1 << 1);
+#else
+#error Host endianness not defined
+#endif
+	/* Check port for UD address vector: */
+	*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |= cpu_to_be32(1);
+
+	/* We leave wqe_quota, responder_exu, etc as 0 (default) */
+
+	/* QPC/EEC/CQC/EQC/RDB attributes */
+
+	MTHCA_PUT(inbox, param->qpc_base,     INIT_HCA_QPC_BASE_OFFSET);
+	MTHCA_PUT(inbox, param->log_num_qps,  INIT_HCA_LOG_QP_OFFSET);
+	MTHCA_PUT(inbox, param->eec_base,     INIT_HCA_EEC_BASE_OFFSET);
+	MTHCA_PUT(inbox, param->log_num_eecs, INIT_HCA_LOG_EEC_OFFSET);
+	MTHCA_PUT(inbox, param->srqc_base,    INIT_HCA_SRQC_BASE_OFFSET);
+	MTHCA_PUT(inbox, param->log_num_srqs, INIT_HCA_LOG_SRQ_OFFSET);
+	MTHCA_PUT(inbox, param->cqc_base,     INIT_HCA_CQC_BASE_OFFSET);
+	MTHCA_PUT(inbox, param->log_num_cqs,  INIT_HCA_LOG_CQ_OFFSET);
+	MTHCA_PUT(inbox, param->eqpc_base,    INIT_HCA_EQPC_BASE_OFFSET);
+	MTHCA_PUT(inbox, param->eeec_base,    INIT_HCA_EEEC_BASE_OFFSET);
+	MTHCA_PUT(inbox, param->eqc_base,     INIT_HCA_EQC_BASE_OFFSET);
+	MTHCA_PUT(inbox, param->log_num_eqs,  INIT_HCA_LOG_EQ_OFFSET);
+	MTHCA_PUT(inbox, param->rdb_base,     INIT_HCA_RDB_BASE_OFFSET);
+
+	/* UD AV attributes */
+
+	/* multicast attributes */
+
+	MTHCA_PUT(inbox, param->mc_base,         INIT_HCA_MC_BASE_OFFSET);
+	MTHCA_PUT(inbox, param->log_mc_entry_sz, INIT_HCA_LOG_MC_ENTRY_SZ_OFFSET);
+	MTHCA_PUT(inbox, param->mc_hash_sz,      INIT_HCA_MC_HASH_SZ_OFFSET);
+	MTHCA_PUT(inbox, param->log_mc_table_sz, INIT_HCA_LOG_MC_TABLE_SZ_OFFSET);
+
+	/* TPT attributes */
+
+	MTHCA_PUT(inbox, param->mpt_base,   INIT_HCA_MPT_BASE_OFFSET);
+	MTHCA_PUT(inbox, param->mtt_seg_sz, INIT_HCA_MTT_SEG_SZ_OFFSET);
+	MTHCA_PUT(inbox, param->log_mpt_sz, INIT_HCA_LOG_MPT_SZ_OFFSET);
+	MTHCA_PUT(inbox, param->mtt_base,   INIT_HCA_MTT_BASE_OFFSET);
+
+	/* UAR attributes */
+	{
+		u8 uar_page_sz = PAGE_SHIFT - 12;
+		MTHCA_PUT(inbox, uar_page_sz, INIT_HCA_UAR_PAGE_SZ_OFFSET);
+		MTHCA_PUT(inbox, param->uar_scratch_base, INIT_HCA_UAR_SCATCH_BASE_OFFSET);
+	}
+
+	err = mthca_cmd(dev, indma, 0, 0, CMD_INIT_HCA,
+			HZ, status);
+
+	pci_free_consistent(dev->pdev, INIT_HCA_IN_SIZE, inbox, indma);
+	return err;
+}
+
+int mthca_INIT_IB(struct mthca_dev *dev,
+		  struct mthca_init_ib_param *param,
+		  int port, u8 *status)
+{
+	u32 *inbox;
+	dma_addr_t indma;
+	int err;
+	u32 flags;
+
+#define INIT_IB_IN_SIZE          56
+#define INIT_IB_FLAGS_OFFSET     0x00
+#define INIT_IB_FLAG_SIG         (1 << 18)
+#define INIT_IB_FLAG_NG          (1 << 17)
+#define INIT_IB_FLAG_G0          (1 << 16)
+#define INIT_IB_FLAG_1X          (1 << 8)
+#define INIT_IB_FLAG_4X          (1 << 9)
+#define INIT_IB_FLAG_12X         (1 << 11)
+#define INIT_IB_VL_SHIFT         4
+#define INIT_IB_MTU_SHIFT        12
+#define INIT_IB_MAX_GID_OFFSET   0x06
+#define INIT_IB_MAX_PKEY_OFFSET  0x0a
+#define INIT_IB_GUID0_OFFSET     0x10
+#define INIT_IB_NODE_GUID_OFFSET 0x18
+#define INIT_IB_SI_GUID_OFFSET   0x20
+
+	inbox = pci_alloc_consistent(dev->pdev, INIT_IB_IN_SIZE, &indma);
+	if (!inbox)
+		return -ENOMEM;
+
+	memset(inbox, 0, INIT_IB_IN_SIZE);
+
+	flags = 0;
+	flags |= param->enable_1x     ? INIT_IB_FLAG_1X  : 0;
+	flags |= param->enable_4x     ? INIT_IB_FLAG_4X  : 0;
+	flags |= param->set_guid0     ? INIT_IB_FLAG_G0  : 0;
+	flags |= param->set_node_guid ? INIT_IB_FLAG_NG  : 0;
+	flags |= param->set_si_guid   ? INIT_IB_FLAG_SIG : 0;
+	flags |= param->vl_cap << INIT_IB_VL_SHIFT;
+	flags |= param->mtu_cap << INIT_IB_MTU_SHIFT;
+	MTHCA_PUT(inbox, flags, INIT_IB_FLAGS_OFFSET);
+
+	MTHCA_PUT(inbox, param->gid_cap,   INIT_IB_MAX_GID_OFFSET);
+	MTHCA_PUT(inbox, param->pkey_cap,  INIT_IB_MAX_PKEY_OFFSET);
+	MTHCA_PUT(inbox, param->guid0,     INIT_IB_GUID0_OFFSET);
+	MTHCA_PUT(inbox, param->node_guid, INIT_IB_NODE_GUID_OFFSET);
+	MTHCA_PUT(inbox, param->si_guid,   INIT_IB_SI_GUID_OFFSET);
+
+	err = mthca_cmd(dev, indma, port, 0, CMD_INIT_IB,
+			CMD_TIME_CLASS_A, status);
+
+	pci_free_consistent(dev->pdev, INIT_HCA_IN_SIZE, inbox, indma);
+	return err;
+}
+
+int mthca_CLOSE_IB(struct mthca_dev *dev, int port, u8 *status)
+{
+	return mthca_cmd(dev, 0, port, 0, CMD_CLOSE_IB, HZ, status);
+}
+
+int mthca_CLOSE_HCA(struct mthca_dev *dev, int panic, u8 *status)
+{
+	return mthca_cmd(dev, 0, 0, panic, CMD_CLOSE_HCA, HZ, status);
+}
+
+int mthca_SW2HW_MPT(struct mthca_dev *dev, void *mpt_entry,
+		    int mpt_index, u8 *status)
+{
+	dma_addr_t indma;
+	int err;
+
+	indma = pci_map_single(dev->pdev, mpt_entry,
+			       MTHCA_MPT_ENTRY_SIZE,
+			       PCI_DMA_TODEVICE);
+	if (pci_dma_mapping_error(indma))
+		return -ENOMEM;
+
+	err = mthca_cmd(dev, indma, mpt_index, 0, CMD_SW2HW_MPT,
+			CMD_TIME_CLASS_B, status);
+
+	pci_unmap_single(dev->pdev, indma,
+			 MTHCA_MPT_ENTRY_SIZE, PCI_DMA_TODEVICE);
+	return err;
+}
+
+int mthca_HW2SW_MPT(struct mthca_dev *dev, void *mpt_entry,
+		    int mpt_index, u8 *status)
+{
+	dma_addr_t outdma = 0;
+	int err;
+
+	if (mpt_entry) {
+		outdma = pci_map_single(dev->pdev, mpt_entry,
+					MTHCA_MPT_ENTRY_SIZE,
+					PCI_DMA_FROMDEVICE);
+		if (pci_dma_mapping_error(outdma))
+			return -ENOMEM;
+	}
+
+	err = mthca_cmd_box(dev, 0, outdma, mpt_index, !mpt_entry,
+			    CMD_HW2SW_MPT,
+			    CMD_TIME_CLASS_B, status);
+
+	if (mpt_entry)
+		pci_unmap_single(dev->pdev, outdma,
+				 MTHCA_MPT_ENTRY_SIZE,
+				 PCI_DMA_FROMDEVICE);
+	return err;
+}
+
+int mthca_WRITE_MTT(struct mthca_dev *dev, u64 *mtt_entry,
+		    int num_mtt, u8 *status)
+{
+	dma_addr_t indma;
+	int err;
+
+	indma = pci_map_single(dev->pdev, mtt_entry,
+			       (num_mtt + 2) * 8,
+			       PCI_DMA_TODEVICE);
+	if (pci_dma_mapping_error(indma))
+		return -ENOMEM;
+
+	err = mthca_cmd(dev, indma, num_mtt, 0, CMD_WRITE_MTT,
+			CMD_TIME_CLASS_B, status);
+
+	pci_unmap_single(dev->pdev, indma,
+			 (num_mtt + 2) * 8, PCI_DMA_TODEVICE);
+	return err;
+}
+
+int mthca_MAP_EQ(struct mthca_dev *dev, u64 event_mask, int unmap,
+		 int eq_num, u8 *status)
+{
+	mthca_dbg(dev, "%s mask %016llx for eqn %d\n",
+		  unmap ? "Clearing" : "Setting",
+		  (unsigned long long) event_mask, eq_num);
+	return mthca_cmd(dev, event_mask, (unmap << 31) | eq_num,
+			 0, CMD_MAP_EQ, CMD_TIME_CLASS_B, status);
+}
+
+int mthca_SW2HW_EQ(struct mthca_dev *dev, void *eq_context,
+		   int eq_num, u8 *status)
+{
+	dma_addr_t indma;
+	int err;
+
+	indma = pci_map_single(dev->pdev, eq_context,
+			       MTHCA_EQ_CONTEXT_SIZE,
+			       PCI_DMA_TODEVICE);
+	if (pci_dma_mapping_error(indma))
+		return -ENOMEM;
+
+	err = mthca_cmd(dev, indma, eq_num, 0, CMD_SW2HW_EQ,
+			CMD_TIME_CLASS_A, status);
+
+	pci_unmap_single(dev->pdev, indma,
+			 MTHCA_EQ_CONTEXT_SIZE, PCI_DMA_TODEVICE);
+	return err;
+}
+
+int mthca_HW2SW_EQ(struct mthca_dev *dev, void *eq_context,
+		   int eq_num, u8 *status)
+{
+	dma_addr_t outdma = 0;
+	int err;
+
+	outdma = pci_map_single(dev->pdev, eq_context,
+				MTHCA_EQ_CONTEXT_SIZE,
+				PCI_DMA_FROMDEVICE);
+	if (pci_dma_mapping_error(outdma))
+		return -ENOMEM;
+
+	err = mthca_cmd_box(dev, 0, outdma, eq_num, 0,
+			    CMD_HW2SW_EQ,
+			    CMD_TIME_CLASS_A, status);
+
+	pci_unmap_single(dev->pdev, outdma,
+			 MTHCA_EQ_CONTEXT_SIZE,
+			 PCI_DMA_FROMDEVICE);
+	return err;
+}
+
+int mthca_SW2HW_CQ(struct mthca_dev *dev, void *cq_context,
+		   int cq_num, u8 *status)
+{
+	dma_addr_t indma;
+	int err;
+
+	indma = pci_map_single(dev->pdev, cq_context,
+			       MTHCA_CQ_CONTEXT_SIZE,
+			       PCI_DMA_TODEVICE);
+	if (pci_dma_mapping_error(indma))
+		return -ENOMEM;
+
+	err = mthca_cmd(dev, indma, cq_num, 0, CMD_SW2HW_CQ,
+			CMD_TIME_CLASS_A, status);
+
+	pci_unmap_single(dev->pdev, indma,
+			 MTHCA_CQ_CONTEXT_SIZE, PCI_DMA_TODEVICE);
+	return err;
+}
+
+int mthca_HW2SW_CQ(struct mthca_dev *dev, void *cq_context,
+		   int cq_num, u8 *status)
+{
+	dma_addr_t outdma = 0;
+	int err;
+
+	outdma = pci_map_single(dev->pdev, cq_context,
+				MTHCA_CQ_CONTEXT_SIZE,
+				PCI_DMA_FROMDEVICE);
+	if (pci_dma_mapping_error(outdma))
+		return -ENOMEM;
+
+	err = mthca_cmd_box(dev, 0, outdma, cq_num, 0,
+			    CMD_HW2SW_CQ,
+			    CMD_TIME_CLASS_A, status);
+
+	pci_unmap_single(dev->pdev, outdma,
+			 MTHCA_CQ_CONTEXT_SIZE,
+			 PCI_DMA_FROMDEVICE);
+	return err;
+}
+
+int mthca_MODIFY_QP(struct mthca_dev *dev, int trans, u32 num,
+		    int is_ee, void *qp_context, u32 optmask,
+		    u8 *status)
+{
+	static const u16 op[] = {
+		[MTHCA_TRANS_RST2INIT]  = CMD_RST2INIT_QPEE,
+		[MTHCA_TRANS_INIT2INIT] = CMD_INIT2INIT_QPEE,
+		[MTHCA_TRANS_INIT2RTR]  = CMD_INIT2RTR_QPEE,
+		[MTHCA_TRANS_RTR2RTS]   = CMD_RTR2RTS_QPEE,
+		[MTHCA_TRANS_RTS2RTS]   = CMD_RTS2RTS_QPEE,
+		[MTHCA_TRANS_SQERR2RTS] = CMD_SQERR2RTS_QPEE,
+		[MTHCA_TRANS_ANY2ERR]   = CMD_2ERR_QPEE,
+		[MTHCA_TRANS_RTS2SQD]   = CMD_RTS2SQD_QPEE,
+		[MTHCA_TRANS_SQD2SQD]   = CMD_SQD2SQD_QPEE,
+		[MTHCA_TRANS_SQD2RTS]   = CMD_SQD2RTS_QPEE,
+		[MTHCA_TRANS_ANY2RST]   = CMD_ERR2RST_QPEE
+	};
+	u8 op_mod = 0;
+
+	dma_addr_t indma;
+	int err;
+
+	if (trans < 0 || trans >= ARRAY_SIZE(op))
+		return -EINVAL;
+
+	if (trans == MTHCA_TRANS_ANY2RST) {
+		indma  = 0;
+		op_mod = 3;	/* don't write outbox, any->reset */
+
+		/* For debugging */
+		qp_context = pci_alloc_consistent(dev->pdev, MTHCA_QP_CONTEXT_SIZE,
+						  &indma);
+		op_mod = 2;	/* write outbox, any->reset */
+	} else {
+		indma = pci_map_single(dev->pdev, qp_context,
+				       MTHCA_QP_CONTEXT_SIZE,
+				       PCI_DMA_TODEVICE);
+		if (pci_dma_mapping_error(indma))
+			return -ENOMEM;
+
+		if (0) {
+			int i;
+			mthca_dbg(dev, "Dumping QP context:\n");
+			printk(" %08x\n", be32_to_cpup(qp_context));
+			for (i = 0; i < 0x100 / 4; ++i) {
+				if (i % 8 == 0)
+					printk("[%02x] ", i * 4);
+				printk(" %08x", be32_to_cpu(((u32 *) qp_context)[i + 2]));
+				if ((i + 1) % 8 == 0)
+					printk("\n");
+			}
+		}
+	}
+
+	if (trans == MTHCA_TRANS_ANY2RST) {
+		err = mthca_cmd_box(dev, 0, indma, (!!is_ee << 24) | num,
+				    op_mod, op[trans], CMD_TIME_CLASS_C, status);
+
+		if (0) {
+			int i;
+			mthca_dbg(dev, "Dumping QP context:\n");
+			printk(" %08x\n", be32_to_cpup(qp_context));
+			for (i = 0; i < 0x100 / 4; ++i) {
+				if (i % 8 == 0)
+					printk("[%02x] ", i * 4);
+				printk(" %08x", be32_to_cpu(((u32 *) qp_context)[i + 2]));
+				if ((i + 1) % 8 == 0)
+					printk("\n");
+			}
+		}
+
+	} else
+		err = mthca_cmd(dev, indma, (!!is_ee << 24) | num,
+				op_mod, op[trans], CMD_TIME_CLASS_C, status);
+
+	if (trans != MTHCA_TRANS_ANY2RST)
+		pci_unmap_single(dev->pdev, indma,
+				 MTHCA_QP_CONTEXT_SIZE, PCI_DMA_TODEVICE);
+	else
+		pci_free_consistent(dev->pdev, MTHCA_QP_CONTEXT_SIZE,
+				    qp_context, indma);
+	return err;
+}
+
+int mthca_QUERY_QP(struct mthca_dev *dev, u32 num, int is_ee,
+		   void *qp_context, u8 *status)
+{
+	dma_addr_t outdma = 0;
+	int err;
+
+	outdma = pci_map_single(dev->pdev, qp_context,
+				MTHCA_QP_CONTEXT_SIZE,
+				PCI_DMA_FROMDEVICE);
+	if (pci_dma_mapping_error(outdma))
+		return -ENOMEM;
+
+	err = mthca_cmd_box(dev, 0, outdma, (!!is_ee << 24) | num, 0,
+			    CMD_QUERY_QPEE,
+			    CMD_TIME_CLASS_A, status);
+
+	pci_unmap_single(dev->pdev, outdma,
+			 MTHCA_QP_CONTEXT_SIZE,
+			 PCI_DMA_FROMDEVICE);
+	return err;
+}
+
+int mthca_CONF_SPECIAL_QP(struct mthca_dev *dev, int type, u32 qpn,
+			  u8 *status)
+{
+	u8 op_mod;
+
+	switch (type) {
+	case IB_QPT_SMI:
+		op_mod = 0;
+		break;
+	case IB_QPT_GSI:
+		op_mod = 1;
+		break;
+	case IB_QPT_RAW_IPV6:
+		op_mod = 2;
+		break;
+	case IB_QPT_RAW_ETY:
+		op_mod = 3;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return mthca_cmd(dev, 0, qpn, op_mod, CMD_CONF_SPECIAL_QP,
+			 CMD_TIME_CLASS_B, status);
+}
+
+int mthca_MAD_IFC(struct mthca_dev *dev, int ignore_mkey, int port,
+		  void *in_mad, void *response_mad, u8 *status) {
+	void *box;
+	dma_addr_t dma;
+	int err;
+
+#define MAD_IFC_BOX_SIZE 512
+
+	box = pci_alloc_consistent(dev->pdev, MAD_IFC_BOX_SIZE, &dma);
+	if (!box)
+		return -ENOMEM;
+
+	memcpy(box, in_mad, 256);
+
+	err = mthca_cmd_box(dev, dma, dma + 256, port, !!ignore_mkey,
+			    CMD_MAD_IFC, CMD_TIME_CLASS_C, status);
+
+	if (!err && !*status)
+		memcpy(response_mad, box + 256, 256);
+
+	pci_free_consistent(dev->pdev, MAD_IFC_BOX_SIZE, box, dma);
+	return err;
+}
+
+int mthca_READ_MGM(struct mthca_dev *dev, int index, void *mgm,
+		   u8 *status)
+{
+	dma_addr_t outdma = 0;
+	int err;
+
+	outdma = pci_map_single(dev->pdev, mgm,
+				MTHCA_MGM_ENTRY_SIZE,
+				PCI_DMA_FROMDEVICE);
+	if (pci_dma_mapping_error(outdma))
+		return -ENOMEM;
+
+	err = mthca_cmd_box(dev, 0, outdma, index, 0,
+			    CMD_READ_MGM,
+			    CMD_TIME_CLASS_A, status);
+
+	pci_unmap_single(dev->pdev, outdma,
+			 MTHCA_MGM_ENTRY_SIZE,
+			 PCI_DMA_FROMDEVICE);
+	return err;
+}
+
+int mthca_WRITE_MGM(struct mthca_dev *dev, int index, void *mgm,
+		    u8 *status)
+{
+	dma_addr_t indma;
+	int err;
+
+	indma = pci_map_single(dev->pdev, mgm,
+			       MTHCA_MGM_ENTRY_SIZE,
+			       PCI_DMA_TODEVICE);
+	if (pci_dma_mapping_error(indma))
+		return -ENOMEM;
+
+	err = mthca_cmd(dev, indma, index, 0, CMD_WRITE_MGM,
+			CMD_TIME_CLASS_A, status);
+
+	pci_unmap_single(dev->pdev, indma,
+			 MTHCA_MGM_ENTRY_SIZE, PCI_DMA_TODEVICE);
+	return err;
+}
+
+int mthca_MGID_HASH(struct mthca_dev *dev, void *gid, u16 *hash,
+		    u8 *status)
+{
+	dma_addr_t indma;
+	u64 imm;
+	int err;
+
+	indma = pci_map_single(dev->pdev, gid, 16, PCI_DMA_TODEVICE);
+	if (pci_dma_mapping_error(indma))
+		return -ENOMEM;
+
+	err = mthca_cmd_imm(dev, indma, &imm, 0, 0, CMD_MGID_HASH,
+			    CMD_TIME_CLASS_A, status);
+	*hash = imm;
+
+	pci_unmap_single(dev->pdev, indma, 16, PCI_DMA_TODEVICE);
+	return err;
+}
drivers/infiniband/hw/mthca/mthca_cmd.h new file +276 -0

276 lines not on this page · raw message

#12 lore raw

PATCH v4 12/24 Add Mellanox HCA low-level driver (EQ)

Add event queue code for Mellanox HCA driver.

drivers/infiniband/hw/mthca/mthca_eq.c new file +689 -0
@@ -0,0 +1,689 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: mthca_eq.c 1349 2004-12-16 21:09:43Z roland $
+ */
+
+#include <linux/init.h>
+#include <linux/errno.h>
+#include <linux/interrupt.h>
+#include <linux/pci.h>
+
+#include "mthca_dev.h"
+#include "mthca_cmd.h"
+#include "mthca_config_reg.h"
+
+enum {
+	MTHCA_NUM_ASYNC_EQE = 0x80,
+	MTHCA_NUM_CMD_EQE   = 0x80,
+	MTHCA_EQ_ENTRY_SIZE = 0x20
+};
+
+/*
+ * Must be packed because start is 64 bits but only aligned to 32 bits.
+ */
+struct mthca_eq_context {
+	u32 flags;
+	u64 start;
+	u32 logsize_usrpage;
+	u32 pd;
+	u8  reserved1[3];
+	u8  intr;
+	u32 lost_count;
+	u32 lkey;
+	u32 reserved2[2];
+	u32 consumer_index;
+	u32 producer_index;
+	u32 reserved3[4];
+} __attribute__((packed));
+
+#define MTHCA_EQ_STATUS_OK          ( 0 << 28)
+#define MTHCA_EQ_STATUS_OVERFLOW    ( 9 << 28)
+#define MTHCA_EQ_STATUS_WRITE_FAIL  (10 << 28)
+#define MTHCA_EQ_OWNER_SW           ( 0 << 24)
+#define MTHCA_EQ_OWNER_HW           ( 1 << 24)
+#define MTHCA_EQ_FLAG_TR            ( 1 << 18)
+#define MTHCA_EQ_FLAG_OI            ( 1 << 17)
+#define MTHCA_EQ_STATE_ARMED        ( 1 <<  8)
+#define MTHCA_EQ_STATE_FIRED        ( 2 <<  8)
+#define MTHCA_EQ_STATE_ALWAYS_ARMED ( 3 <<  8)
+
+enum {
+	MTHCA_EVENT_TYPE_COMP       	    = 0x00,
+	MTHCA_EVENT_TYPE_PATH_MIG   	    = 0x01,
+	MTHCA_EVENT_TYPE_COMM_EST   	    = 0x02,
+	MTHCA_EVENT_TYPE_SQ_DRAINED 	    = 0x03,
+	MTHCA_EVENT_TYPE_SRQ_LAST_WQE       = 0x13,
+	MTHCA_EVENT_TYPE_CQ_ERROR   	    = 0x04,
+	MTHCA_EVENT_TYPE_WQ_CATAS_ERROR     = 0x05,
+	MTHCA_EVENT_TYPE_EEC_CATAS_ERROR    = 0x06,
+	MTHCA_EVENT_TYPE_PATH_MIG_FAILED    = 0x07,
+	MTHCA_EVENT_TYPE_WQ_INVAL_REQ_ERROR = 0x10,
+	MTHCA_EVENT_TYPE_WQ_ACCESS_ERROR    = 0x11,
+	MTHCA_EVENT_TYPE_SRQ_CATAS_ERROR    = 0x12,
+	MTHCA_EVENT_TYPE_LOCAL_CATAS_ERROR  = 0x08,
+	MTHCA_EVENT_TYPE_PORT_CHANGE        = 0x09,
+	MTHCA_EVENT_TYPE_EQ_OVERFLOW        = 0x0f,
+	MTHCA_EVENT_TYPE_ECC_DETECT         = 0x0e,
+	MTHCA_EVENT_TYPE_CMD                = 0x0a
+};
+
+#define MTHCA_ASYNC_EVENT_MASK ((1ULL << MTHCA_EVENT_TYPE_PATH_MIG)           | \
+				(1ULL << MTHCA_EVENT_TYPE_COMM_EST)           | \
+				(1ULL << MTHCA_EVENT_TYPE_SQ_DRAINED)         | \
+				(1ULL << MTHCA_EVENT_TYPE_CQ_ERROR)           | \
+				(1ULL << MTHCA_EVENT_TYPE_WQ_CATAS_ERROR)     | \
+				(1ULL << MTHCA_EVENT_TYPE_EEC_CATAS_ERROR)    | \
+				(1ULL << MTHCA_EVENT_TYPE_PATH_MIG_FAILED)    | \
+				(1ULL << MTHCA_EVENT_TYPE_WQ_INVAL_REQ_ERROR) | \
+				(1ULL << MTHCA_EVENT_TYPE_WQ_ACCESS_ERROR)    | \
+				(1ULL << MTHCA_EVENT_TYPE_LOCAL_CATAS_ERROR)  | \
+				(1ULL << MTHCA_EVENT_TYPE_PORT_CHANGE)        | \
+				(1ULL << MTHCA_EVENT_TYPE_ECC_DETECT))
+#define MTHCA_SRQ_EVENT_MASK    (1ULL << MTHCA_EVENT_TYPE_SRQ_CATAS_ERROR)    | \
+				(1ULL << MTHCA_EVENT_TYPE_SRQ_LAST_WQE)
+#define MTHCA_CMD_EVENT_MASK    (1ULL << MTHCA_EVENT_TYPE_CMD)
+
+#define MTHCA_EQ_DB_INC_CI     (1 << 24)
+#define MTHCA_EQ_DB_REQ_NOT    (2 << 24)
+#define MTHCA_EQ_DB_DISARM_CQ  (3 << 24)
+#define MTHCA_EQ_DB_SET_CI     (4 << 24)
+#define MTHCA_EQ_DB_ALWAYS_ARM (5 << 24)
+
+struct mthca_eqe {
+	u8 reserved1;
+	u8 type;
+	u8 reserved2;
+	u8 subtype;
+	union {
+		u32 raw[6];
+		struct {
+			u32 cqn;
+		} __attribute__((packed)) comp;
+		struct {
+			u16 reserved1;
+			u16 token;
+			u32 reserved2;
+			u8  reserved3[3];
+			u8  status;
+			u64 out_param;
+		} __attribute__((packed)) cmd;
+		struct {
+			u32 qpn;
+		} __attribute__((packed)) qp;
+		struct {
+			u32 cqn;
+			u32 reserved1;
+			u8  reserved2[3];
+			u8  syndrome;
+		} __attribute__((packed)) cq_err;
+		struct {
+			u32 reserved1[2];
+			u32 port;
+		} __attribute__((packed)) port_change;
+	} event;
+	u8 reserved3[3];
+	u8 owner;
+} __attribute__((packed));
+
+#define  MTHCA_EQ_ENTRY_OWNER_SW      (0 << 7)
+#define  MTHCA_EQ_ENTRY_OWNER_HW      (1 << 7)
+
+static inline u64 async_mask(struct mthca_dev *dev)
+{
+	return dev->mthca_flags & MTHCA_FLAG_SRQ ?
+		MTHCA_ASYNC_EVENT_MASK | MTHCA_SRQ_EVENT_MASK :
+		MTHCA_ASYNC_EVENT_MASK;
+}
+
+static inline void set_eq_ci(struct mthca_dev *dev, int eqn, int ci)
+{
+	u32 doorbell[2];
+
+	doorbell[0] = cpu_to_be32(MTHCA_EQ_DB_SET_CI | eqn);
+	doorbell[1] = cpu_to_be32(ci);
+
+	mthca_write64(doorbell,
+		      dev->kar + MTHCA_EQ_DOORBELL,
+		      MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock));
+}
+
+static inline void eq_req_not(struct mthca_dev *dev, int eqn)
+{
+	u32 doorbell[2];
+
+	doorbell[0] = cpu_to_be32(MTHCA_EQ_DB_REQ_NOT | eqn);
+	doorbell[1] = 0;
+
+	mthca_write64(doorbell,
+		      dev->kar + MTHCA_EQ_DOORBELL,
+		      MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock));
+}
+
+static inline void disarm_cq(struct mthca_dev *dev, int eqn, int cqn)
+{
+	u32 doorbell[2];
+
+	doorbell[0] = cpu_to_be32(MTHCA_EQ_DB_DISARM_CQ | eqn);
+	doorbell[1] = cpu_to_be32(cqn);
+
+	mthca_write64(doorbell,
+		      dev->kar + MTHCA_EQ_DOORBELL,
+		      MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock));
+}
+
+static inline struct mthca_eqe *get_eqe(struct mthca_eq *eq, int entry)
+{
+	return eq->page_list[entry * MTHCA_EQ_ENTRY_SIZE / PAGE_SIZE].buf
+		+ (entry * MTHCA_EQ_ENTRY_SIZE) % PAGE_SIZE;
+}
+
+static inline int next_eqe_sw(struct mthca_eq *eq)
+{
+	return !(MTHCA_EQ_ENTRY_OWNER_HW &
+		 get_eqe(eq, eq->cons_index)->owner);
+}
+
+static inline void set_eqe_hw(struct mthca_eq *eq, int entry)
+{
+	get_eqe(eq, entry)->owner =  MTHCA_EQ_ENTRY_OWNER_HW;
+}
+
+static void port_change(struct mthca_dev *dev, int port, int active)
+{
+	struct ib_event record;
+
+	mthca_dbg(dev, "Port change to %s for port %d\n",
+		  active ? "active" : "down", port);
+
+	record.device = &dev->ib_dev;
+	record.event  = active ? IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
+	record.element.port_num = port;
+
+	ib_dispatch_event(&record);
+}
+
+static void mthca_eq_int(struct mthca_dev *dev, struct mthca_eq *eq)
+{
+	struct mthca_eqe *eqe;
+	int disarm_cqn;
+
+	while (next_eqe_sw(eq)) {
+		int set_ci = 0;
+		eqe = get_eqe(eq, eq->cons_index);
+
+		switch (eqe->type) {
+		case MTHCA_EVENT_TYPE_COMP:
+			disarm_cqn = be32_to_cpu(eqe->event.comp.cqn) & 0xffffff;
+			disarm_cq(dev, eq->eqn, disarm_cqn);
+			mthca_cq_event(dev, disarm_cqn);
+			break;
+
+		case MTHCA_EVENT_TYPE_PATH_MIG:
+			mthca_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) & 0xffffff,
+				       IB_EVENT_PATH_MIG);
+			break;
+
+		case MTHCA_EVENT_TYPE_COMM_EST:
+			mthca_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) & 0xffffff,
+				       IB_EVENT_COMM_EST);
+			break;
+
+		case MTHCA_EVENT_TYPE_SQ_DRAINED:
+			mthca_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) & 0xffffff,
+				       IB_EVENT_SQ_DRAINED);
+			break;
+
+		case MTHCA_EVENT_TYPE_WQ_CATAS_ERROR:
+			mthca_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) & 0xffffff,
+				       IB_EVENT_QP_FATAL);
+			break;
+
+		case MTHCA_EVENT_TYPE_PATH_MIG_FAILED:
+			mthca_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) & 0xffffff,
+				       IB_EVENT_PATH_MIG_ERR);
+			break;
+
+		case MTHCA_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
+			mthca_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) & 0xffffff,
+				       IB_EVENT_QP_REQ_ERR);
+			break;
+
+		case MTHCA_EVENT_TYPE_WQ_ACCESS_ERROR:
+			mthca_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) & 0xffffff,
+				       IB_EVENT_QP_ACCESS_ERR);
+			break;
+
+		case MTHCA_EVENT_TYPE_CMD:
+			mthca_cmd_event(dev,
+					be16_to_cpu(eqe->event.cmd.token),
+					eqe->event.cmd.status,
+					be64_to_cpu(eqe->event.cmd.out_param));
+			/* cmd_event() may add more commands.
+			 * The card will think the queue has overflowed if
+			 * we don't tell it we've been processing events.
+			 */
+			set_ci = 1;
+			break;
+
+		case MTHCA_EVENT_TYPE_PORT_CHANGE:
+			port_change(dev,
+				    (be32_to_cpu(eqe->event.port_change.port) >> 28) & 3,
+				    eqe->subtype == 0x4);
+			break;
+
+		case MTHCA_EVENT_TYPE_CQ_ERROR:
+			mthca_warn(dev, "CQ %s on CQN %08x\n",
+				   eqe->event.cq_err.syndrome == 1 ?
+				   "overrun" : "access violation",
+				   be32_to_cpu(eqe->event.cq_err.cqn));
+			break;
+
+		case MTHCA_EVENT_TYPE_EQ_OVERFLOW:
+			mthca_warn(dev, "EQ overrun on EQN %d\n", eq->eqn);
+			break;
+
+		case MTHCA_EVENT_TYPE_EEC_CATAS_ERROR:
+		case MTHCA_EVENT_TYPE_SRQ_CATAS_ERROR:
+		case MTHCA_EVENT_TYPE_LOCAL_CATAS_ERROR:
+		case MTHCA_EVENT_TYPE_ECC_DETECT:
+		default:
+			mthca_warn(dev, "Unhandled event %02x(%02x) on EQ %d\n",
+				   eqe->type, eqe->subtype, eq->eqn);
+			break;
+		};
+
+		set_eqe_hw(eq, eq->cons_index);
+		eq->cons_index = (eq->cons_index + 1) & (eq->nent - 1);
+
+		if (set_ci) {
+			wmb(); /* see comment below */
+			set_eq_ci(dev, eq->eqn, eq->cons_index);
+			set_ci = 0;
+		}
+	}
+
+	/*
+	 * This barrier makes sure that all updates to
+	 * ownership bits done by set_eqe_hw() hit memory
+	 * before the consumer index is updated.  set_eq_ci()
+	 * allows the HCA to possibly write more EQ entries,
+	 * and we want to avoid the exceedingly unlikely
+	 * possibility of the HCA writing an entry and then
+	 * having set_eqe_hw() overwrite the owner field.
+	 */
+	wmb();
+	set_eq_ci(dev, eq->eqn, eq->cons_index);
+	eq_req_not(dev, eq->eqn);
+}
+
+static irqreturn_t mthca_interrupt(int irq, void *dev_ptr, struct pt_regs *regs)
+{
+	struct mthca_dev *dev = dev_ptr;
+	u32 ecr;
+	int work = 0;
+	int i;
+
+	if (dev->eq_table.clr_mask)
+		writel(dev->eq_table.clr_mask, dev->eq_table.clr_int);
+
+	while ((ecr = readl(dev->hcr + MTHCA_ECR_OFFSET + 4)) != 0) {
+		work = 1;
+
+		writel(ecr, dev->hcr + MTHCA_ECR_CLR_OFFSET + 4);
+
+		for (i = 0; i < MTHCA_NUM_EQ; ++i)
+			if (ecr & dev->eq_table.eq[i].ecr_mask)
+				mthca_eq_int(dev, &dev->eq_table.eq[i]);
+	}
+
+	return IRQ_RETVAL(work);
+}
+
+static irqreturn_t mthca_msi_x_interrupt(int irq, void *eq_ptr,
+					 struct pt_regs *regs)
+{
+	struct mthca_eq  *eq  = eq_ptr;
+	struct mthca_dev *dev = eq->dev;
+
+	writel(eq->ecr_mask, dev->hcr + MTHCA_ECR_CLR_OFFSET + 4);
+	mthca_eq_int(dev, eq);
+
+	/* MSI-X vectors always belong to us */
+	return IRQ_HANDLED;
+}
+
+static int __devinit mthca_create_eq(struct mthca_dev *dev,
+				     int nent,
+				     u8 intr,
+				     struct mthca_eq *eq)
+{
+	int npages = (nent * MTHCA_EQ_ENTRY_SIZE + PAGE_SIZE - 1) /
+		PAGE_SIZE;
+	u64 *dma_list = NULL;
+	dma_addr_t t;
+	void *mailbox = NULL;
+	struct mthca_eq_context *eq_context;
+	int err = -ENOMEM;
+	int i;
+	u8 status;
+
+	/* Make sure EQ size is aligned to a power of 2 size. */
+	for (i = 1; i < nent; i <<= 1)
+		; /* nothing */
+	nent = i;
+
+	eq->dev = dev;
+
+	eq->page_list = kmalloc(npages * sizeof *eq->page_list,
+				GFP_KERNEL);
+	if (!eq->page_list)
+		goto err_out;
+
+	for (i = 0; i < npages; ++i)
+		eq->page_list[i].buf = NULL;
+
+	dma_list = kmalloc(npages * sizeof *dma_list, GFP_KERNEL);
+	if (!dma_list)
+		goto err_out_free;
+
+	mailbox = kmalloc(sizeof *eq_context + MTHCA_CMD_MAILBOX_EXTRA,
+			  GFP_KERNEL);
+	if (!mailbox)
+		goto err_out_free;
+	eq_context = MAILBOX_ALIGN(mailbox);
+
+	for (i = 0; i < npages; ++i) {
+		eq->page_list[i].buf = pci_alloc_consistent(dev->pdev,
+							    PAGE_SIZE, &t);
+		if (!eq->page_list[i].buf)
+			goto err_out_free;
+
+		dma_list[i] = t;
+		pci_unmap_addr_set(&eq->page_list[i], mapping, t);
+
+		memset(eq->page_list[i].buf, 0, PAGE_SIZE);
+	}
+
+	for (i = 0; i < nent; ++i)
+		set_eqe_hw(eq, i);
+
+	eq->eqn = mthca_alloc(&dev->eq_table.alloc);
+	if (eq->eqn == -1)
+		goto err_out_free;
+
+	err = mthca_mr_alloc_phys(dev, dev->driver_pd.pd_num,
+				  dma_list, PAGE_SHIFT, npages,
+				  0, npages * PAGE_SIZE,
+				  MTHCA_MPT_FLAG_LOCAL_WRITE |
+				  MTHCA_MPT_FLAG_LOCAL_READ,
+				  &eq->mr);
+	if (err)
+		goto err_out_free_eq;
+
+	eq->nent = nent;
+
+	memset(eq_context, 0, sizeof *eq_context);
+	eq_context->flags           = cpu_to_be32(MTHCA_EQ_STATUS_OK   |
+						  MTHCA_EQ_OWNER_HW    |
+						  MTHCA_EQ_STATE_ARMED |
+						  MTHCA_EQ_FLAG_TR);
+	eq_context->start           = cpu_to_be64(0);
+	eq_context->logsize_usrpage = cpu_to_be32((ffs(nent) - 1) << 24 |
+						  MTHCA_KAR_PAGE);
+	eq_context->pd              = cpu_to_be32(dev->driver_pd.pd_num);
+	eq_context->intr            = intr;
+	eq_context->lkey            = cpu_to_be32(eq->mr.ibmr.lkey);
+
+	err = mthca_SW2HW_EQ(dev, eq_context, eq->eqn, &status);
+	if (err) {
+		mthca_warn(dev, "SW2HW_EQ failed (%d)\n", err);
+		goto err_out_free_mr;
+	}
+	if (status) {
+		mthca_warn(dev, "SW2HW_EQ returned status 0x%02x\n",
+			   status);
+		err = -EINVAL;
+		goto err_out_free_mr;
+	}
+
+	kfree(dma_list);
+	kfree(mailbox);
+
+	eq->ecr_mask   = swab32(1 << eq->eqn);
+	eq->cons_index = 0;
+
+	eq_req_not(dev, eq->eqn);
+
+	mthca_dbg(dev, "Allocated EQ %d with %d entries\n",
+		  eq->eqn, nent);
+
+	return err;
+
+ err_out_free_mr:
+	mthca_free_mr(dev, &eq->mr);
+
+ err_out_free_eq:
+	mthca_free(&dev->eq_table.alloc, eq->eqn);
+
+ err_out_free:
+	for (i = 0; i < npages; ++i)
+		if (eq->page_list[i].buf)
+			pci_free_consistent(dev->pdev, PAGE_SIZE,
+					    eq->page_list[i].buf,
+					    pci_unmap_addr(&eq->page_list[i],
+							   mapping));
+
+	kfree(eq->page_list);
+	kfree(dma_list);
+	kfree(mailbox);
+
+ err_out:
+	return err;
+}
+
+static void mthca_free_eq(struct mthca_dev *dev,
+			  struct mthca_eq *eq)
+{
+	void *mailbox = NULL;
+	int err;
+	u8 status;
+	int npages = (eq->nent * MTHCA_EQ_ENTRY_SIZE + PAGE_SIZE - 1) /
+		PAGE_SIZE;
+	int i;
+
+	mailbox = kmalloc(sizeof (struct mthca_eq_context) + MTHCA_CMD_MAILBOX_EXTRA,
+			  GFP_KERNEL);
+	if (!mailbox)
+		return;
+
+	err = mthca_HW2SW_EQ(dev, MAILBOX_ALIGN(mailbox),
+			     eq->eqn, &status);
+	if (err)
+		mthca_warn(dev, "HW2SW_EQ failed (%d)\n", err);
+	if (status)
+		mthca_warn(dev, "HW2SW_EQ returned status 0x%02x\n",
+			   status);
+
+	if (0) {
+		mthca_dbg(dev, "Dumping EQ context %02x:\n", eq->eqn);
+		for (i = 0; i < sizeof (struct mthca_eq_context) / 4; ++i) {
+			if (i % 4 == 0)
+				printk("[%02x] ", i * 4);
+			printk(" %08x", be32_to_cpup(MAILBOX_ALIGN(mailbox) + i * 4));
+			if ((i + 1) % 4 == 0)
+				printk("\n");
+		}
+	}
+
+
+	mthca_free_mr(dev, &eq->mr);
+	for (i = 0; i < npages; ++i)
+		pci_free_consistent(dev->pdev, PAGE_SIZE,
+				    eq->page_list[i].buf,
+				    pci_unmap_addr(&eq->page_list[i], mapping));
+
+	kfree(eq->page_list);
+	kfree(mailbox);
+}
+
+static void mthca_free_irqs(struct mthca_dev *dev)
+{
+	int i;
+
+	if (dev->eq_table.have_irq)
+		free_irq(dev->pdev->irq, dev);
+	for (i = 0; i < MTHCA_NUM_EQ; ++i)
+		if (dev->eq_table.eq[i].have_irq)
+			free_irq(dev->eq_table.eq[i].msi_x_vector,
+				 dev->eq_table.eq + i);
+}
+
+int __devinit mthca_init_eq_table(struct mthca_dev *dev)
+{
+	int err;
+	u8 status;
+	u8 intr;
+	int i;
+
+	err = mthca_alloc_init(&dev->eq_table.alloc,
+			       dev->limits.num_eqs,
+			       dev->limits.num_eqs - 1,
+			       dev->limits.reserved_eqs);
+	if (err)
+		return err;
+
+	if (dev->mthca_flags & MTHCA_FLAG_MSI ||
+	    dev->mthca_flags & MTHCA_FLAG_MSI_X) {
+		dev->eq_table.clr_mask = 0;
+	} else {
+		dev->eq_table.clr_mask =
+			swab32(1 << (dev->eq_table.inta_pin & 31));
+		dev->eq_table.clr_int  = dev->clr_base +
+			(dev->eq_table.inta_pin < 31 ? 4 : 0);
+	}
+
+	intr = (dev->mthca_flags & MTHCA_FLAG_MSI) ?
+		128 : dev->eq_table.inta_pin;
+
+	err = mthca_create_eq(dev, dev->limits.num_cqs,
+			      (dev->mthca_flags & MTHCA_FLAG_MSI_X) ? 128 : intr,
+			      &dev->eq_table.eq[MTHCA_EQ_COMP]);
+	if (err)
+		goto err_out_free;
+
+	err = mthca_create_eq(dev, MTHCA_NUM_ASYNC_EQE,
+			      (dev->mthca_flags & MTHCA_FLAG_MSI_X) ? 129 : intr,
+			      &dev->eq_table.eq[MTHCA_EQ_ASYNC]);
+	if (err)
+		goto err_out_comp;
+
+	err = mthca_create_eq(dev, MTHCA_NUM_CMD_EQE,
+			      (dev->mthca_flags & MTHCA_FLAG_MSI_X) ? 130 : intr,
+			      &dev->eq_table.eq[MTHCA_EQ_CMD]);
+	if (err)
+		goto err_out_async;
+
+	if (dev->mthca_flags & MTHCA_FLAG_MSI_X) {
+		static const char *eq_name[] = {
+			[MTHCA_EQ_COMP]  = DRV_NAME " (comp)",
+			[MTHCA_EQ_ASYNC] = DRV_NAME " (async)",
+			[MTHCA_EQ_CMD]   = DRV_NAME " (cmd)"
+		};
+
+		for (i = 0; i < MTHCA_NUM_EQ; ++i) {
+			err = request_irq(dev->eq_table.eq[i].msi_x_vector,
+					  mthca_msi_x_interrupt, 0,
+					  eq_name[i], dev->eq_table.eq + i);
+			if (err)
+				goto err_out_cmd;
+			dev->eq_table.eq[i].have_irq = 1;
+		}
+	} else {
+		err = request_irq(dev->pdev->irq, mthca_interrupt, SA_SHIRQ,
+				  DRV_NAME, dev);
+		if (err)
+			goto err_out_cmd;
+		dev->eq_table.have_irq = 1;
+	}
+
+	err = mthca_MAP_EQ(dev, async_mask(dev),
+			   0, dev->eq_table.eq[MTHCA_EQ_ASYNC].eqn, &status);
+	if (err)
+		mthca_warn(dev, "MAP_EQ for async EQ %d failed (%d)\n",
+			   dev->eq_table.eq[MTHCA_EQ_ASYNC].eqn, err);
+	if (status)
+		mthca_warn(dev, "MAP_EQ for async EQ %d returned status 0x%02x\n",
+			   dev->eq_table.eq[MTHCA_EQ_ASYNC].eqn, status);
+
+	err = mthca_MAP_EQ(dev, MTHCA_CMD_EVENT_MASK,
+			   0, dev->eq_table.eq[MTHCA_EQ_CMD].eqn, &status);
+	if (err)
+		mthca_warn(dev, "MAP_EQ for cmd EQ %d failed (%d)\n",
+			   dev->eq_table.eq[MTHCA_EQ_CMD].eqn, err);
+	if (status)
+		mthca_warn(dev, "MAP_EQ for cmd EQ %d returned status 0x%02x\n",
+			   dev->eq_table.eq[MTHCA_EQ_CMD].eqn, status);
+
+	return 0;
+
+err_out_cmd:
+	mthca_free_irqs(dev);
+	mthca_free_eq(dev, &dev->eq_table.eq[MTHCA_EQ_CMD]);
+
+err_out_async:
+	mthca_free_eq(dev, &dev->eq_table.eq[MTHCA_EQ_ASYNC]);
+
+err_out_comp:
+	mthca_free_eq(dev, &dev->eq_table.eq[MTHCA_EQ_COMP]);
+
+err_out_free:
+	mthca_alloc_cleanup(&dev->eq_table.alloc);
+	return err;
+}
+
+void __devexit mthca_cleanup_eq_table(struct mthca_dev *dev)
+{
+	u8 status;
+	int i;
+
+	mthca_free_irqs(dev);
+
+	mthca_MAP_EQ(dev, async_mask(dev),
+		     1, dev->eq_table.eq[MTHCA_EQ_ASYNC].eqn, &status);
+	mthca_MAP_EQ(dev, MTHCA_CMD_EVENT_MASK,
+		     1, dev->eq_table.eq[MTHCA_EQ_CMD].eqn, &status);
+
+	for (i = 0; i < MTHCA_NUM_EQ; ++i)
+		mthca_free_eq(dev, &dev->eq_table.eq[i]);
+
+	mthca_alloc_cleanup(&dev->eq_table.alloc);
+}
#13 lore raw

PATCH v4 13/24 Add Mellanox HCA low-level driver (initialization)

Add device initializaton code for Mellanox HCA driver.

drivers/infiniband/hw/mthca/mthca_profile.c new file +226 -0
@@ -0,0 +1,226 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: mthca_profile.c 1349 2004-12-16 21:09:43Z roland $
+ */
+
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+
+#include "mthca_profile.h"
+
+static int default_profile[MTHCA_RES_NUM] = {
+	[MTHCA_RES_QP]    = 1 << 16,
+	[MTHCA_RES_EQP]   = 1 << 16,
+	[MTHCA_RES_CQ]    = 1 << 16,
+	[MTHCA_RES_EQ]    = 32,
+	[MTHCA_RES_RDB]   = 1 << 18,
+	[MTHCA_RES_MCG]   = 1 << 13,
+	[MTHCA_RES_MPT]   = 1 << 17,
+	[MTHCA_RES_MTT]   = 1 << 20,
+	[MTHCA_RES_UDAV]  = 1 << 15
+};
+
+enum {
+	MTHCA_RDB_ENTRY_SIZE = 32,
+	MTHCA_MTT_SEG_SIZE   = 64
+};
+
+enum {
+	MTHCA_NUM_PDS = 1 << 15
+};
+
+int mthca_make_profile(struct mthca_dev *dev,
+		       struct mthca_dev_lim *dev_lim,
+		       struct mthca_init_hca_param *init_hca)
+{
+	/* just use default profile for now */
+	struct mthca_resource {
+		u64 size;
+		u64 start;
+		int type;
+		int num;
+		int log_num;
+	};
+
+	u64 total_size = 0;
+	struct mthca_resource *profile;
+	struct mthca_resource tmp;
+	int i, j;
+
+	default_profile[MTHCA_RES_UAR] = dev_lim->uar_size / PAGE_SIZE;
+
+	profile = kmalloc(MTHCA_RES_NUM * sizeof *profile, GFP_KERNEL);
+	if (!profile)
+		return -ENOMEM;
+
+	profile[MTHCA_RES_QP].size   = dev_lim->qpc_entry_sz;
+	profile[MTHCA_RES_EEC].size  = dev_lim->eec_entry_sz;
+	profile[MTHCA_RES_SRQ].size  = dev_lim->srq_entry_sz;
+	profile[MTHCA_RES_CQ].size   = dev_lim->cqc_entry_sz;
+	profile[MTHCA_RES_EQP].size  = dev_lim->eqpc_entry_sz;
+	profile[MTHCA_RES_EEEC].size = dev_lim->eeec_entry_sz;
+	profile[MTHCA_RES_EQ].size   = dev_lim->eqc_entry_sz;
+	profile[MTHCA_RES_RDB].size  = MTHCA_RDB_ENTRY_SIZE;
+	profile[MTHCA_RES_MCG].size  = MTHCA_MGM_ENTRY_SIZE;
+	profile[MTHCA_RES_MPT].size  = MTHCA_MPT_ENTRY_SIZE;
+	profile[MTHCA_RES_MTT].size  = MTHCA_MTT_SEG_SIZE;
+	profile[MTHCA_RES_UAR].size  = dev_lim->uar_scratch_entry_sz;
+	profile[MTHCA_RES_UDAV].size = MTHCA_AV_SIZE;
+
+	for (i = 0; i < MTHCA_RES_NUM; ++i) {
+		profile[i].type     = i;
+		profile[i].num      = default_profile[i];
+		profile[i].log_num  = max(ffs(default_profile[i]) - 1, 0);
+		profile[i].size    *= default_profile[i];
+	}
+
+	/*
+	 * Sort the resources in decreasing order of size.  Since they
+	 * all have sizes that are powers of 2, we'll be able to keep
+	 * resources aligned to their size and pack them without gaps
+	 * using the sorted order.
+	 */
+	for (i = MTHCA_RES_NUM; i > 0; --i)
+		for (j = 1; j < i; ++j) {
+			if (profile[j].size > profile[j - 1].size) {
+				tmp            = profile[j];
+				profile[j]     = profile[j - 1];
+				profile[j - 1] = tmp;
+			}
+		}
+
+	for (i = 0; i < MTHCA_RES_NUM; ++i) {
+		if (profile[i].size) {
+			profile[i].start = dev->ddr_start + total_size;
+			total_size      += profile[i].size;
+		}
+		if (total_size > dev->fw.tavor.fw_start - dev->ddr_start) {
+			mthca_err(dev, "Profile requires 0x%llx bytes; "
+				  "won't fit between DDR start at 0x%016llx "
+				  "and FW start at 0x%016llx.\n",
+				  (unsigned long long) total_size,
+				  (unsigned long long) dev->ddr_start,
+				  (unsigned long long) dev->fw.tavor.fw_start);
+			kfree(profile);
+			return -ENOMEM;
+		}
+
+		if (profile[i].size)
+			mthca_dbg(dev, "profile[%2d]--%2d/%2d @ 0x%16llx "
+				  "(size 0x%8llx)\n",
+				  i, profile[i].type, profile[i].log_num,
+				  (unsigned long long) profile[i].start,
+				  (unsigned long long) profile[i].size);
+	}
+
+	mthca_dbg(dev, "HCA memory: allocated %d KB/%d KB (%d KB free)\n",
+		  (int) (total_size >> 10),
+		  (int) ((dev->fw.tavor.fw_start - dev->ddr_start) >> 10),
+		  (int) ((dev->fw.tavor.fw_start - dev->ddr_start - total_size) >> 10));
+
+	for (i = 0; i < MTHCA_RES_NUM; ++i) {
+		switch (profile[i].type) {
+		case MTHCA_RES_QP:
+			dev->limits.num_qps   = profile[i].num;
+			init_hca->qpc_base    = profile[i].start;
+			init_hca->log_num_qps = profile[i].log_num;
+			break;
+		case MTHCA_RES_EEC:
+			dev->limits.num_eecs   = profile[i].num;
+			init_hca->eec_base     = profile[i].start;
+			init_hca->log_num_eecs = profile[i].log_num;
+			break;
+		case MTHCA_RES_SRQ:
+			dev->limits.num_srqs   = profile[i].num;
+			init_hca->srqc_base    = profile[i].start;
+			init_hca->log_num_srqs = profile[i].log_num;
+			break;
+		case MTHCA_RES_CQ:
+			dev->limits.num_cqs   = profile[i].num;
+			init_hca->cqc_base    = profile[i].start;
+			init_hca->log_num_cqs = profile[i].log_num;
+			break;
+		case MTHCA_RES_EQP:
+			init_hca->eqpc_base = profile[i].start;
+			break;
+		case MTHCA_RES_EEEC:
+			init_hca->eeec_base = profile[i].start;
+			break;
+		case MTHCA_RES_EQ:
+			dev->limits.num_eqs   = profile[i].num;
+			init_hca->eqc_base    = profile[i].start;
+			init_hca->log_num_eqs = profile[i].log_num;
+			break;
+		case MTHCA_RES_RDB:
+			dev->limits.num_rdbs = profile[i].num;
+			init_hca->rdb_base   = profile[i].start;
+			break;
+		case MTHCA_RES_MCG:
+			dev->limits.num_mgms      = profile[i].num >> 1;
+			dev->limits.num_amgms     = profile[i].num >> 1;
+			init_hca->mc_base         = profile[i].start;
+			init_hca->log_mc_entry_sz = ffs(MTHCA_MGM_ENTRY_SIZE) - 1;
+			init_hca->log_mc_table_sz = profile[i].log_num;
+			init_hca->mc_hash_sz      = 1 << (profile[i].log_num - 1);
+			break;
+		case MTHCA_RES_MPT:
+			dev->limits.num_mpts = profile[i].num;
+			init_hca->mpt_base   = profile[i].start;
+			init_hca->log_mpt_sz = profile[i].log_num;
+			break;
+		case MTHCA_RES_MTT:
+			dev->limits.num_mtt_segs = profile[i].num;
+			dev->limits.mtt_seg_size = MTHCA_MTT_SEG_SIZE;
+			dev->mr_table.mtt_base   = profile[i].start;
+			init_hca->mtt_base       = profile[i].start;
+			init_hca->mtt_seg_sz     = ffs(MTHCA_MTT_SEG_SIZE) - 7;
+			break;
+		case MTHCA_RES_UAR:
+			init_hca->uar_scratch_base = profile[i].start;
+			break;
+		case MTHCA_RES_UDAV:
+			dev->av_table.ddr_av_base = profile[i].start;
+			dev->av_table.num_ddr_avs = profile[i].num;
+		default:
+			break;
+		}
+	}
+
+	/*
+	 * PDs don't take any HCA memory, but we assign them as part
+	 * of the HCA profile anyway.
+	 */
+	dev->limits.num_pds = MTHCA_NUM_PDS;
+
+	kfree(profile);
+	return 0;
+}
drivers/infiniband/hw/mthca/mthca_profile.h new file +62 -0

62 lines not on this page · raw message

drivers/infiniband/hw/mthca/mthca_reset.c new file +232 -0

232 lines not on this page · raw message

#14 lore raw

PATCH v4 14/24 Add Mellanox HCA low-level driver (QP/CQ)

Add CQ (completion queue) and QP (queue pair) code for Mellanox HCA driver.

drivers/infiniband/hw/mthca/mthca_cq.c new file +831 -0
@@ -0,0 +1,831 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: mthca_cq.c 1349 2004-12-16 21:09:43Z roland $
+ */
+
+#include <linux/init.h>
+
+#include <ib_pack.h>
+
+#include "mthca_dev.h"
+#include "mthca_cmd.h"
+
+enum {
+	MTHCA_MAX_DIRECT_CQ_SIZE = 4 * PAGE_SIZE
+};
+
+enum {
+	MTHCA_CQ_ENTRY_SIZE = 0x20
+};
+
+/*
+ * Must be packed because start is 64 bits but only aligned to 32 bits.
+ */
+struct mthca_cq_context {
+	u32 flags;
+	u64 start;
+	u32 logsize_usrpage;
+	u32 error_eqn;
+	u32 comp_eqn;
+	u32 pd;
+	u32 lkey;
+	u32 last_notified_index;
+	u32 solicit_producer_index;
+	u32 consumer_index;
+	u32 producer_index;
+	u32 cqn;
+	u32 reserved[3];
+} __attribute__((packed));
+
+#define MTHCA_CQ_STATUS_OK          ( 0 << 28)
+#define MTHCA_CQ_STATUS_OVERFLOW    ( 9 << 28)
+#define MTHCA_CQ_STATUS_WRITE_FAIL  (10 << 28)
+#define MTHCA_CQ_FLAG_TR            ( 1 << 18)
+#define MTHCA_CQ_FLAG_OI            ( 1 << 17)
+#define MTHCA_CQ_STATE_DISARMED     ( 0 <<  8)
+#define MTHCA_CQ_STATE_ARMED        ( 1 <<  8)
+#define MTHCA_CQ_STATE_ARMED_SOL    ( 4 <<  8)
+#define MTHCA_EQ_STATE_FIRED        (10 <<  8)
+
+enum {
+	MTHCA_ERROR_CQE_OPCODE_MASK = 0xfe
+};
+
+enum {
+	SYNDROME_LOCAL_LENGTH_ERR 	 = 0x01,
+	SYNDROME_LOCAL_QP_OP_ERR  	 = 0x02,
+	SYNDROME_LOCAL_EEC_OP_ERR 	 = 0x03,
+	SYNDROME_LOCAL_PROT_ERR   	 = 0x04,
+	SYNDROME_WR_FLUSH_ERR     	 = 0x05,
+	SYNDROME_MW_BIND_ERR      	 = 0x06,
+	SYNDROME_BAD_RESP_ERR     	 = 0x10,
+	SYNDROME_LOCAL_ACCESS_ERR 	 = 0x11,
+	SYNDROME_REMOTE_INVAL_REQ_ERR 	 = 0x12,
+	SYNDROME_REMOTE_ACCESS_ERR 	 = 0x13,
+	SYNDROME_REMOTE_OP_ERR     	 = 0x14,
+	SYNDROME_RETRY_EXC_ERR 		 = 0x15,
+	SYNDROME_RNR_RETRY_EXC_ERR 	 = 0x16,
+	SYNDROME_LOCAL_RDD_VIOL_ERR 	 = 0x20,
+	SYNDROME_REMOTE_INVAL_RD_REQ_ERR = 0x21,
+	SYNDROME_REMOTE_ABORTED_ERR 	 = 0x22,
+	SYNDROME_INVAL_EECN_ERR 	 = 0x23,
+	SYNDROME_INVAL_EEC_STATE_ERR 	 = 0x24
+};
+
+struct mthca_cqe {
+	u32 my_qpn;
+	u32 my_ee;
+	u32 rqpn;
+	u16 sl_g_mlpath;
+	u16 rlid;
+	u32 imm_etype_pkey_eec;
+	u32 byte_cnt;
+	u32 wqe;
+	u8  opcode;
+	u8  is_send;
+	u8  reserved;
+	u8  owner;
+};
+
+struct mthca_err_cqe {
+	u32 my_qpn;
+	u32 reserved1[3];
+	u8  syndrome;
+	u8  reserved2;
+	u16 db_cnt;
+	u32 reserved3;
+	u32 wqe;
+	u8  opcode;
+	u8  reserved4[2];
+	u8  owner;
+};
+
+#define MTHCA_CQ_ENTRY_OWNER_SW      (0 << 7)
+#define MTHCA_CQ_ENTRY_OWNER_HW      (1 << 7)
+
+#define MTHCA_CQ_DB_INC_CI       (1 << 24)
+#define MTHCA_CQ_DB_REQ_NOT      (2 << 24)
+#define MTHCA_CQ_DB_REQ_NOT_SOL  (3 << 24)
+#define MTHCA_CQ_DB_SET_CI       (4 << 24)
+#define MTHCA_CQ_DB_REQ_NOT_MULT (5 << 24)
+
+static inline struct mthca_cqe *get_cqe(struct mthca_cq *cq, int entry)
+{
+	if (cq->is_direct)
+		return cq->queue.direct.buf + (entry * MTHCA_CQ_ENTRY_SIZE);
+	else
+		return cq->queue.page_list[entry * MTHCA_CQ_ENTRY_SIZE / PAGE_SIZE].buf
+			+ (entry * MTHCA_CQ_ENTRY_SIZE) % PAGE_SIZE;
+}
+
+static inline int cqe_sw(struct mthca_cq *cq, int i)
+{
+	return !(MTHCA_CQ_ENTRY_OWNER_HW &
+		 get_cqe(cq, i)->owner);
+}
+
+static inline int next_cqe_sw(struct mthca_cq *cq)
+{
+	return cqe_sw(cq, cq->cons_index);
+}
+
+static inline void set_cqe_hw(struct mthca_cq *cq, int entry)
+{
+	get_cqe(cq, entry)->owner = MTHCA_CQ_ENTRY_OWNER_HW;
+}
+
+static inline void inc_cons_index(struct mthca_dev *dev, struct mthca_cq *cq,
+				  int nent)
+{
+	u32 doorbell[2];
+
+	doorbell[0] = cpu_to_be32(MTHCA_CQ_DB_INC_CI | cq->cqn);
+	doorbell[1] = cpu_to_be32(nent - 1);
+
+	mthca_write64(doorbell,
+		      dev->kar + MTHCA_CQ_DOORBELL,
+		      MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock));
+}
+
+void mthca_cq_event(struct mthca_dev *dev, u32 cqn)
+{
+	struct mthca_cq *cq;
+
+	spin_lock(&dev->cq_table.lock);
+	cq = mthca_array_get(&dev->cq_table.cq, cqn & (dev->limits.num_cqs - 1));
+	if (cq)
+		atomic_inc(&cq->refcount);
+	spin_unlock(&dev->cq_table.lock);
+
+	if (!cq) {
+		mthca_warn(dev, "Completion event for bogus CQ %08x\n", cqn);
+		return;
+	}
+
+	cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
+
+	if (atomic_dec_and_test(&cq->refcount))
+		wake_up(&cq->wait);
+}
+
+void mthca_cq_clean(struct mthca_dev *dev, u32 cqn, u32 qpn)
+{
+	struct mthca_cq *cq;
+	struct mthca_cqe *cqe;
+	int prod_index;
+	int nfreed = 0;
+
+	spin_lock_irq(&dev->cq_table.lock);
+	cq = mthca_array_get(&dev->cq_table.cq, cqn & (dev->limits.num_cqs - 1));
+	if (cq)
+		atomic_inc(&cq->refcount);
+	spin_unlock_irq(&dev->cq_table.lock);
+
+	if (!cq)
+		return;
+
+	spin_lock_irq(&cq->lock);
+
+	/*
+	 * First we need to find the current producer index, so we
+	 * know where to start cleaning from.  It doesn't matter if HW
+	 * adds new entries after this loop -- the QP we're worried
+	 * about is already in RESET, so the new entries won't come
+	 * from our QP and therefore don't need to be checked.
+	 */
+	for (prod_index = cq->cons_index;
+	     cqe_sw(cq, prod_index & cq->ibcq.cqe);
+	     ++prod_index)
+		if (prod_index == cq->cons_index + cq->ibcq.cqe)
+			break;
+
+	if (0)
+		mthca_dbg(dev, "Cleaning QPN %06x from CQN %06x; ci %d, pi %d\n",
+			  qpn, cqn, cq->cons_index, prod_index);
+
+	/*
+	 * Now sweep backwards through the CQ, removing CQ entries
+	 * that match our QP by copying older entries on top of them.
+	 */
+	while (prod_index > cq->cons_index) {
+		cqe = get_cqe(cq, (prod_index - 1) & cq->ibcq.cqe);
+		if (cqe->my_qpn == cpu_to_be32(qpn))
+			++nfreed;
+		else if (nfreed)
+			memcpy(get_cqe(cq, (prod_index - 1 + nfreed) &
+				       cq->ibcq.cqe),
+			       cqe,
+			       MTHCA_CQ_ENTRY_SIZE);
+		--prod_index;
+	}
+
+	if (nfreed) {
+		wmb();
+		inc_cons_index(dev, cq, nfreed);
+		cq->cons_index = (cq->cons_index + nfreed) & cq->ibcq.cqe;
+	}
+
+	spin_unlock_irq(&cq->lock);
+	if (atomic_dec_and_test(&cq->refcount))
+		wake_up(&cq->wait);
+}
+
+static int handle_error_cqe(struct mthca_dev *dev, struct mthca_cq *cq,
+			    struct mthca_qp *qp, int wqe_index, int is_send,
+			    struct mthca_err_cqe *cqe,
+			    struct ib_wc *entry, int *free_cqe)
+{
+	int err;
+	int dbd;
+	u32 new_wqe;
+
+	if (1 && cqe->syndrome != SYNDROME_WR_FLUSH_ERR) {
+		int j;
+
+		mthca_dbg(dev, "%x/%d: error CQE -> QPN %06x, WQE @ %08x\n",
+			  cq->cqn, cq->cons_index, be32_to_cpu(cqe->my_qpn),
+			  be32_to_cpu(cqe->wqe));
+
+		for (j = 0; j < 8; ++j)
+			printk(KERN_DEBUG "  [%2x] %08x\n",
+			       j * 4, be32_to_cpu(((u32 *) cqe)[j]));
+	}
+
+	/*
+	 * For completions in error, only work request ID, status (and
+	 * freed resource count for RD) have to be set.
+	 */
+	switch (cqe->syndrome) {
+	case SYNDROME_LOCAL_LENGTH_ERR:
+		entry->status = IB_WC_LOC_LEN_ERR;
+		break;
+	case SYNDROME_LOCAL_QP_OP_ERR:
+		entry->status = IB_WC_LOC_QP_OP_ERR;
+		break;
+	case SYNDROME_LOCAL_EEC_OP_ERR:
+		entry->status = IB_WC_LOC_EEC_OP_ERR;
+		break;
+	case SYNDROME_LOCAL_PROT_ERR:
+		entry->status = IB_WC_LOC_PROT_ERR;
+		break;
+	case SYNDROME_WR_FLUSH_ERR:
+		entry->status = IB_WC_WR_FLUSH_ERR;
+		break;
+	case SYNDROME_MW_BIND_ERR:
+		entry->status = IB_WC_MW_BIND_ERR;
+		break;
+	case SYNDROME_BAD_RESP_ERR:
+		entry->status = IB_WC_BAD_RESP_ERR;
+		break;
+	case SYNDROME_LOCAL_ACCESS_ERR:
+		entry->status = IB_WC_LOC_ACCESS_ERR;
+		break;
+	case SYNDROME_REMOTE_INVAL_REQ_ERR:
+		entry->status = IB_WC_REM_INV_REQ_ERR;
+		break;
+	case SYNDROME_REMOTE_ACCESS_ERR:
+		entry->status = IB_WC_REM_ACCESS_ERR;
+		break;
+	case SYNDROME_REMOTE_OP_ERR:
+		entry->status = IB_WC_REM_OP_ERR;
+		break;
+	case SYNDROME_RETRY_EXC_ERR:
+		entry->status = IB_WC_RETRY_EXC_ERR;
+		break;
+	case SYNDROME_RNR_RETRY_EXC_ERR:
+		entry->status = IB_WC_RNR_RETRY_EXC_ERR;
+		break;
+	case SYNDROME_LOCAL_RDD_VIOL_ERR:
+		entry->status = IB_WC_LOC_RDD_VIOL_ERR;
+		break;
+	case SYNDROME_REMOTE_INVAL_RD_REQ_ERR:
+		entry->status = IB_WC_REM_INV_RD_REQ_ERR;
+		break;
+	case SYNDROME_REMOTE_ABORTED_ERR:
+		entry->status = IB_WC_REM_ABORT_ERR;
+		break;
+	case SYNDROME_INVAL_EECN_ERR:
+		entry->status = IB_WC_INV_EECN_ERR;
+		break;
+	case SYNDROME_INVAL_EEC_STATE_ERR:
+		entry->status = IB_WC_INV_EEC_STATE_ERR;
+		break;
+	default:
+		entry->status = IB_WC_GENERAL_ERR;
+		break;
+	}
+
+	err = mthca_free_err_wqe(qp, is_send, wqe_index, &dbd, &new_wqe);
+	if (err)
+		return err;
+
+	/*
+	 * If we're at the end of the WQE chain, or we've used up our
+	 * doorbell count, free the CQE.  Otherwise just update it for
+	 * the next poll operation.
+	 */
+	if (!(new_wqe & cpu_to_be32(0x3f)) || (!cqe->db_cnt && dbd))
+		return 0;
+
+	cqe->db_cnt   = cpu_to_be16(be16_to_cpu(cqe->db_cnt) - dbd);
+	cqe->wqe      = new_wqe;
+	cqe->syndrome = SYNDROME_WR_FLUSH_ERR;
+
+	*free_cqe = 0;
+
+	return 0;
+}
+
+static void dump_cqe(struct mthca_cqe *cqe)
+{
+	int j;
+
+	for (j = 0; j < 8; ++j)
+		printk(KERN_DEBUG "  [%2x] %08x\n",
+		       j * 4, be32_to_cpu(((u32 *) cqe)[j]));
+}
+
+static inline int mthca_poll_one(struct mthca_dev *dev,
+				 struct mthca_cq *cq,
+				 struct mthca_qp **cur_qp,
+				 int *freed,
+				 struct ib_wc *entry)
+{
+	struct mthca_wq *wq;
+	struct mthca_cqe *cqe;
+	int wqe_index;
+	int is_error = 0;
+	int is_send;
+	int free_cqe = 1;
+	int err = 0;
+
+	if (!next_cqe_sw(cq))
+		return -EAGAIN;
+
+	rmb();
+
+	cqe = get_cqe(cq, cq->cons_index);
+
+	if (0) {
+		mthca_dbg(dev, "%x/%d: CQE -> QPN %06x, WQE @ %08x\n",
+			  cq->cqn, cq->cons_index, be32_to_cpu(cqe->my_qpn),
+			  be32_to_cpu(cqe->wqe));
+
+		dump_cqe(cqe);
+	}
+
+	if ((cqe->opcode & MTHCA_ERROR_CQE_OPCODE_MASK) ==
+	    MTHCA_ERROR_CQE_OPCODE_MASK) {
+		is_error = 1;
+		is_send = cqe->opcode & 1;
+	} else
+		is_send = cqe->is_send & 0x80;
+
+	if (!*cur_qp || be32_to_cpu(cqe->my_qpn) != (*cur_qp)->qpn) {
+		if (*cur_qp) {
+			spin_unlock(&(*cur_qp)->lock);
+			if (atomic_dec_and_test(&(*cur_qp)->refcount))
+				wake_up(&(*cur_qp)->wait);
+		}
+
+		spin_lock(&dev->qp_table.lock);
+		*cur_qp = mthca_array_get(&dev->qp_table.qp,
+					  be32_to_cpu(cqe->my_qpn) &
+					  (dev->limits.num_qps - 1));
+		if (*cur_qp)
+			atomic_inc(&(*cur_qp)->refcount);
+		spin_unlock(&dev->qp_table.lock);
+
+		if (!*cur_qp) {
+			mthca_warn(dev, "CQ entry for unknown QP %06x\n",
+				   be32_to_cpu(cqe->my_qpn) & 0xffffff);
+			err = -EINVAL;
+			goto out;
+		}
+
+		spin_lock(&(*cur_qp)->lock);
+	}
+
+	if (is_send) {
+		wq = &(*cur_qp)->sq;
+		wqe_index = ((be32_to_cpu(cqe->wqe) - (*cur_qp)->send_wqe_offset)
+			     >> wq->wqe_shift);
+		entry->wr_id = (*cur_qp)->wrid[wqe_index +
+					       (*cur_qp)->rq.max];
+	} else {
+		wq = &(*cur_qp)->rq;
+		wqe_index = be32_to_cpu(cqe->wqe) >> wq->wqe_shift;
+		entry->wr_id = (*cur_qp)->wrid[wqe_index];
+	}
+
+	if (wq->last_comp < wqe_index)
+		wq->cur -= wqe_index - wq->last_comp;
+	else
+		wq->cur -= wq->max - wq->last_comp + wqe_index;
+
+	wq->last_comp = wqe_index;
+
+	if (0)
+		mthca_dbg(dev, "%s completion for QP %06x, index %d (nr %d)\n",
+			  is_send ? "Send" : "Receive",
+			  (*cur_qp)->qpn, wqe_index, wq->max);
+
+	if (is_error) {
+		err = handle_error_cqe(dev, cq, *cur_qp, wqe_index, is_send,
+				       (struct mthca_err_cqe *) cqe,
+				       entry, &free_cqe);
+		goto out;
+	}
+
+	if (is_send) {
+		entry->opcode = IB_WC_SEND; /* XXX */
+	} else {
+		entry->byte_len = be32_to_cpu(cqe->byte_cnt);
+		switch (cqe->opcode & 0x1f) {
+		case IB_OPCODE_SEND_LAST_WITH_IMMEDIATE:
+		case IB_OPCODE_SEND_ONLY_WITH_IMMEDIATE:
+			entry->wc_flags = IB_WC_WITH_IMM;
+			entry->imm_data = cqe->imm_etype_pkey_eec;
+			entry->opcode = IB_WC_RECV;
+			break;
+		case IB_OPCODE_RDMA_WRITE_LAST_WITH_IMMEDIATE:
+		case IB_OPCODE_RDMA_WRITE_ONLY_WITH_IMMEDIATE:
+			entry->wc_flags = IB_WC_WITH_IMM;
+			entry->imm_data = cqe->imm_etype_pkey_eec;
+			entry->opcode = IB_WC_RECV_RDMA_WITH_IMM;
+			break;
+		default:
+			entry->wc_flags = 0;
+			entry->opcode = IB_WC_RECV;
+			break;
+		}
+		entry->slid 	   = be16_to_cpu(cqe->rlid);
+		entry->sl   	   = be16_to_cpu(cqe->sl_g_mlpath) >> 12;
+		entry->src_qp 	   = be32_to_cpu(cqe->rqpn) & 0xffffff;
+		entry->dlid_path_bits = be16_to_cpu(cqe->sl_g_mlpath) & 0x7f;
+		entry->pkey_index  = be32_to_cpu(cqe->imm_etype_pkey_eec) >> 16;
+		entry->wc_flags   |= be16_to_cpu(cqe->sl_g_mlpath) & 0x80 ?
+					IB_WC_GRH : 0;
+	}
+
+	entry->status = IB_WC_SUCCESS;
+
+ out:
+	if (free_cqe) {
+		set_cqe_hw(cq, cq->cons_index);
+		++(*freed);
+		cq->cons_index = (cq->cons_index + 1) & cq->ibcq.cqe;
+	}
+
+	return err;
+}
+
+int mthca_poll_cq(struct ib_cq *ibcq, int num_entries,
+		  struct ib_wc *entry)
+{
+	struct mthca_dev *dev = to_mdev(ibcq->device);
+	struct mthca_cq *cq = to_mcq(ibcq);
+	struct mthca_qp *qp = NULL;
+	unsigned long flags;
+	int err = 0;
+	int freed = 0;
+	int npolled;
+
+	spin_lock_irqsave(&cq->lock, flags);
+
+	for (npolled = 0; npolled < num_entries; ++npolled) {
+		err = mthca_poll_one(dev, cq, &qp,
+				     &freed, entry + npolled);
+		if (err)
+			break;
+	}
+
+	if (qp) {
+		spin_unlock(&qp->lock);
+		if (atomic_dec_and_test(&qp->refcount))
+			wake_up(&qp->wait);
+	}
+
+	if (freed) {
+		wmb();
+		while (freed--)
+			inc_cons_index(dev, cq, 1);
+	}
+
+	spin_unlock_irqrestore(&cq->lock, flags);
+
+	return err == 0 || err == -EAGAIN ? npolled : err;
+}
+
+void mthca_arm_cq(struct mthca_dev *dev, struct mthca_cq *cq,
+		  int solicited)
+{
+	u32 doorbell[2];
+
+	doorbell[0] =  cpu_to_be32((solicited ?
+				    MTHCA_CQ_DB_REQ_NOT_SOL :
+				    MTHCA_CQ_DB_REQ_NOT)      |
+				   cq->cqn);
+	doorbell[1] = 0xffffffff;
+
+	mthca_write64(doorbell,
+		      dev->kar + MTHCA_CQ_DOORBELL,
+		      MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock));
+}
+
+int mthca_init_cq(struct mthca_dev *dev, int nent,
+		  struct mthca_cq *cq)
+{
+	int size = nent * MTHCA_CQ_ENTRY_SIZE;
+	dma_addr_t t;
+	void *mailbox = NULL;
+	int npages, shift;
+	u64 *dma_list = NULL;
+	struct mthca_cq_context *cq_context;
+	int err = -ENOMEM;
+	u8 status;
+	int i;
+
+	might_sleep();
+
+	mailbox = kmalloc(sizeof (struct mthca_cq_context) + MTHCA_CMD_MAILBOX_EXTRA,
+			  GFP_KERNEL);
+	if (!mailbox)
+		goto err_out;
+
+	cq_context = MAILBOX_ALIGN(mailbox);
+
+	if (size <= MTHCA_MAX_DIRECT_CQ_SIZE) {
+		if (0)
+			mthca_dbg(dev, "Creating direct CQ of size %d\n", size);
+
+		cq->is_direct = 1;
+		npages        = 1;
+		shift         = get_order(size) + PAGE_SHIFT;
+
+		cq->queue.direct.buf = pci_alloc_consistent(dev->pdev,
+							    size, &t);
+		if (!cq->queue.direct.buf)
+			goto err_out;
+
+		pci_unmap_addr_set(&cq->queue.direct, mapping, t);
+
+		memset(cq->queue.direct.buf, 0, size);
+
+		while (t & ((1 << shift) - 1)) {
+			--shift;
+			npages *= 2;
+		}
+
+		dma_list = kmalloc(npages * sizeof *dma_list, GFP_KERNEL);
+		if (!dma_list)
+			goto err_out_free;
+
+		for (i = 0; i < npages; ++i)
+			dma_list[i] = t + i * (1 << shift);
+	} else {
+		cq->is_direct = 0;
+		npages        = (size + PAGE_SIZE - 1) / PAGE_SIZE;
+		shift         = PAGE_SHIFT;
+
+		if (0)
+			mthca_dbg(dev, "Creating indirect CQ with %d pages\n", npages);
+
+		dma_list = kmalloc(npages * sizeof *dma_list, GFP_KERNEL);
+		if (!dma_list)
+			goto err_out;
+
+		cq->queue.page_list = kmalloc(npages * sizeof *cq->queue.page_list,
+					      GFP_KERNEL);
+		if (!cq->queue.page_list)
+			goto err_out;
+
+		for (i = 0; i < npages; ++i)
+			cq->queue.page_list[i].buf = NULL;
+
+		for (i = 0; i < npages; ++i) {
+			cq->queue.page_list[i].buf =
+				pci_alloc_consistent(dev->pdev, PAGE_SIZE, &t);
+			if (!cq->queue.page_list[i].buf)
+				goto err_out_free;
+
+			dma_list[i] = t;
+			pci_unmap_addr_set(&cq->queue.page_list[i], mapping, t);
+
+			memset(cq->queue.page_list[i].buf, 0, PAGE_SIZE);
+		}
+	}
+
+	for (i = 0; i < nent; ++i)
+		set_cqe_hw(cq, i);
+
+	cq->cqn = mthca_alloc(&dev->cq_table.alloc);
+	if (cq->cqn == -1)
+		goto err_out_free;
+
+	err = mthca_mr_alloc_phys(dev, dev->driver_pd.pd_num,
+				  dma_list, shift, npages,
+				  0, size,
+				  MTHCA_MPT_FLAG_LOCAL_WRITE |
+				  MTHCA_MPT_FLAG_LOCAL_READ,
+				  &cq->mr);
+	if (err)
+		goto err_out_free_cq;
+
+	spin_lock_init(&cq->lock);
+	atomic_set(&cq->refcount, 1);
+	init_waitqueue_head(&cq->wait);
+
+	memset(cq_context, 0, sizeof *cq_context);
+	cq_context->flags           = cpu_to_be32(MTHCA_CQ_STATUS_OK      |
+						  MTHCA_CQ_STATE_DISARMED |
+						  MTHCA_CQ_FLAG_TR);
+	cq_context->start           = cpu_to_be64(0);
+	cq_context->logsize_usrpage = cpu_to_be32((ffs(nent) - 1) << 24 |
+						  MTHCA_KAR_PAGE);
+	cq_context->error_eqn       = cpu_to_be32(dev->eq_table.eq[MTHCA_EQ_ASYNC].eqn);
+	cq_context->comp_eqn        = cpu_to_be32(dev->eq_table.eq[MTHCA_EQ_COMP].eqn);
+	cq_context->pd              = cpu_to_be32(dev->driver_pd.pd_num);
+	cq_context->lkey            = cpu_to_be32(cq->mr.ibmr.lkey);
+	cq_context->cqn             = cpu_to_be32(cq->cqn);
+
+	err = mthca_SW2HW_CQ(dev, cq_context, cq->cqn, &status);
+	if (err) {
+		mthca_warn(dev, "SW2HW_CQ failed (%d)\n", err);
+		goto err_out_free_mr;
+	}
+
+	if (status) {
+		mthca_warn(dev, "SW2HW_CQ returned status 0x%02x\n",
+			   status);
+		err = -EINVAL;
+		goto err_out_free_mr;
+	}
+
+	spin_lock_irq(&dev->cq_table.lock);
+	if (mthca_array_set(&dev->cq_table.cq,
+			    cq->cqn & (dev->limits.num_cqs - 1),
+			    cq)) {
+		spin_unlock_irq(&dev->cq_table.lock);
+		goto err_out_free_mr;
+	}
+	spin_unlock_irq(&dev->cq_table.lock);
+
+	cq->cons_index = 0;
+
+	kfree(dma_list);
+	kfree(mailbox);
+
+	return 0;
+
+ err_out_free_mr:
+	mthca_free_mr(dev, &cq->mr);
+
+ err_out_free_cq:
+	mthca_free(&dev->cq_table.alloc, cq->cqn);
+
+ err_out_free:
+	if (cq->is_direct)
+		pci_free_consistent(dev->pdev, size,
+				    cq->queue.direct.buf,
+				    pci_unmap_addr(&cq->queue.direct, mapping));
+	else {
+		for (i = 0; i < npages; ++i)
+			if (cq->queue.page_list[i].buf)
+				pci_free_consistent(dev->pdev, PAGE_SIZE,
+						    cq->queue.page_list[i].buf,
+						    pci_unmap_addr(&cq->queue.page_list[i],
+								   mapping));
+
+		kfree(cq->queue.page_list);
+	}
+
+ err_out:
+	kfree(dma_list);
+	kfree(mailbox);
+
+	return err;
+}
+
+void mthca_free_cq(struct mthca_dev *dev,
+		   struct mthca_cq *cq)
+{
+	void *mailbox;
+	int err;
+	u8 status;
+
+	might_sleep();
+
+	mailbox = kmalloc(sizeof (struct mthca_cq_context) + MTHCA_CMD_MAILBOX_EXTRA,
+			  GFP_KERNEL);
+	if (!mailbox) {
+		mthca_warn(dev, "No memory for mailbox to free CQ.\n");
+		return;
+	}
+
+	err = mthca_HW2SW_CQ(dev, MAILBOX_ALIGN(mailbox), cq->cqn, &status);
+	if (err)
+		mthca_warn(dev, "HW2SW_CQ failed (%d)\n", err);
+	else if (status)
+		mthca_warn(dev, "HW2SW_CQ returned status 0x%02x\n",
+			   status);
+
+	if (0) {
+		u32 *ctx = MAILBOX_ALIGN(mailbox);
+		int j;
+
+		printk(KERN_ERR "context for CQN %x\n", cq->cqn);
+		for (j = 0; j < 16; ++j)
+			printk(KERN_ERR "[%2x] %08x\n", j * 4, be32_to_cpu(ctx[j]));
+	}
+
+	spin_lock_irq(&dev->cq_table.lock);
+	mthca_array_clear(&dev->cq_table.cq,
+			  cq->cqn & (dev->limits.num_cqs - 1));
+	spin_unlock_irq(&dev->cq_table.lock);
+
+	atomic_dec(&cq->refcount);
+	wait_event(cq->wait, !atomic_read(&cq->refcount));
+
+	mthca_free_mr(dev, &cq->mr);
+
+	if (cq->is_direct)
+		pci_free_consistent(dev->pdev,
+				    (cq->ibcq.cqe + 1) * MTHCA_CQ_ENTRY_SIZE,
+				    cq->queue.direct.buf,
+				    pci_unmap_addr(&cq->queue.direct,
+						   mapping));
+	else {
+		int i;
+
+		for (i = 0;
+		     i < ((cq->ibcq.cqe + 1) * MTHCA_CQ_ENTRY_SIZE + PAGE_SIZE - 1) /
+			     PAGE_SIZE;
+		     ++i)
+			pci_free_consistent(dev->pdev, PAGE_SIZE,
+					    cq->queue.page_list[i].buf,
+					    pci_unmap_addr(&cq->queue.page_list[i],
+							   mapping));
+
+		kfree(cq->queue.page_list);
+	}
+
+	mthca_free(&dev->cq_table.alloc, cq->cqn);
+	kfree(mailbox);
+}
+
+int __devinit mthca_init_cq_table(struct mthca_dev *dev)
+{
+	int err;
+
+	spin_lock_init(&dev->cq_table.lock);
+
+	err = mthca_alloc_init(&dev->cq_table.alloc,
+			       dev->limits.num_cqs,
+			       (1 << 24) - 1,
+			       dev->limits.reserved_cqs);
+	if (err)
+		return err;
+
+	err = mthca_array_init(&dev->cq_table.cq,
+			       dev->limits.num_cqs);
+	if (err)
+		mthca_alloc_cleanup(&dev->cq_table.alloc);
+
+	return err;
+}
+
+void __devexit mthca_cleanup_cq_table(struct mthca_dev *dev)
+{
+	mthca_array_cleanup(&dev->cq_table.cq, dev->limits.num_cqs);
+	mthca_alloc_cleanup(&dev->cq_table.alloc);
+}
drivers/infiniband/hw/mthca/mthca_qp.c new file +1536 -0

1,536 lines not on this page · raw message

#15 lore raw

PATCH v4 15/24 Add Mellanox HCA low-level driver (last bits)

Add code for remaining InfiniBand objects (address vectors, multicast groups, memory regions and protection domains)

drivers/infiniband/hw/mthca/mthca_av.c new file +219 -0
@@ -0,0 +1,219 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: mthca_av.c 1349 2004-12-16 21:09:43Z roland $
+ */
+
+#include <linux/init.h>
+
+#include <ib_verbs.h>
+#include <ib_cache.h>
+
+#include "mthca_dev.h"
+
+struct mthca_av {
+	u32 port_pd;
+	u8  reserved1;
+	u8  g_slid;
+	u16 dlid;
+	u8  reserved2;
+	u8  gid_index;
+	u8  msg_sr;
+	u8  hop_limit;
+	u32 sl_tclass_flowlabel;
+	u32 dgid[4];
+};
+
+int mthca_create_ah(struct mthca_dev *dev,
+		    struct mthca_pd *pd,
+		    struct ib_ah_attr *ah_attr,
+		    struct mthca_ah *ah)
+{
+	u32 index = -1;
+	struct mthca_av *av = NULL;
+
+	ah->on_hca = 0;
+
+	if (!atomic_read(&pd->sqp_count) &&
+	    !(dev->mthca_flags & MTHCA_FLAG_DDR_HIDDEN)) {
+		index = mthca_alloc(&dev->av_table.alloc);
+
+		/* fall back to allocate in host memory */
+		if (index == -1)
+			goto host_alloc;
+
+		av = kmalloc(sizeof *av, GFP_KERNEL);
+		if (!av)
+			goto host_alloc;
+
+		ah->on_hca = 1;
+		ah->avdma  = dev->av_table.ddr_av_base +
+			index * MTHCA_AV_SIZE;
+	}
+
+ host_alloc:
+	if (!ah->on_hca) {
+		ah->av = pci_pool_alloc(dev->av_table.pool,
+					SLAB_KERNEL, &ah->avdma);
+		if (!ah->av)
+			return -ENOMEM;
+
+		av = ah->av;
+	}
+
+	ah->key = pd->ntmr.ibmr.lkey;
+
+	memset(av, 0, MTHCA_AV_SIZE);
+
+	av->port_pd = cpu_to_be32(pd->pd_num | (ah_attr->port_num << 24));
+	av->g_slid  = ah_attr->src_path_bits;
+	av->dlid    = cpu_to_be16(ah_attr->dlid);
+	av->msg_sr  = (3 << 4) | /* 2K message */
+		ah_attr->static_rate;
+	av->sl_tclass_flowlabel = cpu_to_be32(ah_attr->sl << 28);
+	if (ah_attr->ah_flags & IB_AH_GRH) {
+		av->g_slid |= 0x80;
+		av->gid_index = (ah_attr->port_num - 1) * dev->limits.gid_table_len +
+			ah_attr->grh.sgid_index;
+		av->hop_limit = ah_attr->grh.hop_limit;
+		av->sl_tclass_flowlabel |=
+			cpu_to_be32((ah_attr->grh.traffic_class << 20) |
+				    ah_attr->grh.flow_label);
+		memcpy(av->dgid, ah_attr->grh.dgid.raw, 16);
+	} else {
+		/* Arbel workaround -- low byte of GID must be 2 */
+		av->dgid[3] = cpu_to_be32(2);
+	}
+
+	if (0) {
+		int j;
+
+		mthca_dbg(dev, "Created UDAV at %p/%08lx:\n",
+			  av, (unsigned long) ah->avdma);
+		for (j = 0; j < 8; ++j)
+			printk(KERN_DEBUG "  [%2x] %08x\n",
+			       j * 4, be32_to_cpu(((u32 *) av)[j]));
+	}
+
+	if (ah->on_hca) {
+		memcpy_toio(dev->av_table.av_map + index * MTHCA_AV_SIZE,
+			    av, MTHCA_AV_SIZE);
+		kfree(av);
+	}
+
+	return 0;
+}
+
+int mthca_destroy_ah(struct mthca_dev *dev, struct mthca_ah *ah)
+{
+	if (ah->on_hca)
+		mthca_free(&dev->av_table.alloc,
+ 			   (ah->avdma - dev->av_table.ddr_av_base) /
+			   MTHCA_AV_SIZE);
+	else
+		pci_pool_free(dev->av_table.pool, ah->av, ah->avdma);
+
+	return 0;
+}
+
+int mthca_read_ah(struct mthca_dev *dev, struct mthca_ah *ah,
+		  struct ib_ud_header *header)
+{
+	if (ah->on_hca)
+		return -EINVAL;
+
+	header->lrh.service_level   = be32_to_cpu(ah->av->sl_tclass_flowlabel) >> 28;
+	header->lrh.destination_lid = ah->av->dlid;
+	header->lrh.source_lid      = ah->av->g_slid & 0x7f;
+	if (ah->av->g_slid & 0x80) {
+		header->grh_present = 1;
+		header->grh.traffic_class =
+			(be32_to_cpu(ah->av->sl_tclass_flowlabel) >> 20) & 0xff;
+		header->grh.flow_label    =
+			ah->av->sl_tclass_flowlabel & cpu_to_be32(0xfffff);
+		ib_cached_gid_get(&dev->ib_dev,
+				  be32_to_cpu(ah->av->port_pd) >> 24,
+				  ah->av->gid_index,
+				  &header->grh.source_gid);
+		memcpy(header->grh.destination_gid.raw,
+		       ah->av->dgid, 16);
+	} else {
+		header->grh_present = 0;
+	}
+
+	return 0;
+}
+
+int __devinit mthca_init_av_table(struct mthca_dev *dev)
+{
+	int err;
+
+	err = mthca_alloc_init(&dev->av_table.alloc,
+			       dev->av_table.num_ddr_avs,
+			       dev->av_table.num_ddr_avs - 1,
+			       0);
+	if (err)
+		return err;
+
+	dev->av_table.pool = pci_pool_create("mthca_av", dev->pdev,
+					     MTHCA_AV_SIZE,
+					     MTHCA_AV_SIZE, 0);
+	if (!dev->av_table.pool)
+		goto out_free_alloc;
+
+	if (!(dev->mthca_flags & MTHCA_FLAG_DDR_HIDDEN)) {
+		dev->av_table.av_map = ioremap(pci_resource_start(dev->pdev, 4) +
+					       dev->av_table.ddr_av_base -
+					       dev->ddr_start,
+					       dev->av_table.num_ddr_avs *
+					       MTHCA_AV_SIZE);
+		if (!dev->av_table.av_map)
+			goto out_free_pool;
+	} else
+		dev->av_table.av_map = NULL;
+
+	return 0;
+
+ out_free_pool:
+	pci_pool_destroy(dev->av_table.pool);
+
+ out_free_alloc:
+	mthca_alloc_cleanup(&dev->av_table.alloc);
+	return -ENOMEM;
+}
+
+void __devexit mthca_cleanup_av_table(struct mthca_dev *dev)
+{
+	if (dev->av_table.av_map)
+		iounmap(dev->av_table.av_map);
+	pci_pool_destroy(dev->av_table.pool);
+	mthca_alloc_cleanup(&dev->av_table.alloc);
+}
drivers/infiniband/hw/mthca/mthca_mcg.c new file +376 -0

376 lines not on this page · raw message

drivers/infiniband/hw/mthca/mthca_mr.c new file +396 -0

396 lines not on this page · raw message

drivers/infiniband/hw/mthca/mthca_pd.c new file +80 -0

80 lines not on this page · raw message

#16 lore raw

PATCH v4 16/24 Add Mellanox HCA low-level driver (MAD)

Add MAD (management datagram) code for Mellanox HCA driver.

drivers/infiniband/hw/mthca/mthca_mad.c new file +320 -0
@@ -0,0 +1,320 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: mthca_mad.c 1349 2004-12-16 21:09:43Z roland $
+ */
+
+#include <ib_verbs.h>
+#include <ib_mad.h>
+#include <ib_smi.h>
+
+#include "mthca_dev.h"
+#include "mthca_cmd.h"
+
+enum {
+	MTHCA_VENDOR_CLASS1 = 0x9,
+	MTHCA_VENDOR_CLASS2 = 0xa
+};
+
+struct mthca_trap_mad {
+	struct ib_mad *mad;
+	DECLARE_PCI_UNMAP_ADDR(mapping)
+};
+
+static void update_sm_ah(struct mthca_dev *dev,
+			 u8 port_num, u16 lid, u8 sl)
+{
+	struct ib_ah *new_ah;
+	struct ib_ah_attr ah_attr;
+	unsigned long flags;
+
+	if (!dev->send_agent[port_num - 1][0])
+		return;
+
+	memset(&ah_attr, 0, sizeof ah_attr);
+	ah_attr.dlid     = lid;
+	ah_attr.sl       = sl;
+	ah_attr.port_num = port_num;
+
+	new_ah = ib_create_ah(dev->send_agent[port_num - 1][0]->qp->pd,
+			      &ah_attr);
+	if (IS_ERR(new_ah))
+		return;
+
+	spin_lock_irqsave(&dev->sm_lock, flags);
+	if (dev->sm_ah[port_num - 1])
+		ib_destroy_ah(dev->sm_ah[port_num - 1]);
+	dev->sm_ah[port_num - 1] = new_ah;
+	spin_unlock_irqrestore(&dev->sm_lock, flags);
+}
+
+/*
+ * Snoop SM MADs for port info and P_Key table sets, so we can
+ * synthesize LID change and P_Key change events.
+ */
+static void smp_snoop(struct ib_device *ibdev,
+		      u8 port_num,
+		      struct ib_mad *mad)
+{
+	struct ib_event event;
+
+	if ((mad->mad_hdr.mgmt_class  == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
+	     mad->mad_hdr.mgmt_class  == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) &&
+	    mad->mad_hdr.method     == IB_MGMT_METHOD_SET) {
+		if (mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO) {
+			update_sm_ah(to_mdev(ibdev), port_num,
+				     be16_to_cpup((__be16 *) (mad->data + 58)),
+				     (*(u8 *) (mad->data + 76)) & 0xf);
+
+			event.device           = ibdev;
+			event.event            = IB_EVENT_LID_CHANGE;
+			event.element.port_num = port_num;
+			ib_dispatch_event(&event);
+		}
+
+		if (mad->mad_hdr.attr_id == IB_SMP_ATTR_PKEY_TABLE) {
+			event.device           = ibdev;
+			event.event            = IB_EVENT_PKEY_CHANGE;
+			event.element.port_num = port_num;
+			ib_dispatch_event(&event);
+		}
+	}
+}
+
+static void forward_trap(struct mthca_dev *dev,
+			 u8 port_num,
+			 struct ib_mad *mad)
+{
+	int qpn = mad->mad_hdr.mgmt_class != IB_MGMT_CLASS_SUBN_LID_ROUTED;
+	struct mthca_trap_mad *tmad;
+	struct ib_sge      gather_list;
+	struct ib_send_wr *bad_wr, wr = {
+		.opcode      = IB_WR_SEND,
+		.sg_list     = &gather_list,
+		.num_sge     = 1,
+		.send_flags  = IB_SEND_SIGNALED,
+		.wr	     = {
+			 .ud = {
+				 .remote_qpn  = qpn,
+				 .remote_qkey = qpn ? IB_QP1_QKEY : 0,
+				 .timeout_ms  = 0
+			 }
+		 }
+	};
+	struct ib_mad_agent *agent = dev->send_agent[port_num - 1][qpn];
+	int ret;
+	unsigned long flags;
+
+	if (agent) {
+		tmad = kmalloc(sizeof *tmad, GFP_KERNEL);
+		if (!tmad)
+			return;
+
+		tmad->mad = kmalloc(sizeof *tmad->mad, GFP_KERNEL);
+		if (!tmad->mad) {
+			kfree(tmad);
+			return;
+		}
+
+		memcpy(tmad->mad, mad, sizeof *mad);
+
+		wr.wr.ud.mad_hdr = &tmad->mad->mad_hdr;
+		wr.wr_id         = (unsigned long) tmad;
+
+		gather_list.addr   = dma_map_single(agent->device->dma_device,
+						    tmad->mad,
+						    sizeof *tmad->mad,
+						    DMA_TO_DEVICE);
+		gather_list.length = sizeof *tmad->mad;
+		gather_list.lkey   = to_mpd(agent->qp->pd)->ntmr.ibmr.lkey;
+		pci_unmap_addr_set(tmad, mapping, gather_list.addr);
+
+		/*
+		 * We rely here on the fact that MLX QPs don't use the
+		 * address handle after the send is posted (this is
+		 * wrong following the IB spec strictly, but we know
+		 * it's OK for our devices).
+		 */
+		spin_lock_irqsave(&dev->sm_lock, flags);
+		wr.wr.ud.ah      = dev->sm_ah[port_num - 1];
+		if (wr.wr.ud.ah)
+			ret = ib_post_send_mad(agent, &wr, &bad_wr);
+		else
+			ret = -EINVAL;
+		spin_unlock_irqrestore(&dev->sm_lock, flags);
+
+		if (ret) {
+			dma_unmap_single(agent->device->dma_device,
+					 pci_unmap_addr(tmad, mapping),
+					 sizeof *tmad->mad,
+					 DMA_TO_DEVICE);
+			kfree(tmad->mad);
+			kfree(tmad);
+		}
+	}
+}
+
+int mthca_process_mad(struct ib_device *ibdev,
+		      int mad_flags,
+		      u8 port_num,
+		      u16 slid,
+		      struct ib_mad *in_mad,
+		      struct ib_mad *out_mad)
+{
+	int err;
+	u8 status;
+
+	/* Forward locally generated traps to the SM */
+	if (in_mad->mad_hdr.method == IB_MGMT_METHOD_TRAP &&
+	    slid == 0) {
+		forward_trap(to_mdev(ibdev), port_num, in_mad);
+		return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
+	}
+
+	/*
+	 * Only handle SM gets, sets and trap represses for SM class
+	 *
+	 * Only handle PMA and Mellanox vendor-specific class gets and
+	 * sets for other classes.
+	 */
+	if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
+	    in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
+		if (in_mad->mad_hdr.method   != IB_MGMT_METHOD_GET &&
+		    in_mad->mad_hdr.method   != IB_MGMT_METHOD_SET &&
+		    in_mad->mad_hdr.method   != IB_MGMT_METHOD_TRAP_REPRESS)
+			return IB_MAD_RESULT_SUCCESS;
+
+		/*
+		 * Don't process SMInfo queries or vendor-specific
+		 * MADs -- the SMA can't handle them.
+		 */
+		if (in_mad->mad_hdr.attr_id == IB_SMP_ATTR_SM_INFO ||
+		    ((in_mad->mad_hdr.attr_id & IB_SMP_ATTR_VENDOR_MASK) ==
+		     IB_SMP_ATTR_VENDOR_MASK))
+			return IB_MAD_RESULT_SUCCESS;
+	} else if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_PERF_MGMT ||
+		   in_mad->mad_hdr.mgmt_class == MTHCA_VENDOR_CLASS1     ||
+		   in_mad->mad_hdr.mgmt_class == MTHCA_VENDOR_CLASS2) {
+		if (in_mad->mad_hdr.method  != IB_MGMT_METHOD_GET &&
+		    in_mad->mad_hdr.method  != IB_MGMT_METHOD_SET)
+			return IB_MAD_RESULT_SUCCESS;
+	} else
+		return IB_MAD_RESULT_SUCCESS;
+
+	err = mthca_MAD_IFC(to_mdev(ibdev),
+			    !!(mad_flags & IB_MAD_IGNORE_MKEY),
+			    port_num, in_mad, out_mad,
+			    &status);
+	if (err) {
+		mthca_err(to_mdev(ibdev), "MAD_IFC failed\n");
+		return IB_MAD_RESULT_FAILURE;
+	}
+	if (status == MTHCA_CMD_STAT_BAD_PKT)
+		return IB_MAD_RESULT_SUCCESS;
+	if (status) {
+		mthca_err(to_mdev(ibdev), "MAD_IFC returned status %02x\n",
+			  status);
+		return IB_MAD_RESULT_FAILURE;
+	}
+
+	if (!out_mad->mad_hdr.status)
+		smp_snoop(ibdev, port_num, in_mad);
+
+	/* set return bit in status of directed route responses */
+	if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
+		out_mad->mad_hdr.status |= cpu_to_be16(1 << 15);
+
+	if (in_mad->mad_hdr.method == IB_MGMT_METHOD_TRAP_REPRESS)
+		/* no response for trap repress */
+		return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
+
+	return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
+}
+
+static void send_handler(struct ib_mad_agent *agent,
+			 struct ib_mad_send_wc *mad_send_wc)
+{
+	struct mthca_trap_mad *tmad =
+		(void *) (unsigned long) mad_send_wc->wr_id;
+
+	dma_unmap_single(agent->device->dma_device,
+			 pci_unmap_addr(tmad, mapping),
+			 sizeof *tmad->mad,
+			 DMA_TO_DEVICE);
+	kfree(tmad->mad);
+	kfree(tmad);
+}
+
+int mthca_create_agents(struct mthca_dev *dev)
+{
+	struct ib_mad_agent *agent;
+	int p, q;
+
+	spin_lock_init(&dev->sm_lock);
+
+	for (p = 0; p < dev->limits.num_ports; ++p)
+		for (q = 0; q <= 1; ++q) {
+			agent = ib_register_mad_agent(&dev->ib_dev, p + 1,
+						      q ? IB_QPT_GSI : IB_QPT_SMI,
+						      NULL, 0, send_handler,
+						      NULL, NULL);
+			if (IS_ERR(agent))
+				goto err;
+			dev->send_agent[p][q] = agent;
+		}
+
+	return 0;
+
+err:
+	for (p = 0; p < dev->limits.num_ports; ++p)
+		for (q = 0; q <= 1; ++q)
+			if (dev->send_agent[p][q])
+				ib_unregister_mad_agent(dev->send_agent[p][q]);
+
+	return PTR_ERR(agent);
+}
+
+void mthca_free_agents(struct mthca_dev *dev)
+{
+	struct ib_mad_agent *agent;
+	int p, q;
+
+	for (p = 0; p < dev->limits.num_ports; ++p) {
+		for (q = 0; q <= 1; ++q) {
+			agent = dev->send_agent[p][q];
+			dev->send_agent[p][q] = NULL;
+			ib_unregister_mad_agent(agent);
+		}
+
+		if (dev->sm_ah[p])
+			ib_destroy_ah(dev->sm_ah[p]);
+	}
+}
#17 lore raw

PATCH v4 17/24 IPoIB IPv4 multicast

Add ip_ib_mc_map() to convert IPv4 multicast addresses to IPoIB hardware addresses. Also add <linux/if_infiniband.h> so INFINIBAND_ALEN has a home.

The mapping for multicast addresses is described in
http://www.ietf.org/internet-drafts/draft-ietf-ipoib-ip-over-infiniband-08.txt

include/linux/if_infiniband.h new file +29 -0
@@ -0,0 +1,29 @@
+/*
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available at
+ * <http://www.fsf.org/copyleft/gpl.html>, or the OpenIB.org BSD
+ * license, available in the LICENSE.TXT file accompanying this
+ * software.  These details are also available at
+ * <http://openib.org/license.html>.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * $Id$
+ */
+
+#ifndef _LINUX_IF_INFINIBAND_H
+#define _LINUX_IF_INFINIBAND_H
+
+#define INFINIBAND_ALEN		20	/* Octets in IPoIB HW addr	*/
+
+#endif /* _LINUX_IF_INFINIBAND_H */
include/net/ip.h +33 -0
@@ -229,6 +229,39 @@
 	buf[3]=addr&0x7F;
 }
 
+/*
+ *	Map a multicast IP onto multicast MAC for type IP-over-InfiniBand.
+ *	Leave P_Key as 0 to be filled in by driver.
+ */
+
+static inline void ip_ib_mc_map(u32 addr, char *buf)
+{
+	buf[0]  = 0;		/* Reserved */
+	buf[1]  = 0xff;		/* Multicast QPN */
+	buf[2]  = 0xff;
+	buf[3]  = 0xff;
+	addr    = ntohl(addr);
+	buf[4]  = 0xff;
+	buf[5]  = 0x12;		/* link local scope */
+	buf[6]  = 0x40;		/* IPv4 signature */
+	buf[7]  = 0x1b;
+	buf[8]  = 0;		/* P_Key */
+	buf[9]  = 0;
+	buf[10] = 0;
+	buf[11] = 0;
+	buf[12] = 0;
+	buf[13] = 0;
+	buf[14] = 0;
+	buf[15] = 0;
+	buf[19] = addr & 0xff;
+	addr  >>= 8;
+	buf[18] = addr & 0xff;
+	addr  >>= 8;
+	buf[17] = addr & 0xff;
+	addr  >>= 8;
+	buf[16] = addr & 0x0f;
+}
+
 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 #include <linux/ipv6.h>
 #endif
net/ipv4/arp.c +3 -0
@@ -213,6 +213,9 @@
 	case ARPHRD_IEEE802_TR:
 		ip_tr_mc_map(addr, haddr);
 		return 0;
+	case ARPHRD_INFINIBAND:
+		ip_ib_mc_map(addr, haddr);
+		return 0;
 	default:
 		if (dir) {
 			memcpy(haddr, dev->broadcast, dev->addr_len);
#18 lore raw

PATCH v4 18/24 IPoIB IPv6 support

Add ipv6_ib_mc_map() to convert IPv6 multicast addresses to IPoIB hardware addresses, and add support for autoconfiguration for devices with type ARPHRD_INFINIBAND.

The mapping for multicast addresses is described in
http://www.ietf.org/internet-drafts/draft-ietf-ipoib-ip-over-infiniband-08.txt

include/net/if_inet6.h +15 -0
@@ -266,5 +266,20 @@
 {
 	buf[0] = 0x00;
 }
+
+static inline void ipv6_ib_mc_map(struct in6_addr *addr, char *buf)
+{
+	buf[0]  = 0;		/* Reserved */
+	buf[1]  = 0xff;		/* Multicast QPN */
+	buf[2]  = 0xff;
+	buf[3]  = 0xff;
+	buf[4]  = 0xff;
+	buf[5]  = 0x12;		/* link local scope */
+	buf[6]  = 0x60;		/* IPv6 signature */
+	buf[7]  = 0x1b;
+	buf[8]  = 0;		/* P_Key */
+	buf[9]  = 0;
+	memcpy(buf + 10, addr->s6_addr + 6, 10);
+}
 #endif
 #endif
net/ipv6/addrconf.c +9 -1
@@ -48,6 +48,7 @@
 #include <linux/netdevice.h>
 #include <linux/if_arp.h>
 #include <linux/if_arcnet.h>
+#include <linux/if_infiniband.h>
 #include <linux/route.h>
 #include <linux/inetdevice.h>
 #include <linux/init.h>
@@ -1095,6 +1096,12 @@
 		memset(eui, 0, 7);
 		eui[7] = *(u8*)dev->dev_addr;
 		return 0;
+	case ARPHRD_INFINIBAND:
+		if (dev->addr_len != INFINIBAND_ALEN)
+			return -1;
+		memcpy(eui, dev->dev_addr + 12, 8);
+		eui[0] |= 2;
+		return 0;
 	}
 	return -1;
 }
@@ -1794,7 +1801,8 @@
 	if ((dev->type != ARPHRD_ETHER) && 
 	    (dev->type != ARPHRD_FDDI) &&
 	    (dev->type != ARPHRD_IEEE802_TR) &&
-	    (dev->type != ARPHRD_ARCNET)) {
+	    (dev->type != ARPHRD_ARCNET) &&
+	    (dev->type != ARPHRD_INFINIBAND)) {
 		/* Alas, we support only Ethernet autoconfiguration. */
 		return;
 	}
net/ipv6/ndisc.c +3 -0
@@ -260,6 +260,9 @@
 	case ARPHRD_ARCNET:
 		ipv6_arcnet_mc_map(addr, buf);
 		return 0;
+	case ARPHRD_INFINIBAND:
+		ipv6_ib_mc_map(addr, buf);
+		return 0;
 	default:
 		if (dir) {
 			memcpy(buf, dev->broadcast, dev->addr_len);
#19 lore raw

PATCH v4 19/24 Add IPoIB (IP-over-InfiniBand) driver

Add a driver that implements the (IPoIB) IP-over-InfiniBand protocol. This is a network device driver of type ARPHRD_INFINIBAND (and addr_len INFINIBAND_ALEN bytes).

The ARP/ND implementation for this driver is not completely straightforward, because InfiniBand requires an additional path lookup be performed (through an IB-specific mechanism) after a remote hardware address has been resolved. We are very open to suggestions of a better way to handle this than the current implementation.

Although IB has a special multicast group join mode intended to support IP multicast routing (non member join), no means to identify different multicast styles has yet been determined, so all joins by the driver are currently full member joins. We are looking for guidance in how to solve this.

The IPoIB protocol/encapsulation is described in the Internet-Drafts
http://www.ietf.org/internet-drafts/draft-ietf-ipoib-architecture-04.txt
http://www.ietf.org/internet-drafts/draft-ietf-ipoib-ip-over-infiniband-08.txt

drivers/infiniband/Kconfig +2 -0
@@ -9,4 +9,6 @@
 
 source "drivers/infiniband/hw/mthca/Kconfig"
 
+source "drivers/infiniband/ulp/ipoib/Kconfig"
+
 endmenu
drivers/infiniband/Makefile +1 -0
@@ -1,2 +1,3 @@
 obj-$(CONFIG_INFINIBAND)		+= core/
 obj-$(CONFIG_INFINIBAND_MTHCA)		+= hw/mthca/
+obj-$(CONFIG_INFINIBAND_IPOIB)		+= ulp/ipoib/
drivers/infiniband/ulp/ipoib/Kconfig new file +33 -0
@@ -0,0 +1,33 @@
+config INFINIBAND_IPOIB
+	tristate "IP-over-InfiniBand"
+	depends on INFINIBAND && NETDEVICES && INET
+	---help---
+	  Support for the IP-over-InfiniBand protocol (IPoIB). This
+	  transports IP packets over InfiniBand so you can use your IB
+	  device as a fancy NIC.
+
+	  The IPoIB protocol is defined by the IETF ipoib working
+	  group: <http://www.ietf.org/html.charters/ipoib-charter.html>.
+
+config INFINIBAND_IPOIB_DEBUG
+	bool "IP-over-InfiniBand debugging"
+	depends on INFINIBAND_IPOIB
+	---help---
+	  This option causes debugging code to be compiled into the
+	  IPoIB driver.  The output can be turned on via the
+	  debug_level and mcast_debug_level module parameters (which
+	  can also be set after the driver is loaded through sysfs).
+
+	  This option also creates an "ipoib_debugfs," which can be
+	  mounted to expose debugging information about IB multicast
+	  groups used by the IPoIB driver.
+
+config INFINIBAND_IPOIB_DEBUG_DATA
+	bool "IP-over-InfiniBand data path debugging"
+	depends on INFINIBAND_IPOIB_DEBUG
+	---help---
+	  This option compiles debugging code into the the data path
+	  of the IPoIB driver.  The output can be turned on via the
+	  data_debug_level module parameter; however, even with output
+	  turned off, this debugging code will have some performance
+	  impact.
drivers/infiniband/ulp/ipoib/Makefile new file +11 -0
@@ -0,0 +1,11 @@
+EXTRA_CFLAGS += -Idrivers/infiniband/include
+
+obj-$(CONFIG_INFINIBAND_IPOIB)			+= ib_ipoib.o
+
+ib_ipoib-y					:= ipoib_main.o \
+						   ipoib_ib.o \
+						   ipoib_multicast.o \
+						   ipoib_verbs.o \
+						   ipoib_vlan.o
+ib_ipoib-$(CONFIG_INFINIBAND_IPOIB_DEBUG)	+= ipoib_fs.o
+
drivers/infiniband/ulp/ipoib/ipoib.h new file +350 -0

350 lines not on this page · raw message

drivers/infiniband/ulp/ipoib/ipoib_fs.c new file +287 -0

287 lines not on this page · raw message

drivers/infiniband/ulp/ipoib/ipoib_ib.c new file +632 -0

632 lines not on this page · raw message

drivers/infiniband/ulp/ipoib/ipoib_main.c new file +1084 -0

1,084 lines not on this page · raw message

drivers/infiniband/ulp/ipoib/ipoib_verbs.c new file +254 -0

254 lines not on this page · raw message

#20 lore raw

PATCH v4 20/24 Add IPoIB multicast & partition code

Add functions for handling IPoIB multicast and multiple partitions.

drivers/infiniband/ulp/ipoib/ipoib_multicast.c new file +981 -0
@@ -0,0 +1,981 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: ipoib_multicast.c 1362 2004-12-18 15:56:29Z roland $
+ */
+
+#include <linux/skbuff.h>
+#include <linux/rtnetlink.h>
+#include <linux/ip.h>
+#include <linux/in.h>
+#include <linux/igmp.h>
+#include <linux/inetdevice.h>
+#include <linux/delay.h>
+#include <linux/completion.h>
+
+#include "ipoib.h"
+
+#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
+int mcast_debug_level;
+
+module_param(mcast_debug_level, int, 0644);
+MODULE_PARM_DESC(mcast_debug_level,
+		 "Enable multicast debug tracing if > 0");
+#endif
+
+static DECLARE_MUTEX(mcast_mutex);
+
+/* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */
+struct ipoib_mcast {
+	struct ib_sa_mcmember_rec mcmember;
+	struct ipoib_ah          *ah;
+
+	struct rb_node    rb_node;
+	struct list_head  list;
+	struct completion done;
+
+	int                 query_id;
+	struct ib_sa_query *query;
+
+	unsigned long created;
+	unsigned long backoff;
+
+	unsigned long flags;
+	unsigned char logcount;
+
+	struct list_head  neigh_list;
+
+	struct sk_buff_head pkt_queue;
+
+	struct net_device *dev;
+};
+
+struct ipoib_mcast_iter {
+	struct net_device *dev;
+	union ib_gid       mgid;
+	unsigned long      created;
+	unsigned int       queuelen;
+	unsigned int       complete;
+	unsigned int       send_only;
+};
+
+static void ipoib_mcast_free(struct ipoib_mcast *mcast)
+{
+	struct net_device *dev = mcast->dev;
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_neigh *neigh, *tmp;
+	unsigned long flags;
+
+	ipoib_dbg_mcast(netdev_priv(dev),
+			"deleting multicast group " IPOIB_GID_FMT "\n",
+			IPOIB_GID_ARG(mcast->mcmember.mgid));
+
+	spin_lock_irqsave(&priv->lock, flags);
+
+	list_for_each_entry_safe(neigh, tmp, &mcast->neigh_list, list) {
+		ipoib_put_ah(neigh->ah);
+		*to_ipoib_neigh(neigh->neighbour) = NULL;
+		neigh->neighbour->ops->destructor = NULL;
+		kfree(neigh);
+	}
+
+	spin_unlock_irqrestore(&priv->lock, flags);
+
+	if (mcast->ah)
+		ipoib_put_ah(mcast->ah);
+
+	while (!skb_queue_empty(&mcast->pkt_queue)) {
+		struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
+
+		skb->dev = dev;
+		dev_kfree_skb_any(skb);
+	}
+
+	kfree(mcast);
+}
+
+static struct ipoib_mcast *ipoib_mcast_alloc(struct net_device *dev,
+					     int can_sleep)
+{
+	struct ipoib_mcast *mcast;
+
+	mcast = kmalloc(sizeof (*mcast), can_sleep ? GFP_KERNEL : GFP_ATOMIC);
+	if (!mcast)
+		return NULL;
+
+	memset(mcast, 0, sizeof (*mcast));
+
+	init_completion(&mcast->done);
+
+	mcast->dev = dev;
+	mcast->created = jiffies;
+	mcast->backoff = HZ;
+	mcast->logcount = 0;
+
+	INIT_LIST_HEAD(&mcast->list);
+	INIT_LIST_HEAD(&mcast->neigh_list);
+	skb_queue_head_init(&mcast->pkt_queue);
+
+	mcast->ah    = NULL;
+	mcast->query = NULL;
+
+	return mcast;
+}
+
+static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, union ib_gid *mgid)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct rb_node *n = priv->multicast_tree.rb_node;
+
+	while (n) {
+		struct ipoib_mcast *mcast;
+		int ret;
+
+		mcast = rb_entry(n, struct ipoib_mcast, rb_node);
+
+		ret = memcmp(mgid->raw, mcast->mcmember.mgid.raw,
+			     sizeof (union ib_gid));
+		if (ret < 0)
+			n = n->rb_left;
+		else if (ret > 0)
+			n = n->rb_right;
+		else
+			return mcast;
+	}
+
+	return NULL;
+}
+
+static int __ipoib_mcast_add(struct net_device *dev, struct ipoib_mcast *mcast)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL;
+
+	while (*n) {
+		struct ipoib_mcast *tmcast;
+		int ret;
+
+		pn = *n;
+		tmcast = rb_entry(pn, struct ipoib_mcast, rb_node);
+
+		ret = memcmp(mcast->mcmember.mgid.raw, tmcast->mcmember.mgid.raw,
+			     sizeof (union ib_gid));
+		if (ret < 0)
+			n = &pn->rb_left;
+		else if (ret > 0)
+			n = &pn->rb_right;
+		else
+			return -EEXIST;
+	}
+
+	rb_link_node(&mcast->rb_node, pn, n);
+	rb_insert_color(&mcast->rb_node, &priv->multicast_tree);
+
+	return 0;
+}
+
+static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
+				   struct ib_sa_mcmember_rec *mcmember)
+{
+	struct net_device *dev = mcast->dev;
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	int ret;
+
+	mcast->mcmember = *mcmember;
+
+	/* Set the cached Q_Key before we attach if it's the broadcast group */
+	if (!memcmp(mcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
+		    sizeof (union ib_gid)))
+		priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey);
+
+	if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
+		if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
+			ipoib_warn(priv, "multicast group " IPOIB_GID_FMT
+				   " already attached\n",
+				   IPOIB_GID_ARG(mcast->mcmember.mgid));
+
+			return 0;
+		}
+
+		ret = ipoib_mcast_attach(dev, be16_to_cpu(mcast->mcmember.mlid),
+					 &mcast->mcmember.mgid);
+		if (ret < 0) {
+			ipoib_warn(priv, "couldn't attach QP to multicast group "
+				   IPOIB_GID_FMT "\n",
+				   IPOIB_GID_ARG(mcast->mcmember.mgid));
+
+			clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags);
+			return ret;
+		}
+	}
+
+	{
+		/*
+		 * For now we set static_rate to 0.  This is not
+		 * really correct: we should look at the rate
+		 * component of the MC member record, compare it with
+		 * the rate of our local port (calculated from the
+		 * active link speed and link width) and set an
+		 * inter-packet delay appropriately.
+		 */
+		struct ib_ah_attr av = {
+			.dlid	       = be16_to_cpu(mcast->mcmember.mlid),
+			.port_num      = priv->port,
+			.sl	       = mcast->mcmember.sl,
+			.static_rate   = 0,
+			.ah_flags      = IB_AH_GRH,
+			.grh	       = {
+				.flow_label    = be32_to_cpu(mcast->mcmember.flow_label),
+				.hop_limit     = mcast->mcmember.hop_limit,
+				.sgid_index    = 0,
+				.traffic_class = mcast->mcmember.traffic_class
+			}
+		};
+
+		av.grh.dgid = mcast->mcmember.mgid;
+
+		mcast->ah = ipoib_create_ah(dev, priv->pd, &av);
+		if (!mcast->ah) {
+			ipoib_warn(priv, "ib_address_create failed\n");
+		} else {
+			ipoib_dbg_mcast(priv, "MGID " IPOIB_GID_FMT
+					" AV %p, LID 0x%04x, SL %d\n",
+					IPOIB_GID_ARG(mcast->mcmember.mgid),
+					mcast->ah->ah,
+					be16_to_cpu(mcast->mcmember.mlid),
+					mcast->mcmember.sl);
+		}
+	}
+
+	/* actually send any queued packets */
+	while (!skb_queue_empty(&mcast->pkt_queue)) {
+		struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
+
+		skb->dev = dev;
+
+		if (!skb->dst || !skb->dst->neighbour) {
+			/* put pseudoheader back on for next time */
+			skb_push(skb, sizeof (struct ipoib_pseudoheader));
+		}
+
+		if (dev_queue_xmit(skb))
+			ipoib_warn(priv, "dev_queue_xmit failed to requeue packet\n");
+	}
+
+	return 0;
+}
+
+static void
+ipoib_mcast_sendonly_join_complete(int status,
+				   struct ib_sa_mcmember_rec *mcmember,
+				   void *mcast_ptr)
+{
+	struct ipoib_mcast *mcast = mcast_ptr;
+	struct net_device *dev = mcast->dev;
+
+	if (!status)
+		ipoib_mcast_join_finish(mcast, mcmember);
+	else {
+		if (mcast->logcount++ < 20)
+			ipoib_dbg_mcast(netdev_priv(dev), "multicast join failed for "
+					IPOIB_GID_FMT ", status %d\n",
+					IPOIB_GID_ARG(mcast->mcmember.mgid), status);
+
+		/* Flush out any queued packets */
+		while (!skb_queue_empty(&mcast->pkt_queue)) {
+			struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
+
+			skb->dev = dev;
+
+			dev_kfree_skb_any(skb);
+		}
+
+		/* Clear the busy flag so we try again */
+		clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
+	}
+
+	complete(&mcast->done);
+}
+
+static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast)
+{
+	struct net_device *dev = mcast->dev;
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ib_sa_mcmember_rec rec = {
+#if 0				/* Some SMs don't support send-only yet */
+		.join_state = 4
+#else
+		.join_state = 1
+#endif
+	};
+	int ret = 0;
+
+	if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
+		ipoib_dbg_mcast(priv, "device shutting down, no multicast joins\n");
+		return -ENODEV;
+	}
+
+	if (test_and_set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) {
+		ipoib_dbg_mcast(priv, "multicast entry busy, skipping\n");
+		return -EBUSY;
+	}
+
+	rec.mgid     = mcast->mcmember.mgid;
+	rec.port_gid = priv->local_gid;
+	rec.pkey     = be16_to_cpu(priv->pkey);
+
+	ret = ib_sa_mcmember_rec_set(priv->ca, priv->port, &rec,
+				     IB_SA_MCMEMBER_REC_MGID		|
+				     IB_SA_MCMEMBER_REC_PORT_GID	|
+				     IB_SA_MCMEMBER_REC_PKEY		|
+				     IB_SA_MCMEMBER_REC_JOIN_STATE,
+				     1000, GFP_ATOMIC,
+				     ipoib_mcast_sendonly_join_complete,
+				     mcast, &mcast->query);
+	if (ret < 0) {
+		ipoib_warn(priv, "ib_sa_mcmember_rec_set failed (ret = %d)\n",
+			   ret);
+	} else {
+		ipoib_dbg_mcast(priv, "no multicast record for " IPOIB_GID_FMT
+				", starting join\n",
+				IPOIB_GID_ARG(mcast->mcmember.mgid));
+
+		mcast->query_id = ret;
+	}
+
+	return ret;
+}
+
+static void ipoib_mcast_join_complete(int status,
+				      struct ib_sa_mcmember_rec *mcmember,
+				      void *mcast_ptr)
+{
+	struct ipoib_mcast *mcast = mcast_ptr;
+	struct net_device *dev = mcast->dev;
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+
+	ipoib_dbg_mcast(priv, "join completion for " IPOIB_GID_FMT
+			" (status %d)\n",
+			IPOIB_GID_ARG(mcast->mcmember.mgid), status);
+
+	if (!status && !ipoib_mcast_join_finish(mcast, mcmember)) {
+		mcast->backoff = HZ;
+		down(&mcast_mutex);
+		if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
+			queue_work(ipoib_workqueue, &priv->mcast_task);
+		up(&mcast_mutex);
+		complete(&mcast->done);
+		return;
+	}
+
+	if (status == -EINTR) {
+		complete(&mcast->done);
+		return;
+	}
+
+	if (status && mcast->logcount++ < 20) {
+		if (status == -ETIMEDOUT || status == -EINTR) {
+			ipoib_dbg_mcast(priv, "multicast join failed for " IPOIB_GID_FMT
+					", status %d\n",
+					IPOIB_GID_ARG(mcast->mcmember.mgid),
+					status);
+		} else {
+			ipoib_warn(priv, "multicast join failed for "
+				   IPOIB_GID_FMT ", status %d\n",
+				   IPOIB_GID_ARG(mcast->mcmember.mgid),
+				   status);
+		}
+	}
+
+	mcast->backoff *= 2;
+	if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
+		mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
+
+	mcast->query = NULL;
+
+	down(&mcast_mutex);
+	if (test_bit(IPOIB_MCAST_RUN, &priv->flags)) {
+		if (status == -ETIMEDOUT)
+			queue_work(ipoib_workqueue, &priv->mcast_task);
+		else
+			queue_delayed_work(ipoib_workqueue, &priv->mcast_task,
+					   mcast->backoff * HZ);
+	} else
+		complete(&mcast->done);
+	up(&mcast_mutex);
+
+	return;
+}
+
+static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast,
+			     int create)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ib_sa_mcmember_rec rec = {
+		.join_state = 1
+	};
+	ib_sa_comp_mask comp_mask;
+	int ret = 0;
+
+	ipoib_dbg_mcast(priv, "joining MGID " IPOIB_GID_FMT "\n",
+			IPOIB_GID_ARG(mcast->mcmember.mgid));
+
+	rec.mgid     = mcast->mcmember.mgid;
+	rec.port_gid = priv->local_gid;
+	rec.pkey     = be16_to_cpu(priv->pkey);
+
+	comp_mask =
+		IB_SA_MCMEMBER_REC_MGID		|
+		IB_SA_MCMEMBER_REC_PORT_GID	|
+		IB_SA_MCMEMBER_REC_PKEY		|
+		IB_SA_MCMEMBER_REC_JOIN_STATE;
+
+	if (create) {
+		comp_mask |=
+			IB_SA_MCMEMBER_REC_QKEY		|
+			IB_SA_MCMEMBER_REC_SL		|
+			IB_SA_MCMEMBER_REC_FLOW_LABEL	|
+			IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
+
+		rec.qkey	  = priv->broadcast->mcmember.qkey;
+		rec.sl		  = priv->broadcast->mcmember.sl;
+		rec.flow_label	  = priv->broadcast->mcmember.flow_label;
+		rec.traffic_class = priv->broadcast->mcmember.traffic_class;
+	}
+
+	ret = ib_sa_mcmember_rec_set(priv->ca, priv->port, &rec, comp_mask,
+				     mcast->backoff * 1000, GFP_ATOMIC,
+				     ipoib_mcast_join_complete,
+				     mcast, &mcast->query);
+
+	if (ret < 0) {
+		ipoib_warn(priv, "ib_sa_mcmember_rec_set failed, status %d\n", ret);
+
+		mcast->backoff *= 2;
+		if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
+			mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
+
+		down(&mcast_mutex);
+		if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
+			queue_delayed_work(ipoib_workqueue,
+					   &priv->mcast_task,
+					   mcast->backoff);
+		up(&mcast_mutex);
+	} else
+		mcast->query_id = ret;
+}
+
+void ipoib_mcast_join_task(void *dev_ptr)
+{
+	struct net_device *dev = dev_ptr;
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+
+	if (!test_bit(IPOIB_MCAST_RUN, &priv->flags))
+		return;
+
+	if (ib_query_gid(priv->ca, priv->port, 0, &priv->local_gid))
+		ipoib_warn(priv, "ib_gid_entry_get() failed\n");
+	else
+		memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid));
+
+	if (!priv->broadcast) {
+		priv->broadcast = ipoib_mcast_alloc(dev, 1);
+		if (!priv->broadcast) {
+			ipoib_warn(priv, "failed to allocate broadcast group\n");
+			down(&mcast_mutex);
+			if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
+				queue_delayed_work(ipoib_workqueue,
+						   &priv->mcast_task, HZ);
+			up(&mcast_mutex);
+			return;
+		}
+
+		memcpy(priv->broadcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
+		       sizeof (union ib_gid));
+
+		spin_lock_irq(&priv->lock);
+		__ipoib_mcast_add(dev, priv->broadcast);
+		spin_unlock_irq(&priv->lock);
+	}
+
+	if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
+		ipoib_mcast_join(dev, priv->broadcast, 0);
+		return;
+	}
+
+	while (1) {
+		struct ipoib_mcast *mcast = NULL;
+
+		spin_lock_irq(&priv->lock);
+		list_for_each_entry(mcast, &priv->multicast_list, list) {
+			if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)
+			    && !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)
+			    && !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
+				/* Found the next unjoined group */
+				break;
+			}
+		}
+		spin_unlock_irq(&priv->lock);
+
+		if (&mcast->list == &priv->multicast_list) {
+			/* All done */
+			break;
+		}
+
+		ipoib_mcast_join(dev, mcast, 1);
+		return;
+	}
+
+	{
+		struct ib_port_attr attr;
+
+		if (!ib_query_port(priv->ca, priv->port, &attr))
+			priv->local_lid = attr.lid;
+		else
+			ipoib_warn(priv, "ib_query_port failed\n");
+	}
+
+	priv->mcast_mtu = ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu) -
+		IPOIB_ENCAP_LEN;
+	dev->mtu = min(priv->mcast_mtu, priv->admin_mtu);
+
+	ipoib_dbg_mcast(priv, "successfully joined all multicast groups\n");
+
+	clear_bit(IPOIB_MCAST_RUN, &priv->flags);
+	netif_carrier_on(dev);
+}
+
+int ipoib_mcast_start_thread(struct net_device *dev)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+
+	ipoib_dbg_mcast(priv, "starting multicast thread\n");
+
+	down(&mcast_mutex);
+	if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags))
+		queue_work(ipoib_workqueue, &priv->mcast_task);
+	up(&mcast_mutex);
+
+	return 0;
+}
+
+int ipoib_mcast_stop_thread(struct net_device *dev)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_mcast *mcast;
+
+	ipoib_dbg_mcast(priv, "stopping multicast thread\n");
+
+	down(&mcast_mutex);
+	clear_bit(IPOIB_MCAST_RUN, &priv->flags);
+	cancel_delayed_work(&priv->mcast_task);
+	up(&mcast_mutex);
+
+	flush_workqueue(ipoib_workqueue);
+
+	if (priv->broadcast && priv->broadcast->query) {
+		ib_sa_cancel_query(priv->broadcast->query_id, priv->broadcast->query);
+		priv->broadcast->query = NULL;
+		ipoib_dbg_mcast(priv, "waiting for bcast\n");
+		wait_for_completion(&priv->broadcast->done);
+	}
+
+	list_for_each_entry(mcast, &priv->multicast_list, list) {
+		if (mcast->query) {
+			ib_sa_cancel_query(mcast->query_id, mcast->query);
+			mcast->query = NULL;
+			ipoib_dbg_mcast(priv, "waiting for MGID " IPOIB_GID_FMT "\n",
+					IPOIB_GID_ARG(mcast->mcmember.mgid));
+			wait_for_completion(&mcast->done);
+		}
+	}
+
+	return 0;
+}
+
+int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ib_sa_mcmember_rec rec = {
+		.join_state = 1
+	};
+	int ret = 0;
+
+	if (!test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags))
+		return 0;
+
+	ipoib_dbg_mcast(priv, "leaving MGID " IPOIB_GID_FMT "\n",
+			IPOIB_GID_ARG(mcast->mcmember.mgid));
+
+	rec.mgid     = mcast->mcmember.mgid;
+	rec.port_gid = priv->local_gid;
+	rec.pkey     = be16_to_cpu(priv->pkey);
+
+	/* Remove ourselves from the multicast group */
+	ret = ipoib_mcast_detach(dev, be16_to_cpu(mcast->mcmember.mlid),
+				 &mcast->mcmember.mgid);
+	if (ret)
+		ipoib_warn(priv, "ipoib_mcast_detach failed (result = %d)\n", ret);
+
+	/*
+	 * Just make one shot at leaving and don't wait for a reply;
+	 * if we fail, too bad.
+	 */
+	ret = ib_sa_mcmember_rec_delete(priv->ca, priv->port, &rec,
+					IB_SA_MCMEMBER_REC_MGID		|
+					IB_SA_MCMEMBER_REC_PORT_GID	|
+					IB_SA_MCMEMBER_REC_PKEY		|
+					IB_SA_MCMEMBER_REC_JOIN_STATE,
+					0, GFP_ATOMIC, NULL,
+					mcast, &mcast->query);
+	if (ret < 0)
+		ipoib_warn(priv, "ib_sa_mcmember_rec_delete failed "
+			   "for leave (result = %d)\n", ret);
+
+	return 0;
+}
+
+void ipoib_mcast_send(struct net_device *dev, union ib_gid *mgid,
+		      struct sk_buff *skb)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_mcast *mcast;
+
+	/*
+	 * We can only be called from ipoib_start_xmit, so we're
+	 * inside tx_lock -- no need to save/restore flags.
+	 */
+	spin_lock(&priv->lock);
+
+	mcast = __ipoib_mcast_find(dev, mgid);
+	if (!mcast) {
+		/* Let's create a new send only group now */
+		ipoib_dbg_mcast(priv, "setting up send only multicast group for "
+				IPOIB_GID_FMT "\n", IPOIB_GID_ARG(*mgid));
+
+		mcast = ipoib_mcast_alloc(dev, 0);
+		if (!mcast) {
+			ipoib_warn(priv, "unable to allocate memory for "
+				   "multicast structure\n");
+			dev_kfree_skb_any(skb);
+			goto out;
+		}
+
+		set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags);
+		mcast->mcmember.mgid = *mgid;
+		__ipoib_mcast_add(dev, mcast);
+		list_add_tail(&mcast->list, &priv->multicast_list);
+	}
+
+	if (!mcast->ah) {
+		if (skb_queue_len(&mcast->pkt_queue) < IPOIB_MAX_MCAST_QUEUE)
+			skb_queue_tail(&mcast->pkt_queue, skb);
+		else
+			dev_kfree_skb_any(skb);
+
+		if (mcast->query)
+			ipoib_dbg_mcast(priv, "no address vector, "
+					"but multicast join already started\n");
+		else if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
+			ipoib_mcast_sendonly_join(mcast);
+
+		/*
+		 * If lookup completes between here and out:, don't
+		 * want to send packet twice.
+		 */
+		mcast = NULL;
+	}
+
+out:
+	if (mcast && mcast->ah) {
+		if (skb->dst            &&
+		    skb->dst->neighbour &&
+		    !*to_ipoib_neigh(skb->dst->neighbour)) {
+			struct ipoib_neigh *neigh = kmalloc(sizeof *neigh, GFP_ATOMIC);
+
+			if (neigh) {
+				kref_get(&mcast->ah->ref);
+				neigh->ah  	= mcast->ah;
+				neigh->neighbour = skb->dst->neighbour;
+				*to_ipoib_neigh(skb->dst->neighbour) = neigh;
+				list_add_tail(&neigh->list, &mcast->neigh_list);
+			}
+		}
+
+		ipoib_send(dev, skb, mcast->ah, IB_MULTICAST_QPN);
+	}
+
+	spin_unlock(&priv->lock);
+}
+
+void ipoib_mcast_dev_flush(struct net_device *dev)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	LIST_HEAD(remove_list);
+	struct ipoib_mcast *mcast, *tmcast, *nmcast;
+	unsigned long flags;
+
+	ipoib_dbg_mcast(priv, "flushing multicast list\n");
+
+	spin_lock_irqsave(&priv->lock, flags);
+	list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
+		nmcast = ipoib_mcast_alloc(dev, 0);
+		if (nmcast) {
+			nmcast->flags =
+				mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY);
+
+			nmcast->mcmember.mgid = mcast->mcmember.mgid;
+
+			/* Add the new group in before the to-be-destroyed group */
+			list_add_tail(&nmcast->list, &mcast->list);
+			list_del_init(&mcast->list);
+
+			rb_replace_node(&mcast->rb_node, &nmcast->rb_node,
+					&priv->multicast_tree);
+
+			list_add_tail(&mcast->list, &remove_list);
+		} else {
+			ipoib_warn(priv, "could not reallocate multicast group "
+				   IPOIB_GID_FMT "\n",
+				   IPOIB_GID_ARG(mcast->mcmember.mgid));
+		}
+	}
+
+	if (priv->broadcast) {
+		nmcast = ipoib_mcast_alloc(dev, 0);
+		if (nmcast) {
+			nmcast->mcmember.mgid = priv->broadcast->mcmember.mgid;
+
+			rb_replace_node(&priv->broadcast->rb_node,
+					&nmcast->rb_node,
+					&priv->multicast_tree);
+
+			list_add_tail(&priv->broadcast->list, &remove_list);
+		}
+
+		priv->broadcast = nmcast;
+	}
+
+	spin_unlock_irqrestore(&priv->lock, flags);
+
+	list_for_each_entry(mcast, &remove_list, list) {
+		ipoib_mcast_leave(dev, mcast);
+		ipoib_mcast_free(mcast);
+	}
+}
+
+void ipoib_mcast_dev_down(struct net_device *dev)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	unsigned long flags;
+
+	/* Delete broadcast since it will be recreated */
+	if (priv->broadcast) {
+		ipoib_dbg_mcast(priv, "deleting broadcast group\n");
+
+		spin_lock_irqsave(&priv->lock, flags);
+		rb_erase(&priv->broadcast->rb_node, &priv->multicast_tree);
+		spin_unlock_irqrestore(&priv->lock, flags);
+		ipoib_mcast_leave(dev, priv->broadcast);
+		ipoib_mcast_free(priv->broadcast);
+		priv->broadcast = NULL;
+	}
+}
+
+void ipoib_mcast_restart_task(void *dev_ptr)
+{
+	struct net_device *dev = dev_ptr;
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct dev_mc_list *mclist;
+	struct ipoib_mcast *mcast, *tmcast;
+	LIST_HEAD(remove_list);
+	unsigned long flags;
+
+	ipoib_dbg_mcast(priv, "restarting multicast task\n");
+
+	ipoib_mcast_stop_thread(dev);
+
+	spin_lock_irqsave(&priv->lock, flags);
+
+	/*
+	 * Unfortunately, the networking core only gives us a list of all of
+	 * the multicast hardware addresses. We need to figure out which ones
+	 * are new and which ones have been removed
+	 */
+
+	/* Clear out the found flag */
+	list_for_each_entry(mcast, &priv->multicast_list, list)
+		clear_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
+
+	/* Mark all of the entries that are found or don't exist */
+	for (mclist = dev->mc_list; mclist; mclist = mclist->next) {
+		union ib_gid mgid;
+
+		memcpy(mgid.raw, mclist->dmi_addr + 4, sizeof mgid);
+
+		/* Add in the P_Key */
+		mgid.raw[4] = (priv->pkey >> 8) & 0xff;
+		mgid.raw[5] = priv->pkey & 0xff;
+
+		mcast = __ipoib_mcast_find(dev, &mgid);
+		if (!mcast || test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
+			struct ipoib_mcast *nmcast;
+
+			/* Not found or send-only group, let's add a new entry */
+			ipoib_dbg_mcast(priv, "adding multicast entry for mgid "
+					IPOIB_GID_FMT "\n", IPOIB_GID_ARG(mgid));
+
+			nmcast = ipoib_mcast_alloc(dev, 0);
+			if (!nmcast) {
+				ipoib_warn(priv, "unable to allocate memory for multicast structure\n");
+				continue;
+			}
+
+			set_bit(IPOIB_MCAST_FLAG_FOUND, &nmcast->flags);
+
+			nmcast->mcmember.mgid = mgid;
+
+			if (mcast) {
+				/* Destroy the send only entry */
+				list_del(&mcast->list);
+				list_add_tail(&mcast->list, &remove_list);
+
+				rb_replace_node(&mcast->rb_node,
+						&nmcast->rb_node,
+						&priv->multicast_tree);
+			} else
+				__ipoib_mcast_add(dev, nmcast);
+
+			list_add_tail(&nmcast->list, &priv->multicast_list);
+		}
+
+		if (mcast)
+			set_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
+	}
+
+	/* Remove all of the entries don't exist anymore */
+	list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
+		if (!test_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags) &&
+		    !test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
+			ipoib_dbg_mcast(priv, "deleting multicast group " IPOIB_GID_FMT "\n",
+					IPOIB_GID_ARG(mcast->mcmember.mgid));
+
+			rb_erase(&mcast->rb_node, &priv->multicast_tree);
+
+			/* Move to the remove list */
+			list_del(&mcast->list);
+			list_add_tail(&mcast->list, &remove_list);
+		}
+	}
+	spin_unlock_irqrestore(&priv->lock, flags);
+
+	/* We have to cancel outside of the spinlock */
+	list_for_each_entry(mcast, &remove_list, list) {
+		ipoib_mcast_leave(mcast->dev, mcast);
+		ipoib_mcast_free(mcast);
+	}
+
+	if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
+		ipoib_mcast_start_thread(dev);
+}
+
+struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev)
+{
+	struct ipoib_mcast_iter *iter;
+
+	iter = kmalloc(sizeof *iter, GFP_KERNEL);
+	if (!iter)
+		return NULL;
+
+	iter->dev = dev;
+	memset(iter->mgid.raw, 0, sizeof iter->mgid);
+
+	if (ipoib_mcast_iter_next(iter)) {
+		ipoib_mcast_iter_free(iter);
+		return NULL;
+	}
+
+	return iter;
+}
+
+void ipoib_mcast_iter_free(struct ipoib_mcast_iter *iter)
+{
+	kfree(iter);
+}
+
+int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
+	struct rb_node *n;
+	struct ipoib_mcast *mcast;
+	int ret = 1;
+
+	spin_lock_irq(&priv->lock);
+
+	n = rb_first(&priv->multicast_tree);
+
+	while (n) {
+		mcast = rb_entry(n, struct ipoib_mcast, rb_node);
+
+		if (memcmp(iter->mgid.raw, mcast->mcmember.mgid.raw,
+			   sizeof (union ib_gid)) < 0) {
+			iter->mgid      = mcast->mcmember.mgid;
+			iter->created   = mcast->created;
+			iter->queuelen  = skb_queue_len(&mcast->pkt_queue);
+			iter->complete  = !!mcast->ah;
+			iter->send_only = !!(mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY));
+
+			ret = 0;
+
+			break;
+		}
+
+		n = rb_next(n);
+	}
+
+	spin_unlock_irq(&priv->lock);
+
+	return ret;
+}
+
+void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
+			   union ib_gid *mgid,
+			   unsigned long *created,
+			   unsigned int *queuelen,
+			   unsigned int *complete,
+			   unsigned int *send_only)
+{
+	*mgid      = iter->mgid;
+	*created   = iter->created;
+	*queuelen  = iter->queuelen;
+	*complete  = iter->complete;
+	*send_only = iter->send_only;
+}
drivers/infiniband/ulp/ipoib/ipoib_vlan.c new file +177 -0

177 lines not on this page · raw message

#21 lore raw

PATCH v4 21/24 Add InfiniBand userspace MAD support

Add a driver that provides a character special device for each InfiniBand port. This device allows userspace to send and receive MADs via write() and read() (with some control operations implemented as ioctls).

All operations are 32/64 clean and have been tested with 32-bit userspace running on a ppc64 kernel.

drivers/infiniband/core/Makefile +3 -1
@@ -1,6 +1,6 @@
 EXTRA_CFLAGS += -Idrivers/infiniband/include
 
-obj-$(CONFIG_INFINIBAND) +=	ib_core.o ib_mad.o ib_sa.o
+obj-$(CONFIG_INFINIBAND) +=	ib_core.o ib_mad.o ib_sa.o ib_umad.o
 
 ib_core-y :=			packer.o ud_header.o verbs.o sysfs.o \
 				device.o fmr_pool.o cache.o
@@ -8,3 +8,5 @@
 ib_mad-y :=			mad.o smi.o agent.o
 
 ib_sa-y :=			sa_query.o
+
+ib_umad-y :=			user_mad.o
drivers/infiniband/core/user_mad.c new file +719 -0

719 lines not on this page · raw message

drivers/infiniband/include/ib_user_mad.h new file +123 -0

123 lines not on this page · raw message

#22 lore raw

PATCH v4 22/24 Document InfiniBand ioctl use

Add the 0x1b ioctl magic number used by ib_umad module to Documentation/ioctl-number.txt.

Documentation/ioctl-number.txt +1 -0
@@ -72,6 +72,7 @@
 0x09	all	linux/md.h
 0x12	all	linux/fs.h
 		linux/blkpg.h
+0x1b	all	InfiniBand Subsystem	<http://www.openib.org/>
 0x20	all	drivers/cdrom/cm206.h
 0x22	all	scsi/sg.h
 '#'	00-3F	IEEE 1394 Subsystem	Block for the entire subsystem
#23 lore raw

PATCH v4 23/24 Add InfiniBand Documentation files

Add files to Documentation/infiniband that describe the tree under /sys/class/infiniband, the IPoIB driver and the userspace MAD access driver.

Documentation/infiniband/ipoib.txt new file +56 -0
@@ -0,0 +1,56 @@
+IP OVER INFINIBAND
+
+  The ib_ipoib driver is an implementation of the IP over InfiniBand
+  protocol as specified by the latest Internet-Drafts issued by the
+  IETF ipoib working group.  It is a "native" implementation in the
+  sense of setting the interface type to ARPHRD_INFINIBAND and the
+  hardware address length to 20 (earlier proprietary implementations
+  masqueraded to the kernel as ethernet interfaces).
+
+Partitions and P_Keys
+
+  When the IPoIB driver is loaded, it creates one interface for each
+  port using the P_Key at index 0.  To create an interface with a
+  different P_Key, write the desired P_Key into the main interface's
+  /sys/class/net/<intf name>/create_child file.  For example:
+
+    echo 0x8001 > /sys/class/net/ib0/create_child
+
+  This will create an interface named ib0.8001 with P_Key 0x8001.  To
+  remove a subinterface, use the "delete_child" file:
+
+    echo 0x8001 > /sys/class/net/ib0/delete_child
+
+  The P_Key for any interface is given by the "pkey" file, and the
+  main interface for a subinterface is in "parent."
+
+Debugging Information
+
+  By compiling the IPoIB driver with CONFIG_INFINIBAND_IPOIB_DEBUG set
+  to 'y', tracing messages are compiled into the driver.  They are
+  turned on by setting the module parameters debug_level and
+  mcast_debug_level to 1.  These parameters can be controlled at
+  runtime through files in /sys/module/ib_ipoib/.
+
+  CONFIG_INFINIBAND_IPOIB_DEBUG also enables the "ipoib_debugfs"
+  virtual filesystem.  By mounting this filesystem, for example with
+
+    mkdir -p /ipoib_debugfs
+    mount -t ipoib_debugfs none /ipoib_debufs
+
+  it is possible to get statistics about multicast groups from the
+  files /ipoib_debugfs/ib0_mcg and so on.
+
+  The performance impact of this option is negligible, so it
+  is safe to enable this option with debug_level set to 0 for normal
+  operation.
+
+  CONFIG_INFINIBAND_IPOIB_DEBUG_DATA enables even more debug output in
+  the data path when data_debug_level is set to 1.  However, even with
+  the output disabled, enabling this configuration option will affect
+  performance, because it adds tests to the fast path.
+
+References
+
+  IETF IP over InfiniBand (ipoib) Working Group
+    http://ietf.org/html.charters/ipoib-charter.html
Documentation/infiniband/sysfs.txt new file +63 -0

63 lines not on this page · raw message

Documentation/infiniband/user_mad.txt new file +81 -0

81 lines not on this page · raw message

#24 lore raw

PATCH v4 24/24 InfiniBand MAINTAINERS entry

Add OpenIB maintainers information to MAINTAINERS.

MAINTAINERS +11 -0
@@ -1081,6 +1081,17 @@
 L:	[email protected]
 S:	Maintained
 
+INFINIBAND SUBSYSTEM
+P:	Roland Dreier
+M:	[email protected]
+P:	Sean Hefty
+M:	[email protected]
+P:	Hal Rosenstock
+M:	[email protected]
+L:	[email protected]
+W:	http://www.openib.org/
+S:	Supported
+
 INPUT (KEYBOARD, MOUSE, JOYSTICK) DRIVERS
 P:	Vojtech Pavlik
 M:	[email protected]
#25 lore raw

Re: [PATCH][v4][19/24] Add IPoIB (IP-over-InfiniBand) driver

In reply to Roland Dreier

In article <[email protected]> (at Sun, 19 Dec 2004 22:15:14 -0800), Roland Dreier <[email protected]> says:

drivers/infiniband/ulp/ipoib/ipoib.h line 70

Quoted 13 lines
drivers/infiniband/ulp/ipoib/ipoib.h new file +13 -0
+enum {
+	IPOIB_PACKET_SIZE         = 2048,
+	IPOIB_BUF_SIZE 		  = IPOIB_PACKET_SIZE + IB_GRH_BYTES,
+
+	IPOIB_ENCAP_LEN 	  = 4,
+
+	IPOIB_RX_RING_SIZE 	  = 128,
+	IPOIB_TX_RING_SIZE 	  = 64,
+
+	IPOIB_NUM_WC 		  = 4,
+
+	IPOIB_MAX_PATH_REC_QUEUE  = 3,
+	IPOIB_MAX_MCAST_QUEUE     = 3,

above entries does not seem to appropriate for enum (than #define).

drivers/infiniband/ulp/ipoib/ipoib.h line 78

Quoted 8 lines
drivers/infiniband/ulp/ipoib/ipoib.h new file +8 -0
+
+	IPOIB_FLAG_OPER_UP 	  = 0,
+	IPOIB_FLAG_ADMIN_UP 	  = 1,
+	IPOIB_PKEY_ASSIGNED 	  = 2,
+	IPOIB_PKEY_STOP 	  = 3,
+	IPOIB_FLAG_SUBINTERFACE   = 4,
+	IPOIB_MCAST_RUN 	  = 5,
+	IPOIB_STOP_REAPER         = 6,

this seems ok, but are "xxx_FLAG_xxx" entries really flags?

drivers/infiniband/ulp/ipoib/ipoib.h line 80

Quoted 1 line
drivers/infiniband/ulp/ipoib/ipoib.h new file +1 -0
+	IPOIB_MAX_BACKOFF_SECONDS = 16,

ditto, w/ first one.

drivers/infiniband/ulp/ipoib/ipoib.h line 85

Quoted 4 lines
drivers/infiniband/ulp/ipoib/ipoib.h new file +4 -0
+	IPOIB_MCAST_FLAG_FOUND 	  = 0,	/* used in set_multicast_list */
+	IPOIB_MCAST_FLAG_SENDONLY = 1,
+	IPOIB_MCAST_FLAG_BUSY 	  = 2,	/* joining or already joined */
+	IPOIB_MCAST_FLAG_ATTACHED = 3,

seems fine, but are these really flags?

--yoshfuji

#26 lore raw

Re: [PATCH][v4][19/24] Add IPoIB (IP-over-InfiniBand) driver

Quoted 13 lines · YOSHIFUJI Hideaki / 吉藤英明

Roland Dreier <[email protected]> says:

drivers/infiniband/ulp/ipoib/ipoib.h line 66

Quoted 9 lines
drivers/infiniband/ulp/ipoib/ipoib.h new file +9 -0
+enum {
+     IPOIB_PACKET_SIZE         = 2048,
+     IPOIB_BUF_SIZE            = IPOIB_PACKET_SIZE + IB_GRH_BYTES,
+
+     IPOIB_ENCAP_LEN           = 4,
+
+     IPOIB_RX_RING_SIZE        = 128,
+     IPOIB_TX_RING_SIZE        = 64,
+

above entries does not seem to appropriate for enum (than #define).

According to Documentation/CodingStyle, it actually is preferred like this. See also include/linux/ide.h for another example where this is done.

Arnd <><

#27 lore raw

Re: [PATCH][v4][19/24] Add IPoIB (IP-over-InfiniBand) driver

In article <[email protected]> (at Mon, 20 Dec 2004 13:14:35 +0100), Arnd Bergmann <[email protected]> says:

Quoted 15 lines · YOSHIFUJI Hideaki / 吉藤英明
Quoted 13 lines

Roland Dreier <[email protected]> says:

drivers/infiniband/ulp/ipoib/ipoib.h line 66

Quoted 9 lines
drivers/infiniband/ulp/ipoib/ipoib.h new file +9 -0
+enum {
+     IPOIB_PACKET_SIZE         = 2048,
+     IPOIB_BUF_SIZE            = IPOIB_PACKET_SIZE + IB_GRH_BYTES,
+
+     IPOIB_ENCAP_LEN           = 4,
+
+     IPOIB_RX_RING_SIZE        = 128,
+     IPOIB_TX_RING_SIZE        = 64,
+

above entries does not seem to appropriate for enum (than #define).

Quoted 2 lines

According to Documentation/CodingStyle, it actually is preferred like this. See also include/linux/ide.h for another example where this is done.

No, it is not the similar case.

--yoshfuji

#28 lore raw

Re: [PATCH][v4][19/24] Add IPoIB (IP-over-InfiniBand) driver

Quoted 4 lines · YOSHIFUJI Hideaki / 吉藤英明

In article <[email protected]> (at Mon, 20 Dec 2004 13:14:35 +0100), Arnd Bergmann <[email protected]> says:

Quoted 1 line

See also include/linux/ide.h for another example where this is done.

No, it is not the similar case.

Sorry, it was a typo on my side, I meant include/linux/ata.h.

Arnd <><

#29 lore raw

Re: [PATCH][v4][19/24] Add IPoIB (IP-over-InfiniBand) driver

In reply to YOSHIFUJI Hideaki / 吉藤英明

YOSHIFUJI> above entries does not seem to appropriate for enum YOSHIFUJI> (than #define).

As Arnd mentioned, I thought enum values were preferred to using the preprocessor. What's the advantage of converting to macros (which have no type, are invisible to the compiler, etc)?

Thanks,
Roland

#30 lore raw
In reply to Roland Dreier

In article <[email protected]> (at Sun, 19 Dec 2004 22:14:43 -0800), Roland Dreier <[email protected]> says:

Quoted 5 lines

The following series of patches is the latest version of the OpenIB InfiniBand drivers. We believe that this version is suitable for merging when 2.6.11 opens (or into -mm immediately), although of course we are willing to go through as many more iterations as required to fix any remaining issues.

Maybe, via the net queue. David?

--yoshfuji

maintainer
#31 lore raw
Quoted 9 lines · On Mon, 20 Dec 2004 15:38:45 +0900 (JST) YOSHIFUJI Hideaki / $B5HF#1QL@(B…

In article <[email protected]> (at Sun, 19 Dec 2004 22:14:43 -0800), Roland Dreier <[email protected]> says:

Quoted 5 lines

The following series of patches is the latest version of the OpenIB InfiniBand drivers. We believe that this version is suitable for merging when 2.6.11 opens (or into -mm immediately), although of course we are willing to go through as many more iterations as required to fix any remaining issues.

Maybe, via the net queue. David?

If Roland can resubmit his patch queue to me with the fixes folks have recommended to him, I can start merging this stuff in.

I leave for vacation Wednesday morning (PST time), so if it is submitted after that I'll get to it at the beginning of the new year.

#32 lore raw

David> If Roland can resubmit his patch queue to me with the fixes David> folks have recommended to him, I can start merging this David> stuff in.

Andrew has indicated to me that he has merged the whole InfiniBand patch into -mm now.

Dave, did you want to handle the entire merge of the whole IB stack, or just the net/ parts, which are pretty trivial and stand alone, since AF_INFINIBAND is already defined in the tree?

I'm happy to do whatever it takes to get IB merged as expeditiously as possible so Dave & Andrew, please let me know what seems easiest and best to you.

Thanks,
Roland

maintainer
#33 lore raw
Quoted 3 lines · On Mon, 27 Dec 2004 21:19:26 -0800 Roland Dreier <[email protected]> wrote:

Dave, did you want to handle the entire merge of the whole IB stack, or just the net/ parts, which are pretty trivial and stand alone, since AF_INFINIBAND is already defined in the tree?

Send it all over.

#34 lore raw

v5 cover Latest IB patch queue

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"David" == David S Miller <[email protected]> writes:

David> Send it all over.

OK, you asked for it... here's our latest tree, which should incorporate all the feedback I've seen.

(Individuals trimmed from CC list, since they probably don't want to get all 24 patches over again)

Thanks,
Roland Dreier

#35 lore raw

PATCH v5 1/24 Add core InfiniBand support (public headers)

Add public headers for core InfiniBand support. This can be thought of as a midlayer that provides an abstraction between low-level hardware drivers and upper level protocols (such as IP-over-InfiniBand).

drivers/infiniband/include/ib_cache.h new file +53 -0
@@ -0,0 +1,53 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: ib_cache.h 1349 2004-12-16 21:09:43Z roland $
+ */
+
+#ifndef _IB_CACHE_H
+#define _IB_CACHE_H
+
+#include <ib_verbs.h>
+
+int ib_cached_gid_get(struct ib_device    *device,
+		      u8                   port,
+		      int                  index,
+		      union ib_gid        *gid);
+int ib_cached_pkey_get(struct ib_device    *device_handle,
+		       u8                   port,
+		       int                  index,
+		       u16                 *pkey);
+int ib_cached_pkey_find(struct ib_device    *device,
+			u8                   port,
+			u16                  pkey,
+			u16                 *index);
+
+#endif /* _IB_CACHE_H */
drivers/infiniband/include/ib_fmr_pool.h new file +92 -0

92 lines not on this page · raw message

drivers/infiniband/include/ib_pack.h new file +245 -0

245 lines not on this page · raw message

drivers/infiniband/include/ib_verbs.h new file +1249 -0

1,249 lines not on this page · raw message

#36 lore raw

PATCH v5 2/24 Add core InfiniBand support

Add implementation of core InfiniBand support. This can be thought of as a midlayer that provides an abstraction between low-level hardware drivers and upper level protocols (such as IP-over-InfiniBand).

drivers/infiniband/Kconfig new file +10 -0
@@ -0,0 +1,10 @@
+menu "InfiniBand support"
+
+config INFINIBAND
+	tristate "InfiniBand support"
+	---help---
+	  Core support for InfiniBand (IB).  Make sure to also select
+	  any protocols you wish to use as well as drivers for your
+	  InfiniBand hardware.
+
+endmenu
drivers/infiniband/Makefile new file +1 -0
@@ -0,0 +1 @@
+obj-$(CONFIG_INFINIBAND)		+= core/
drivers/infiniband/core/Makefile new file +6 -0
@@ -0,0 +1,6 @@
+EXTRA_CFLAGS += -Idrivers/infiniband/include
+
+obj-$(CONFIG_INFINIBAND) +=	ib_core.o
+
+ib_core-y :=			packer.o ud_header.o verbs.o sysfs.o \
+				device.o fmr_pool.o cache.o
drivers/infiniband/core/cache.c new file +328 -0

328 lines not on this page · raw message

drivers/infiniband/core/core_priv.h new file +52 -0

52 lines not on this page · raw message

drivers/infiniband/core/device.c new file +614 -0

614 lines not on this page · raw message

drivers/infiniband/core/fmr_pool.c new file +507 -0

507 lines not on this page · raw message

drivers/infiniband/core/packer.c new file +201 -0

201 lines not on this page · raw message

drivers/infiniband/core/sysfs.c new file +725 -0

725 lines not on this page · raw message

drivers/infiniband/core/ud_header.c new file +365 -0

365 lines not on this page · raw message

drivers/infiniband/core/verbs.c new file +433 -0

191 lines not on this page · raw message

The rest of this diff is not shown. View the raw message.

Part of the diff is not shown here. View the raw message.

#37 lore raw

PATCH v5 3/24 Hook up drivers/infiniband

Add the appropriate lines to drivers/Kconfig and drivers/Makefile so that the kernel configuration and build systems know about drivers/infiniband.

drivers/Kconfig +2 -0
@@ -56,4 +56,6 @@
 
 source "drivers/mmc/Kconfig"
 
+source "drivers/infiniband/Kconfig"
+
 endmenu
drivers/Makefile +1 -0
@@ -59,5 +59,6 @@
 obj-$(CONFIG_EISA)		+= eisa/
 obj-$(CONFIG_CPU_FREQ)		+= cpufreq/
 obj-$(CONFIG_MMC)		+= mmc/
+obj-$(CONFIG_INFINIBAND)	+= infiniband/
 obj-y				+= firmware/
 obj-$(CONFIG_CRYPTO)		+= crypto/
#38 lore raw

PATCH v5 4/24 Add InfiniBand MAD (management datagram) support (public headers)

Add public headers for handling InfiniBand MADs (management datagrams), including sending and receiving MADs as well as passing MADs on to local agents.

drivers/infiniband/include/ib_mad.h new file +404 -0
@@ -0,0 +1,404 @@
+/*
+ * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
+ * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
+ * Copyright (c) 2004 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
+ * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: ib_mad.h 1389 2004-12-27 22:56:47Z roland $
+ */
+
+#if !defined( IB_MAD_H )
+#define IB_MAD_H
+
+#include <ib_verbs.h>
+
+/* Management base version */
+#define IB_MGMT_BASE_VERSION			1
+
+/* Management classes */
+#define IB_MGMT_CLASS_SUBN_LID_ROUTED		0x01
+#define IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE	0x81
+#define IB_MGMT_CLASS_SUBN_ADM			0x03
+#define IB_MGMT_CLASS_PERF_MGMT			0x04
+#define IB_MGMT_CLASS_BM			0x05
+#define IB_MGMT_CLASS_DEVICE_MGMT		0x06
+#define IB_MGMT_CLASS_CM			0x07
+#define IB_MGMT_CLASS_SNMP			0x08
+#define IB_MGMT_CLASS_VENDOR_RANGE2_START	0x30
+#define IB_MGMT_CLASS_VENDOR_RANGE2_END		0x4F
+
+/* Management methods */
+#define IB_MGMT_METHOD_GET			0x01
+#define IB_MGMT_METHOD_SET			0x02
+#define IB_MGMT_METHOD_GET_RESP			0x81
+#define IB_MGMT_METHOD_SEND			0x03
+#define IB_MGMT_METHOD_TRAP			0x05
+#define IB_MGMT_METHOD_REPORT			0x06
+#define IB_MGMT_METHOD_REPORT_RESP		0x86
+#define IB_MGMT_METHOD_TRAP_REPRESS		0x07
+
+#define IB_MGMT_METHOD_RESP			0x80
+
+#define IB_MGMT_MAX_METHODS			128
+
+#define IB_QP0		0
+#define IB_QP1		__constant_htonl(1)
+#define IB_QP1_QKEY	0x80010000
+
+struct ib_grh {
+	u32		version_tclass_flow;
+	u16		paylen;
+	u8		next_hdr;
+	u8		hop_limit;
+	union ib_gid	sgid;
+	union ib_gid	dgid;
+} __attribute__ ((packed));
+
+struct ib_mad_hdr {
+	u8	base_version;
+	u8	mgmt_class;
+	u8	class_version;
+	u8	method;
+	u16	status;
+	u16	class_specific;
+	u64	tid;
+	u16	attr_id;
+	u16	resv;
+	u32	attr_mod;
+} __attribute__ ((packed));
+
+struct ib_rmpp_hdr {
+	u8	rmpp_version;
+	u8	rmpp_type;
+	u8	rmpp_rtime_flags;
+	u8	rmpp_status;
+	u32	seg_num;
+	u32	paylen_newwin;
+} __attribute__ ((packed));
+
+struct ib_mad {
+	struct ib_mad_hdr	mad_hdr;
+	u8			data[232];
+} __attribute__ ((packed));
+
+struct ib_rmpp_mad {
+	struct ib_mad_hdr	mad_hdr;
+	struct ib_rmpp_hdr	rmpp_hdr;
+	u8			data[220];
+} __attribute__ ((packed));
+
+struct ib_vendor_mad {
+	struct ib_mad_hdr	mad_hdr;
+	struct ib_rmpp_hdr	rmpp_hdr;
+	u8			reserved;
+	u8			oui[3];
+	u8			data[216];
+} __attribute__ ((packed));
+
+struct ib_mad_agent;
+struct ib_mad_send_wc;
+struct ib_mad_recv_wc;
+
+/**
+ * ib_mad_send_handler - callback handler for a sent MAD.
+ * @mad_agent: MAD agent that sent the MAD.
+ * @mad_send_wc: Send work completion information on the sent MAD.
+ */
+typedef void (*ib_mad_send_handler)(struct ib_mad_agent *mad_agent,
+				    struct ib_mad_send_wc *mad_send_wc);
+
+/**
+ * ib_mad_snoop_handler - Callback handler for snooping sent MADs.
+ * @mad_agent: MAD agent that snooped the MAD.
+ * @send_wr: Work request information on the sent MAD.
+ * @mad_send_wc: Work completion information on the sent MAD.  Valid
+ *   only for snooping that occurs on a send completion.
+ *
+ * Clients snooping MADs should not modify data referenced by the @send_wr
+ * or @mad_send_wc.
+ */
+typedef void (*ib_mad_snoop_handler)(struct ib_mad_agent *mad_agent,
+				     struct ib_send_wr *send_wr,
+				     struct ib_mad_send_wc *mad_send_wc);
+
+/**
+ * ib_mad_recv_handler - callback handler for a received MAD.
+ * @mad_agent: MAD agent requesting the received MAD.
+ * @mad_recv_wc: Received work completion information on the received MAD.
+ *
+ * MADs received in response to a send request operation will be handed to
+ * the user after the send operation completes.  All data buffers given
+ * to registered agents through this routine are owned by the receiving
+ * client, except for snooping agents.  Clients snooping MADs should not
+ * modify the data referenced by @mad_recv_wc.
+ */
+typedef void (*ib_mad_recv_handler)(struct ib_mad_agent *mad_agent,
+				    struct ib_mad_recv_wc *mad_recv_wc);
+
+/**
+ * ib_mad_agent - Used to track MAD registration with the access layer.
+ * @device: Reference to device registration is on.
+ * @qp: Reference to QP used for sending and receiving MADs.
+ * @recv_handler: Callback handler for a received MAD.
+ * @send_handler: Callback handler for a sent MAD.
+ * @snoop_handler: Callback handler for snooped sent MADs.
+ * @context: User-specified context associated with this registration.
+ * @hi_tid: Access layer assigned transaction ID for this client.
+ *   Unsolicited MADs sent by this client will have the upper 32-bits
+ *   of their TID set to this value.
+ * @port_num: Port number on which QP is registered
+ */
+struct ib_mad_agent {
+	struct ib_device	*device;
+	struct ib_qp		*qp;
+	ib_mad_recv_handler	recv_handler;
+	ib_mad_send_handler	send_handler;
+	ib_mad_snoop_handler	snoop_handler;
+	void			*context;
+	u32			hi_tid;
+	u8			port_num;
+};
+
+/**
+ * ib_mad_send_wc - MAD send completion information.
+ * @wr_id: Work request identifier associated with the send MAD request.
+ * @status: Completion status.
+ * @vendor_err: Optional vendor error information returned with a failed
+ *   request.
+ */
+struct ib_mad_send_wc {
+	u64			wr_id;
+	enum ib_wc_status	status;
+	u32			vendor_err;
+};
+
+/**
+ * ib_mad_recv_buf - received MAD buffer information.
+ * @list: Reference to next data buffer for a received RMPP MAD.
+ * @grh: References a data buffer containing the global route header.
+ *   The data refereced by this buffer is only valid if the GRH is
+ *   valid.
+ * @mad: References the start of the received MAD.
+ */
+struct ib_mad_recv_buf {
+	struct list_head	list;
+	struct ib_grh		*grh;
+	struct ib_mad		*mad;
+};
+
+/**
+ * ib_mad_recv_wc - received MAD information.
+ * @wc: Completion information for the received data.
+ * @recv_buf: Specifies the location of the received data buffer(s).
+ * @mad_len: The length of the received MAD, without duplicated headers.
+ *
+ * For received response, the wr_id field of the wc is set to the wr_id
+ *   for the corresponding send request.
+ */
+struct ib_mad_recv_wc {
+	struct ib_wc		*wc;
+	struct ib_mad_recv_buf	recv_buf;
+	int			mad_len;
+};
+
+/**
+ * ib_mad_reg_req - MAD registration request
+ * @mgmt_class: Indicates which management class of MADs should be receive
+ *   by the caller.  This field is only required if the user wishes to
+ *   receive unsolicited MADs, otherwise it should be 0.
+ * @mgmt_class_version: Indicates which version of MADs for the given
+ *   management class to receive.
+ * @oui: Indicates IEEE OUI when mgmt_class is a vendor class
+ *   in the range from 0x30 to 0x4f. Otherwise not used.
+ * @method_mask: The caller will receive unsolicited MADs for any method
+ *   where @method_mask = 1.
+ */
+struct ib_mad_reg_req {
+	u8	mgmt_class;
+	u8	mgmt_class_version;
+	u8	oui[3];
+	DECLARE_BITMAP(method_mask, IB_MGMT_MAX_METHODS);
+};
+
+/**
+ * ib_register_mad_agent - Register to send/receive MADs.
+ * @device: The device to register with.
+ * @port_num: The port on the specified device to use.
+ * @qp_type: Specifies which QP to access.  Must be either
+ *   IB_QPT_SMI or IB_QPT_GSI.
+ * @mad_reg_req: Specifies which unsolicited MADs should be received
+ *   by the caller.  This parameter may be NULL if the caller only
+ *   wishes to receive solicited responses.
+ * @rmpp_version: If set, indicates that the client will send
+ *   and receive MADs that contain the RMPP header for the given version.
+ *   If set to 0, indicates that RMPP is not used by this client.
+ * @send_handler: The completion callback routine invoked after a send
+ *   request has completed.
+ * @recv_handler: The completion callback routine invoked for a received
+ *   MAD.
+ * @context: User specified context associated with the registration.
+ */
+struct ib_mad_agent *ib_register_mad_agent(struct ib_device *device,
+					   u8 port_num,
+					   enum ib_qp_type qp_type,
+					   struct ib_mad_reg_req *mad_reg_req,
+					   u8 rmpp_version,
+					   ib_mad_send_handler send_handler,
+					   ib_mad_recv_handler recv_handler,
+					   void *context);
+
+enum ib_mad_snoop_flags {
+	/*IB_MAD_SNOOP_POSTED_SENDS	   = 1,*/
+	/*IB_MAD_SNOOP_RMPP_SENDS	   = (1<<1),*/
+	IB_MAD_SNOOP_SEND_COMPLETIONS	   = (1<<2),
+	/*IB_MAD_SNOOP_RMPP_SEND_COMPLETIONS = (1<<3),*/
+	IB_MAD_SNOOP_RECVS		   = (1<<4)
+	/*IB_MAD_SNOOP_RMPP_RECVS	   = (1<<5),*/
+	/*IB_MAD_SNOOP_REDIRECTED_QPS	   = (1<<6)*/
+};
+
+/**
+ * ib_register_mad_snoop - Register to snoop sent and received MADs.
+ * @device: The device to register with.
+ * @port_num: The port on the specified device to use.
+ * @qp_type: Specifies which QP traffic to snoop.  Must be either
+ *   IB_QPT_SMI or IB_QPT_GSI.
+ * @mad_snoop_flags: Specifies information where snooping occurs.
+ * @send_handler: The callback routine invoked for a snooped send.
+ * @recv_handler: The callback routine invoked for a snooped receive.
+ * @context: User specified context associated with the registration.
+ */
+struct ib_mad_agent *ib_register_mad_snoop(struct ib_device *device,
+					   u8 port_num,
+					   enum ib_qp_type qp_type,
+					   int mad_snoop_flags,
+					   ib_mad_snoop_handler snoop_handler,
+					   ib_mad_recv_handler recv_handler,
+					   void *context);
+
+/**
+ * ib_unregister_mad_agent - Unregisters a client from using MAD services.
+ * @mad_agent: Corresponding MAD registration request to deregister.
+ *
+ * After invoking this routine, MAD services are no longer usable by the
+ * client on the associated QP.
+ */
+int ib_unregister_mad_agent(struct ib_mad_agent *mad_agent);
+
+/**
+ * ib_post_send_mad - Posts MAD(s) to the send queue of the QP associated
+ *   with the registered client.
+ * @mad_agent: Specifies the associated registration to post the send to.
+ * @send_wr: Specifies the information needed to send the MAD(s).
+ * @bad_send_wr: Specifies the MAD on which an error was encountered.
+ *
+ * Sent MADs are not guaranteed to complete in the order that they were posted.
+ */
+int ib_post_send_mad(struct ib_mad_agent *mad_agent,
+		     struct ib_send_wr *send_wr,
+		     struct ib_send_wr **bad_send_wr);
+
+/**
+ * ib_coalesce_recv_mad - Coalesces received MAD data into a single buffer.
+ * @mad_recv_wc: Work completion information for a received MAD.
+ * @buf: User-provided data buffer to receive the coalesced buffers.  The
+ *   referenced buffer should be at least the size of the mad_len specified
+ *   by @mad_recv_wc.
+ *
+ * This call copies a chain of received RMPP MADs into a single data buffer,
+ * removing duplicated headers.
+ */
+void ib_coalesce_recv_mad(struct ib_mad_recv_wc *mad_recv_wc,
+			  void *buf);
+
+/**
+ * ib_free_recv_mad - Returns data buffers used to receive a MAD to the
+ *   access layer.
+ * @mad_recv_wc: Work completion information for a received MAD.
+ *
+ * Clients receiving MADs through their ib_mad_recv_handler must call this
+ * routine to return the work completion buffers to the access layer.
+ */
+void ib_free_recv_mad(struct ib_mad_recv_wc *mad_recv_wc);
+
+/**
+ * ib_cancel_mad - Cancels an outstanding send MAD operation.
+ * @mad_agent: Specifies the registration associated with sent MAD.
+ * @wr_id: Indicates the work request identifier of the MAD to cancel.
+ *
+ * MADs will be returned to the user through the corresponding
+ * ib_mad_send_handler.
+ */
+void ib_cancel_mad(struct ib_mad_agent *mad_agent,
+		   u64 wr_id);
+
+/**
+ * ib_redirect_mad_qp - Registers a QP for MAD services.
+ * @qp: Reference to a QP that requires MAD services.
+ * @rmpp_version: If set, indicates that the client will send
+ *   and receive MADs that contain the RMPP header for the given version.
+ *   If set to 0, indicates that RMPP is not used by this client.
+ * @send_handler: The completion callback routine invoked after a send
+ *   request has completed.
+ * @recv_handler: The completion callback routine invoked for a received
+ *   MAD.
+ * @context: User specified context associated with the registration.
+ *
+ * Use of this call allows clients to use MAD services, such as RMPP,
+ * on user-owned QPs.  After calling this routine, users may send
+ * MADs on the specified QP by calling ib_mad_post_send.
+ */
+struct ib_mad_agent *ib_redirect_mad_qp(struct ib_qp *qp,
+					u8 rmpp_version,
+					ib_mad_send_handler send_handler,
+					ib_mad_recv_handler recv_handler,
+					void *context);
+
+/**
+ * ib_process_mad_wc - Processes a work completion associated with a
+ *   MAD sent or received on a redirected QP.
+ * @mad_agent: Specifies the registered MAD service using the redirected QP.
+ * @wc: References a work completion associated with a sent or received
+ *   MAD segment.
+ *
+ * This routine is used to complete or continue processing on a MAD request.
+ * If the work completion is associated with a send operation, calling
+ * this routine is required to continue an RMPP transfer or to wait for a
+ * corresponding response, if it is a request.  If the work completion is
+ * associated with a receive operation, calling this routine is required to
+ * process an inbound or outbound RMPP transfer, or to match a response MAD
+ * with its corresponding request.
+ */
+int ib_process_mad_wc(struct ib_mad_agent *mad_agent,
+		      struct ib_wc *wc);
+
+#endif /* IB_MAD_H */
drivers/infiniband/include/ib_smi.h new file +96 -0

96 lines not on this page · raw message

#39 lore raw

PATCH v5 5/24 Add InfiniBand MAD (management datagram) support

Add support for handling InfiniBand MADs (management datagrams), including sending and receiving MADs as well as passing MADs on to local agents.

This is required for an SM (subnet manager) to discover and configure the host, since the SM's query MADs must be passed to the local SMA (subnet management agent). In addition, this support is used by upper level protocols to send queries to and receive responses from the SM.

drivers/infiniband/core/Makefile +3 -1
@@ -1,6 +1,8 @@
 EXTRA_CFLAGS += -Idrivers/infiniband/include
 
-obj-$(CONFIG_INFINIBAND) +=	ib_core.o
+obj-$(CONFIG_INFINIBAND) +=	ib_core.o ib_mad.o
 
 ib_core-y :=			packer.o ud_header.o verbs.o sysfs.o \
 				device.o fmr_pool.o cache.o
+
+ib_mad-y :=			mad.o smi.o agent.o
drivers/infiniband/core/agent.c new file +399 -0

399 lines not on this page · raw message

drivers/infiniband/core/mad.c new file +2632 -0

2,592 lines not on this page · raw message

The rest of this diff is not shown. View the raw message.

Part of the diff is not shown here. View the raw message.

#40 lore raw

PATCH v5 6/24 Add InfiniBand MAD (management datagram) support (private headers)

Add MAD layer private implementation headers.

drivers/infiniband/core/agent.h new file +55 -0
@@ -0,0 +1,55 @@
+/*
+ * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
+ * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
+ * Copyright (c) 2004 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
+ * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: agent.h 1389 2004-12-27 22:56:47Z roland $
+ */
+
+#ifndef __AGENT_H_
+#define __AGENT_H_
+
+extern spinlock_t ib_agent_port_list_lock;
+
+extern int ib_agent_port_open(struct ib_device *device,
+			      int port_num);
+
+extern int ib_agent_port_close(struct ib_device *device, int port_num);
+
+extern int agent_send(struct ib_mad_private *mad,
+		      struct ib_grh *grh,
+		      struct ib_wc *wc,
+		      struct ib_device *device,
+		      int port_num);
+
+#endif	/* __AGENT_H_ */
drivers/infiniband/core/agent_priv.h new file +64 -0

64 lines not on this page · raw message

drivers/infiniband/core/mad_priv.h new file +194 -0

194 lines not on this page · raw message

#41 lore raw

PATCH v5 7/24 Add InfiniBand MAD SMI support

Add MAD layer SMI (Subnet Management Interface) code.

drivers/infiniband/core/smi.c new file +234 -0
@@ -0,0 +1,234 @@
+/*
+ * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
+ * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
+ * Copyright (c) 2004 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
+ * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: smi.c 1389 2004-12-27 22:56:47Z roland $
+ */
+
+#include <ib_smi.h>
+
+
+/*
+ * Fixup a directed route SMP for sending
+ * Return 0 if the SMP should be discarded
+ */
+int smi_handle_dr_smp_send(struct ib_smp *smp,
+			   u8 node_type,
+			   int port_num)
+{
+	u8 hop_ptr, hop_cnt;
+
+	hop_ptr = smp->hop_ptr;
+	hop_cnt = smp->hop_cnt;
+
+	/* See section 14.2.2.2, Vol 1 IB spec */
+	if (!ib_get_smp_direction(smp)) {
+		/* C14-9:1 */
+		if (hop_cnt && hop_ptr == 0) {
+			smp->hop_ptr++;
+			return (smp->initial_path[smp->hop_ptr] ==
+				port_num);
+		}
+
+		/* C14-9:2 */
+		if (hop_ptr && hop_ptr < hop_cnt) {
+			if (node_type != IB_NODE_SWITCH)
+				return 0;
+
+			/* smp->return_path set when received */
+			smp->hop_ptr++;
+			return (smp->initial_path[smp->hop_ptr] ==
+				port_num);
+		}
+
+		/* C14-9:3 -- We're at the end of the DR segment of path */
+		if (hop_ptr == hop_cnt) {
+			/* smp->return_path set when received */
+			smp->hop_ptr++;
+			return (node_type == IB_NODE_SWITCH ||
+				smp->dr_dlid == IB_LID_PERMISSIVE);
+		}
+
+		/* C14-9:4 -- hop_ptr = hop_cnt + 1 -> give to SMA/SM */
+		/* C14-9:5 -- Fail unreasonable hop pointer */
+		return (hop_ptr == hop_cnt + 1);
+
+	} else {
+		/* C14-13:1 */
+		if (hop_cnt && hop_ptr == hop_cnt + 1) {
+			smp->hop_ptr--;
+			return (smp->return_path[smp->hop_ptr] ==
+				port_num);
+		}
+
+		/* C14-13:2 */
+		if (2 <= hop_ptr && hop_ptr <= hop_cnt) {
+			if (node_type != IB_NODE_SWITCH)
+				return 0;
+
+			smp->hop_ptr--;
+			return (smp->return_path[smp->hop_ptr] ==
+				port_num);
+		}
+
+		/* C14-13:3 -- at the end of the DR segment of path */
+		if (hop_ptr == 1) {
+			smp->hop_ptr--;
+			/* C14-13:3 -- SMPs destined for SM shouldn't be here */
+			return (node_type == IB_NODE_SWITCH ||
+				smp->dr_slid == IB_LID_PERMISSIVE);
+		}
+
+		/* C14-13:4 -- hop_ptr = 0 -> should have gone to SM */
+		if (hop_ptr == 0)
+			return 1;
+
+		/* C14-13:5 -- Check for unreasonable hop pointer */
+		return 0;
+	}
+}
+
+/*
+ * Adjust information for a received SMP
+ * Return 0 if the SMP should be dropped
+ */
+int smi_handle_dr_smp_recv(struct ib_smp *smp,
+			   u8 node_type,
+			   int port_num,
+			   int phys_port_cnt)
+{
+	u8 hop_ptr, hop_cnt;
+
+	hop_ptr = smp->hop_ptr;
+	hop_cnt = smp->hop_cnt;
+
+	/* See section 14.2.2.2, Vol 1 IB spec */
+	if (!ib_get_smp_direction(smp)) {
+		/* C14-9:1 -- sender should have incremented hop_ptr */
+		if (hop_cnt && hop_ptr == 0)
+			return 0;
+
+		/* C14-9:2 -- intermediate hop */
+		if (hop_ptr && hop_ptr < hop_cnt) {
+			if (node_type != IB_NODE_SWITCH)
+				return 0;
+
+			smp->return_path[hop_ptr] = port_num;
+			/* smp->hop_ptr updated when sending */
+			return (smp->initial_path[hop_ptr+1] <= phys_port_cnt);
+		}
+
+		/* C14-9:3 -- We're at the end of the DR segment of path */
+		if (hop_ptr == hop_cnt) {
+			if (hop_cnt)
+				smp->return_path[hop_ptr] = port_num;
+			/* smp->hop_ptr updated when sending */
+
+			return (node_type == IB_NODE_SWITCH ||
+				smp->dr_dlid == IB_LID_PERMISSIVE);
+		}
+
+		/* C14-9:4 -- hop_ptr = hop_cnt + 1 -> give to SMA/SM */
+		/* C14-9:5 -- fail unreasonable hop pointer */
+		return (hop_ptr == hop_cnt + 1);
+
+	} else {
+
+		/* C14-13:1 */
+		if (hop_cnt && hop_ptr == hop_cnt + 1) {
+			smp->hop_ptr--;
+			return (smp->return_path[smp->hop_ptr] ==
+				port_num);
+		}
+
+		/* C14-13:2 */
+		if (2 <= hop_ptr && hop_ptr <= hop_cnt) {
+			if (node_type != IB_NODE_SWITCH)
+				return 0;
+
+			/* smp->hop_ptr updated when sending */
+			return (smp->return_path[hop_ptr-1] <= phys_port_cnt);
+		}
+
+		/* C14-13:3 -- We're at the end of the DR segment of path */
+		if (hop_ptr == 1) {
+			if (smp->dr_slid == IB_LID_PERMISSIVE) {
+				/* giving SMP to SM - update hop_ptr */
+				smp->hop_ptr--;
+				return 1;
+			}
+			/* smp->hop_ptr updated when sending */
+			return (node_type == IB_NODE_SWITCH);
+		}
+
+		/* C14-13:4 -- hop_ptr = 0 -> give to SM */
+		/* C14-13:5 -- Check for unreasonable hop pointer */
+		return (hop_ptr == 0);
+	}
+}
+
+/*
+ * Return 1 if the received DR SMP should be forwarded to the send queue
+ * Return 0 if the SMP should be completed up the stack
+ */
+int smi_check_forward_dr_smp(struct ib_smp *smp)
+{
+	u8 hop_ptr, hop_cnt;
+
+	hop_ptr = smp->hop_ptr;
+	hop_cnt = smp->hop_cnt;
+
+	if (!ib_get_smp_direction(smp)) {
+		/* C14-9:2 -- intermediate hop */
+		if (hop_ptr && hop_ptr < hop_cnt)
+			return 1;
+
+		/* C14-9:3 -- at the end of the DR segment of path */
+		if (hop_ptr == hop_cnt)
+			return (smp->dr_dlid == IB_LID_PERMISSIVE);
+
+		/* C14-9:4 -- hop_ptr = hop_cnt + 1 -> give to SMA/SM */
+		if (hop_ptr == hop_cnt + 1)
+			return 1;
+	} else {
+		/* C14-13:2 */
+		if (2 <= hop_ptr && hop_ptr <= hop_cnt)
+			return 1;
+
+		/* C14-13:3 -- at the end of the DR segment of path */
+		if (hop_ptr == 1)
+			return (smp->dr_slid != IB_LID_PERMISSIVE);
+	}
+	return 0;
+}
drivers/infiniband/core/smi.h new file +67 -0

67 lines not on this page · raw message

#42 lore raw

PATCH v5 8/24 Add InfiniBand SA (Subnet Administration) query support

Add support for sending queries to the SA (Subnet Administration). In particular the PathRecord and MCMember (multicast group member) used by the IP-over-InfiniBand driver are implemented.

drivers/infiniband/core/Makefile +3 -1
@@ -1,8 +1,10 @@
 EXTRA_CFLAGS += -Idrivers/infiniband/include
 
-obj-$(CONFIG_INFINIBAND) +=	ib_core.o ib_mad.o
+obj-$(CONFIG_INFINIBAND) +=	ib_core.o ib_mad.o ib_sa.o
 
 ib_core-y :=			packer.o ud_header.o verbs.o sysfs.o \
 				device.o fmr_pool.o cache.o
 
 ib_mad-y :=			mad.o smi.o agent.o
+
+ib_sa-y :=			sa_query.o
drivers/infiniband/core/sa_query.c new file +866 -0

866 lines not on this page · raw message

drivers/infiniband/include/ib_sa.h new file +280 -0

280 lines not on this page · raw message

#43 lore raw

PATCH v5 9/24 Add Mellanox HCA low-level driver

Add a low-level driver for Mellanox MT23108 and MT25208 HCAs. The MT25208 is only fully supported when in MT23108 compatibility mode; only the very beginnings of support for native MT25208 mode (required for HCAs without local memory) is present.

(As a side note, I believe this driver would be the first in-tree consumer of the PCI MSI/MSI-X API)

drivers/infiniband/Kconfig +2 -0
@@ -7,4 +7,6 @@
 	  any protocols you wish to use as well as drivers for your
 	  InfiniBand hardware.
 
+source "drivers/infiniband/hw/mthca/Kconfig"
+
 endmenu
drivers/infiniband/Makefile +1 -0
@@ -1 +1,2 @@
 obj-$(CONFIG_INFINIBAND)		+= core/
+obj-$(CONFIG_INFINIBAND_MTHCA)		+= hw/mthca/
drivers/infiniband/hw/mthca/Kconfig new file +26 -0
@@ -0,0 +1,26 @@
+config INFINIBAND_MTHCA
+	tristate "Mellanox HCA support"
+	depends on PCI && INFINIBAND
+	---help---
+	  This is a low-level driver for Mellanox InfiniHost host
+	  channel adapters (HCAs), including the MT23108 PCI-X HCA
+	  ("Tavor") and the MT25208 PCI Express HCA ("Arbel").
+
+config INFINIBAND_MTHCA_DEBUG
+	bool "Verbose debugging output"
+	depends on INFINIBAND_MTHCA
+	default n
+	---help---
+	  This option causes the mthca driver produce a bunch of debug
+	  messages.  Select this is you are developing the driver or
+	  trying to diagnose a problem.
+
+config INFINIBAND_MTHCA_SSE_DOORBELL
+	bool "SSE doorbell code"
+	depends on INFINIBAND_MTHCA && X86 && !X86_64
+	default n
+	---help---
+	  This option will have the mthca driver use SSE instructions
+	  to ring hardware doorbell registers.  This may improve
+	  performance for some workloads, but the driver will not run
+	  on processors without SSE instructions.
drivers/infiniband/hw/mthca/Makefile new file +12 -0
@@ -0,0 +1,12 @@
+EXTRA_CFLAGS += -Idrivers/infiniband/include
+
+ifdef CONFIG_INFINIBAND_MTHCA_DEBUG
+EXTRA_CFLAGS += -DDEBUG
+endif
+
+obj-$(CONFIG_INFINIBAND_MTHCA) += ib_mthca.o
+
+ib_mthca-y :=	mthca_main.o mthca_cmd.o mthca_profile.o mthca_reset.o \
+		mthca_allocator.o mthca_eq.o mthca_pd.o mthca_cq.o \
+		mthca_mr.o mthca_qp.o mthca_av.o mthca_mcg.o mthca_mad.o \
+		mthca_provider.o
drivers/infiniband/hw/mthca/mthca_allocator.c new file +179 -0

179 lines not on this page · raw message

drivers/infiniband/hw/mthca/mthca_config_reg.h new file +55 -0

55 lines not on this page · raw message

drivers/infiniband/hw/mthca/mthca_dev.h new file +391 -0

391 lines not on this page · raw message

drivers/infiniband/hw/mthca/mthca_doorbell.h new file +123 -0

123 lines not on this page · raw message

drivers/infiniband/hw/mthca/mthca_main.c new file +936 -0

936 lines not on this page · raw message

#44 lore raw

PATCH v5 10/24 Add Mellanox HCA low-level driver (midlayer interface)

Add midlayer interface code for Mellanox HCA driver.

drivers/infiniband/hw/mthca/mthca_provider.c new file +627 -0
@@ -0,0 +1,627 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: mthca_provider.c 1397 2004-12-28 05:09:00Z roland $
+ */
+
+#include <ib_smi.h>
+
+#include "mthca_dev.h"
+#include "mthca_cmd.h"
+
+static int mthca_query_device(struct ib_device *ibdev,
+			      struct ib_device_attr *props)
+{
+	struct ib_smp *in_mad  = NULL;
+	struct ib_smp *out_mad = NULL;
+	int err = -ENOMEM;
+	u8 status;
+
+	in_mad  = kmalloc(sizeof *in_mad, GFP_KERNEL);
+	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	props->fw_ver        = to_mdev(ibdev)->fw_ver;
+
+	memset(in_mad, 0, sizeof *in_mad);
+	in_mad->base_version       = 1;
+	in_mad->mgmt_class     	   = IB_MGMT_CLASS_SUBN_LID_ROUTED;
+	in_mad->class_version  	   = 1;
+	in_mad->method         	   = IB_MGMT_METHOD_GET;
+	in_mad->attr_id   	   = IB_SMP_ATTR_NODE_INFO;
+
+	err = mthca_MAD_IFC(to_mdev(ibdev), 1,
+			    1, in_mad, out_mad,
+			    &status);
+	if (err)
+		goto out;
+	if (status) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	props->vendor_id      = be32_to_cpup((u32 *) (out_mad->data + 36)) &
+		0xffffff;
+	props->vendor_part_id = be16_to_cpup((u16 *) (out_mad->data + 30));
+	props->hw_ver         = be16_to_cpup((u16 *) (out_mad->data + 32));
+	memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
+	memcpy(&props->node_guid,      out_mad->data + 12, 8);
+
+	err = 0;
+ out:
+	kfree(in_mad);
+	kfree(out_mad);
+	return err;
+}
+
+static int mthca_query_port(struct ib_device *ibdev,
+			    u8 port, struct ib_port_attr *props)
+{
+	struct ib_smp *in_mad  = NULL;
+	struct ib_smp *out_mad = NULL;
+	int err = -ENOMEM;
+	u8 status;
+
+	in_mad  = kmalloc(sizeof *in_mad, GFP_KERNEL);
+	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	memset(in_mad, 0, sizeof *in_mad);
+	in_mad->base_version       = 1;
+	in_mad->mgmt_class     	   = IB_MGMT_CLASS_SUBN_LID_ROUTED;
+	in_mad->class_version  	   = 1;
+	in_mad->method         	   = IB_MGMT_METHOD_GET;
+	in_mad->attr_id   	   = IB_SMP_ATTR_PORT_INFO;
+	in_mad->attr_mod           = cpu_to_be32(port);
+
+	err = mthca_MAD_IFC(to_mdev(ibdev), 1,
+			    port, in_mad, out_mad,
+			    &status);
+	if (err)
+		goto out;
+	if (status) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	props->lid               = be16_to_cpup((u16 *) (out_mad->data + 16));
+	props->lmc               = out_mad->data[34] & 0x7;
+	props->sm_lid            = be16_to_cpup((u16 *) (out_mad->data + 18));
+	props->sm_sl             = out_mad->data[36] & 0xf;
+	props->state             = out_mad->data[32] & 0xf;
+	props->port_cap_flags    = be32_to_cpup((u32 *) (out_mad->data + 20));
+	props->gid_tbl_len       = to_mdev(ibdev)->limits.gid_table_len;
+	props->pkey_tbl_len      = to_mdev(ibdev)->limits.pkey_table_len;
+	props->qkey_viol_cntr    = be16_to_cpup((u16 *) (out_mad->data + 48));
+	props->active_width      = out_mad->data[31] & 0xf;
+	props->active_speed      = out_mad->data[35] >> 4;
+
+ out:
+	kfree(in_mad);
+	kfree(out_mad);
+	return err;
+}
+
+static int mthca_modify_port(struct ib_device *ibdev,
+			     u8 port, int port_modify_mask,
+			     struct ib_port_modify *props)
+{
+	return 0;
+}
+
+static int mthca_query_pkey(struct ib_device *ibdev,
+			    u8 port, u16 index, u16 *pkey)
+{
+	struct ib_smp *in_mad  = NULL;
+	struct ib_smp *out_mad = NULL;
+	int err = -ENOMEM;
+	u8 status;
+
+	in_mad  = kmalloc(sizeof *in_mad, GFP_KERNEL);
+	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	memset(in_mad, 0, sizeof *in_mad);
+	in_mad->base_version       = 1;
+	in_mad->mgmt_class     	   = IB_MGMT_CLASS_SUBN_LID_ROUTED;
+	in_mad->class_version  	   = 1;
+	in_mad->method         	   = IB_MGMT_METHOD_GET;
+	in_mad->attr_id   	   = IB_SMP_ATTR_PKEY_TABLE;
+	in_mad->attr_mod           = cpu_to_be32(index / 32);
+
+	err = mthca_MAD_IFC(to_mdev(ibdev), 1,
+			    port, in_mad, out_mad,
+			    &status);
+	if (err)
+		goto out;
+	if (status) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	*pkey = be16_to_cpu(((u16 *) out_mad->data)[index % 32]);
+
+ out:
+	kfree(in_mad);
+	kfree(out_mad);
+	return err;
+}
+
+static int mthca_query_gid(struct ib_device *ibdev, u8 port,
+			   int index, union ib_gid *gid)
+{
+	struct ib_smp *in_mad  = NULL;
+	struct ib_smp *out_mad = NULL;
+	int err = -ENOMEM;
+	u8 status;
+
+	in_mad  = kmalloc(sizeof *in_mad, GFP_KERNEL);
+	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	memset(in_mad, 0, sizeof *in_mad);
+	in_mad->base_version       = 1;
+	in_mad->mgmt_class     	   = IB_MGMT_CLASS_SUBN_LID_ROUTED;
+	in_mad->class_version  	   = 1;
+	in_mad->method         	   = IB_MGMT_METHOD_GET;
+	in_mad->attr_id   	   = IB_SMP_ATTR_PORT_INFO;
+	in_mad->attr_mod           = cpu_to_be32(port);
+
+	err = mthca_MAD_IFC(to_mdev(ibdev), 1,
+			    port, in_mad, out_mad,
+			    &status);
+	if (err)
+		goto out;
+	if (status) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	memcpy(gid->raw, out_mad->data + 8, 8);
+
+	memset(in_mad, 0, sizeof *in_mad);
+	in_mad->base_version       = 1;
+	in_mad->mgmt_class     	   = IB_MGMT_CLASS_SUBN_LID_ROUTED;
+	in_mad->class_version  	   = 1;
+	in_mad->method         	   = IB_MGMT_METHOD_GET;
+	in_mad->attr_id   	   = IB_SMP_ATTR_GUID_INFO;
+	in_mad->attr_mod           = cpu_to_be32(index / 8);
+
+	err = mthca_MAD_IFC(to_mdev(ibdev), 1,
+			    port, in_mad, out_mad,
+			    &status);
+	if (err)
+		goto out;
+	if (status) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 16, 8);
+
+ out:
+	kfree(in_mad);
+	kfree(out_mad);
+	return err;
+}
+
+static struct ib_pd *mthca_alloc_pd(struct ib_device *ibdev)
+{
+	struct mthca_pd *pd;
+	int err;
+
+	pd = kmalloc(sizeof *pd, GFP_KERNEL);
+	if (!pd)
+		return ERR_PTR(-ENOMEM);
+
+	err = mthca_pd_alloc(to_mdev(ibdev), pd);
+	if (err) {
+		kfree(pd);
+		return ERR_PTR(err);
+	}
+
+	return &pd->ibpd;
+}
+
+static int mthca_dealloc_pd(struct ib_pd *pd)
+{
+	mthca_pd_free(to_mdev(pd->device), to_mpd(pd));
+	kfree(pd);
+
+	return 0;
+}
+
+static struct ib_ah *mthca_ah_create(struct ib_pd *pd,
+				     struct ib_ah_attr *ah_attr)
+{
+	int err;
+	struct mthca_ah *ah;
+
+	ah = kmalloc(sizeof *ah, GFP_KERNEL);
+	if (!ah)
+		return ERR_PTR(-ENOMEM);
+
+	err = mthca_create_ah(to_mdev(pd->device), to_mpd(pd), ah_attr, ah);
+	if (err) {
+		kfree(ah);
+		return ERR_PTR(err);
+	}
+
+	return &ah->ibah;
+}
+
+static int mthca_ah_destroy(struct ib_ah *ah)
+{
+	mthca_destroy_ah(to_mdev(ah->device), to_mah(ah));
+	kfree(ah);
+
+	return 0;
+}
+
+static struct ib_qp *mthca_create_qp(struct ib_pd *pd,
+				     struct ib_qp_init_attr *init_attr)
+{
+	struct mthca_qp *qp;
+	int err;
+
+	switch (init_attr->qp_type) {
+	case IB_QPT_RC:
+	case IB_QPT_UC:
+	case IB_QPT_UD:
+	{
+		qp = kmalloc(sizeof *qp, GFP_KERNEL);
+		if (!qp)
+			return ERR_PTR(-ENOMEM);
+
+		qp->sq.max    = init_attr->cap.max_send_wr;
+		qp->rq.max    = init_attr->cap.max_recv_wr;
+		qp->sq.max_gs = init_attr->cap.max_send_sge;
+		qp->rq.max_gs = init_attr->cap.max_recv_sge;
+
+		err = mthca_alloc_qp(to_mdev(pd->device), to_mpd(pd),
+				     to_mcq(init_attr->send_cq),
+				     to_mcq(init_attr->recv_cq),
+				     init_attr->qp_type, init_attr->sq_sig_type,
+				     init_attr->rq_sig_type, qp);
+		qp->ibqp.qp_num = qp->qpn;
+		break;
+	}
+	case IB_QPT_SMI:
+	case IB_QPT_GSI:
+	{
+		qp = kmalloc(sizeof (struct mthca_sqp), GFP_KERNEL);
+		if (!qp)
+			return ERR_PTR(-ENOMEM);
+
+		qp->sq.max    = init_attr->cap.max_send_wr;
+		qp->rq.max    = init_attr->cap.max_recv_wr;
+		qp->sq.max_gs = init_attr->cap.max_send_sge;
+		qp->rq.max_gs = init_attr->cap.max_recv_sge;
+
+		qp->ibqp.qp_num = init_attr->qp_type == IB_QPT_SMI ? 0 : 1;
+
+		err = mthca_alloc_sqp(to_mdev(pd->device), to_mpd(pd),
+				      to_mcq(init_attr->send_cq),
+				      to_mcq(init_attr->recv_cq),
+				      init_attr->sq_sig_type, init_attr->rq_sig_type,
+				      qp->ibqp.qp_num, init_attr->port_num,
+				      to_msqp(qp));
+		break;
+	}
+	default:
+		/* Don't support raw QPs */
+		return ERR_PTR(-ENOSYS);
+	}
+
+	if (err) {
+		kfree(qp);
+		return ERR_PTR(err);
+	}
+
+        init_attr->cap.max_inline_data = 0;
+
+	return &qp->ibqp;
+}
+
+static int mthca_destroy_qp(struct ib_qp *qp)
+{
+	mthca_free_qp(to_mdev(qp->device), to_mqp(qp));
+	kfree(qp);
+	return 0;
+}
+
+static struct ib_cq *mthca_create_cq(struct ib_device *ibdev, int entries)
+{
+	struct mthca_cq *cq;
+	int nent;
+	int err;
+
+	cq = kmalloc(sizeof *cq, GFP_KERNEL);
+	if (!cq)
+		return ERR_PTR(-ENOMEM);
+
+	for (nent = 1; nent <= entries; nent <<= 1)
+		; /* nothing */
+
+	err = mthca_init_cq(to_mdev(ibdev), nent, cq);
+	if (err) {
+		kfree(cq);
+		cq = ERR_PTR(err);
+	} else
+		cq->ibcq.cqe = nent - 1;
+
+	return &cq->ibcq;
+}
+
+static int mthca_destroy_cq(struct ib_cq *cq)
+{
+	mthca_free_cq(to_mdev(cq->device), to_mcq(cq));
+	kfree(cq);
+
+	return 0;
+}
+
+static int mthca_req_notify_cq(struct ib_cq *cq, enum ib_cq_notify notify)
+{
+	mthca_arm_cq(to_mdev(cq->device), to_mcq(cq),
+		     notify == IB_CQ_SOLICITED);
+	return 0;
+}
+
+static inline u32 convert_access(int acc)
+{
+	return (acc & IB_ACCESS_REMOTE_ATOMIC ? MTHCA_MPT_FLAG_ATOMIC       : 0) |
+	       (acc & IB_ACCESS_REMOTE_WRITE  ? MTHCA_MPT_FLAG_REMOTE_WRITE : 0) |
+	       (acc & IB_ACCESS_REMOTE_READ   ? MTHCA_MPT_FLAG_REMOTE_READ  : 0) |
+	       (acc & IB_ACCESS_LOCAL_WRITE   ? MTHCA_MPT_FLAG_LOCAL_WRITE  : 0) |
+	       MTHCA_MPT_FLAG_LOCAL_READ;
+}
+
+static struct ib_mr *mthca_get_dma_mr(struct ib_pd *pd, int acc)
+{
+	struct mthca_mr *mr;
+	int err;
+
+	mr = kmalloc(sizeof *mr, GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(-ENOMEM);
+
+	err = mthca_mr_alloc_notrans(to_mdev(pd->device),
+				     to_mpd(pd)->pd_num,
+				     convert_access(acc), mr);
+
+	if (err) {
+		kfree(mr);
+		return ERR_PTR(err);
+	}
+
+	return &mr->ibmr;
+}
+
+static struct ib_mr *mthca_reg_phys_mr(struct ib_pd       *pd,
+				       struct ib_phys_buf *buffer_list,
+				       int                 num_phys_buf,
+				       int                 acc,
+				       u64                *iova_start)
+{
+	struct mthca_mr *mr;
+	u64 *page_list;
+	u64 total_size;
+	u64 mask;
+	int shift;
+	int npages;
+	int err;
+	int i, j, n;
+
+	/* First check that we have enough alignment */
+	if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK))
+		return ERR_PTR(-EINVAL);
+
+	if (num_phys_buf > 1 &&
+	    ((buffer_list[0].addr + buffer_list[0].size) & ~PAGE_MASK))
+		return ERR_PTR(-EINVAL);
+
+	mask = 0;
+	total_size = 0;
+	for (i = 0; i < num_phys_buf; ++i) {
+		if (buffer_list[i].addr & ~PAGE_MASK)
+			return ERR_PTR(-EINVAL);
+		if (i != 0 && i != num_phys_buf - 1 &&
+		    (buffer_list[i].size & ~PAGE_MASK))
+			return ERR_PTR(-EINVAL);
+
+		total_size += buffer_list[i].size;
+		if (i > 0)
+			mask |= buffer_list[i].addr;
+	}
+
+	/* Find largest page shift we can use to cover buffers */
+	for (shift = PAGE_SHIFT; shift < 31; ++shift)
+		if (num_phys_buf > 1) {
+			if ((1ULL << shift) & mask)
+				break;
+		} else {
+			if (1ULL << shift >=
+			    buffer_list[0].size +
+			    (buffer_list[0].addr & ((1ULL << shift) - 1)))
+				break;
+		}
+
+	buffer_list[0].size += buffer_list[0].addr & ((1ULL << shift) - 1);
+	buffer_list[0].addr &= ~0ull << shift;
+
+	mr = kmalloc(sizeof *mr, GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(-ENOMEM);
+
+	npages = 0;
+	for (i = 0; i < num_phys_buf; ++i)
+		npages += (buffer_list[i].size + (1ULL << shift) - 1) >> shift;
+
+	if (!npages)
+		return &mr->ibmr;
+
+	page_list = kmalloc(npages * sizeof *page_list, GFP_KERNEL);
+	if (!page_list) {
+		kfree(mr);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	n = 0;
+	for (i = 0; i < num_phys_buf; ++i)
+		for (j = 0;
+		     j < (buffer_list[i].size + (1ULL << shift) - 1) >> shift;
+		     ++j)
+			page_list[n++] = buffer_list[i].addr + ((u64) j << shift);
+
+	mthca_dbg(to_mdev(pd->device), "Registering memory at %llx (iova %llx) "
+		  "in PD %x; shift %d, npages %d.\n",
+		  (unsigned long long) buffer_list[0].addr,
+		  (unsigned long long) *iova_start,
+		  to_mpd(pd)->pd_num,
+		  shift, npages);
+
+	err = mthca_mr_alloc_phys(to_mdev(pd->device),
+				  to_mpd(pd)->pd_num,
+				  page_list, shift, npages,
+				  *iova_start, total_size,
+				  convert_access(acc), mr);
+
+	if (err) {
+		kfree(mr);
+		return ERR_PTR(err);
+	}
+
+	kfree(page_list);
+	return &mr->ibmr;
+}
+
+static int mthca_dereg_mr(struct ib_mr *mr)
+{
+	mthca_free_mr(to_mdev(mr->device), to_mmr(mr));
+	kfree(mr);
+	return 0;
+}
+
+static ssize_t show_rev(struct class_device *cdev, char *buf)
+{
+	struct mthca_dev *dev = container_of(cdev, struct mthca_dev, ib_dev.class_dev);
+	return sprintf(buf, "%x\n", dev->rev_id);
+}
+
+static ssize_t show_fw_ver(struct class_device *cdev, char *buf)
+{
+	struct mthca_dev *dev = container_of(cdev, struct mthca_dev, ib_dev.class_dev);
+	return sprintf(buf, "%x.%x.%x\n", (int) (dev->fw_ver >> 32),
+		       (int) (dev->fw_ver >> 16) & 0xffff,
+		       (int) dev->fw_ver & 0xffff);
+}
+
+static ssize_t show_hca(struct class_device *cdev, char *buf)
+{
+	struct mthca_dev *dev = container_of(cdev, struct mthca_dev, ib_dev.class_dev);
+	switch (dev->hca_type) {
+	case TAVOR:        return sprintf(buf, "MT23108\n");
+	case ARBEL_COMPAT: return sprintf(buf, "MT25208 (MT23108 compat mode)\n");
+	case ARBEL_NATIVE: return sprintf(buf, "MT25208\n");
+	default:           return sprintf(buf, "unknown\n");
+	}
+}
+
+static CLASS_DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
+static CLASS_DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
+static CLASS_DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
+
+static struct class_device_attribute *mthca_class_attributes[] = {
+	&class_device_attr_hw_rev,
+	&class_device_attr_fw_ver,
+	&class_device_attr_hca_type
+};
+
+int mthca_register_device(struct mthca_dev *dev)
+{
+	int ret;
+	int i;
+
+	strlcpy(dev->ib_dev.name, "mthca%d", IB_DEVICE_NAME_MAX);
+	dev->ib_dev.node_type            = IB_NODE_CA;
+	dev->ib_dev.phys_port_cnt        = dev->limits.num_ports;
+	dev->ib_dev.dma_device           = &dev->pdev->dev;
+	dev->ib_dev.class_dev.dev        = &dev->pdev->dev;
+	dev->ib_dev.query_device         = mthca_query_device;
+	dev->ib_dev.query_port           = mthca_query_port;
+	dev->ib_dev.modify_port          = mthca_modify_port;
+	dev->ib_dev.query_pkey           = mthca_query_pkey;
+	dev->ib_dev.query_gid            = mthca_query_gid;
+	dev->ib_dev.alloc_pd             = mthca_alloc_pd;
+	dev->ib_dev.dealloc_pd           = mthca_dealloc_pd;
+	dev->ib_dev.create_ah            = mthca_ah_create;
+	dev->ib_dev.destroy_ah           = mthca_ah_destroy;
+	dev->ib_dev.create_qp            = mthca_create_qp;
+	dev->ib_dev.modify_qp            = mthca_modify_qp;
+	dev->ib_dev.destroy_qp           = mthca_destroy_qp;
+	dev->ib_dev.post_send            = mthca_post_send;
+	dev->ib_dev.post_recv            = mthca_post_receive;
+	dev->ib_dev.create_cq            = mthca_create_cq;
+	dev->ib_dev.destroy_cq           = mthca_destroy_cq;
+	dev->ib_dev.poll_cq              = mthca_poll_cq;
+	dev->ib_dev.req_notify_cq        = mthca_req_notify_cq;
+	dev->ib_dev.get_dma_mr           = mthca_get_dma_mr;
+	dev->ib_dev.reg_phys_mr          = mthca_reg_phys_mr;
+	dev->ib_dev.dereg_mr             = mthca_dereg_mr;
+	dev->ib_dev.attach_mcast         = mthca_multicast_attach;
+	dev->ib_dev.detach_mcast         = mthca_multicast_detach;
+	dev->ib_dev.process_mad          = mthca_process_mad;
+
+	ret = ib_register_device(&dev->ib_dev);
+	if (ret)
+		return ret;
+
+	for (i = 0; i < ARRAY_SIZE(mthca_class_attributes); ++i) {
+		ret = class_device_create_file(&dev->ib_dev.class_dev,
+					       mthca_class_attributes[i]);
+		if (ret) {
+			ib_unregister_device(&dev->ib_dev);
+			return ret;
+		}
+	}
+
+	return 0;
+}
+
+void mthca_unregister_device(struct mthca_dev *dev)
+{
+	ib_unregister_device(&dev->ib_dev);
+}
drivers/infiniband/hw/mthca/mthca_provider.h new file +225 -0

225 lines not on this page · raw message

#45 lore raw

PATCH v5 11/24 Add Mellanox HCA low-level driver (FW commands)

Add firmware command processing code for Mellanox HCA driver.

drivers/infiniband/hw/mthca/mthca_cmd.c new file +1573 -0
@@ -0,0 +1,1573 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: mthca_cmd.c 1349 2004-12-16 21:09:43Z roland $
+ */
+
+#include <linux/sched.h>
+#include <linux/pci.h>
+#include <linux/errno.h>
+#include <asm/io.h>
+
+#include "mthca_dev.h"
+#include "mthca_config_reg.h"
+#include "mthca_cmd.h"
+
+#define CMD_POLL_TOKEN 0xffff
+
+enum {
+	HCR_IN_PARAM_OFFSET    = 0x00,
+	HCR_IN_MODIFIER_OFFSET = 0x08,
+	HCR_OUT_PARAM_OFFSET   = 0x0c,
+	HCR_TOKEN_OFFSET       = 0x14,
+	HCR_STATUS_OFFSET      = 0x18,
+
+	HCR_OPMOD_SHIFT        = 12,
+	HCA_E_BIT              = 22,
+	HCR_GO_BIT             = 23
+};
+
+enum {
+	/* initialization and general commands */
+	CMD_SYS_EN          = 0x1,
+	CMD_SYS_DIS         = 0x2,
+	CMD_MAP_FA          = 0xfff,
+	CMD_UNMAP_FA        = 0xffe,
+	CMD_RUN_FW          = 0xff6,
+	CMD_MOD_STAT_CFG    = 0x34,
+	CMD_QUERY_DEV_LIM   = 0x3,
+	CMD_QUERY_FW        = 0x4,
+	CMD_ENABLE_LAM      = 0xff8,
+	CMD_DISABLE_LAM     = 0xff7,
+	CMD_QUERY_DDR       = 0x5,
+	CMD_QUERY_ADAPTER   = 0x6,
+	CMD_INIT_HCA        = 0x7,
+	CMD_CLOSE_HCA       = 0x8,
+	CMD_INIT_IB         = 0x9,
+	CMD_CLOSE_IB        = 0xa,
+	CMD_QUERY_HCA       = 0xb,
+	CMD_SET_IB          = 0xc,
+	CMD_ACCESS_DDR      = 0x2e,
+	CMD_MAP_ICM         = 0xffa,
+	CMD_UNMAP_ICM       = 0xff9,
+	CMD_MAP_ICM_AUX     = 0xffc,
+	CMD_UNMAP_ICM_AUX   = 0xffb,
+	CMD_SET_ICM_SIZE    = 0xffd,
+
+	/* TPT commands */
+	CMD_SW2HW_MPT 	    = 0xd,
+	CMD_QUERY_MPT 	    = 0xe,
+	CMD_HW2SW_MPT 	    = 0xf,
+	CMD_READ_MTT        = 0x10,
+	CMD_WRITE_MTT       = 0x11,
+	CMD_SYNC_TPT        = 0x2f,
+
+	/* EQ commands */
+	CMD_MAP_EQ          = 0x12,
+	CMD_SW2HW_EQ 	    = 0x13,
+	CMD_HW2SW_EQ 	    = 0x14,
+	CMD_QUERY_EQ        = 0x15,
+
+	/* CQ commands */
+	CMD_SW2HW_CQ 	    = 0x16,
+	CMD_HW2SW_CQ 	    = 0x17,
+	CMD_QUERY_CQ 	    = 0x18,
+	CMD_RESIZE_CQ       = 0x2c,
+
+	/* SRQ commands */
+	CMD_SW2HW_SRQ 	    = 0x35,
+	CMD_HW2SW_SRQ 	    = 0x36,
+	CMD_QUERY_SRQ       = 0x37,
+
+	/* QP/EE commands */
+	CMD_RST2INIT_QPEE   = 0x19,
+	CMD_INIT2RTR_QPEE   = 0x1a,
+	CMD_RTR2RTS_QPEE    = 0x1b,
+	CMD_RTS2RTS_QPEE    = 0x1c,
+	CMD_SQERR2RTS_QPEE  = 0x1d,
+	CMD_2ERR_QPEE       = 0x1e,
+	CMD_RTS2SQD_QPEE    = 0x1f,
+	CMD_SQD2SQD_QPEE    = 0x38,
+	CMD_SQD2RTS_QPEE    = 0x20,
+	CMD_ERR2RST_QPEE    = 0x21,
+	CMD_QUERY_QPEE      = 0x22,
+	CMD_INIT2INIT_QPEE  = 0x2d,
+	CMD_SUSPEND_QPEE    = 0x32,
+	CMD_UNSUSPEND_QPEE  = 0x33,
+	/* special QPs and management commands */
+	CMD_CONF_SPECIAL_QP = 0x23,
+	CMD_MAD_IFC         = 0x24,
+
+	/* multicast commands */
+	CMD_READ_MGM        = 0x25,
+	CMD_WRITE_MGM       = 0x26,
+	CMD_MGID_HASH       = 0x27,
+
+	/* miscellaneous commands */
+	CMD_DIAG_RPRT       = 0x30,
+	CMD_NOP             = 0x31,
+
+	/* debug commands */
+	CMD_QUERY_DEBUG_MSG = 0x2a,
+	CMD_SET_DEBUG_MSG   = 0x2b,
+};
+
+/*
+ * According to Mellanox code, FW may be starved and never complete
+ * commands.  So we can't use strict timeouts described in PRM -- we
+ * just arbitrarily select 60 seconds for now.
+ */
+#if 0
+/*
+ * Round up and add 1 to make sure we get the full wait time (since we
+ * will be starting in the middle of a jiffy)
+ */
+enum {
+	CMD_TIME_CLASS_A = (HZ + 999) / 1000 + 1,
+	CMD_TIME_CLASS_B = (HZ +  99) /  100 + 1,
+	CMD_TIME_CLASS_C = (HZ +   9) /   10 + 1
+};
+#else
+enum {
+	CMD_TIME_CLASS_A = 60 * HZ,
+	CMD_TIME_CLASS_B = 60 * HZ,
+	CMD_TIME_CLASS_C = 60 * HZ
+};
+#endif
+
+enum {
+	GO_BIT_TIMEOUT = HZ * 10
+};
+
+struct mthca_cmd_context {
+	struct completion done;
+	struct timer_list timer;
+	int               result;
+	int               next;
+	u64               out_param;
+	u16               token;
+	u8                status;
+};
+
+static inline int go_bit(struct mthca_dev *dev)
+{
+	return readl(dev->hcr + HCR_STATUS_OFFSET) &
+		swab32(1 << HCR_GO_BIT);
+}
+
+static int mthca_cmd_post(struct mthca_dev *dev,
+			  u64 in_param,
+			  u64 out_param,
+			  u32 in_modifier,
+			  u8 op_modifier,
+			  u16 op,
+			  u16 token,
+			  int event)
+{
+	int err = 0;
+
+	if (down_interruptible(&dev->cmd.hcr_sem))
+		return -EINTR;
+
+	if (event) {
+		unsigned long end = jiffies + GO_BIT_TIMEOUT;
+
+		while (go_bit(dev) && time_before(jiffies, end)) {
+			set_current_state(TASK_RUNNING);
+			schedule();
+		}
+	}
+
+	if (go_bit(dev)) {
+		err = -EAGAIN;
+		goto out;
+	}
+
+	/*
+	 * We use writel (instead of something like memcpy_toio)
+	 * because writes of less than 32 bits to the HCR don't work
+	 * (and some architectures such as ia64 implement memcpy_toio
+	 * in terms of writeb).
+	 */
+	__raw_writel(cpu_to_be32(in_param >> 32),           dev->hcr + 0 * 4);
+	__raw_writel(cpu_to_be32(in_param & 0xfffffffful),  dev->hcr + 1 * 4);
+	__raw_writel(cpu_to_be32(in_modifier),              dev->hcr + 2 * 4);
+	__raw_writel(cpu_to_be32(out_param >> 32),          dev->hcr + 3 * 4);
+	__raw_writel(cpu_to_be32(out_param & 0xfffffffful), dev->hcr + 4 * 4);
+	__raw_writel(cpu_to_be32(token << 16),              dev->hcr + 5 * 4);
+
+	/* __raw_writel may not order writes. */
+	wmb();
+
+	__raw_writel(cpu_to_be32((1 << HCR_GO_BIT)                |
+				 (event ? (1 << HCA_E_BIT) : 0)   |
+				 (op_modifier << HCR_OPMOD_SHIFT) |
+				 op),                       dev->hcr + 6 * 4);
+
+out:
+	up(&dev->cmd.hcr_sem);
+	return err;
+}
+
+static int mthca_cmd_poll(struct mthca_dev *dev,
+			  u64 in_param,
+			  u64 *out_param,
+			  int out_is_imm,
+			  u32 in_modifier,
+			  u8 op_modifier,
+			  u16 op,
+			  unsigned long timeout,
+			  u8 *status)
+{
+	int err = 0;
+	unsigned long end;
+
+	if (down_interruptible(&dev->cmd.poll_sem))
+		return -EINTR;
+
+	err = mthca_cmd_post(dev, in_param,
+			     out_param ? *out_param : 0,
+			     in_modifier, op_modifier,
+			     op, CMD_POLL_TOKEN, 0);
+	if (err)
+		goto out;
+
+	end = timeout + jiffies;
+	while (go_bit(dev) && time_before(jiffies, end)) {
+		set_current_state(TASK_RUNNING);
+		schedule();
+	}
+
+	if (go_bit(dev)) {
+		err = -EBUSY;
+		goto out;
+	}
+
+	if (out_is_imm) {
+		memcpy_fromio(out_param, dev->hcr + HCR_OUT_PARAM_OFFSET, sizeof (u64));
+		be64_to_cpus(out_param);
+	}
+
+	*status = be32_to_cpu(__raw_readl(dev->hcr + HCR_STATUS_OFFSET)) >> 24;
+
+out:
+	up(&dev->cmd.poll_sem);
+	return err;
+}
+
+void mthca_cmd_event(struct mthca_dev *dev,
+		     u16 token,
+		     u8  status,
+		     u64 out_param)
+{
+	struct mthca_cmd_context *context =
+		&dev->cmd.context[token & dev->cmd.token_mask];
+
+	/* previously timed out command completing at long last */
+	if (token != context->token)
+		return;
+
+	context->result    = 0;
+	context->status    = status;
+	context->out_param = out_param;
+
+	context->token += dev->cmd.token_mask + 1;
+
+	complete(&context->done);
+}
+
+static void event_timeout(unsigned long context_ptr)
+{
+	struct mthca_cmd_context *context =
+		(struct mthca_cmd_context *) context_ptr;
+
+	context->result = -EBUSY;
+	complete(&context->done);
+}
+
+static int mthca_cmd_wait(struct mthca_dev *dev,
+			  u64 in_param,
+			  u64 *out_param,
+			  int out_is_imm,
+			  u32 in_modifier,
+			  u8 op_modifier,
+			  u16 op,
+			  unsigned long timeout,
+			  u8 *status)
+{
+	int err = 0;
+	struct mthca_cmd_context *context;
+
+	if (down_interruptible(&dev->cmd.event_sem))
+		return -EINTR;
+
+	spin_lock(&dev->cmd.context_lock);
+	BUG_ON(dev->cmd.free_head < 0);
+	context = &dev->cmd.context[dev->cmd.free_head];
+	dev->cmd.free_head = context->next;
+	spin_unlock(&dev->cmd.context_lock);
+
+	init_completion(&context->done);
+
+	err = mthca_cmd_post(dev, in_param,
+			     out_param ? *out_param : 0,
+			     in_modifier, op_modifier,
+			     op, context->token, 1);
+	if (err)
+		goto out;
+
+	context->timer.expires  = jiffies + timeout;
+	add_timer(&context->timer);
+
+	wait_for_completion(&context->done);
+	del_timer_sync(&context->timer);
+
+	err = context->result;
+	if (err)
+		goto out;
+
+	*status = context->status;
+	if (*status)
+		mthca_dbg(dev, "Command %02x completed with status %02x\n",
+			  op, *status);
+
+	if (out_is_imm)
+		*out_param = context->out_param;
+
+out:
+	spin_lock(&dev->cmd.context_lock);
+	context->next = dev->cmd.free_head;
+	dev->cmd.free_head = context - dev->cmd.context;
+	spin_unlock(&dev->cmd.context_lock);
+
+	up(&dev->cmd.event_sem);
+	return err;
+}
+
+/* Invoke a command with an output mailbox */
+static int mthca_cmd_box(struct mthca_dev *dev,
+			 u64 in_param,
+			 u64 out_param,
+			 u32 in_modifier,
+			 u8 op_modifier,
+			 u16 op,
+			 unsigned long timeout,
+			 u8 *status)
+{
+	if (dev->cmd.use_events)
+		return mthca_cmd_wait(dev, in_param, &out_param, 0,
+				      in_modifier, op_modifier, op,
+				      timeout, status);
+	else
+		return mthca_cmd_poll(dev, in_param, &out_param, 0,
+				      in_modifier, op_modifier, op,
+				      timeout, status);
+}
+
+/* Invoke a command with no output parameter */
+static int mthca_cmd(struct mthca_dev *dev,
+		     u64 in_param,
+		     u32 in_modifier,
+		     u8 op_modifier,
+		     u16 op,
+		     unsigned long timeout,
+		     u8 *status)
+{
+	return mthca_cmd_box(dev, in_param, 0, in_modifier,
+			     op_modifier, op, timeout, status);
+}
+
+/*
+ * Invoke a command with an immediate output parameter (and copy the
+ * output into the caller's out_param pointer after the command
+ * executes).
+ */
+static int mthca_cmd_imm(struct mthca_dev *dev,
+			 u64 in_param,
+			 u64 *out_param,
+			 u32 in_modifier,
+			 u8 op_modifier,
+			 u16 op,
+			 unsigned long timeout,
+			 u8 *status)
+{
+	if (dev->cmd.use_events)
+		return mthca_cmd_wait(dev, in_param, out_param, 1,
+				      in_modifier, op_modifier, op,
+				      timeout, status);
+	else
+		return mthca_cmd_poll(dev, in_param, out_param, 1,
+				      in_modifier, op_modifier, op,
+				      timeout, status);
+}
+
+/*
+ * Switch to using events to issue FW commands (should be called after
+ * event queue to command events has been initialized).
+ */
+int mthca_cmd_use_events(struct mthca_dev *dev)
+{
+	int i;
+
+	dev->cmd.context = kmalloc(dev->cmd.max_cmds *
+				   sizeof (struct mthca_cmd_context),
+				   GFP_KERNEL);
+	if (!dev->cmd.context)
+		return -ENOMEM;
+
+	for (i = 0; i < dev->cmd.max_cmds; ++i) {
+		dev->cmd.context[i].token = i;
+		dev->cmd.context[i].next = i + 1;
+		init_timer(&dev->cmd.context[i].timer);
+		dev->cmd.context[i].timer.data     =
+			(unsigned long) &dev->cmd.context[i];
+		dev->cmd.context[i].timer.function = event_timeout;
+	}
+
+	dev->cmd.context[dev->cmd.max_cmds - 1].next = -1;
+	dev->cmd.free_head = 0;
+
+	sema_init(&dev->cmd.event_sem, dev->cmd.max_cmds);
+	spin_lock_init(&dev->cmd.context_lock);
+
+	for (dev->cmd.token_mask = 1;
+	     dev->cmd.token_mask < dev->cmd.max_cmds;
+	     dev->cmd.token_mask <<= 1)
+		; /* nothing */
+	--dev->cmd.token_mask;
+
+	dev->cmd.use_events = 1;
+	down(&dev->cmd.poll_sem);
+
+	return 0;
+}
+
+/*
+ * Switch back to polling (used when shutting down the device)
+ */
+void mthca_cmd_use_polling(struct mthca_dev *dev)
+{
+	int i;
+
+	dev->cmd.use_events = 0;
+
+	for (i = 0; i < dev->cmd.max_cmds; ++i)
+		down(&dev->cmd.event_sem);
+
+	kfree(dev->cmd.context);
+
+	up(&dev->cmd.poll_sem);
+}
+
+int mthca_SYS_EN(struct mthca_dev *dev, u8 *status)
+{
+	u64 out;
+	int ret;
+
+	ret = mthca_cmd_imm(dev, 0, &out, 0, 0, CMD_SYS_EN, HZ, status);
+
+	if (*status == MTHCA_CMD_STAT_DDR_MEM_ERR)
+		mthca_warn(dev, "SYS_EN DDR error: syn=%x, sock=%d, "
+			   "sladdr=%d, SPD source=%s\n",
+			   (int) (out >> 6) & 0xf, (int) (out >> 4) & 3,
+			   (int) (out >> 1) & 7, (int) out & 1 ? "NVMEM" : "DIMM");
+
+	return ret;
+}
+
+int mthca_SYS_DIS(struct mthca_dev *dev, u8 *status)
+{
+	return mthca_cmd(dev, 0, 0, 0, CMD_SYS_DIS, HZ, status);
+}
+
+int mthca_MAP_FA(struct mthca_dev *dev, int count,
+		 struct scatterlist *sglist, u8 *status)
+{
+	u32 *inbox;
+	dma_addr_t indma;
+	int lg;
+	int nent = 0;
+	int i, j;
+	int err = 0;
+	int ts = 0;
+
+	inbox = pci_alloc_consistent(dev->pdev, PAGE_SIZE, &indma);
+	memset(inbox, 0, PAGE_SIZE);
+
+	for (i = 0; i < count; ++i) {
+		/*
+		 * We have to pass pages that are aligned to their
+		 * size, so find the least significant 1 in the
+		 * address or size and use that as our log2 size.
+		 */
+		lg = ffs(sg_dma_address(sglist + i) | sg_dma_len(sglist + i)) - 1;
+		if (lg < 12) {
+			mthca_warn(dev, "Got FW area not aligned to 4K (%llx/%x).\n",
+				   (unsigned long long) sg_dma_address(sglist + i),
+				   sg_dma_len(sglist + i));
+			err = -EINVAL;
+			goto out;
+		}
+		for (j = 0; j < sg_dma_len(sglist + i) / (1 << lg); ++j, ++nent) {
+			*((__be64 *) (inbox + nent * 4 + 2)) =
+				cpu_to_be64((sg_dma_address(sglist + i) +
+					     (j << lg)) |
+					    (lg - 12));
+			ts += 1 << (lg - 10);
+			if (nent == PAGE_SIZE / 16) {
+				err = mthca_cmd(dev, indma, nent, 0, CMD_MAP_FA,
+						CMD_TIME_CLASS_B, status);
+				if (err || *status)
+					goto out;
+				nent = 0;
+			}
+		}
+	}
+
+	if (nent) {
+		err = mthca_cmd(dev, indma, nent, 0, CMD_MAP_FA,
+				CMD_TIME_CLASS_B, status);
+	}
+
+	mthca_dbg(dev, "Mapped %d KB of host memory for FW.\n", ts);
+
+out:
+	pci_free_consistent(dev->pdev, PAGE_SIZE, inbox, indma);
+	return err;
+}
+
+int mthca_UNMAP_FA(struct mthca_dev *dev, u8 *status)
+{
+	return mthca_cmd(dev, 0, 0, 0, CMD_UNMAP_FA, CMD_TIME_CLASS_B, status);
+}
+
+int mthca_RUN_FW(struct mthca_dev *dev, u8 *status)
+{
+	return mthca_cmd(dev, 0, 0, 0, CMD_RUN_FW, CMD_TIME_CLASS_A, status);
+}
+
+int mthca_QUERY_FW(struct mthca_dev *dev, u8 *status)
+{
+	u32 *outbox;
+	dma_addr_t outdma;
+	int err = 0;
+	u8 lg;
+
+#define QUERY_FW_OUT_SIZE             0x100
+#define QUERY_FW_VER_OFFSET            0x00
+#define QUERY_FW_MAX_CMD_OFFSET        0x0f
+#define QUERY_FW_ERR_START_OFFSET      0x30
+#define QUERY_FW_ERR_SIZE_OFFSET       0x38
+
+#define QUERY_FW_START_OFFSET          0x20
+#define QUERY_FW_END_OFFSET            0x28
+
+#define QUERY_FW_SIZE_OFFSET           0x00
+#define QUERY_FW_CLR_INT_BASE_OFFSET   0x20
+#define QUERY_FW_EQ_ARM_BASE_OFFSET    0x40
+#define QUERY_FW_EQ_SET_CI_BASE_OFFSET 0x48
+
+	outbox = pci_alloc_consistent(dev->pdev, QUERY_FW_OUT_SIZE, &outdma);
+	if (!outbox) {
+		return -ENOMEM;
+	}
+
+	err = mthca_cmd_box(dev, 0, outdma, 0, 0, CMD_QUERY_FW,
+			    CMD_TIME_CLASS_A, status);
+
+	if (err)
+		goto out;
+
+	MTHCA_GET(dev->fw_ver,   outbox, QUERY_FW_VER_OFFSET);
+	/*
+	 * FW subminor version is at more signifant bits than minor
+	 * version, so swap here.
+	 */
+	dev->fw_ver = (dev->fw_ver & 0xffff00000000ull) |
+		((dev->fw_ver & 0xffff0000ull) >> 16) |
+		((dev->fw_ver & 0x0000ffffull) << 16);
+
+	MTHCA_GET(lg, outbox, QUERY_FW_MAX_CMD_OFFSET);
+	dev->cmd.max_cmds = 1 << lg;
+
+	mthca_dbg(dev, "FW version %012llx, max commands %d\n",
+		  (unsigned long long) dev->fw_ver, dev->cmd.max_cmds);
+
+	if (dev->hca_type == ARBEL_NATIVE) {
+		MTHCA_GET(dev->fw.arbel.fw_pages,       outbox, QUERY_FW_SIZE_OFFSET);
+		MTHCA_GET(dev->fw.arbel.clr_int_base,   outbox, QUERY_FW_CLR_INT_BASE_OFFSET);
+		MTHCA_GET(dev->fw.arbel.eq_arm_base,    outbox, QUERY_FW_EQ_ARM_BASE_OFFSET);
+		MTHCA_GET(dev->fw.arbel.eq_set_ci_base, outbox, QUERY_FW_EQ_SET_CI_BASE_OFFSET);
+		mthca_dbg(dev, "FW size %d KB\n", dev->fw.arbel.fw_pages << 2);
+
+		/*
+		 * Arbel page size is always 4 KB; round up number of
+		 * system pages needed.
+		 */
+		dev->fw.arbel.fw_pages =
+			(dev->fw.arbel.fw_pages + (1 << (PAGE_SHIFT - 12)) - 1) >>
+			(PAGE_SHIFT - 12);
+
+		mthca_dbg(dev, "Clear int @ %llx, EQ arm @ %llx, EQ set CI @ %llx\n",
+			  (unsigned long long) dev->fw.arbel.clr_int_base,
+			  (unsigned long long) dev->fw.arbel.eq_arm_base,
+			  (unsigned long long) dev->fw.arbel.eq_set_ci_base);
+	} else {
+		MTHCA_GET(dev->fw.tavor.fw_start, outbox, QUERY_FW_START_OFFSET);
+		MTHCA_GET(dev->fw.tavor.fw_end,   outbox, QUERY_FW_END_OFFSET);
+
+		mthca_dbg(dev, "FW size %d KB (start %llx, end %llx)\n",
+			  (int) ((dev->fw.tavor.fw_end - dev->fw.tavor.fw_start) >> 10),
+			  (unsigned long long) dev->fw.tavor.fw_start,
+			  (unsigned long long) dev->fw.tavor.fw_end);
+	}
+
+out:
+	pci_free_consistent(dev->pdev, QUERY_FW_OUT_SIZE, outbox, outdma);
+	return err;
+}
+
+int mthca_ENABLE_LAM(struct mthca_dev *dev, u8 *status)
+{
+	u8 info;
+	u32 *outbox;
+	dma_addr_t outdma;
+	int err = 0;
+
+#define ENABLE_LAM_OUT_SIZE         0x100
+#define ENABLE_LAM_START_OFFSET     0x00
+#define ENABLE_LAM_END_OFFSET       0x08
+#define ENABLE_LAM_INFO_OFFSET      0x13
+
+#define ENABLE_LAM_INFO_HIDDEN_FLAG (1 << 4)
+#define ENABLE_LAM_INFO_ECC_MASK    0x3
+
+	outbox = pci_alloc_consistent(dev->pdev, ENABLE_LAM_OUT_SIZE, &outdma);
+	if (!outbox)
+		return -ENOMEM;
+
+	err = mthca_cmd_box(dev, 0, outdma, 0, 0, CMD_ENABLE_LAM,
+			    CMD_TIME_CLASS_C, status);
+
+	if (err)
+		goto out;
+
+	if (*status == MTHCA_CMD_STAT_LAM_NOT_PRE)
+		goto out;
+
+	MTHCA_GET(dev->ddr_start, outbox, ENABLE_LAM_START_OFFSET);
+	MTHCA_GET(dev->ddr_end,   outbox, ENABLE_LAM_END_OFFSET);
+	MTHCA_GET(info,           outbox, ENABLE_LAM_INFO_OFFSET);
+
+	if (!!(info & ENABLE_LAM_INFO_HIDDEN_FLAG) !=
+	    !!(dev->mthca_flags & MTHCA_FLAG_DDR_HIDDEN)) {
+		mthca_info(dev, "FW reports that HCA-attached memory "
+			   "is %s hidden; does not match PCI config\n",
+			   (info & ENABLE_LAM_INFO_HIDDEN_FLAG) ?
+			   "" : "not");
+	}
+	if (info & ENABLE_LAM_INFO_HIDDEN_FLAG)
+		mthca_dbg(dev, "HCA-attached memory is hidden.\n");
+
+	mthca_dbg(dev, "HCA memory size %d KB (start %llx, end %llx)\n",
+		  (int) ((dev->ddr_end - dev->ddr_start) >> 10),
+		  (unsigned long long) dev->ddr_start,
+		  (unsigned long long) dev->ddr_end);
+
+out:
+	pci_free_consistent(dev->pdev, ENABLE_LAM_OUT_SIZE, outbox, outdma);
+	return err;
+}
+
+int mthca_DISABLE_LAM(struct mthca_dev *dev, u8 *status)
+{
+	return mthca_cmd(dev, 0, 0, 0, CMD_SYS_DIS, CMD_TIME_CLASS_C, status);
+}
+
+int mthca_QUERY_DDR(struct mthca_dev *dev, u8 *status)
+{
+	u8 info;
+	u32 *outbox;
+	dma_addr_t outdma;
+	int err = 0;
+
+#define QUERY_DDR_OUT_SIZE         0x100
+#define QUERY_DDR_START_OFFSET     0x00
+#define QUERY_DDR_END_OFFSET       0x08
+#define QUERY_DDR_INFO_OFFSET      0x13
+
+#define QUERY_DDR_INFO_HIDDEN_FLAG (1 << 4)
+#define QUERY_DDR_INFO_ECC_MASK    0x3
+
+	outbox = pci_alloc_consistent(dev->pdev, QUERY_DDR_OUT_SIZE, &outdma);
+	if (!outbox)
+		return -ENOMEM;
+
+	err = mthca_cmd_box(dev, 0, outdma, 0, 0, CMD_QUERY_DDR,
+			    CMD_TIME_CLASS_A, status);
+
+	if (err)
+		goto out;
+
+	MTHCA_GET(dev->ddr_start, outbox, QUERY_DDR_START_OFFSET);
+	MTHCA_GET(dev->ddr_end,   outbox, QUERY_DDR_END_OFFSET);
+	MTHCA_GET(info,           outbox, QUERY_DDR_INFO_OFFSET);
+
+	if (!!(info & QUERY_DDR_INFO_HIDDEN_FLAG) !=
+	    !!(dev->mthca_flags & MTHCA_FLAG_DDR_HIDDEN)) {
+		mthca_info(dev, "FW reports that HCA-attached memory "
+			   "is %s hidden; does not match PCI config\n",
+			   (info & QUERY_DDR_INFO_HIDDEN_FLAG) ?
+			   "" : "not");
+	}
+	if (info & QUERY_DDR_INFO_HIDDEN_FLAG)
+		mthca_dbg(dev, "HCA-attached memory is hidden.\n");
+
+	mthca_dbg(dev, "HCA memory size %d KB (start %llx, end %llx)\n",
+		  (int) ((dev->ddr_end - dev->ddr_start) >> 10),
+		  (unsigned long long) dev->ddr_start,
+		  (unsigned long long) dev->ddr_end);
+
+out:
+	pci_free_consistent(dev->pdev, QUERY_DDR_OUT_SIZE, outbox, outdma);
+	return err;
+}
+
+int mthca_QUERY_DEV_LIM(struct mthca_dev *dev,
+			struct mthca_dev_lim *dev_lim, u8 *status)
+{
+	u32 *outbox;
+	dma_addr_t outdma;
+	u8 field;
+	u16 size;
+	int err;
+
+#define QUERY_DEV_LIM_OUT_SIZE             0x100
+#define QUERY_DEV_LIM_MAX_SRQ_SZ_OFFSET     0x10
+#define QUERY_DEV_LIM_MAX_QP_SZ_OFFSET      0x11
+#define QUERY_DEV_LIM_RSVD_QP_OFFSET        0x12
+#define QUERY_DEV_LIM_MAX_QP_OFFSET         0x13
+#define QUERY_DEV_LIM_RSVD_SRQ_OFFSET       0x14
+#define QUERY_DEV_LIM_MAX_SRQ_OFFSET        0x15
+#define QUERY_DEV_LIM_RSVD_EEC_OFFSET       0x16
+#define QUERY_DEV_LIM_MAX_EEC_OFFSET        0x17
+#define QUERY_DEV_LIM_MAX_CQ_SZ_OFFSET      0x19
+#define QUERY_DEV_LIM_RSVD_CQ_OFFSET        0x1a
+#define QUERY_DEV_LIM_MAX_CQ_OFFSET         0x1b
+#define QUERY_DEV_LIM_MAX_MPT_OFFSET        0x1d
+#define QUERY_DEV_LIM_RSVD_EQ_OFFSET        0x1e
+#define QUERY_DEV_LIM_MAX_EQ_OFFSET         0x1f
+#define QUERY_DEV_LIM_RSVD_MTT_OFFSET       0x20
+#define QUERY_DEV_LIM_MAX_MRW_SZ_OFFSET     0x21
+#define QUERY_DEV_LIM_RSVD_MRW_OFFSET       0x22
+#define QUERY_DEV_LIM_MAX_MTT_SEG_OFFSET    0x23
+#define QUERY_DEV_LIM_MAX_AV_OFFSET         0x27
+#define QUERY_DEV_LIM_MAX_REQ_QP_OFFSET     0x29
+#define QUERY_DEV_LIM_MAX_RES_QP_OFFSET     0x2b
+#define QUERY_DEV_LIM_MAX_RDMA_OFFSET       0x2f
+#define QUERY_DEV_LIM_RSZ_SRQ_OFFSET        0x33
+#define QUERY_DEV_LIM_ACK_DELAY_OFFSET      0x35
+#define QUERY_DEV_LIM_MTU_WIDTH_OFFSET      0x36
+#define QUERY_DEV_LIM_VL_PORT_OFFSET        0x37
+#define QUERY_DEV_LIM_MAX_GID_OFFSET        0x3b
+#define QUERY_DEV_LIM_MAX_PKEY_OFFSET       0x3f
+#define QUERY_DEV_LIM_FLAGS_OFFSET          0x44
+#define QUERY_DEV_LIM_RSVD_UAR_OFFSET       0x48
+#define QUERY_DEV_LIM_UAR_SZ_OFFSET         0x49
+#define QUERY_DEV_LIM_PAGE_SZ_OFFSET        0x4b
+#define QUERY_DEV_LIM_MAX_SG_OFFSET         0x51
+#define QUERY_DEV_LIM_MAX_DESC_SZ_OFFSET    0x52
+#define QUERY_DEV_LIM_MAX_SG_RQ_OFFSET      0x55
+#define QUERY_DEV_LIM_MAX_DESC_SZ_RQ_OFFSET 0x56
+#define QUERY_DEV_LIM_MAX_QP_MCG_OFFSET     0x61
+#define QUERY_DEV_LIM_RSVD_MCG_OFFSET       0x62
+#define QUERY_DEV_LIM_MAX_MCG_OFFSET        0x63
+#define QUERY_DEV_LIM_RSVD_PD_OFFSET        0x64
+#define QUERY_DEV_LIM_MAX_PD_OFFSET         0x65
+#define QUERY_DEV_LIM_RSVD_RDD_OFFSET       0x66
+#define QUERY_DEV_LIM_MAX_RDD_OFFSET        0x67
+#define QUERY_DEV_LIM_EEC_ENTRY_SZ_OFFSET   0x80
+#define QUERY_DEV_LIM_QPC_ENTRY_SZ_OFFSET   0x82
+#define QUERY_DEV_LIM_EEEC_ENTRY_SZ_OFFSET  0x84
+#define QUERY_DEV_LIM_EQPC_ENTRY_SZ_OFFSET  0x86
+#define QUERY_DEV_LIM_EQC_ENTRY_SZ_OFFSET   0x88
+#define QUERY_DEV_LIM_CQC_ENTRY_SZ_OFFSET   0x8a
+#define QUERY_DEV_LIM_SRQ_ENTRY_SZ_OFFSET   0x8c
+#define QUERY_DEV_LIM_UAR_ENTRY_SZ_OFFSET   0x8e
+#define QUERY_DEV_LIM_MTT_ENTRY_SZ_OFFSET   0x90
+#define QUERY_DEV_LIM_MPT_ENTRY_SZ_OFFSET   0x92
+#define QUERY_DEV_LIM_PBL_SZ_OFFSET         0x96
+#define QUERY_DEV_LIM_BMME_FLAGS_OFFSET     0x97
+#define QUERY_DEV_LIM_RSVD_LKEY_OFFSET      0x98
+#define QUERY_DEV_LIM_LAMR_OFFSET           0x9f
+#define QUERY_DEV_LIM_MAX_ICM_SZ_OFFSET     0xa0
+
+	outbox = pci_alloc_consistent(dev->pdev, QUERY_DEV_LIM_OUT_SIZE, &outdma);
+	if (!outbox)
+		return -ENOMEM;
+
+	err = mthca_cmd_box(dev, 0, outdma, 0, 0, CMD_QUERY_DEV_LIM,
+			    CMD_TIME_CLASS_A, status);
+
+	if (err)
+		goto out;
+
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_SRQ_SZ_OFFSET);
+	dev_lim->max_srq_sz = 1 << field;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_QP_SZ_OFFSET);
+	dev_lim->max_qp_sz = 1 << field;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_QP_OFFSET);
+	dev_lim->reserved_qps = 1 << (field & 0xf);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_QP_OFFSET);
+	dev_lim->max_qps = 1 << (field & 0x1f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_SRQ_OFFSET);
+	dev_lim->reserved_srqs = 1 << (field >> 4);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_SRQ_OFFSET);
+	dev_lim->max_srqs = 1 << (field & 0x1f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_EEC_OFFSET);
+	dev_lim->reserved_eecs = 1 << (field & 0xf);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_EEC_OFFSET);
+	dev_lim->max_eecs = 1 << (field & 0x1f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_CQ_SZ_OFFSET);
+	dev_lim->max_cq_sz = 1 << field;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_CQ_OFFSET);
+	dev_lim->reserved_cqs = 1 << (field & 0xf);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_CQ_OFFSET);
+	dev_lim->max_cqs = 1 << (field & 0x1f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_MPT_OFFSET);
+	dev_lim->max_mpts = 1 << (field & 0x3f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_EQ_OFFSET);
+	dev_lim->reserved_eqs = 1 << (field & 0xf);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_EQ_OFFSET);
+	dev_lim->max_eqs = 1 << (field & 0x7);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_MTT_OFFSET);
+	dev_lim->reserved_mtts = 1 << (field >> 4);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_MRW_SZ_OFFSET);
+	dev_lim->max_mrw_sz = 1 << field;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_MRW_OFFSET);
+	dev_lim->reserved_mrws = 1 << (field & 0xf);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_MTT_SEG_OFFSET);
+	dev_lim->max_mtt_seg = 1 << (field & 0x3f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_REQ_QP_OFFSET);
+	dev_lim->max_requester_per_qp = 1 << (field & 0x3f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_RES_QP_OFFSET);
+	dev_lim->max_responder_per_qp = 1 << (field & 0x3f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_RDMA_OFFSET);
+	dev_lim->max_rdma_global = 1 << (field & 0x3f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_ACK_DELAY_OFFSET);
+	dev_lim->local_ca_ack_delay = field & 0x1f;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MTU_WIDTH_OFFSET);
+	dev_lim->max_mtu        = field >> 4;
+	dev_lim->max_port_width = field & 0xf;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_VL_PORT_OFFSET);
+	dev_lim->max_vl    = field >> 4;
+	dev_lim->num_ports = field & 0xf;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_GID_OFFSET);
+	dev_lim->max_gids = 1 << (field & 0xf);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_PKEY_OFFSET);
+	dev_lim->max_pkeys = 1 << (field & 0xf);
+	MTHCA_GET(dev_lim->flags, outbox, QUERY_DEV_LIM_FLAGS_OFFSET);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_UAR_OFFSET);
+	dev_lim->reserved_uars = field >> 4;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_UAR_SZ_OFFSET);
+	dev_lim->uar_size = 1 << ((field & 0x3f) + 20);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_PAGE_SZ_OFFSET);
+	dev_lim->min_page_sz = 1 << field;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_SG_OFFSET);
+	dev_lim->max_sg = field;
+
+	MTHCA_GET(size, outbox, QUERY_DEV_LIM_MAX_DESC_SZ_OFFSET);
+	dev_lim->max_desc_sz = size;
+
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_QP_MCG_OFFSET);
+	dev_lim->max_qp_per_mcg = 1 << field;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_MCG_OFFSET);
+	dev_lim->reserved_mgms = field & 0xf;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_MCG_OFFSET);
+	dev_lim->max_mcgs = 1 << field;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_PD_OFFSET);
+	dev_lim->reserved_pds = field >> 4;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_PD_OFFSET);
+	dev_lim->max_pds = 1 << (field & 0x3f);
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_RDD_OFFSET);
+	dev_lim->reserved_rdds = field >> 4;
+	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_RDD_OFFSET);
+	dev_lim->max_rdds = 1 << (field & 0x3f);
+
+	MTHCA_GET(size, outbox, QUERY_DEV_LIM_EEC_ENTRY_SZ_OFFSET);
+	dev_lim->eec_entry_sz = size;
+	MTHCA_GET(size, outbox, QUERY_DEV_LIM_QPC_ENTRY_SZ_OFFSET);
+	dev_lim->qpc_entry_sz = size;
+	MTHCA_GET(size, outbox, QUERY_DEV_LIM_EEEC_ENTRY_SZ_OFFSET);
+	dev_lim->eeec_entry_sz = size;
+	MTHCA_GET(size, outbox, QUERY_DEV_LIM_EQPC_ENTRY_SZ_OFFSET);
+	dev_lim->eqpc_entry_sz = size;
+	MTHCA_GET(size, outbox, QUERY_DEV_LIM_EQC_ENTRY_SZ_OFFSET);
+	dev_lim->eqc_entry_sz = size;
+	MTHCA_GET(size, outbox, QUERY_DEV_LIM_CQC_ENTRY_SZ_OFFSET);
+	dev_lim->cqc_entry_sz = size;
+	MTHCA_GET(size, outbox, QUERY_DEV_LIM_SRQ_ENTRY_SZ_OFFSET);
+	dev_lim->srq_entry_sz = size;
+	MTHCA_GET(size, outbox, QUERY_DEV_LIM_UAR_ENTRY_SZ_OFFSET);
+	dev_lim->uar_scratch_entry_sz = size;
+
+	mthca_dbg(dev, "Max QPs: %d, reserved QPs: %d, entry size: %d\n",
+		  dev_lim->max_qps, dev_lim->reserved_qps, dev_lim->qpc_entry_sz);
+	mthca_dbg(dev, "Max CQs: %d, reserved CQs: %d, entry size: %d\n",
+		  dev_lim->max_cqs, dev_lim->reserved_cqs, dev_lim->cqc_entry_sz);
+	mthca_dbg(dev, "Max EQs: %d, reserved EQs: %d, entry size: %d\n",
+		  dev_lim->max_eqs, dev_lim->reserved_eqs, dev_lim->eqc_entry_sz);
+	mthca_dbg(dev, "reserved MPTs: %d, reserved MTTs: %d\n",
+		  dev_lim->reserved_mrws, dev_lim->reserved_mtts);
+	mthca_dbg(dev, "Max PDs: %d, reserved PDs: %d, reserved UARs: %d\n",
+		  dev_lim->max_pds, dev_lim->reserved_pds, dev_lim->reserved_uars);
+	mthca_dbg(dev, "Max QP/MCG: %d, reserved MGMs: %d\n",
+		  dev_lim->max_pds, dev_lim->reserved_mgms);
+
+	mthca_dbg(dev, "Flags: %08x\n", dev_lim->flags);
+
+	if (dev->hca_type == ARBEL_NATIVE) {
+		MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSZ_SRQ_OFFSET);
+		dev_lim->hca.arbel.resize_srq = field & 1;
+		MTHCA_GET(size, outbox, QUERY_DEV_LIM_MTT_ENTRY_SZ_OFFSET);
+		dev_lim->hca.arbel.mtt_entry_sz = size;
+		MTHCA_GET(size, outbox, QUERY_DEV_LIM_MPT_ENTRY_SZ_OFFSET);
+		dev_lim->hca.arbel.mpt_entry_sz = size;
+		MTHCA_GET(field, outbox, QUERY_DEV_LIM_PBL_SZ_OFFSET);
+		dev_lim->hca.arbel.max_pbl_sz = 1 << (field & 0x3f);
+		MTHCA_GET(dev_lim->hca.arbel.bmme_flags, outbox,
+			  QUERY_DEV_LIM_BMME_FLAGS_OFFSET);
+		MTHCA_GET(dev_lim->hca.arbel.reserved_lkey, outbox,
+			  QUERY_DEV_LIM_RSVD_LKEY_OFFSET);
+		MTHCA_GET(field, outbox, QUERY_DEV_LIM_LAMR_OFFSET);
+		dev_lim->hca.arbel.lam_required = field & 1;
+		MTHCA_GET(dev_lim->hca.arbel.max_icm_sz, outbox,
+			  QUERY_DEV_LIM_MAX_ICM_SZ_OFFSET);
+
+		if (dev_lim->hca.arbel.bmme_flags & 1)
+			mthca_dbg(dev, "Base MM extensions: yes "
+				  "(flags %d, max PBL %d, rsvd L_Key %08x)\n",
+				  dev_lim->hca.arbel.bmme_flags,
+				  dev_lim->hca.arbel.max_pbl_sz,
+				  dev_lim->hca.arbel.reserved_lkey);
+		else
+			mthca_dbg(dev, "Base MM extensions: no\n");
+
+		mthca_dbg(dev, "Max ICM size %lld MB\n",
+			  (unsigned long long) dev_lim->hca.arbel.max_icm_sz >> 20);
+	} else {
+		MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_AV_OFFSET);
+		dev_lim->hca.tavor.max_avs = 1 << (field & 0x3f);
+	}
+
+out:
+	pci_free_consistent(dev->pdev, QUERY_DEV_LIM_OUT_SIZE, outbox, outdma);
+	return err;
+}
+
+int mthca_QUERY_ADAPTER(struct mthca_dev *dev,
+			struct mthca_adapter *adapter, u8 *status)
+{
+	u32 *outbox;
+	dma_addr_t outdma;
+	int err;
+
+#define QUERY_ADAPTER_OUT_SIZE             0x100
+#define QUERY_ADAPTER_VENDOR_ID_OFFSET     0x00
+#define QUERY_ADAPTER_DEVICE_ID_OFFSET     0x04
+#define QUERY_ADAPTER_REVISION_ID_OFFSET   0x08
+#define QUERY_ADAPTER_INTA_PIN_OFFSET      0x10
+
+	outbox = pci_alloc_consistent(dev->pdev, QUERY_ADAPTER_OUT_SIZE, &outdma);
+	if (!outbox)
+		return -ENOMEM;
+
+	err = mthca_cmd_box(dev, 0, outdma, 0, 0, CMD_QUERY_ADAPTER,
+			    CMD_TIME_CLASS_A, status);
+
+	if (err)
+		goto out;
+
+	MTHCA_GET(adapter->vendor_id, outbox, QUERY_ADAPTER_VENDOR_ID_OFFSET);
+	MTHCA_GET(adapter->device_id, outbox, QUERY_ADAPTER_DEVICE_ID_OFFSET);
+	MTHCA_GET(adapter->revision_id, outbox, QUERY_ADAPTER_REVISION_ID_OFFSET);
+	MTHCA_GET(adapter->inta_pin, outbox, QUERY_ADAPTER_INTA_PIN_OFFSET);
+
+out:
+	pci_free_consistent(dev->pdev, QUERY_DEV_LIM_OUT_SIZE, outbox, outdma);
+	return err;
+}
+
+int mthca_INIT_HCA(struct mthca_dev *dev,
+		   struct mthca_init_hca_param *param,
+		   u8 *status)
+{
+	u32 *inbox;
+	dma_addr_t indma;
+	int err;
+
+#define INIT_HCA_IN_SIZE             	 0x200
+#define INIT_HCA_FLAGS_OFFSET        	 0x014
+#define INIT_HCA_QPC_OFFSET          	 0x020
+#define  INIT_HCA_QPC_BASE_OFFSET    	 (INIT_HCA_QPC_OFFSET + 0x10)
+#define  INIT_HCA_LOG_QP_OFFSET      	 (INIT_HCA_QPC_OFFSET + 0x17)
+#define  INIT_HCA_EEC_BASE_OFFSET    	 (INIT_HCA_QPC_OFFSET + 0x20)
+#define  INIT_HCA_LOG_EEC_OFFSET     	 (INIT_HCA_QPC_OFFSET + 0x27)
+#define  INIT_HCA_SRQC_BASE_OFFSET   	 (INIT_HCA_QPC_OFFSET + 0x28)
+#define  INIT_HCA_LOG_SRQ_OFFSET     	 (INIT_HCA_QPC_OFFSET + 0x2f)
+#define  INIT_HCA_CQC_BASE_OFFSET    	 (INIT_HCA_QPC_OFFSET + 0x30)
+#define  INIT_HCA_LOG_CQ_OFFSET      	 (INIT_HCA_QPC_OFFSET + 0x37)
+#define  INIT_HCA_EQPC_BASE_OFFSET   	 (INIT_HCA_QPC_OFFSET + 0x40)
+#define  INIT_HCA_EEEC_BASE_OFFSET   	 (INIT_HCA_QPC_OFFSET + 0x50)
+#define  INIT_HCA_EQC_BASE_OFFSET    	 (INIT_HCA_QPC_OFFSET + 0x60)
+#define  INIT_HCA_LOG_EQ_OFFSET      	 (INIT_HCA_QPC_OFFSET + 0x67)
+#define  INIT_HCA_RDB_BASE_OFFSET    	 (INIT_HCA_QPC_OFFSET + 0x70)
+#define INIT_HCA_UDAV_OFFSET         	 0x0b0
+#define  INIT_HCA_UDAV_LKEY_OFFSET   	 (INIT_HCA_UDAV_OFFSET + 0x0)
+#define  INIT_HCA_UDAV_PD_OFFSET     	 (INIT_HCA_UDAV_OFFSET + 0x4)
+#define INIT_HCA_MCAST_OFFSET        	 0x0c0
+#define  INIT_HCA_MC_BASE_OFFSET         (INIT_HCA_MCAST_OFFSET + 0x00)
+#define  INIT_HCA_LOG_MC_ENTRY_SZ_OFFSET (INIT_HCA_MCAST_OFFSET + 0x12)
+#define  INIT_HCA_MC_HASH_SZ_OFFSET      (INIT_HCA_MCAST_OFFSET + 0x16)
+#define  INIT_HCA_LOG_MC_TABLE_SZ_OFFSET (INIT_HCA_MCAST_OFFSET + 0x1b)
+#define INIT_HCA_TPT_OFFSET              0x0f0
+#define  INIT_HCA_MPT_BASE_OFFSET        (INIT_HCA_TPT_OFFSET + 0x00)
+#define  INIT_HCA_MTT_SEG_SZ_OFFSET      (INIT_HCA_TPT_OFFSET + 0x09)
+#define  INIT_HCA_LOG_MPT_SZ_OFFSET      (INIT_HCA_TPT_OFFSET + 0x0b)
+#define  INIT_HCA_MTT_BASE_OFFSET        (INIT_HCA_TPT_OFFSET + 0x10)
+#define INIT_HCA_UAR_OFFSET              0x120
+#define  INIT_HCA_UAR_BASE_OFFSET        (INIT_HCA_UAR_OFFSET + 0x00)
+#define  INIT_HCA_UAR_PAGE_SZ_OFFSET     (INIT_HCA_UAR_OFFSET + 0x0b)
+#define  INIT_HCA_UAR_SCATCH_BASE_OFFSET (INIT_HCA_UAR_OFFSET + 0x10)
+
+	inbox = pci_alloc_consistent(dev->pdev, INIT_HCA_IN_SIZE, &indma);
+	if (!inbox)
+		return -ENOMEM;
+
+	memset(inbox, 0, INIT_HCA_IN_SIZE);
+
+#if defined(__LITTLE_ENDIAN)
+	*(inbox + INIT_HCA_FLAGS_OFFSET / 4) &= ~cpu_to_be32(1 << 1);
+#elif defined(__BIG_ENDIAN)
+	*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |= cpu_to_be32(1 << 1);
+#else
+#error Host endianness not defined
+#endif
+	/* Check port for UD address vector: */
+	*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |= cpu_to_be32(1);
+
+	/* We leave wqe_quota, responder_exu, etc as 0 (default) */
+
+	/* QPC/EEC/CQC/EQC/RDB attributes */
+
+	MTHCA_PUT(inbox, param->qpc_base,     INIT_HCA_QPC_BASE_OFFSET);
+	MTHCA_PUT(inbox, param->log_num_qps,  INIT_HCA_LOG_QP_OFFSET);
+	MTHCA_PUT(inbox, param->eec_base,     INIT_HCA_EEC_BASE_OFFSET);
+	MTHCA_PUT(inbox, param->log_num_eecs, INIT_HCA_LOG_EEC_OFFSET);
+	MTHCA_PUT(inbox, param->srqc_base,    INIT_HCA_SRQC_BASE_OFFSET);
+	MTHCA_PUT(inbox, param->log_num_srqs, INIT_HCA_LOG_SRQ_OFFSET);
+	MTHCA_PUT(inbox, param->cqc_base,     INIT_HCA_CQC_BASE_OFFSET);
+	MTHCA_PUT(inbox, param->log_num_cqs,  INIT_HCA_LOG_CQ_OFFSET);
+	MTHCA_PUT(inbox, param->eqpc_base,    INIT_HCA_EQPC_BASE_OFFSET);
+	MTHCA_PUT(inbox, param->eeec_base,    INIT_HCA_EEEC_BASE_OFFSET);
+	MTHCA_PUT(inbox, param->eqc_base,     INIT_HCA_EQC_BASE_OFFSET);
+	MTHCA_PUT(inbox, param->log_num_eqs,  INIT_HCA_LOG_EQ_OFFSET);
+	MTHCA_PUT(inbox, param->rdb_base,     INIT_HCA_RDB_BASE_OFFSET);
+
+	/* UD AV attributes */
+
+	/* multicast attributes */
+
+	MTHCA_PUT(inbox, param->mc_base,         INIT_HCA_MC_BASE_OFFSET);
+	MTHCA_PUT(inbox, param->log_mc_entry_sz, INIT_HCA_LOG_MC_ENTRY_SZ_OFFSET);
+	MTHCA_PUT(inbox, param->mc_hash_sz,      INIT_HCA_MC_HASH_SZ_OFFSET);
+	MTHCA_PUT(inbox, param->log_mc_table_sz, INIT_HCA_LOG_MC_TABLE_SZ_OFFSET);
+
+	/* TPT attributes */
+
+	MTHCA_PUT(inbox, param->mpt_base,   INIT_HCA_MPT_BASE_OFFSET);
+	MTHCA_PUT(inbox, param->mtt_seg_sz, INIT_HCA_MTT_SEG_SZ_OFFSET);
+	MTHCA_PUT(inbox, param->log_mpt_sz, INIT_HCA_LOG_MPT_SZ_OFFSET);
+	MTHCA_PUT(inbox, param->mtt_base,   INIT_HCA_MTT_BASE_OFFSET);
+
+	/* UAR attributes */
+	{
+		u8 uar_page_sz = PAGE_SHIFT - 12;
+		MTHCA_PUT(inbox, uar_page_sz, INIT_HCA_UAR_PAGE_SZ_OFFSET);
+		MTHCA_PUT(inbox, param->uar_scratch_base, INIT_HCA_UAR_SCATCH_BASE_OFFSET);
+	}
+
+	err = mthca_cmd(dev, indma, 0, 0, CMD_INIT_HCA,
+			HZ, status);
+
+	pci_free_consistent(dev->pdev, INIT_HCA_IN_SIZE, inbox, indma);
+	return err;
+}
+
+int mthca_INIT_IB(struct mthca_dev *dev,
+		  struct mthca_init_ib_param *param,
+		  int port, u8 *status)
+{
+	u32 *inbox;
+	dma_addr_t indma;
+	int err;
+	u32 flags;
+
+#define INIT_IB_IN_SIZE          56
+#define INIT_IB_FLAGS_OFFSET     0x00
+#define INIT_IB_FLAG_SIG         (1 << 18)
+#define INIT_IB_FLAG_NG          (1 << 17)
+#define INIT_IB_FLAG_G0          (1 << 16)
+#define INIT_IB_FLAG_1X          (1 << 8)
+#define INIT_IB_FLAG_4X          (1 << 9)
+#define INIT_IB_FLAG_12X         (1 << 11)
+#define INIT_IB_VL_SHIFT         4
+#define INIT_IB_MTU_SHIFT        12
+#define INIT_IB_MAX_GID_OFFSET   0x06
+#define INIT_IB_MAX_PKEY_OFFSET  0x0a
+#define INIT_IB_GUID0_OFFSET     0x10
+#define INIT_IB_NODE_GUID_OFFSET 0x18
+#define INIT_IB_SI_GUID_OFFSET   0x20
+
+	inbox = pci_alloc_consistent(dev->pdev, INIT_IB_IN_SIZE, &indma);
+	if (!inbox)
+		return -ENOMEM;
+
+	memset(inbox, 0, INIT_IB_IN_SIZE);
+
+	flags = 0;
+	flags |= param->enable_1x     ? INIT_IB_FLAG_1X  : 0;
+	flags |= param->enable_4x     ? INIT_IB_FLAG_4X  : 0;
+	flags |= param->set_guid0     ? INIT_IB_FLAG_G0  : 0;
+	flags |= param->set_node_guid ? INIT_IB_FLAG_NG  : 0;
+	flags |= param->set_si_guid   ? INIT_IB_FLAG_SIG : 0;
+	flags |= param->vl_cap << INIT_IB_VL_SHIFT;
+	flags |= param->mtu_cap << INIT_IB_MTU_SHIFT;
+	MTHCA_PUT(inbox, flags, INIT_IB_FLAGS_OFFSET);
+
+	MTHCA_PUT(inbox, param->gid_cap,   INIT_IB_MAX_GID_OFFSET);
+	MTHCA_PUT(inbox, param->pkey_cap,  INIT_IB_MAX_PKEY_OFFSET);
+	MTHCA_PUT(inbox, param->guid0,     INIT_IB_GUID0_OFFSET);
+	MTHCA_PUT(inbox, param->node_guid, INIT_IB_NODE_GUID_OFFSET);
+	MTHCA_PUT(inbox, param->si_guid,   INIT_IB_SI_GUID_OFFSET);
+
+	err = mthca_cmd(dev, indma, port, 0, CMD_INIT_IB,
+			CMD_TIME_CLASS_A, status);
+
+	pci_free_consistent(dev->pdev, INIT_HCA_IN_SIZE, inbox, indma);
+	return err;
+}
+
+int mthca_CLOSE_IB(struct mthca_dev *dev, int port, u8 *status)
+{
+	return mthca_cmd(dev, 0, port, 0, CMD_CLOSE_IB, HZ, status);
+}
+
+int mthca_CLOSE_HCA(struct mthca_dev *dev, int panic, u8 *status)
+{
+	return mthca_cmd(dev, 0, 0, panic, CMD_CLOSE_HCA, HZ, status);
+}
+
+int mthca_SW2HW_MPT(struct mthca_dev *dev, void *mpt_entry,
+		    int mpt_index, u8 *status)
+{
+	dma_addr_t indma;
+	int err;
+
+	indma = pci_map_single(dev->pdev, mpt_entry,
+			       MTHCA_MPT_ENTRY_SIZE,
+			       PCI_DMA_TODEVICE);
+	if (pci_dma_mapping_error(indma))
+		return -ENOMEM;
+
+	err = mthca_cmd(dev, indma, mpt_index, 0, CMD_SW2HW_MPT,
+			CMD_TIME_CLASS_B, status);
+
+	pci_unmap_single(dev->pdev, indma,
+			 MTHCA_MPT_ENTRY_SIZE, PCI_DMA_TODEVICE);
+	return err;
+}
+
+int mthca_HW2SW_MPT(struct mthca_dev *dev, void *mpt_entry,
+		    int mpt_index, u8 *status)
+{
+	dma_addr_t outdma = 0;
+	int err;
+
+	if (mpt_entry) {
+		outdma = pci_map_single(dev->pdev, mpt_entry,
+					MTHCA_MPT_ENTRY_SIZE,
+					PCI_DMA_FROMDEVICE);
+		if (pci_dma_mapping_error(outdma))
+			return -ENOMEM;
+	}
+
+	err = mthca_cmd_box(dev, 0, outdma, mpt_index, !mpt_entry,
+			    CMD_HW2SW_MPT,
+			    CMD_TIME_CLASS_B, status);
+
+	if (mpt_entry)
+		pci_unmap_single(dev->pdev, outdma,
+				 MTHCA_MPT_ENTRY_SIZE,
+				 PCI_DMA_FROMDEVICE);
+	return err;
+}
+
+int mthca_WRITE_MTT(struct mthca_dev *dev, u64 *mtt_entry,
+		    int num_mtt, u8 *status)
+{
+	dma_addr_t indma;
+	int err;
+
+	indma = pci_map_single(dev->pdev, mtt_entry,
+			       (num_mtt + 2) * 8,
+			       PCI_DMA_TODEVICE);
+	if (pci_dma_mapping_error(indma))
+		return -ENOMEM;
+
+	err = mthca_cmd(dev, indma, num_mtt, 0, CMD_WRITE_MTT,
+			CMD_TIME_CLASS_B, status);
+
+	pci_unmap_single(dev->pdev, indma,
+			 (num_mtt + 2) * 8, PCI_DMA_TODEVICE);
+	return err;
+}
+
+int mthca_MAP_EQ(struct mthca_dev *dev, u64 event_mask, int unmap,
+		 int eq_num, u8 *status)
+{
+	mthca_dbg(dev, "%s mask %016llx for eqn %d\n",
+		  unmap ? "Clearing" : "Setting",
+		  (unsigned long long) event_mask, eq_num);
+	return mthca_cmd(dev, event_mask, (unmap << 31) | eq_num,
+			 0, CMD_MAP_EQ, CMD_TIME_CLASS_B, status);
+}
+
+int mthca_SW2HW_EQ(struct mthca_dev *dev, void *eq_context,
+		   int eq_num, u8 *status)
+{
+	dma_addr_t indma;
+	int err;
+
+	indma = pci_map_single(dev->pdev, eq_context,
+			       MTHCA_EQ_CONTEXT_SIZE,
+			       PCI_DMA_TODEVICE);
+	if (pci_dma_mapping_error(indma))
+		return -ENOMEM;
+
+	err = mthca_cmd(dev, indma, eq_num, 0, CMD_SW2HW_EQ,
+			CMD_TIME_CLASS_A, status);
+
+	pci_unmap_single(dev->pdev, indma,
+			 MTHCA_EQ_CONTEXT_SIZE, PCI_DMA_TODEVICE);
+	return err;
+}
+
+int mthca_HW2SW_EQ(struct mthca_dev *dev, void *eq_context,
+		   int eq_num, u8 *status)
+{
+	dma_addr_t outdma = 0;
+	int err;
+
+	outdma = pci_map_single(dev->pdev, eq_context,
+				MTHCA_EQ_CONTEXT_SIZE,
+				PCI_DMA_FROMDEVICE);
+	if (pci_dma_mapping_error(outdma))
+		return -ENOMEM;
+
+	err = mthca_cmd_box(dev, 0, outdma, eq_num, 0,
+			    CMD_HW2SW_EQ,
+			    CMD_TIME_CLASS_A, status);
+
+	pci_unmap_single(dev->pdev, outdma,
+			 MTHCA_EQ_CONTEXT_SIZE,
+			 PCI_DMA_FROMDEVICE);
+	return err;
+}
+
+int mthca_SW2HW_CQ(struct mthca_dev *dev, void *cq_context,
+		   int cq_num, u8 *status)
+{
+	dma_addr_t indma;
+	int err;
+
+	indma = pci_map_single(dev->pdev, cq_context,
+			       MTHCA_CQ_CONTEXT_SIZE,
+			       PCI_DMA_TODEVICE);
+	if (pci_dma_mapping_error(indma))
+		return -ENOMEM;
+
+	err = mthca_cmd(dev, indma, cq_num, 0, CMD_SW2HW_CQ,
+			CMD_TIME_CLASS_A, status);
+
+	pci_unmap_single(dev->pdev, indma,
+			 MTHCA_CQ_CONTEXT_SIZE, PCI_DMA_TODEVICE);
+	return err;
+}
+
+int mthca_HW2SW_CQ(struct mthca_dev *dev, void *cq_context,
+		   int cq_num, u8 *status)
+{
+	dma_addr_t outdma = 0;
+	int err;
+
+	outdma = pci_map_single(dev->pdev, cq_context,
+				MTHCA_CQ_CONTEXT_SIZE,
+				PCI_DMA_FROMDEVICE);
+	if (pci_dma_mapping_error(outdma))
+		return -ENOMEM;
+
+	err = mthca_cmd_box(dev, 0, outdma, cq_num, 0,
+			    CMD_HW2SW_CQ,
+			    CMD_TIME_CLASS_A, status);
+
+	pci_unmap_single(dev->pdev, outdma,
+			 MTHCA_CQ_CONTEXT_SIZE,
+			 PCI_DMA_FROMDEVICE);
+	return err;
+}
+
+int mthca_MODIFY_QP(struct mthca_dev *dev, int trans, u32 num,
+		    int is_ee, void *qp_context, u32 optmask,
+		    u8 *status)
+{
+	static const u16 op[] = {
+		[MTHCA_TRANS_RST2INIT]  = CMD_RST2INIT_QPEE,
+		[MTHCA_TRANS_INIT2INIT] = CMD_INIT2INIT_QPEE,
+		[MTHCA_TRANS_INIT2RTR]  = CMD_INIT2RTR_QPEE,
+		[MTHCA_TRANS_RTR2RTS]   = CMD_RTR2RTS_QPEE,
+		[MTHCA_TRANS_RTS2RTS]   = CMD_RTS2RTS_QPEE,
+		[MTHCA_TRANS_SQERR2RTS] = CMD_SQERR2RTS_QPEE,
+		[MTHCA_TRANS_ANY2ERR]   = CMD_2ERR_QPEE,
+		[MTHCA_TRANS_RTS2SQD]   = CMD_RTS2SQD_QPEE,
+		[MTHCA_TRANS_SQD2SQD]   = CMD_SQD2SQD_QPEE,
+		[MTHCA_TRANS_SQD2RTS]   = CMD_SQD2RTS_QPEE,
+		[MTHCA_TRANS_ANY2RST]   = CMD_ERR2RST_QPEE
+	};
+	u8 op_mod = 0;
+
+	dma_addr_t indma;
+	int err;
+
+	if (trans < 0 || trans >= ARRAY_SIZE(op))
+		return -EINVAL;
+
+	if (trans == MTHCA_TRANS_ANY2RST) {
+		indma  = 0;
+		op_mod = 3;	/* don't write outbox, any->reset */
+
+		/* For debugging */
+		qp_context = pci_alloc_consistent(dev->pdev, MTHCA_QP_CONTEXT_SIZE,
+						  &indma);
+		op_mod = 2;	/* write outbox, any->reset */
+	} else {
+		indma = pci_map_single(dev->pdev, qp_context,
+				       MTHCA_QP_CONTEXT_SIZE,
+				       PCI_DMA_TODEVICE);
+		if (pci_dma_mapping_error(indma))
+			return -ENOMEM;
+
+		if (0) {
+			int i;
+			mthca_dbg(dev, "Dumping QP context:\n");
+			printk(" %08x\n", be32_to_cpup(qp_context));
+			for (i = 0; i < 0x100 / 4; ++i) {
+				if (i % 8 == 0)
+					printk("[%02x] ", i * 4);
+				printk(" %08x", be32_to_cpu(((u32 *) qp_context)[i + 2]));
+				if ((i + 1) % 8 == 0)
+					printk("\n");
+			}
+		}
+	}
+
+	if (trans == MTHCA_TRANS_ANY2RST) {
+		err = mthca_cmd_box(dev, 0, indma, (!!is_ee << 24) | num,
+				    op_mod, op[trans], CMD_TIME_CLASS_C, status);
+
+		if (0) {
+			int i;
+			mthca_dbg(dev, "Dumping QP context:\n");
+			printk(" %08x\n", be32_to_cpup(qp_context));
+			for (i = 0; i < 0x100 / 4; ++i) {
+				if (i % 8 == 0)
+					printk("[%02x] ", i * 4);
+				printk(" %08x", be32_to_cpu(((u32 *) qp_context)[i + 2]));
+				if ((i + 1) % 8 == 0)
+					printk("\n");
+			}
+		}
+
+	} else
+		err = mthca_cmd(dev, indma, (!!is_ee << 24) | num,
+				op_mod, op[trans], CMD_TIME_CLASS_C, status);
+
+	if (trans != MTHCA_TRANS_ANY2RST)
+		pci_unmap_single(dev->pdev, indma,
+				 MTHCA_QP_CONTEXT_SIZE, PCI_DMA_TODEVICE);
+	else
+		pci_free_consistent(dev->pdev, MTHCA_QP_CONTEXT_SIZE,
+				    qp_context, indma);
+	return err;
+}
+
+int mthca_QUERY_QP(struct mthca_dev *dev, u32 num, int is_ee,
+		   void *qp_context, u8 *status)
+{
+	dma_addr_t outdma = 0;
+	int err;
+
+	outdma = pci_map_single(dev->pdev, qp_context,
+				MTHCA_QP_CONTEXT_SIZE,
+				PCI_DMA_FROMDEVICE);
+	if (pci_dma_mapping_error(outdma))
+		return -ENOMEM;
+
+	err = mthca_cmd_box(dev, 0, outdma, (!!is_ee << 24) | num, 0,
+			    CMD_QUERY_QPEE,
+			    CMD_TIME_CLASS_A, status);
+
+	pci_unmap_single(dev->pdev, outdma,
+			 MTHCA_QP_CONTEXT_SIZE,
+			 PCI_DMA_FROMDEVICE);
+	return err;
+}
+
+int mthca_CONF_SPECIAL_QP(struct mthca_dev *dev, int type, u32 qpn,
+			  u8 *status)
+{
+	u8 op_mod;
+
+	switch (type) {
+	case IB_QPT_SMI:
+		op_mod = 0;
+		break;
+	case IB_QPT_GSI:
+		op_mod = 1;
+		break;
+	case IB_QPT_RAW_IPV6:
+		op_mod = 2;
+		break;
+	case IB_QPT_RAW_ETY:
+		op_mod = 3;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return mthca_cmd(dev, 0, qpn, op_mod, CMD_CONF_SPECIAL_QP,
+			 CMD_TIME_CLASS_B, status);
+}
+
+int mthca_MAD_IFC(struct mthca_dev *dev, int ignore_mkey, int port,
+		  void *in_mad, void *response_mad, u8 *status) {
+	void *box;
+	dma_addr_t dma;
+	int err;
+
+#define MAD_IFC_BOX_SIZE 512
+
+	box = pci_alloc_consistent(dev->pdev, MAD_IFC_BOX_SIZE, &dma);
+	if (!box)
+		return -ENOMEM;
+
+	memcpy(box, in_mad, 256);
+
+	err = mthca_cmd_box(dev, dma, dma + 256, port, !!ignore_mkey,
+			    CMD_MAD_IFC, CMD_TIME_CLASS_C, status);
+
+	if (!err && !*status)
+		memcpy(response_mad, box + 256, 256);
+
+	pci_free_consistent(dev->pdev, MAD_IFC_BOX_SIZE, box, dma);
+	return err;
+}
+
+int mthca_READ_MGM(struct mthca_dev *dev, int index, void *mgm,
+		   u8 *status)
+{
+	dma_addr_t outdma = 0;
+	int err;
+
+	outdma = pci_map_single(dev->pdev, mgm,
+				MTHCA_MGM_ENTRY_SIZE,
+				PCI_DMA_FROMDEVICE);
+	if (pci_dma_mapping_error(outdma))
+		return -ENOMEM;
+
+	err = mthca_cmd_box(dev, 0, outdma, index, 0,
+			    CMD_READ_MGM,
+			    CMD_TIME_CLASS_A, status);
+
+	pci_unmap_single(dev->pdev, outdma,
+			 MTHCA_MGM_ENTRY_SIZE,
+			 PCI_DMA_FROMDEVICE);
+	return err;
+}
+
+int mthca_WRITE_MGM(struct mthca_dev *dev, int index, void *mgm,
+		    u8 *status)
+{
+	dma_addr_t indma;
+	int err;
+
+	indma = pci_map_single(dev->pdev, mgm,
+			       MTHCA_MGM_ENTRY_SIZE,
+			       PCI_DMA_TODEVICE);
+	if (pci_dma_mapping_error(indma))
+		return -ENOMEM;
+
+	err = mthca_cmd(dev, indma, index, 0, CMD_WRITE_MGM,
+			CMD_TIME_CLASS_A, status);
+
+	pci_unmap_single(dev->pdev, indma,
+			 MTHCA_MGM_ENTRY_SIZE, PCI_DMA_TODEVICE);
+	return err;
+}
+
+int mthca_MGID_HASH(struct mthca_dev *dev, void *gid, u16 *hash,
+		    u8 *status)
+{
+	dma_addr_t indma;
+	u64 imm;
+	int err;
+
+	indma = pci_map_single(dev->pdev, gid, 16, PCI_DMA_TODEVICE);
+	if (pci_dma_mapping_error(indma))
+		return -ENOMEM;
+
+	err = mthca_cmd_imm(dev, indma, &imm, 0, 0, CMD_MGID_HASH,
+			    CMD_TIME_CLASS_A, status);
+	*hash = imm;
+
+	pci_unmap_single(dev->pdev, indma, 16, PCI_DMA_TODEVICE);
+	return err;
+}
drivers/infiniband/hw/mthca/mthca_cmd.h new file +276 -0

276 lines not on this page · raw message

#46 lore raw

PATCH v5 12/24 Add Mellanox HCA low-level driver (EQ)

Add event queue code for Mellanox HCA driver.

drivers/infiniband/hw/mthca/mthca_eq.c new file +690 -0
@@ -0,0 +1,690 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: mthca_eq.c 1382 2004-12-24 02:21:02Z roland $
+ */
+
+#include <linux/init.h>
+#include <linux/errno.h>
+#include <linux/interrupt.h>
+#include <linux/pci.h>
+
+#include "mthca_dev.h"
+#include "mthca_cmd.h"
+#include "mthca_config_reg.h"
+
+enum {
+	MTHCA_NUM_ASYNC_EQE = 0x80,
+	MTHCA_NUM_CMD_EQE   = 0x80,
+	MTHCA_EQ_ENTRY_SIZE = 0x20
+};
+
+/*
+ * Must be packed because start is 64 bits but only aligned to 32 bits.
+ */
+struct mthca_eq_context {
+	u32 flags;
+	u64 start;
+	u32 logsize_usrpage;
+	u32 pd;
+	u8  reserved1[3];
+	u8  intr;
+	u32 lost_count;
+	u32 lkey;
+	u32 reserved2[2];
+	u32 consumer_index;
+	u32 producer_index;
+	u32 reserved3[4];
+} __attribute__((packed));
+
+#define MTHCA_EQ_STATUS_OK          ( 0 << 28)
+#define MTHCA_EQ_STATUS_OVERFLOW    ( 9 << 28)
+#define MTHCA_EQ_STATUS_WRITE_FAIL  (10 << 28)
+#define MTHCA_EQ_OWNER_SW           ( 0 << 24)
+#define MTHCA_EQ_OWNER_HW           ( 1 << 24)
+#define MTHCA_EQ_FLAG_TR            ( 1 << 18)
+#define MTHCA_EQ_FLAG_OI            ( 1 << 17)
+#define MTHCA_EQ_STATE_ARMED        ( 1 <<  8)
+#define MTHCA_EQ_STATE_FIRED        ( 2 <<  8)
+#define MTHCA_EQ_STATE_ALWAYS_ARMED ( 3 <<  8)
+
+enum {
+	MTHCA_EVENT_TYPE_COMP       	    = 0x00,
+	MTHCA_EVENT_TYPE_PATH_MIG   	    = 0x01,
+	MTHCA_EVENT_TYPE_COMM_EST   	    = 0x02,
+	MTHCA_EVENT_TYPE_SQ_DRAINED 	    = 0x03,
+	MTHCA_EVENT_TYPE_SRQ_LAST_WQE       = 0x13,
+	MTHCA_EVENT_TYPE_CQ_ERROR   	    = 0x04,
+	MTHCA_EVENT_TYPE_WQ_CATAS_ERROR     = 0x05,
+	MTHCA_EVENT_TYPE_EEC_CATAS_ERROR    = 0x06,
+	MTHCA_EVENT_TYPE_PATH_MIG_FAILED    = 0x07,
+	MTHCA_EVENT_TYPE_WQ_INVAL_REQ_ERROR = 0x10,
+	MTHCA_EVENT_TYPE_WQ_ACCESS_ERROR    = 0x11,
+	MTHCA_EVENT_TYPE_SRQ_CATAS_ERROR    = 0x12,
+	MTHCA_EVENT_TYPE_LOCAL_CATAS_ERROR  = 0x08,
+	MTHCA_EVENT_TYPE_PORT_CHANGE        = 0x09,
+	MTHCA_EVENT_TYPE_EQ_OVERFLOW        = 0x0f,
+	MTHCA_EVENT_TYPE_ECC_DETECT         = 0x0e,
+	MTHCA_EVENT_TYPE_CMD                = 0x0a
+};
+
+#define MTHCA_ASYNC_EVENT_MASK ((1ULL << MTHCA_EVENT_TYPE_PATH_MIG)           | \
+				(1ULL << MTHCA_EVENT_TYPE_COMM_EST)           | \
+				(1ULL << MTHCA_EVENT_TYPE_SQ_DRAINED)         | \
+				(1ULL << MTHCA_EVENT_TYPE_CQ_ERROR)           | \
+				(1ULL << MTHCA_EVENT_TYPE_WQ_CATAS_ERROR)     | \
+				(1ULL << MTHCA_EVENT_TYPE_EEC_CATAS_ERROR)    | \
+				(1ULL << MTHCA_EVENT_TYPE_PATH_MIG_FAILED)    | \
+				(1ULL << MTHCA_EVENT_TYPE_WQ_INVAL_REQ_ERROR) | \
+				(1ULL << MTHCA_EVENT_TYPE_WQ_ACCESS_ERROR)    | \
+				(1ULL << MTHCA_EVENT_TYPE_LOCAL_CATAS_ERROR)  | \
+				(1ULL << MTHCA_EVENT_TYPE_PORT_CHANGE)        | \
+				(1ULL << MTHCA_EVENT_TYPE_ECC_DETECT))
+#define MTHCA_SRQ_EVENT_MASK    (1ULL << MTHCA_EVENT_TYPE_SRQ_CATAS_ERROR)    | \
+				(1ULL << MTHCA_EVENT_TYPE_SRQ_LAST_WQE)
+#define MTHCA_CMD_EVENT_MASK    (1ULL << MTHCA_EVENT_TYPE_CMD)
+
+#define MTHCA_EQ_DB_INC_CI     (1 << 24)
+#define MTHCA_EQ_DB_REQ_NOT    (2 << 24)
+#define MTHCA_EQ_DB_DISARM_CQ  (3 << 24)
+#define MTHCA_EQ_DB_SET_CI     (4 << 24)
+#define MTHCA_EQ_DB_ALWAYS_ARM (5 << 24)
+
+struct mthca_eqe {
+	u8 reserved1;
+	u8 type;
+	u8 reserved2;
+	u8 subtype;
+	union {
+		u32 raw[6];
+		struct {
+			u32 cqn;
+		} __attribute__((packed)) comp;
+		struct {
+			u16 reserved1;
+			u16 token;
+			u32 reserved2;
+			u8  reserved3[3];
+			u8  status;
+			u64 out_param;
+		} __attribute__((packed)) cmd;
+		struct {
+			u32 qpn;
+		} __attribute__((packed)) qp;
+		struct {
+			u32 cqn;
+			u32 reserved1;
+			u8  reserved2[3];
+			u8  syndrome;
+		} __attribute__((packed)) cq_err;
+		struct {
+			u32 reserved1[2];
+			u32 port;
+		} __attribute__((packed)) port_change;
+	} event;
+	u8 reserved3[3];
+	u8 owner;
+} __attribute__((packed));
+
+#define  MTHCA_EQ_ENTRY_OWNER_SW      (0 << 7)
+#define  MTHCA_EQ_ENTRY_OWNER_HW      (1 << 7)
+
+static inline u64 async_mask(struct mthca_dev *dev)
+{
+	return dev->mthca_flags & MTHCA_FLAG_SRQ ?
+		MTHCA_ASYNC_EVENT_MASK | MTHCA_SRQ_EVENT_MASK :
+		MTHCA_ASYNC_EVENT_MASK;
+}
+
+static inline void set_eq_ci(struct mthca_dev *dev, int eqn, int ci)
+{
+	u32 doorbell[2];
+
+	doorbell[0] = cpu_to_be32(MTHCA_EQ_DB_SET_CI | eqn);
+	doorbell[1] = cpu_to_be32(ci);
+
+	mthca_write64(doorbell,
+		      dev->kar + MTHCA_EQ_DOORBELL,
+		      MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock));
+}
+
+static inline void eq_req_not(struct mthca_dev *dev, int eqn)
+{
+	u32 doorbell[2];
+
+	doorbell[0] = cpu_to_be32(MTHCA_EQ_DB_REQ_NOT | eqn);
+	doorbell[1] = 0;
+
+	mthca_write64(doorbell,
+		      dev->kar + MTHCA_EQ_DOORBELL,
+		      MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock));
+}
+
+static inline void disarm_cq(struct mthca_dev *dev, int eqn, int cqn)
+{
+	u32 doorbell[2];
+
+	doorbell[0] = cpu_to_be32(MTHCA_EQ_DB_DISARM_CQ | eqn);
+	doorbell[1] = cpu_to_be32(cqn);
+
+	mthca_write64(doorbell,
+		      dev->kar + MTHCA_EQ_DOORBELL,
+		      MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock));
+}
+
+static inline struct mthca_eqe *get_eqe(struct mthca_eq *eq, int entry)
+{
+	return eq->page_list[entry * MTHCA_EQ_ENTRY_SIZE / PAGE_SIZE].buf
+		+ (entry * MTHCA_EQ_ENTRY_SIZE) % PAGE_SIZE;
+}
+
+static inline int next_eqe_sw(struct mthca_eq *eq)
+{
+	return !(MTHCA_EQ_ENTRY_OWNER_HW &
+		 get_eqe(eq, eq->cons_index)->owner);
+}
+
+static inline void set_eqe_hw(struct mthca_eq *eq, int entry)
+{
+	get_eqe(eq, entry)->owner =  MTHCA_EQ_ENTRY_OWNER_HW;
+}
+
+static void port_change(struct mthca_dev *dev, int port, int active)
+{
+	struct ib_event record;
+
+	mthca_dbg(dev, "Port change to %s for port %d\n",
+		  active ? "active" : "down", port);
+
+	record.device = &dev->ib_dev;
+	record.event  = active ? IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
+	record.element.port_num = port;
+
+	ib_dispatch_event(&record);
+}
+
+static void mthca_eq_int(struct mthca_dev *dev, struct mthca_eq *eq)
+{
+	struct mthca_eqe *eqe;
+	int disarm_cqn;
+
+	while (next_eqe_sw(eq)) {
+		int set_ci = 0;
+		eqe = get_eqe(eq, eq->cons_index);
+
+		switch (eqe->type) {
+		case MTHCA_EVENT_TYPE_COMP:
+			disarm_cqn = be32_to_cpu(eqe->event.comp.cqn) & 0xffffff;
+			disarm_cq(dev, eq->eqn, disarm_cqn);
+			mthca_cq_event(dev, disarm_cqn);
+			break;
+
+		case MTHCA_EVENT_TYPE_PATH_MIG:
+			mthca_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) & 0xffffff,
+				       IB_EVENT_PATH_MIG);
+			break;
+
+		case MTHCA_EVENT_TYPE_COMM_EST:
+			mthca_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) & 0xffffff,
+				       IB_EVENT_COMM_EST);
+			break;
+
+		case MTHCA_EVENT_TYPE_SQ_DRAINED:
+			mthca_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) & 0xffffff,
+				       IB_EVENT_SQ_DRAINED);
+			break;
+
+		case MTHCA_EVENT_TYPE_WQ_CATAS_ERROR:
+			mthca_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) & 0xffffff,
+				       IB_EVENT_QP_FATAL);
+			break;
+
+		case MTHCA_EVENT_TYPE_PATH_MIG_FAILED:
+			mthca_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) & 0xffffff,
+				       IB_EVENT_PATH_MIG_ERR);
+			break;
+
+		case MTHCA_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
+			mthca_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) & 0xffffff,
+				       IB_EVENT_QP_REQ_ERR);
+			break;
+
+		case MTHCA_EVENT_TYPE_WQ_ACCESS_ERROR:
+			mthca_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) & 0xffffff,
+				       IB_EVENT_QP_ACCESS_ERR);
+			break;
+
+		case MTHCA_EVENT_TYPE_CMD:
+			mthca_cmd_event(dev,
+					be16_to_cpu(eqe->event.cmd.token),
+					eqe->event.cmd.status,
+					be64_to_cpu(eqe->event.cmd.out_param));
+			/*
+			 * cmd_event() may add more commands.
+			 * The card will think the queue has overflowed if
+			 * we don't tell it we've been processing events.
+			 */
+			set_ci = 1;
+			break;
+
+		case MTHCA_EVENT_TYPE_PORT_CHANGE:
+			port_change(dev,
+				    (be32_to_cpu(eqe->event.port_change.port) >> 28) & 3,
+				    eqe->subtype == 0x4);
+			break;
+
+		case MTHCA_EVENT_TYPE_CQ_ERROR:
+			mthca_warn(dev, "CQ %s on CQN %08x\n",
+				   eqe->event.cq_err.syndrome == 1 ?
+				   "overrun" : "access violation",
+				   be32_to_cpu(eqe->event.cq_err.cqn));
+			break;
+
+		case MTHCA_EVENT_TYPE_EQ_OVERFLOW:
+			mthca_warn(dev, "EQ overrun on EQN %d\n", eq->eqn);
+			break;
+
+		case MTHCA_EVENT_TYPE_EEC_CATAS_ERROR:
+		case MTHCA_EVENT_TYPE_SRQ_CATAS_ERROR:
+		case MTHCA_EVENT_TYPE_LOCAL_CATAS_ERROR:
+		case MTHCA_EVENT_TYPE_ECC_DETECT:
+		default:
+			mthca_warn(dev, "Unhandled event %02x(%02x) on EQ %d\n",
+				   eqe->type, eqe->subtype, eq->eqn);
+			break;
+		};
+
+		set_eqe_hw(eq, eq->cons_index);
+		eq->cons_index = (eq->cons_index + 1) & (eq->nent - 1);
+
+		if (set_ci) {
+			wmb(); /* see comment below */
+			set_eq_ci(dev, eq->eqn, eq->cons_index);
+			set_ci = 0;
+		}
+	}
+
+	/*
+	 * This barrier makes sure that all updates to
+	 * ownership bits done by set_eqe_hw() hit memory
+	 * before the consumer index is updated.  set_eq_ci()
+	 * allows the HCA to possibly write more EQ entries,
+	 * and we want to avoid the exceedingly unlikely
+	 * possibility of the HCA writing an entry and then
+	 * having set_eqe_hw() overwrite the owner field.
+	 */
+	wmb();
+	set_eq_ci(dev, eq->eqn, eq->cons_index);
+	eq_req_not(dev, eq->eqn);
+}
+
+static irqreturn_t mthca_interrupt(int irq, void *dev_ptr, struct pt_regs *regs)
+{
+	struct mthca_dev *dev = dev_ptr;
+	u32 ecr;
+	int work = 0;
+	int i;
+
+	if (dev->eq_table.clr_mask)
+		writel(dev->eq_table.clr_mask, dev->eq_table.clr_int);
+
+	while ((ecr = readl(dev->hcr + MTHCA_ECR_OFFSET + 4)) != 0) {
+		work = 1;
+
+		writel(ecr, dev->hcr + MTHCA_ECR_CLR_OFFSET + 4);
+
+		for (i = 0; i < MTHCA_NUM_EQ; ++i)
+			if (ecr & dev->eq_table.eq[i].ecr_mask)
+				mthca_eq_int(dev, &dev->eq_table.eq[i]);
+	}
+
+	return IRQ_RETVAL(work);
+}
+
+static irqreturn_t mthca_msi_x_interrupt(int irq, void *eq_ptr,
+					 struct pt_regs *regs)
+{
+	struct mthca_eq  *eq  = eq_ptr;
+	struct mthca_dev *dev = eq->dev;
+
+	writel(eq->ecr_mask, dev->hcr + MTHCA_ECR_CLR_OFFSET + 4);
+	mthca_eq_int(dev, eq);
+
+	/* MSI-X vectors always belong to us */
+	return IRQ_HANDLED;
+}
+
+static int __devinit mthca_create_eq(struct mthca_dev *dev,
+				     int nent,
+				     u8 intr,
+				     struct mthca_eq *eq)
+{
+	int npages = (nent * MTHCA_EQ_ENTRY_SIZE + PAGE_SIZE - 1) /
+		PAGE_SIZE;
+	u64 *dma_list = NULL;
+	dma_addr_t t;
+	void *mailbox = NULL;
+	struct mthca_eq_context *eq_context;
+	int err = -ENOMEM;
+	int i;
+	u8 status;
+
+	/* Make sure EQ size is aligned to a power of 2 size. */
+	for (i = 1; i < nent; i <<= 1)
+		; /* nothing */
+	nent = i;
+
+	eq->dev = dev;
+
+	eq->page_list = kmalloc(npages * sizeof *eq->page_list,
+				GFP_KERNEL);
+	if (!eq->page_list)
+		goto err_out;
+
+	for (i = 0; i < npages; ++i)
+		eq->page_list[i].buf = NULL;
+
+	dma_list = kmalloc(npages * sizeof *dma_list, GFP_KERNEL);
+	if (!dma_list)
+		goto err_out_free;
+
+	mailbox = kmalloc(sizeof *eq_context + MTHCA_CMD_MAILBOX_EXTRA,
+			  GFP_KERNEL);
+	if (!mailbox)
+		goto err_out_free;
+	eq_context = MAILBOX_ALIGN(mailbox);
+
+	for (i = 0; i < npages; ++i) {
+		eq->page_list[i].buf = pci_alloc_consistent(dev->pdev,
+							    PAGE_SIZE, &t);
+		if (!eq->page_list[i].buf)
+			goto err_out_free;
+
+		dma_list[i] = t;
+		pci_unmap_addr_set(&eq->page_list[i], mapping, t);
+
+		memset(eq->page_list[i].buf, 0, PAGE_SIZE);
+	}
+
+	for (i = 0; i < nent; ++i)
+		set_eqe_hw(eq, i);
+
+	eq->eqn = mthca_alloc(&dev->eq_table.alloc);
+	if (eq->eqn == -1)
+		goto err_out_free;
+
+	err = mthca_mr_alloc_phys(dev, dev->driver_pd.pd_num,
+				  dma_list, PAGE_SHIFT, npages,
+				  0, npages * PAGE_SIZE,
+				  MTHCA_MPT_FLAG_LOCAL_WRITE |
+				  MTHCA_MPT_FLAG_LOCAL_READ,
+				  &eq->mr);
+	if (err)
+		goto err_out_free_eq;
+
+	eq->nent = nent;
+
+	memset(eq_context, 0, sizeof *eq_context);
+	eq_context->flags           = cpu_to_be32(MTHCA_EQ_STATUS_OK   |
+						  MTHCA_EQ_OWNER_HW    |
+						  MTHCA_EQ_STATE_ARMED |
+						  MTHCA_EQ_FLAG_TR);
+	eq_context->start           = cpu_to_be64(0);
+	eq_context->logsize_usrpage = cpu_to_be32((ffs(nent) - 1) << 24 |
+						  MTHCA_KAR_PAGE);
+	eq_context->pd              = cpu_to_be32(dev->driver_pd.pd_num);
+	eq_context->intr            = intr;
+	eq_context->lkey            = cpu_to_be32(eq->mr.ibmr.lkey);
+
+	err = mthca_SW2HW_EQ(dev, eq_context, eq->eqn, &status);
+	if (err) {
+		mthca_warn(dev, "SW2HW_EQ failed (%d)\n", err);
+		goto err_out_free_mr;
+	}
+	if (status) {
+		mthca_warn(dev, "SW2HW_EQ returned status 0x%02x\n",
+			   status);
+		err = -EINVAL;
+		goto err_out_free_mr;
+	}
+
+	kfree(dma_list);
+	kfree(mailbox);
+
+	eq->ecr_mask   = swab32(1 << eq->eqn);
+	eq->cons_index = 0;
+
+	eq_req_not(dev, eq->eqn);
+
+	mthca_dbg(dev, "Allocated EQ %d with %d entries\n",
+		  eq->eqn, nent);
+
+	return err;
+
+ err_out_free_mr:
+	mthca_free_mr(dev, &eq->mr);
+
+ err_out_free_eq:
+	mthca_free(&dev->eq_table.alloc, eq->eqn);
+
+ err_out_free:
+	for (i = 0; i < npages; ++i)
+		if (eq->page_list[i].buf)
+			pci_free_consistent(dev->pdev, PAGE_SIZE,
+					    eq->page_list[i].buf,
+					    pci_unmap_addr(&eq->page_list[i],
+							   mapping));
+
+	kfree(eq->page_list);
+	kfree(dma_list);
+	kfree(mailbox);
+
+ err_out:
+	return err;
+}
+
+static void mthca_free_eq(struct mthca_dev *dev,
+			  struct mthca_eq *eq)
+{
+	void *mailbox = NULL;
+	int err;
+	u8 status;
+	int npages = (eq->nent * MTHCA_EQ_ENTRY_SIZE + PAGE_SIZE - 1) /
+		PAGE_SIZE;
+	int i;
+
+	mailbox = kmalloc(sizeof (struct mthca_eq_context) + MTHCA_CMD_MAILBOX_EXTRA,
+			  GFP_KERNEL);
+	if (!mailbox)
+		return;
+
+	err = mthca_HW2SW_EQ(dev, MAILBOX_ALIGN(mailbox),
+			     eq->eqn, &status);
+	if (err)
+		mthca_warn(dev, "HW2SW_EQ failed (%d)\n", err);
+	if (status)
+		mthca_warn(dev, "HW2SW_EQ returned status 0x%02x\n",
+			   status);
+
+	if (0) {
+		mthca_dbg(dev, "Dumping EQ context %02x:\n", eq->eqn);
+		for (i = 0; i < sizeof (struct mthca_eq_context) / 4; ++i) {
+			if (i % 4 == 0)
+				printk("[%02x] ", i * 4);
+			printk(" %08x", be32_to_cpup(MAILBOX_ALIGN(mailbox) + i * 4));
+			if ((i + 1) % 4 == 0)
+				printk("\n");
+		}
+	}
+
+
+	mthca_free_mr(dev, &eq->mr);
+	for (i = 0; i < npages; ++i)
+		pci_free_consistent(dev->pdev, PAGE_SIZE,
+				    eq->page_list[i].buf,
+				    pci_unmap_addr(&eq->page_list[i], mapping));
+
+	kfree(eq->page_list);
+	kfree(mailbox);
+}
+
+static void mthca_free_irqs(struct mthca_dev *dev)
+{
+	int i;
+
+	if (dev->eq_table.have_irq)
+		free_irq(dev->pdev->irq, dev);
+	for (i = 0; i < MTHCA_NUM_EQ; ++i)
+		if (dev->eq_table.eq[i].have_irq)
+			free_irq(dev->eq_table.eq[i].msi_x_vector,
+				 dev->eq_table.eq + i);
+}
+
+int __devinit mthca_init_eq_table(struct mthca_dev *dev)
+{
+	int err;
+	u8 status;
+	u8 intr;
+	int i;
+
+	err = mthca_alloc_init(&dev->eq_table.alloc,
+			       dev->limits.num_eqs,
+			       dev->limits.num_eqs - 1,
+			       dev->limits.reserved_eqs);
+	if (err)
+		return err;
+
+	if (dev->mthca_flags & MTHCA_FLAG_MSI ||
+	    dev->mthca_flags & MTHCA_FLAG_MSI_X) {
+		dev->eq_table.clr_mask = 0;
+	} else {
+		dev->eq_table.clr_mask =
+			swab32(1 << (dev->eq_table.inta_pin & 31));
+		dev->eq_table.clr_int  = dev->clr_base +
+			(dev->eq_table.inta_pin < 31 ? 4 : 0);
+	}
+
+	intr = (dev->mthca_flags & MTHCA_FLAG_MSI) ?
+		128 : dev->eq_table.inta_pin;
+
+	err = mthca_create_eq(dev, dev->limits.num_cqs,
+			      (dev->mthca_flags & MTHCA_FLAG_MSI_X) ? 128 : intr,
+			      &dev->eq_table.eq[MTHCA_EQ_COMP]);
+	if (err)
+		goto err_out_free;
+
+	err = mthca_create_eq(dev, MTHCA_NUM_ASYNC_EQE,
+			      (dev->mthca_flags & MTHCA_FLAG_MSI_X) ? 129 : intr,
+			      &dev->eq_table.eq[MTHCA_EQ_ASYNC]);
+	if (err)
+		goto err_out_comp;
+
+	err = mthca_create_eq(dev, MTHCA_NUM_CMD_EQE,
+			      (dev->mthca_flags & MTHCA_FLAG_MSI_X) ? 130 : intr,
+			      &dev->eq_table.eq[MTHCA_EQ_CMD]);
+	if (err)
+		goto err_out_async;
+
+	if (dev->mthca_flags & MTHCA_FLAG_MSI_X) {
+		static const char *eq_name[] = {
+			[MTHCA_EQ_COMP]  = DRV_NAME " (comp)",
+			[MTHCA_EQ_ASYNC] = DRV_NAME " (async)",
+			[MTHCA_EQ_CMD]   = DRV_NAME " (cmd)"
+		};
+
+		for (i = 0; i < MTHCA_NUM_EQ; ++i) {
+			err = request_irq(dev->eq_table.eq[i].msi_x_vector,
+					  mthca_msi_x_interrupt, 0,
+					  eq_name[i], dev->eq_table.eq + i);
+			if (err)
+				goto err_out_cmd;
+			dev->eq_table.eq[i].have_irq = 1;
+		}
+	} else {
+		err = request_irq(dev->pdev->irq, mthca_interrupt, SA_SHIRQ,
+				  DRV_NAME, dev);
+		if (err)
+			goto err_out_cmd;
+		dev->eq_table.have_irq = 1;
+	}
+
+	err = mthca_MAP_EQ(dev, async_mask(dev),
+			   0, dev->eq_table.eq[MTHCA_EQ_ASYNC].eqn, &status);
+	if (err)
+		mthca_warn(dev, "MAP_EQ for async EQ %d failed (%d)\n",
+			   dev->eq_table.eq[MTHCA_EQ_ASYNC].eqn, err);
+	if (status)
+		mthca_warn(dev, "MAP_EQ for async EQ %d returned status 0x%02x\n",
+			   dev->eq_table.eq[MTHCA_EQ_ASYNC].eqn, status);
+
+	err = mthca_MAP_EQ(dev, MTHCA_CMD_EVENT_MASK,
+			   0, dev->eq_table.eq[MTHCA_EQ_CMD].eqn, &status);
+	if (err)
+		mthca_warn(dev, "MAP_EQ for cmd EQ %d failed (%d)\n",
+			   dev->eq_table.eq[MTHCA_EQ_CMD].eqn, err);
+	if (status)
+		mthca_warn(dev, "MAP_EQ for cmd EQ %d returned status 0x%02x\n",
+			   dev->eq_table.eq[MTHCA_EQ_CMD].eqn, status);
+
+	return 0;
+
+err_out_cmd:
+	mthca_free_irqs(dev);
+	mthca_free_eq(dev, &dev->eq_table.eq[MTHCA_EQ_CMD]);
+
+err_out_async:
+	mthca_free_eq(dev, &dev->eq_table.eq[MTHCA_EQ_ASYNC]);
+
+err_out_comp:
+	mthca_free_eq(dev, &dev->eq_table.eq[MTHCA_EQ_COMP]);
+
+err_out_free:
+	mthca_alloc_cleanup(&dev->eq_table.alloc);
+	return err;
+}
+
+void __devexit mthca_cleanup_eq_table(struct mthca_dev *dev)
+{
+	u8 status;
+	int i;
+
+	mthca_free_irqs(dev);
+
+	mthca_MAP_EQ(dev, async_mask(dev),
+		     1, dev->eq_table.eq[MTHCA_EQ_ASYNC].eqn, &status);
+	mthca_MAP_EQ(dev, MTHCA_CMD_EVENT_MASK,
+		     1, dev->eq_table.eq[MTHCA_EQ_CMD].eqn, &status);
+
+	for (i = 0; i < MTHCA_NUM_EQ; ++i)
+		mthca_free_eq(dev, &dev->eq_table.eq[i]);
+
+	mthca_alloc_cleanup(&dev->eq_table.alloc);
+}
#47 lore raw

PATCH v5 13/24 Add Mellanox HCA low-level driver (initialization)

Add device initializaton code for Mellanox HCA driver.

drivers/infiniband/hw/mthca/mthca_profile.c new file +226 -0
@@ -0,0 +1,226 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: mthca_profile.c 1349 2004-12-16 21:09:43Z roland $
+ */
+
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+
+#include "mthca_profile.h"
+
+static int default_profile[MTHCA_RES_NUM] = {
+	[MTHCA_RES_QP]    = 1 << 16,
+	[MTHCA_RES_EQP]   = 1 << 16,
+	[MTHCA_RES_CQ]    = 1 << 16,
+	[MTHCA_RES_EQ]    = 32,
+	[MTHCA_RES_RDB]   = 1 << 18,
+	[MTHCA_RES_MCG]   = 1 << 13,
+	[MTHCA_RES_MPT]   = 1 << 17,
+	[MTHCA_RES_MTT]   = 1 << 20,
+	[MTHCA_RES_UDAV]  = 1 << 15
+};
+
+enum {
+	MTHCA_RDB_ENTRY_SIZE = 32,
+	MTHCA_MTT_SEG_SIZE   = 64
+};
+
+enum {
+	MTHCA_NUM_PDS = 1 << 15
+};
+
+int mthca_make_profile(struct mthca_dev *dev,
+		       struct mthca_dev_lim *dev_lim,
+		       struct mthca_init_hca_param *init_hca)
+{
+	/* just use default profile for now */
+	struct mthca_resource {
+		u64 size;
+		u64 start;
+		int type;
+		int num;
+		int log_num;
+	};
+
+	u64 total_size = 0;
+	struct mthca_resource *profile;
+	struct mthca_resource tmp;
+	int i, j;
+
+	default_profile[MTHCA_RES_UAR] = dev_lim->uar_size / PAGE_SIZE;
+
+	profile = kmalloc(MTHCA_RES_NUM * sizeof *profile, GFP_KERNEL);
+	if (!profile)
+		return -ENOMEM;
+
+	profile[MTHCA_RES_QP].size   = dev_lim->qpc_entry_sz;
+	profile[MTHCA_RES_EEC].size  = dev_lim->eec_entry_sz;
+	profile[MTHCA_RES_SRQ].size  = dev_lim->srq_entry_sz;
+	profile[MTHCA_RES_CQ].size   = dev_lim->cqc_entry_sz;
+	profile[MTHCA_RES_EQP].size  = dev_lim->eqpc_entry_sz;
+	profile[MTHCA_RES_EEEC].size = dev_lim->eeec_entry_sz;
+	profile[MTHCA_RES_EQ].size   = dev_lim->eqc_entry_sz;
+	profile[MTHCA_RES_RDB].size  = MTHCA_RDB_ENTRY_SIZE;
+	profile[MTHCA_RES_MCG].size  = MTHCA_MGM_ENTRY_SIZE;
+	profile[MTHCA_RES_MPT].size  = MTHCA_MPT_ENTRY_SIZE;
+	profile[MTHCA_RES_MTT].size  = MTHCA_MTT_SEG_SIZE;
+	profile[MTHCA_RES_UAR].size  = dev_lim->uar_scratch_entry_sz;
+	profile[MTHCA_RES_UDAV].size = MTHCA_AV_SIZE;
+
+	for (i = 0; i < MTHCA_RES_NUM; ++i) {
+		profile[i].type     = i;
+		profile[i].num      = default_profile[i];
+		profile[i].log_num  = max(ffs(default_profile[i]) - 1, 0);
+		profile[i].size    *= default_profile[i];
+	}
+
+	/*
+	 * Sort the resources in decreasing order of size.  Since they
+	 * all have sizes that are powers of 2, we'll be able to keep
+	 * resources aligned to their size and pack them without gaps
+	 * using the sorted order.
+	 */
+	for (i = MTHCA_RES_NUM; i > 0; --i)
+		for (j = 1; j < i; ++j) {
+			if (profile[j].size > profile[j - 1].size) {
+				tmp            = profile[j];
+				profile[j]     = profile[j - 1];
+				profile[j - 1] = tmp;
+			}
+		}
+
+	for (i = 0; i < MTHCA_RES_NUM; ++i) {
+		if (profile[i].size) {
+			profile[i].start = dev->ddr_start + total_size;
+			total_size      += profile[i].size;
+		}
+		if (total_size > dev->fw.tavor.fw_start - dev->ddr_start) {
+			mthca_err(dev, "Profile requires 0x%llx bytes; "
+				  "won't fit between DDR start at 0x%016llx "
+				  "and FW start at 0x%016llx.\n",
+				  (unsigned long long) total_size,
+				  (unsigned long long) dev->ddr_start,
+				  (unsigned long long) dev->fw.tavor.fw_start);
+			kfree(profile);
+			return -ENOMEM;
+		}
+
+		if (profile[i].size)
+			mthca_dbg(dev, "profile[%2d]--%2d/%2d @ 0x%16llx "
+				  "(size 0x%8llx)\n",
+				  i, profile[i].type, profile[i].log_num,
+				  (unsigned long long) profile[i].start,
+				  (unsigned long long) profile[i].size);
+	}
+
+	mthca_dbg(dev, "HCA memory: allocated %d KB/%d KB (%d KB free)\n",
+		  (int) (total_size >> 10),
+		  (int) ((dev->fw.tavor.fw_start - dev->ddr_start) >> 10),
+		  (int) ((dev->fw.tavor.fw_start - dev->ddr_start - total_size) >> 10));
+
+	for (i = 0; i < MTHCA_RES_NUM; ++i) {
+		switch (profile[i].type) {
+		case MTHCA_RES_QP:
+			dev->limits.num_qps   = profile[i].num;
+			init_hca->qpc_base    = profile[i].start;
+			init_hca->log_num_qps = profile[i].log_num;
+			break;
+		case MTHCA_RES_EEC:
+			dev->limits.num_eecs   = profile[i].num;
+			init_hca->eec_base     = profile[i].start;
+			init_hca->log_num_eecs = profile[i].log_num;
+			break;
+		case MTHCA_RES_SRQ:
+			dev->limits.num_srqs   = profile[i].num;
+			init_hca->srqc_base    = profile[i].start;
+			init_hca->log_num_srqs = profile[i].log_num;
+			break;
+		case MTHCA_RES_CQ:
+			dev->limits.num_cqs   = profile[i].num;
+			init_hca->cqc_base    = profile[i].start;
+			init_hca->log_num_cqs = profile[i].log_num;
+			break;
+		case MTHCA_RES_EQP:
+			init_hca->eqpc_base = profile[i].start;
+			break;
+		case MTHCA_RES_EEEC:
+			init_hca->eeec_base = profile[i].start;
+			break;
+		case MTHCA_RES_EQ:
+			dev->limits.num_eqs   = profile[i].num;
+			init_hca->eqc_base    = profile[i].start;
+			init_hca->log_num_eqs = profile[i].log_num;
+			break;
+		case MTHCA_RES_RDB:
+			dev->limits.num_rdbs = profile[i].num;
+			init_hca->rdb_base   = profile[i].start;
+			break;
+		case MTHCA_RES_MCG:
+			dev->limits.num_mgms      = profile[i].num >> 1;
+			dev->limits.num_amgms     = profile[i].num >> 1;
+			init_hca->mc_base         = profile[i].start;
+			init_hca->log_mc_entry_sz = ffs(MTHCA_MGM_ENTRY_SIZE) - 1;
+			init_hca->log_mc_table_sz = profile[i].log_num;
+			init_hca->mc_hash_sz      = 1 << (profile[i].log_num - 1);
+			break;
+		case MTHCA_RES_MPT:
+			dev->limits.num_mpts = profile[i].num;
+			init_hca->mpt_base   = profile[i].start;
+			init_hca->log_mpt_sz = profile[i].log_num;
+			break;
+		case MTHCA_RES_MTT:
+			dev->limits.num_mtt_segs = profile[i].num;
+			dev->limits.mtt_seg_size = MTHCA_MTT_SEG_SIZE;
+			dev->mr_table.mtt_base   = profile[i].start;
+			init_hca->mtt_base       = profile[i].start;
+			init_hca->mtt_seg_sz     = ffs(MTHCA_MTT_SEG_SIZE) - 7;
+			break;
+		case MTHCA_RES_UAR:
+			init_hca->uar_scratch_base = profile[i].start;
+			break;
+		case MTHCA_RES_UDAV:
+			dev->av_table.ddr_av_base = profile[i].start;
+			dev->av_table.num_ddr_avs = profile[i].num;
+		default:
+			break;
+		}
+	}
+
+	/*
+	 * PDs don't take any HCA memory, but we assign them as part
+	 * of the HCA profile anyway.
+	 */
+	dev->limits.num_pds = MTHCA_NUM_PDS;
+
+	kfree(profile);
+	return 0;
+}
drivers/infiniband/hw/mthca/mthca_profile.h new file +62 -0

62 lines not on this page · raw message

drivers/infiniband/hw/mthca/mthca_reset.c new file +232 -0

232 lines not on this page · raw message

#48 lore raw

PATCH v5 14/24 Add Mellanox HCA low-level driver (QP/CQ)

Add CQ (completion queue) and QP (queue pair) code for Mellanox HCA driver.

drivers/infiniband/hw/mthca/mthca_cq.c new file +836 -0
@@ -0,0 +1,836 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: mthca_cq.c 1369 2004-12-20 16:17:07Z roland $
+ */
+
+#include <linux/init.h>
+
+#include <ib_pack.h>
+
+#include "mthca_dev.h"
+#include "mthca_cmd.h"
+
+enum {
+	MTHCA_MAX_DIRECT_CQ_SIZE = 4 * PAGE_SIZE
+};
+
+enum {
+	MTHCA_CQ_ENTRY_SIZE = 0x20
+};
+
+/*
+ * Must be packed because start is 64 bits but only aligned to 32 bits.
+ */
+struct mthca_cq_context {
+	u32 flags;
+	u64 start;
+	u32 logsize_usrpage;
+	u32 error_eqn;
+	u32 comp_eqn;
+	u32 pd;
+	u32 lkey;
+	u32 last_notified_index;
+	u32 solicit_producer_index;
+	u32 consumer_index;
+	u32 producer_index;
+	u32 cqn;
+	u32 reserved[3];
+} __attribute__((packed));
+
+#define MTHCA_CQ_STATUS_OK          ( 0 << 28)
+#define MTHCA_CQ_STATUS_OVERFLOW    ( 9 << 28)
+#define MTHCA_CQ_STATUS_WRITE_FAIL  (10 << 28)
+#define MTHCA_CQ_FLAG_TR            ( 1 << 18)
+#define MTHCA_CQ_FLAG_OI            ( 1 << 17)
+#define MTHCA_CQ_STATE_DISARMED     ( 0 <<  8)
+#define MTHCA_CQ_STATE_ARMED        ( 1 <<  8)
+#define MTHCA_CQ_STATE_ARMED_SOL    ( 4 <<  8)
+#define MTHCA_EQ_STATE_FIRED        (10 <<  8)
+
+enum {
+	MTHCA_ERROR_CQE_OPCODE_MASK = 0xfe
+};
+
+enum {
+	SYNDROME_LOCAL_LENGTH_ERR 	 = 0x01,
+	SYNDROME_LOCAL_QP_OP_ERR  	 = 0x02,
+	SYNDROME_LOCAL_EEC_OP_ERR 	 = 0x03,
+	SYNDROME_LOCAL_PROT_ERR   	 = 0x04,
+	SYNDROME_WR_FLUSH_ERR     	 = 0x05,
+	SYNDROME_MW_BIND_ERR      	 = 0x06,
+	SYNDROME_BAD_RESP_ERR     	 = 0x10,
+	SYNDROME_LOCAL_ACCESS_ERR 	 = 0x11,
+	SYNDROME_REMOTE_INVAL_REQ_ERR 	 = 0x12,
+	SYNDROME_REMOTE_ACCESS_ERR 	 = 0x13,
+	SYNDROME_REMOTE_OP_ERR     	 = 0x14,
+	SYNDROME_RETRY_EXC_ERR 		 = 0x15,
+	SYNDROME_RNR_RETRY_EXC_ERR 	 = 0x16,
+	SYNDROME_LOCAL_RDD_VIOL_ERR 	 = 0x20,
+	SYNDROME_REMOTE_INVAL_RD_REQ_ERR = 0x21,
+	SYNDROME_REMOTE_ABORTED_ERR 	 = 0x22,
+	SYNDROME_INVAL_EECN_ERR 	 = 0x23,
+	SYNDROME_INVAL_EEC_STATE_ERR 	 = 0x24
+};
+
+struct mthca_cqe {
+	u32 my_qpn;
+	u32 my_ee;
+	u32 rqpn;
+	u16 sl_g_mlpath;
+	u16 rlid;
+	u32 imm_etype_pkey_eec;
+	u32 byte_cnt;
+	u32 wqe;
+	u8  opcode;
+	u8  is_send;
+	u8  reserved;
+	u8  owner;
+};
+
+struct mthca_err_cqe {
+	u32 my_qpn;
+	u32 reserved1[3];
+	u8  syndrome;
+	u8  reserved2;
+	u16 db_cnt;
+	u32 reserved3;
+	u32 wqe;
+	u8  opcode;
+	u8  reserved4[2];
+	u8  owner;
+};
+
+#define MTHCA_CQ_ENTRY_OWNER_SW      (0 << 7)
+#define MTHCA_CQ_ENTRY_OWNER_HW      (1 << 7)
+
+#define MTHCA_CQ_DB_INC_CI       (1 << 24)
+#define MTHCA_CQ_DB_REQ_NOT      (2 << 24)
+#define MTHCA_CQ_DB_REQ_NOT_SOL  (3 << 24)
+#define MTHCA_CQ_DB_SET_CI       (4 << 24)
+#define MTHCA_CQ_DB_REQ_NOT_MULT (5 << 24)
+
+static inline struct mthca_cqe *get_cqe(struct mthca_cq *cq, int entry)
+{
+	if (cq->is_direct)
+		return cq->queue.direct.buf + (entry * MTHCA_CQ_ENTRY_SIZE);
+	else
+		return cq->queue.page_list[entry * MTHCA_CQ_ENTRY_SIZE / PAGE_SIZE].buf
+			+ (entry * MTHCA_CQ_ENTRY_SIZE) % PAGE_SIZE;
+}
+
+static inline int cqe_sw(struct mthca_cq *cq, int i)
+{
+	return !(MTHCA_CQ_ENTRY_OWNER_HW &
+		 get_cqe(cq, i)->owner);
+}
+
+static inline int next_cqe_sw(struct mthca_cq *cq)
+{
+	return cqe_sw(cq, cq->cons_index);
+}
+
+static inline void set_cqe_hw(struct mthca_cq *cq, int entry)
+{
+	get_cqe(cq, entry)->owner = MTHCA_CQ_ENTRY_OWNER_HW;
+}
+
+static inline void inc_cons_index(struct mthca_dev *dev, struct mthca_cq *cq,
+				  int nent)
+{
+	u32 doorbell[2];
+
+	doorbell[0] = cpu_to_be32(MTHCA_CQ_DB_INC_CI | cq->cqn);
+	doorbell[1] = cpu_to_be32(nent - 1);
+
+	mthca_write64(doorbell,
+		      dev->kar + MTHCA_CQ_DOORBELL,
+		      MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock));
+}
+
+void mthca_cq_event(struct mthca_dev *dev, u32 cqn)
+{
+	struct mthca_cq *cq;
+
+	spin_lock(&dev->cq_table.lock);
+	cq = mthca_array_get(&dev->cq_table.cq, cqn & (dev->limits.num_cqs - 1));
+	if (cq)
+		atomic_inc(&cq->refcount);
+	spin_unlock(&dev->cq_table.lock);
+
+	if (!cq) {
+		mthca_warn(dev, "Completion event for bogus CQ %08x\n", cqn);
+		return;
+	}
+
+	cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
+
+	if (atomic_dec_and_test(&cq->refcount))
+		wake_up(&cq->wait);
+}
+
+void mthca_cq_clean(struct mthca_dev *dev, u32 cqn, u32 qpn)
+{
+	struct mthca_cq *cq;
+	struct mthca_cqe *cqe;
+	int prod_index;
+	int nfreed = 0;
+
+	spin_lock_irq(&dev->cq_table.lock);
+	cq = mthca_array_get(&dev->cq_table.cq, cqn & (dev->limits.num_cqs - 1));
+	if (cq)
+		atomic_inc(&cq->refcount);
+	spin_unlock_irq(&dev->cq_table.lock);
+
+	if (!cq)
+		return;
+
+	spin_lock_irq(&cq->lock);
+
+	/*
+	 * First we need to find the current producer index, so we
+	 * know where to start cleaning from.  It doesn't matter if HW
+	 * adds new entries after this loop -- the QP we're worried
+	 * about is already in RESET, so the new entries won't come
+	 * from our QP and therefore don't need to be checked.
+	 */
+	for (prod_index = cq->cons_index;
+	     cqe_sw(cq, prod_index & cq->ibcq.cqe);
+	     ++prod_index)
+		if (prod_index == cq->cons_index + cq->ibcq.cqe)
+			break;
+
+	if (0)
+		mthca_dbg(dev, "Cleaning QPN %06x from CQN %06x; ci %d, pi %d\n",
+			  qpn, cqn, cq->cons_index, prod_index);
+
+	/*
+	 * Now sweep backwards through the CQ, removing CQ entries
+	 * that match our QP by copying older entries on top of them.
+	 */
+	while (prod_index > cq->cons_index) {
+		cqe = get_cqe(cq, (prod_index - 1) & cq->ibcq.cqe);
+		if (cqe->my_qpn == cpu_to_be32(qpn))
+			++nfreed;
+		else if (nfreed)
+			memcpy(get_cqe(cq, (prod_index - 1 + nfreed) &
+				       cq->ibcq.cqe),
+			       cqe,
+			       MTHCA_CQ_ENTRY_SIZE);
+		--prod_index;
+	}
+
+	if (nfreed) {
+		wmb();
+		inc_cons_index(dev, cq, nfreed);
+		cq->cons_index = (cq->cons_index + nfreed) & cq->ibcq.cqe;
+	}
+
+	spin_unlock_irq(&cq->lock);
+	if (atomic_dec_and_test(&cq->refcount))
+		wake_up(&cq->wait);
+}
+
+static int handle_error_cqe(struct mthca_dev *dev, struct mthca_cq *cq,
+			    struct mthca_qp *qp, int wqe_index, int is_send,
+			    struct mthca_err_cqe *cqe,
+			    struct ib_wc *entry, int *free_cqe)
+{
+	int err;
+	int dbd;
+	u32 new_wqe;
+
+	if (1 && cqe->syndrome != SYNDROME_WR_FLUSH_ERR) {
+		int j;
+
+		mthca_dbg(dev, "%x/%d: error CQE -> QPN %06x, WQE @ %08x\n",
+			  cq->cqn, cq->cons_index, be32_to_cpu(cqe->my_qpn),
+			  be32_to_cpu(cqe->wqe));
+
+		for (j = 0; j < 8; ++j)
+			printk(KERN_DEBUG "  [%2x] %08x\n",
+			       j * 4, be32_to_cpu(((u32 *) cqe)[j]));
+	}
+
+	/*
+	 * For completions in error, only work request ID, status (and
+	 * freed resource count for RD) have to be set.
+	 */
+	switch (cqe->syndrome) {
+	case SYNDROME_LOCAL_LENGTH_ERR:
+		entry->status = IB_WC_LOC_LEN_ERR;
+		break;
+	case SYNDROME_LOCAL_QP_OP_ERR:
+		entry->status = IB_WC_LOC_QP_OP_ERR;
+		break;
+	case SYNDROME_LOCAL_EEC_OP_ERR:
+		entry->status = IB_WC_LOC_EEC_OP_ERR;
+		break;
+	case SYNDROME_LOCAL_PROT_ERR:
+		entry->status = IB_WC_LOC_PROT_ERR;
+		break;
+	case SYNDROME_WR_FLUSH_ERR:
+		entry->status = IB_WC_WR_FLUSH_ERR;
+		break;
+	case SYNDROME_MW_BIND_ERR:
+		entry->status = IB_WC_MW_BIND_ERR;
+		break;
+	case SYNDROME_BAD_RESP_ERR:
+		entry->status = IB_WC_BAD_RESP_ERR;
+		break;
+	case SYNDROME_LOCAL_ACCESS_ERR:
+		entry->status = IB_WC_LOC_ACCESS_ERR;
+		break;
+	case SYNDROME_REMOTE_INVAL_REQ_ERR:
+		entry->status = IB_WC_REM_INV_REQ_ERR;
+		break;
+	case SYNDROME_REMOTE_ACCESS_ERR:
+		entry->status = IB_WC_REM_ACCESS_ERR;
+		break;
+	case SYNDROME_REMOTE_OP_ERR:
+		entry->status = IB_WC_REM_OP_ERR;
+		break;
+	case SYNDROME_RETRY_EXC_ERR:
+		entry->status = IB_WC_RETRY_EXC_ERR;
+		break;
+	case SYNDROME_RNR_RETRY_EXC_ERR:
+		entry->status = IB_WC_RNR_RETRY_EXC_ERR;
+		break;
+	case SYNDROME_LOCAL_RDD_VIOL_ERR:
+		entry->status = IB_WC_LOC_RDD_VIOL_ERR;
+		break;
+	case SYNDROME_REMOTE_INVAL_RD_REQ_ERR:
+		entry->status = IB_WC_REM_INV_RD_REQ_ERR;
+		break;
+	case SYNDROME_REMOTE_ABORTED_ERR:
+		entry->status = IB_WC_REM_ABORT_ERR;
+		break;
+	case SYNDROME_INVAL_EECN_ERR:
+		entry->status = IB_WC_INV_EECN_ERR;
+		break;
+	case SYNDROME_INVAL_EEC_STATE_ERR:
+		entry->status = IB_WC_INV_EEC_STATE_ERR;
+		break;
+	default:
+		entry->status = IB_WC_GENERAL_ERR;
+		break;
+	}
+
+	err = mthca_free_err_wqe(qp, is_send, wqe_index, &dbd, &new_wqe);
+	if (err)
+		return err;
+
+	/*
+	 * If we're at the end of the WQE chain, or we've used up our
+	 * doorbell count, free the CQE.  Otherwise just update it for
+	 * the next poll operation.
+	 */
+	if (!(new_wqe & cpu_to_be32(0x3f)) || (!cqe->db_cnt && dbd))
+		return 0;
+
+	cqe->db_cnt   = cpu_to_be16(be16_to_cpu(cqe->db_cnt) - dbd);
+	cqe->wqe      = new_wqe;
+	cqe->syndrome = SYNDROME_WR_FLUSH_ERR;
+
+	*free_cqe = 0;
+
+	return 0;
+}
+
+static void dump_cqe(struct mthca_cqe *cqe)
+{
+	int j;
+
+	for (j = 0; j < 8; ++j)
+		printk(KERN_DEBUG "  [%2x] %08x\n",
+		       j * 4, be32_to_cpu(((u32 *) cqe)[j]));
+}
+
+static inline int mthca_poll_one(struct mthca_dev *dev,
+				 struct mthca_cq *cq,
+				 struct mthca_qp **cur_qp,
+				 int *freed,
+				 struct ib_wc *entry)
+{
+	struct mthca_wq *wq;
+	struct mthca_cqe *cqe;
+	int wqe_index;
+	int is_error = 0;
+	int is_send;
+	int free_cqe = 1;
+	int err = 0;
+
+	if (!next_cqe_sw(cq))
+		return -EAGAIN;
+
+	rmb();
+
+	cqe = get_cqe(cq, cq->cons_index);
+
+	if (0) {
+		mthca_dbg(dev, "%x/%d: CQE -> QPN %06x, WQE @ %08x\n",
+			  cq->cqn, cq->cons_index, be32_to_cpu(cqe->my_qpn),
+			  be32_to_cpu(cqe->wqe));
+
+		dump_cqe(cqe);
+	}
+
+	if ((cqe->opcode & MTHCA_ERROR_CQE_OPCODE_MASK) ==
+	    MTHCA_ERROR_CQE_OPCODE_MASK) {
+		is_error = 1;
+		is_send = cqe->opcode & 1;
+	} else
+		is_send = cqe->is_send & 0x80;
+
+	if (!*cur_qp || be32_to_cpu(cqe->my_qpn) != (*cur_qp)->qpn) {
+		if (*cur_qp) {
+			if (*freed) {
+				wmb();
+				inc_cons_index(dev, cq, *freed);
+				*freed = 0;
+			}
+			spin_unlock(&(*cur_qp)->lock);
+			if (atomic_dec_and_test(&(*cur_qp)->refcount))
+				wake_up(&(*cur_qp)->wait);
+		}
+
+		spin_lock(&dev->qp_table.lock);
+		*cur_qp = mthca_array_get(&dev->qp_table.qp,
+					  be32_to_cpu(cqe->my_qpn) &
+					  (dev->limits.num_qps - 1));
+		if (*cur_qp)
+			atomic_inc(&(*cur_qp)->refcount);
+		spin_unlock(&dev->qp_table.lock);
+
+		if (!*cur_qp) {
+			mthca_warn(dev, "CQ entry for unknown QP %06x\n",
+				   be32_to_cpu(cqe->my_qpn) & 0xffffff);
+			err = -EINVAL;
+			goto out;
+		}
+
+		spin_lock(&(*cur_qp)->lock);
+	}
+
+	if (is_send) {
+		wq = &(*cur_qp)->sq;
+		wqe_index = ((be32_to_cpu(cqe->wqe) - (*cur_qp)->send_wqe_offset)
+			     >> wq->wqe_shift);
+		entry->wr_id = (*cur_qp)->wrid[wqe_index +
+					       (*cur_qp)->rq.max];
+	} else {
+		wq = &(*cur_qp)->rq;
+		wqe_index = be32_to_cpu(cqe->wqe) >> wq->wqe_shift;
+		entry->wr_id = (*cur_qp)->wrid[wqe_index];
+	}
+
+	if (wq->last_comp < wqe_index)
+		wq->cur -= wqe_index - wq->last_comp;
+	else
+		wq->cur -= wq->max - wq->last_comp + wqe_index;
+
+	wq->last_comp = wqe_index;
+
+	if (0)
+		mthca_dbg(dev, "%s completion for QP %06x, index %d (nr %d)\n",
+			  is_send ? "Send" : "Receive",
+			  (*cur_qp)->qpn, wqe_index, wq->max);
+
+	if (is_error) {
+		err = handle_error_cqe(dev, cq, *cur_qp, wqe_index, is_send,
+				       (struct mthca_err_cqe *) cqe,
+				       entry, &free_cqe);
+		goto out;
+	}
+
+	if (is_send) {
+		entry->opcode = IB_WC_SEND; /* XXX */
+	} else {
+		entry->byte_len = be32_to_cpu(cqe->byte_cnt);
+		switch (cqe->opcode & 0x1f) {
+		case IB_OPCODE_SEND_LAST_WITH_IMMEDIATE:
+		case IB_OPCODE_SEND_ONLY_WITH_IMMEDIATE:
+			entry->wc_flags = IB_WC_WITH_IMM;
+			entry->imm_data = cqe->imm_etype_pkey_eec;
+			entry->opcode = IB_WC_RECV;
+			break;
+		case IB_OPCODE_RDMA_WRITE_LAST_WITH_IMMEDIATE:
+		case IB_OPCODE_RDMA_WRITE_ONLY_WITH_IMMEDIATE:
+			entry->wc_flags = IB_WC_WITH_IMM;
+			entry->imm_data = cqe->imm_etype_pkey_eec;
+			entry->opcode = IB_WC_RECV_RDMA_WITH_IMM;
+			break;
+		default:
+			entry->wc_flags = 0;
+			entry->opcode = IB_WC_RECV;
+			break;
+		}
+		entry->slid 	   = be16_to_cpu(cqe->rlid);
+		entry->sl   	   = be16_to_cpu(cqe->sl_g_mlpath) >> 12;
+		entry->src_qp 	   = be32_to_cpu(cqe->rqpn) & 0xffffff;
+		entry->dlid_path_bits = be16_to_cpu(cqe->sl_g_mlpath) & 0x7f;
+		entry->pkey_index  = be32_to_cpu(cqe->imm_etype_pkey_eec) >> 16;
+		entry->wc_flags   |= be16_to_cpu(cqe->sl_g_mlpath) & 0x80 ?
+					IB_WC_GRH : 0;
+	}
+
+	entry->status = IB_WC_SUCCESS;
+
+ out:
+	if (free_cqe) {
+		set_cqe_hw(cq, cq->cons_index);
+		++(*freed);
+		cq->cons_index = (cq->cons_index + 1) & cq->ibcq.cqe;
+	}
+
+	return err;
+}
+
+int mthca_poll_cq(struct ib_cq *ibcq, int num_entries,
+		  struct ib_wc *entry)
+{
+	struct mthca_dev *dev = to_mdev(ibcq->device);
+	struct mthca_cq *cq = to_mcq(ibcq);
+	struct mthca_qp *qp = NULL;
+	unsigned long flags;
+	int err = 0;
+	int freed = 0;
+	int npolled;
+
+	spin_lock_irqsave(&cq->lock, flags);
+
+	for (npolled = 0; npolled < num_entries; ++npolled) {
+		err = mthca_poll_one(dev, cq, &qp,
+				     &freed, entry + npolled);
+		if (err)
+			break;
+	}
+
+	if (freed) {
+		wmb();
+		inc_cons_index(dev, cq, freed);
+	}
+
+	if (qp) {
+		spin_unlock(&qp->lock);
+		if (atomic_dec_and_test(&qp->refcount))
+			wake_up(&qp->wait);
+	}
+
+
+	spin_unlock_irqrestore(&cq->lock, flags);
+
+	return err == 0 || err == -EAGAIN ? npolled : err;
+}
+
+void mthca_arm_cq(struct mthca_dev *dev, struct mthca_cq *cq,
+		  int solicited)
+{
+	u32 doorbell[2];
+
+	doorbell[0] =  cpu_to_be32((solicited ?
+				    MTHCA_CQ_DB_REQ_NOT_SOL :
+				    MTHCA_CQ_DB_REQ_NOT)      |
+				   cq->cqn);
+	doorbell[1] = 0xffffffff;
+
+	mthca_write64(doorbell,
+		      dev->kar + MTHCA_CQ_DOORBELL,
+		      MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock));
+}
+
+int mthca_init_cq(struct mthca_dev *dev, int nent,
+		  struct mthca_cq *cq)
+{
+	int size = nent * MTHCA_CQ_ENTRY_SIZE;
+	dma_addr_t t;
+	void *mailbox = NULL;
+	int npages, shift;
+	u64 *dma_list = NULL;
+	struct mthca_cq_context *cq_context;
+	int err = -ENOMEM;
+	u8 status;
+	int i;
+
+	might_sleep();
+
+	mailbox = kmalloc(sizeof (struct mthca_cq_context) + MTHCA_CMD_MAILBOX_EXTRA,
+			  GFP_KERNEL);
+	if (!mailbox)
+		goto err_out;
+
+	cq_context = MAILBOX_ALIGN(mailbox);
+
+	if (size <= MTHCA_MAX_DIRECT_CQ_SIZE) {
+		if (0)
+			mthca_dbg(dev, "Creating direct CQ of size %d\n", size);
+
+		cq->is_direct = 1;
+		npages        = 1;
+		shift         = get_order(size) + PAGE_SHIFT;
+
+		cq->queue.direct.buf = pci_alloc_consistent(dev->pdev,
+							    size, &t);
+		if (!cq->queue.direct.buf)
+			goto err_out;
+
+		pci_unmap_addr_set(&cq->queue.direct, mapping, t);
+
+		memset(cq->queue.direct.buf, 0, size);
+
+		while (t & ((1 << shift) - 1)) {
+			--shift;
+			npages *= 2;
+		}
+
+		dma_list = kmalloc(npages * sizeof *dma_list, GFP_KERNEL);
+		if (!dma_list)
+			goto err_out_free;
+
+		for (i = 0; i < npages; ++i)
+			dma_list[i] = t + i * (1 << shift);
+	} else {
+		cq->is_direct = 0;
+		npages        = (size + PAGE_SIZE - 1) / PAGE_SIZE;
+		shift         = PAGE_SHIFT;
+
+		if (0)
+			mthca_dbg(dev, "Creating indirect CQ with %d pages\n", npages);
+
+		dma_list = kmalloc(npages * sizeof *dma_list, GFP_KERNEL);
+		if (!dma_list)
+			goto err_out;
+
+		cq->queue.page_list = kmalloc(npages * sizeof *cq->queue.page_list,
+					      GFP_KERNEL);
+		if (!cq->queue.page_list)
+			goto err_out;
+
+		for (i = 0; i < npages; ++i)
+			cq->queue.page_list[i].buf = NULL;
+
+		for (i = 0; i < npages; ++i) {
+			cq->queue.page_list[i].buf =
+				pci_alloc_consistent(dev->pdev, PAGE_SIZE, &t);
+			if (!cq->queue.page_list[i].buf)
+				goto err_out_free;
+
+			dma_list[i] = t;
+			pci_unmap_addr_set(&cq->queue.page_list[i], mapping, t);
+
+			memset(cq->queue.page_list[i].buf, 0, PAGE_SIZE);
+		}
+	}
+
+	for (i = 0; i < nent; ++i)
+		set_cqe_hw(cq, i);
+
+	cq->cqn = mthca_alloc(&dev->cq_table.alloc);
+	if (cq->cqn == -1)
+		goto err_out_free;
+
+	err = mthca_mr_alloc_phys(dev, dev->driver_pd.pd_num,
+				  dma_list, shift, npages,
+				  0, size,
+				  MTHCA_MPT_FLAG_LOCAL_WRITE |
+				  MTHCA_MPT_FLAG_LOCAL_READ,
+				  &cq->mr);
+	if (err)
+		goto err_out_free_cq;
+
+	spin_lock_init(&cq->lock);
+	atomic_set(&cq->refcount, 1);
+	init_waitqueue_head(&cq->wait);
+
+	memset(cq_context, 0, sizeof *cq_context);
+	cq_context->flags           = cpu_to_be32(MTHCA_CQ_STATUS_OK      |
+						  MTHCA_CQ_STATE_DISARMED |
+						  MTHCA_CQ_FLAG_TR);
+	cq_context->start           = cpu_to_be64(0);
+	cq_context->logsize_usrpage = cpu_to_be32((ffs(nent) - 1) << 24 |
+						  MTHCA_KAR_PAGE);
+	cq_context->error_eqn       = cpu_to_be32(dev->eq_table.eq[MTHCA_EQ_ASYNC].eqn);
+	cq_context->comp_eqn        = cpu_to_be32(dev->eq_table.eq[MTHCA_EQ_COMP].eqn);
+	cq_context->pd              = cpu_to_be32(dev->driver_pd.pd_num);
+	cq_context->lkey            = cpu_to_be32(cq->mr.ibmr.lkey);
+	cq_context->cqn             = cpu_to_be32(cq->cqn);
+
+	err = mthca_SW2HW_CQ(dev, cq_context, cq->cqn, &status);
+	if (err) {
+		mthca_warn(dev, "SW2HW_CQ failed (%d)\n", err);
+		goto err_out_free_mr;
+	}
+
+	if (status) {
+		mthca_warn(dev, "SW2HW_CQ returned status 0x%02x\n",
+			   status);
+		err = -EINVAL;
+		goto err_out_free_mr;
+	}
+
+	spin_lock_irq(&dev->cq_table.lock);
+	if (mthca_array_set(&dev->cq_table.cq,
+			    cq->cqn & (dev->limits.num_cqs - 1),
+			    cq)) {
+		spin_unlock_irq(&dev->cq_table.lock);
+		goto err_out_free_mr;
+	}
+	spin_unlock_irq(&dev->cq_table.lock);
+
+	cq->cons_index = 0;
+
+	kfree(dma_list);
+	kfree(mailbox);
+
+	return 0;
+
+ err_out_free_mr:
+	mthca_free_mr(dev, &cq->mr);
+
+ err_out_free_cq:
+	mthca_free(&dev->cq_table.alloc, cq->cqn);
+
+ err_out_free:
+	if (cq->is_direct)
+		pci_free_consistent(dev->pdev, size,
+				    cq->queue.direct.buf,
+				    pci_unmap_addr(&cq->queue.direct, mapping));
+	else {
+		for (i = 0; i < npages; ++i)
+			if (cq->queue.page_list[i].buf)
+				pci_free_consistent(dev->pdev, PAGE_SIZE,
+						    cq->queue.page_list[i].buf,
+						    pci_unmap_addr(&cq->queue.page_list[i],
+								   mapping));
+
+		kfree(cq->queue.page_list);
+	}
+
+ err_out:
+	kfree(dma_list);
+	kfree(mailbox);
+
+	return err;
+}
+
+void mthca_free_cq(struct mthca_dev *dev,
+		   struct mthca_cq *cq)
+{
+	void *mailbox;
+	int err;
+	u8 status;
+
+	might_sleep();
+
+	mailbox = kmalloc(sizeof (struct mthca_cq_context) + MTHCA_CMD_MAILBOX_EXTRA,
+			  GFP_KERNEL);
+	if (!mailbox) {
+		mthca_warn(dev, "No memory for mailbox to free CQ.\n");
+		return;
+	}
+
+	err = mthca_HW2SW_CQ(dev, MAILBOX_ALIGN(mailbox), cq->cqn, &status);
+	if (err)
+		mthca_warn(dev, "HW2SW_CQ failed (%d)\n", err);
+	else if (status)
+		mthca_warn(dev, "HW2SW_CQ returned status 0x%02x\n",
+			   status);
+
+	if (0) {
+		u32 *ctx = MAILBOX_ALIGN(mailbox);
+		int j;
+
+		printk(KERN_ERR "context for CQN %x\n", cq->cqn);
+		for (j = 0; j < 16; ++j)
+			printk(KERN_ERR "[%2x] %08x\n", j * 4, be32_to_cpu(ctx[j]));
+	}
+
+	spin_lock_irq(&dev->cq_table.lock);
+	mthca_array_clear(&dev->cq_table.cq,
+			  cq->cqn & (dev->limits.num_cqs - 1));
+	spin_unlock_irq(&dev->cq_table.lock);
+
+	atomic_dec(&cq->refcount);
+	wait_event(cq->wait, !atomic_read(&cq->refcount));
+
+	mthca_free_mr(dev, &cq->mr);
+
+	if (cq->is_direct)
+		pci_free_consistent(dev->pdev,
+				    (cq->ibcq.cqe + 1) * MTHCA_CQ_ENTRY_SIZE,
+				    cq->queue.direct.buf,
+				    pci_unmap_addr(&cq->queue.direct,
+						   mapping));
+	else {
+		int i;
+
+		for (i = 0;
+		     i < ((cq->ibcq.cqe + 1) * MTHCA_CQ_ENTRY_SIZE + PAGE_SIZE - 1) /
+			     PAGE_SIZE;
+		     ++i)
+			pci_free_consistent(dev->pdev, PAGE_SIZE,
+					    cq->queue.page_list[i].buf,
+					    pci_unmap_addr(&cq->queue.page_list[i],
+							   mapping));
+
+		kfree(cq->queue.page_list);
+	}
+
+	mthca_free(&dev->cq_table.alloc, cq->cqn);
+	kfree(mailbox);
+}
+
+int __devinit mthca_init_cq_table(struct mthca_dev *dev)
+{
+	int err;
+
+	spin_lock_init(&dev->cq_table.lock);
+
+	err = mthca_alloc_init(&dev->cq_table.alloc,
+			       dev->limits.num_cqs,
+			       (1 << 24) - 1,
+			       dev->limits.reserved_cqs);
+	if (err)
+		return err;
+
+	err = mthca_array_init(&dev->cq_table.cq,
+			       dev->limits.num_cqs);
+	if (err)
+		mthca_alloc_cleanup(&dev->cq_table.alloc);
+
+	return err;
+}
+
+void __devexit mthca_cleanup_cq_table(struct mthca_dev *dev)
+{
+	mthca_array_cleanup(&dev->cq_table.cq, dev->limits.num_cqs);
+	mthca_alloc_cleanup(&dev->cq_table.alloc);
+}
drivers/infiniband/hw/mthca/mthca_qp.c new file +1536 -0

1,536 lines not on this page · raw message

#49 lore raw

PATCH v5 15/24 Add Mellanox HCA low-level driver (last bits)

Add code for remaining InfiniBand objects (address vectors, multicast groups, memory regions and protection domains)

drivers/infiniband/hw/mthca/mthca_av.c new file +219 -0
@@ -0,0 +1,219 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: mthca_av.c 1349 2004-12-16 21:09:43Z roland $
+ */
+
+#include <linux/init.h>
+
+#include <ib_verbs.h>
+#include <ib_cache.h>
+
+#include "mthca_dev.h"
+
+struct mthca_av {
+	u32 port_pd;
+	u8  reserved1;
+	u8  g_slid;
+	u16 dlid;
+	u8  reserved2;
+	u8  gid_index;
+	u8  msg_sr;
+	u8  hop_limit;
+	u32 sl_tclass_flowlabel;
+	u32 dgid[4];
+};
+
+int mthca_create_ah(struct mthca_dev *dev,
+		    struct mthca_pd *pd,
+		    struct ib_ah_attr *ah_attr,
+		    struct mthca_ah *ah)
+{
+	u32 index = -1;
+	struct mthca_av *av = NULL;
+
+	ah->on_hca = 0;
+
+	if (!atomic_read(&pd->sqp_count) &&
+	    !(dev->mthca_flags & MTHCA_FLAG_DDR_HIDDEN)) {
+		index = mthca_alloc(&dev->av_table.alloc);
+
+		/* fall back to allocate in host memory */
+		if (index == -1)
+			goto host_alloc;
+
+		av = kmalloc(sizeof *av, GFP_KERNEL);
+		if (!av)
+			goto host_alloc;
+
+		ah->on_hca = 1;
+		ah->avdma  = dev->av_table.ddr_av_base +
+			index * MTHCA_AV_SIZE;
+	}
+
+ host_alloc:
+	if (!ah->on_hca) {
+		ah->av = pci_pool_alloc(dev->av_table.pool,
+					SLAB_KERNEL, &ah->avdma);
+		if (!ah->av)
+			return -ENOMEM;
+
+		av = ah->av;
+	}
+
+	ah->key = pd->ntmr.ibmr.lkey;
+
+	memset(av, 0, MTHCA_AV_SIZE);
+
+	av->port_pd = cpu_to_be32(pd->pd_num | (ah_attr->port_num << 24));
+	av->g_slid  = ah_attr->src_path_bits;
+	av->dlid    = cpu_to_be16(ah_attr->dlid);
+	av->msg_sr  = (3 << 4) | /* 2K message */
+		ah_attr->static_rate;
+	av->sl_tclass_flowlabel = cpu_to_be32(ah_attr->sl << 28);
+	if (ah_attr->ah_flags & IB_AH_GRH) {
+		av->g_slid |= 0x80;
+		av->gid_index = (ah_attr->port_num - 1) * dev->limits.gid_table_len +
+			ah_attr->grh.sgid_index;
+		av->hop_limit = ah_attr->grh.hop_limit;
+		av->sl_tclass_flowlabel |=
+			cpu_to_be32((ah_attr->grh.traffic_class << 20) |
+				    ah_attr->grh.flow_label);
+		memcpy(av->dgid, ah_attr->grh.dgid.raw, 16);
+	} else {
+		/* Arbel workaround -- low byte of GID must be 2 */
+		av->dgid[3] = cpu_to_be32(2);
+	}
+
+	if (0) {
+		int j;
+
+		mthca_dbg(dev, "Created UDAV at %p/%08lx:\n",
+			  av, (unsigned long) ah->avdma);
+		for (j = 0; j < 8; ++j)
+			printk(KERN_DEBUG "  [%2x] %08x\n",
+			       j * 4, be32_to_cpu(((u32 *) av)[j]));
+	}
+
+	if (ah->on_hca) {
+		memcpy_toio(dev->av_table.av_map + index * MTHCA_AV_SIZE,
+			    av, MTHCA_AV_SIZE);
+		kfree(av);
+	}
+
+	return 0;
+}
+
+int mthca_destroy_ah(struct mthca_dev *dev, struct mthca_ah *ah)
+{
+	if (ah->on_hca)
+		mthca_free(&dev->av_table.alloc,
+ 			   (ah->avdma - dev->av_table.ddr_av_base) /
+			   MTHCA_AV_SIZE);
+	else
+		pci_pool_free(dev->av_table.pool, ah->av, ah->avdma);
+
+	return 0;
+}
+
+int mthca_read_ah(struct mthca_dev *dev, struct mthca_ah *ah,
+		  struct ib_ud_header *header)
+{
+	if (ah->on_hca)
+		return -EINVAL;
+
+	header->lrh.service_level   = be32_to_cpu(ah->av->sl_tclass_flowlabel) >> 28;
+	header->lrh.destination_lid = ah->av->dlid;
+	header->lrh.source_lid      = ah->av->g_slid & 0x7f;
+	if (ah->av->g_slid & 0x80) {
+		header->grh_present = 1;
+		header->grh.traffic_class =
+			(be32_to_cpu(ah->av->sl_tclass_flowlabel) >> 20) & 0xff;
+		header->grh.flow_label    =
+			ah->av->sl_tclass_flowlabel & cpu_to_be32(0xfffff);
+		ib_cached_gid_get(&dev->ib_dev,
+				  be32_to_cpu(ah->av->port_pd) >> 24,
+				  ah->av->gid_index,
+				  &header->grh.source_gid);
+		memcpy(header->grh.destination_gid.raw,
+		       ah->av->dgid, 16);
+	} else {
+		header->grh_present = 0;
+	}
+
+	return 0;
+}
+
+int __devinit mthca_init_av_table(struct mthca_dev *dev)
+{
+	int err;
+
+	err = mthca_alloc_init(&dev->av_table.alloc,
+			       dev->av_table.num_ddr_avs,
+			       dev->av_table.num_ddr_avs - 1,
+			       0);
+	if (err)
+		return err;
+
+	dev->av_table.pool = pci_pool_create("mthca_av", dev->pdev,
+					     MTHCA_AV_SIZE,
+					     MTHCA_AV_SIZE, 0);
+	if (!dev->av_table.pool)
+		goto out_free_alloc;
+
+	if (!(dev->mthca_flags & MTHCA_FLAG_DDR_HIDDEN)) {
+		dev->av_table.av_map = ioremap(pci_resource_start(dev->pdev, 4) +
+					       dev->av_table.ddr_av_base -
+					       dev->ddr_start,
+					       dev->av_table.num_ddr_avs *
+					       MTHCA_AV_SIZE);
+		if (!dev->av_table.av_map)
+			goto out_free_pool;
+	} else
+		dev->av_table.av_map = NULL;
+
+	return 0;
+
+ out_free_pool:
+	pci_pool_destroy(dev->av_table.pool);
+
+ out_free_alloc:
+	mthca_alloc_cleanup(&dev->av_table.alloc);
+	return -ENOMEM;
+}
+
+void __devexit mthca_cleanup_av_table(struct mthca_dev *dev)
+{
+	if (dev->av_table.av_map)
+		iounmap(dev->av_table.av_map);
+	pci_pool_destroy(dev->av_table.pool);
+	mthca_alloc_cleanup(&dev->av_table.alloc);
+}
drivers/infiniband/hw/mthca/mthca_mcg.c new file +376 -0

376 lines not on this page · raw message

drivers/infiniband/hw/mthca/mthca_mr.c new file +396 -0

396 lines not on this page · raw message

drivers/infiniband/hw/mthca/mthca_pd.c new file +80 -0

80 lines not on this page · raw message

#50 lore raw

PATCH v5 16/24 Add Mellanox HCA low-level driver (MAD)

Add MAD (management datagram) code for Mellanox HCA driver.

drivers/infiniband/hw/mthca/mthca_mad.c new file +320 -0
@@ -0,0 +1,320 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: mthca_mad.c 1349 2004-12-16 21:09:43Z roland $
+ */
+
+#include <ib_verbs.h>
+#include <ib_mad.h>
+#include <ib_smi.h>
+
+#include "mthca_dev.h"
+#include "mthca_cmd.h"
+
+enum {
+	MTHCA_VENDOR_CLASS1 = 0x9,
+	MTHCA_VENDOR_CLASS2 = 0xa
+};
+
+struct mthca_trap_mad {
+	struct ib_mad *mad;
+	DECLARE_PCI_UNMAP_ADDR(mapping)
+};
+
+static void update_sm_ah(struct mthca_dev *dev,
+			 u8 port_num, u16 lid, u8 sl)
+{
+	struct ib_ah *new_ah;
+	struct ib_ah_attr ah_attr;
+	unsigned long flags;
+
+	if (!dev->send_agent[port_num - 1][0])
+		return;
+
+	memset(&ah_attr, 0, sizeof ah_attr);
+	ah_attr.dlid     = lid;
+	ah_attr.sl       = sl;
+	ah_attr.port_num = port_num;
+
+	new_ah = ib_create_ah(dev->send_agent[port_num - 1][0]->qp->pd,
+			      &ah_attr);
+	if (IS_ERR(new_ah))
+		return;
+
+	spin_lock_irqsave(&dev->sm_lock, flags);
+	if (dev->sm_ah[port_num - 1])
+		ib_destroy_ah(dev->sm_ah[port_num - 1]);
+	dev->sm_ah[port_num - 1] = new_ah;
+	spin_unlock_irqrestore(&dev->sm_lock, flags);
+}
+
+/*
+ * Snoop SM MADs for port info and P_Key table sets, so we can
+ * synthesize LID change and P_Key change events.
+ */
+static void smp_snoop(struct ib_device *ibdev,
+		      u8 port_num,
+		      struct ib_mad *mad)
+{
+	struct ib_event event;
+
+	if ((mad->mad_hdr.mgmt_class  == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
+	     mad->mad_hdr.mgmt_class  == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) &&
+	    mad->mad_hdr.method     == IB_MGMT_METHOD_SET) {
+		if (mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO) {
+			update_sm_ah(to_mdev(ibdev), port_num,
+				     be16_to_cpup((__be16 *) (mad->data + 58)),
+				     (*(u8 *) (mad->data + 76)) & 0xf);
+
+			event.device           = ibdev;
+			event.event            = IB_EVENT_LID_CHANGE;
+			event.element.port_num = port_num;
+			ib_dispatch_event(&event);
+		}
+
+		if (mad->mad_hdr.attr_id == IB_SMP_ATTR_PKEY_TABLE) {
+			event.device           = ibdev;
+			event.event            = IB_EVENT_PKEY_CHANGE;
+			event.element.port_num = port_num;
+			ib_dispatch_event(&event);
+		}
+	}
+}
+
+static void forward_trap(struct mthca_dev *dev,
+			 u8 port_num,
+			 struct ib_mad *mad)
+{
+	int qpn = mad->mad_hdr.mgmt_class != IB_MGMT_CLASS_SUBN_LID_ROUTED;
+	struct mthca_trap_mad *tmad;
+	struct ib_sge      gather_list;
+	struct ib_send_wr *bad_wr, wr = {
+		.opcode      = IB_WR_SEND,
+		.sg_list     = &gather_list,
+		.num_sge     = 1,
+		.send_flags  = IB_SEND_SIGNALED,
+		.wr	     = {
+			 .ud = {
+				 .remote_qpn  = qpn,
+				 .remote_qkey = qpn ? IB_QP1_QKEY : 0,
+				 .timeout_ms  = 0
+			 }
+		 }
+	};
+	struct ib_mad_agent *agent = dev->send_agent[port_num - 1][qpn];
+	int ret;
+	unsigned long flags;
+
+	if (agent) {
+		tmad = kmalloc(sizeof *tmad, GFP_KERNEL);
+		if (!tmad)
+			return;
+
+		tmad->mad = kmalloc(sizeof *tmad->mad, GFP_KERNEL);
+		if (!tmad->mad) {
+			kfree(tmad);
+			return;
+		}
+
+		memcpy(tmad->mad, mad, sizeof *mad);
+
+		wr.wr.ud.mad_hdr = &tmad->mad->mad_hdr;
+		wr.wr_id         = (unsigned long) tmad;
+
+		gather_list.addr   = dma_map_single(agent->device->dma_device,
+						    tmad->mad,
+						    sizeof *tmad->mad,
+						    DMA_TO_DEVICE);
+		gather_list.length = sizeof *tmad->mad;
+		gather_list.lkey   = to_mpd(agent->qp->pd)->ntmr.ibmr.lkey;
+		pci_unmap_addr_set(tmad, mapping, gather_list.addr);
+
+		/*
+		 * We rely here on the fact that MLX QPs don't use the
+		 * address handle after the send is posted (this is
+		 * wrong following the IB spec strictly, but we know
+		 * it's OK for our devices).
+		 */
+		spin_lock_irqsave(&dev->sm_lock, flags);
+		wr.wr.ud.ah      = dev->sm_ah[port_num - 1];
+		if (wr.wr.ud.ah)
+			ret = ib_post_send_mad(agent, &wr, &bad_wr);
+		else
+			ret = -EINVAL;
+		spin_unlock_irqrestore(&dev->sm_lock, flags);
+
+		if (ret) {
+			dma_unmap_single(agent->device->dma_device,
+					 pci_unmap_addr(tmad, mapping),
+					 sizeof *tmad->mad,
+					 DMA_TO_DEVICE);
+			kfree(tmad->mad);
+			kfree(tmad);
+		}
+	}
+}
+
+int mthca_process_mad(struct ib_device *ibdev,
+		      int mad_flags,
+		      u8 port_num,
+		      u16 slid,
+		      struct ib_mad *in_mad,
+		      struct ib_mad *out_mad)
+{
+	int err;
+	u8 status;
+
+	/* Forward locally generated traps to the SM */
+	if (in_mad->mad_hdr.method == IB_MGMT_METHOD_TRAP &&
+	    slid == 0) {
+		forward_trap(to_mdev(ibdev), port_num, in_mad);
+		return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
+	}
+
+	/*
+	 * Only handle SM gets, sets and trap represses for SM class
+	 *
+	 * Only handle PMA and Mellanox vendor-specific class gets and
+	 * sets for other classes.
+	 */
+	if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
+	    in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
+		if (in_mad->mad_hdr.method   != IB_MGMT_METHOD_GET &&
+		    in_mad->mad_hdr.method   != IB_MGMT_METHOD_SET &&
+		    in_mad->mad_hdr.method   != IB_MGMT_METHOD_TRAP_REPRESS)
+			return IB_MAD_RESULT_SUCCESS;
+
+		/*
+		 * Don't process SMInfo queries or vendor-specific
+		 * MADs -- the SMA can't handle them.
+		 */
+		if (in_mad->mad_hdr.attr_id == IB_SMP_ATTR_SM_INFO ||
+		    ((in_mad->mad_hdr.attr_id & IB_SMP_ATTR_VENDOR_MASK) ==
+		     IB_SMP_ATTR_VENDOR_MASK))
+			return IB_MAD_RESULT_SUCCESS;
+	} else if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_PERF_MGMT ||
+		   in_mad->mad_hdr.mgmt_class == MTHCA_VENDOR_CLASS1     ||
+		   in_mad->mad_hdr.mgmt_class == MTHCA_VENDOR_CLASS2) {
+		if (in_mad->mad_hdr.method  != IB_MGMT_METHOD_GET &&
+		    in_mad->mad_hdr.method  != IB_MGMT_METHOD_SET)
+			return IB_MAD_RESULT_SUCCESS;
+	} else
+		return IB_MAD_RESULT_SUCCESS;
+
+	err = mthca_MAD_IFC(to_mdev(ibdev),
+			    !!(mad_flags & IB_MAD_IGNORE_MKEY),
+			    port_num, in_mad, out_mad,
+			    &status);
+	if (err) {
+		mthca_err(to_mdev(ibdev), "MAD_IFC failed\n");
+		return IB_MAD_RESULT_FAILURE;
+	}
+	if (status == MTHCA_CMD_STAT_BAD_PKT)
+		return IB_MAD_RESULT_SUCCESS;
+	if (status) {
+		mthca_err(to_mdev(ibdev), "MAD_IFC returned status %02x\n",
+			  status);
+		return IB_MAD_RESULT_FAILURE;
+	}
+
+	if (!out_mad->mad_hdr.status)
+		smp_snoop(ibdev, port_num, in_mad);
+
+	/* set return bit in status of directed route responses */
+	if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
+		out_mad->mad_hdr.status |= cpu_to_be16(1 << 15);
+
+	if (in_mad->mad_hdr.method == IB_MGMT_METHOD_TRAP_REPRESS)
+		/* no response for trap repress */
+		return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
+
+	return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
+}
+
+static void send_handler(struct ib_mad_agent *agent,
+			 struct ib_mad_send_wc *mad_send_wc)
+{
+	struct mthca_trap_mad *tmad =
+		(void *) (unsigned long) mad_send_wc->wr_id;
+
+	dma_unmap_single(agent->device->dma_device,
+			 pci_unmap_addr(tmad, mapping),
+			 sizeof *tmad->mad,
+			 DMA_TO_DEVICE);
+	kfree(tmad->mad);
+	kfree(tmad);
+}
+
+int mthca_create_agents(struct mthca_dev *dev)
+{
+	struct ib_mad_agent *agent;
+	int p, q;
+
+	spin_lock_init(&dev->sm_lock);
+
+	for (p = 0; p < dev->limits.num_ports; ++p)
+		for (q = 0; q <= 1; ++q) {
+			agent = ib_register_mad_agent(&dev->ib_dev, p + 1,
+						      q ? IB_QPT_GSI : IB_QPT_SMI,
+						      NULL, 0, send_handler,
+						      NULL, NULL);
+			if (IS_ERR(agent))
+				goto err;
+			dev->send_agent[p][q] = agent;
+		}
+
+	return 0;
+
+err:
+	for (p = 0; p < dev->limits.num_ports; ++p)
+		for (q = 0; q <= 1; ++q)
+			if (dev->send_agent[p][q])
+				ib_unregister_mad_agent(dev->send_agent[p][q]);
+
+	return PTR_ERR(agent);
+}
+
+void mthca_free_agents(struct mthca_dev *dev)
+{
+	struct ib_mad_agent *agent;
+	int p, q;
+
+	for (p = 0; p < dev->limits.num_ports; ++p) {
+		for (q = 0; q <= 1; ++q) {
+			agent = dev->send_agent[p][q];
+			dev->send_agent[p][q] = NULL;
+			ib_unregister_mad_agent(agent);
+		}
+
+		if (dev->sm_ah[p])
+			ib_destroy_ah(dev->sm_ah[p]);
+	}
+}
#51 lore raw

PATCH v5 17/24 IPoIB IPv4 multicast

Add ip_ib_mc_map() to convert IPv4 multicast addresses to IPoIB hardware addresses. Also add <linux/if_infiniband.h> so INFINIBAND_ALEN has a home.

The mapping for multicast addresses is described in
http://www.ietf.org/internet-drafts/draft-ietf-ipoib-ip-over-infiniband-08.txt

include/linux/if_infiniband.h new file +29 -0
@@ -0,0 +1,29 @@
+/*
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available at
+ * <http://www.fsf.org/copyleft/gpl.html>, or the OpenIB.org BSD
+ * license, available in the LICENSE.TXT file accompanying this
+ * software.  These details are also available at
+ * <http://openib.org/license.html>.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * $Id$
+ */
+
+#ifndef _LINUX_IF_INFINIBAND_H
+#define _LINUX_IF_INFINIBAND_H
+
+#define INFINIBAND_ALEN		20	/* Octets in IPoIB HW addr	*/
+
+#endif /* _LINUX_IF_INFINIBAND_H */
include/net/ip.h +33 -0
@@ -229,6 +229,39 @@
 	buf[3]=addr&0x7F;
 }
 
+/*
+ *	Map a multicast IP onto multicast MAC for type IP-over-InfiniBand.
+ *	Leave P_Key as 0 to be filled in by driver.
+ */
+
+static inline void ip_ib_mc_map(u32 addr, char *buf)
+{
+	buf[0]  = 0;		/* Reserved */
+	buf[1]  = 0xff;		/* Multicast QPN */
+	buf[2]  = 0xff;
+	buf[3]  = 0xff;
+	addr    = ntohl(addr);
+	buf[4]  = 0xff;
+	buf[5]  = 0x12;		/* link local scope */
+	buf[6]  = 0x40;		/* IPv4 signature */
+	buf[7]  = 0x1b;
+	buf[8]  = 0;		/* P_Key */
+	buf[9]  = 0;
+	buf[10] = 0;
+	buf[11] = 0;
+	buf[12] = 0;
+	buf[13] = 0;
+	buf[14] = 0;
+	buf[15] = 0;
+	buf[19] = addr & 0xff;
+	addr  >>= 8;
+	buf[18] = addr & 0xff;
+	addr  >>= 8;
+	buf[17] = addr & 0xff;
+	addr  >>= 8;
+	buf[16] = addr & 0x0f;
+}
+
 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 #include <linux/ipv6.h>
 #endif
net/ipv4/arp.c +3 -0
@@ -213,6 +213,9 @@
 	case ARPHRD_IEEE802_TR:
 		ip_tr_mc_map(addr, haddr);
 		return 0;
+	case ARPHRD_INFINIBAND:
+		ip_ib_mc_map(addr, haddr);
+		return 0;
 	default:
 		if (dir) {
 			memcpy(haddr, dev->broadcast, dev->addr_len);
#52 lore raw

PATCH v5 18/24 IPoIB IPv6 support

Add ipv6_ib_mc_map() to convert IPv6 multicast addresses to IPoIB hardware addresses, and add support for autoconfiguration for devices with type ARPHRD_INFINIBAND.

The mapping for multicast addresses is described in
http://www.ietf.org/internet-drafts/draft-ietf-ipoib-ip-over-infiniband-08.txt

include/net/if_inet6.h +15 -0
@@ -266,5 +266,20 @@
 {
 	buf[0] = 0x00;
 }
+
+static inline void ipv6_ib_mc_map(struct in6_addr *addr, char *buf)
+{
+	buf[0]  = 0;		/* Reserved */
+	buf[1]  = 0xff;		/* Multicast QPN */
+	buf[2]  = 0xff;
+	buf[3]  = 0xff;
+	buf[4]  = 0xff;
+	buf[5]  = 0x12;		/* link local scope */
+	buf[6]  = 0x60;		/* IPv6 signature */
+	buf[7]  = 0x1b;
+	buf[8]  = 0;		/* P_Key */
+	buf[9]  = 0;
+	memcpy(buf + 10, addr->s6_addr + 6, 10);
+}
 #endif
 #endif
net/ipv6/addrconf.c +9 -1
@@ -48,6 +48,7 @@
 #include <linux/netdevice.h>
 #include <linux/if_arp.h>
 #include <linux/if_arcnet.h>
+#include <linux/if_infiniband.h>
 #include <linux/route.h>
 #include <linux/inetdevice.h>
 #include <linux/init.h>
@@ -1095,6 +1096,12 @@
 		memset(eui, 0, 7);
 		eui[7] = *(u8*)dev->dev_addr;
 		return 0;
+	case ARPHRD_INFINIBAND:
+		if (dev->addr_len != INFINIBAND_ALEN)
+			return -1;
+		memcpy(eui, dev->dev_addr + 12, 8);
+		eui[0] |= 2;
+		return 0;
 	}
 	return -1;
 }
@@ -1794,7 +1801,8 @@
 	if ((dev->type != ARPHRD_ETHER) && 
 	    (dev->type != ARPHRD_FDDI) &&
 	    (dev->type != ARPHRD_IEEE802_TR) &&
-	    (dev->type != ARPHRD_ARCNET)) {
+	    (dev->type != ARPHRD_ARCNET) &&
+	    (dev->type != ARPHRD_INFINIBAND)) {
 		/* Alas, we support only Ethernet autoconfiguration. */
 		return;
 	}
net/ipv6/ndisc.c +3 -0
@@ -260,6 +260,9 @@
 	case ARPHRD_ARCNET:
 		ipv6_arcnet_mc_map(addr, buf);
 		return 0;
+	case ARPHRD_INFINIBAND:
+		ipv6_ib_mc_map(addr, buf);
+		return 0;
 	default:
 		if (dir) {
 			memcpy(buf, dev->broadcast, dev->addr_len);
#53 lore raw

PATCH v5 19/24 Add IPoIB (IP-over-InfiniBand) driver

Add a driver that implements the (IPoIB) IP-over-InfiniBand protocol. This is a network device driver of type ARPHRD_INFINIBAND (and addr_len INFINIBAND_ALEN bytes).

The ARP/ND implementation for this driver is not completely straightforward, because InfiniBand requires an additional path lookup be performed (through an IB-specific mechanism) after a remote hardware address has been resolved. We are very open to suggestions of a better way to handle this than the current implementation.

Although IB has a special multicast group join mode intended to support IP multicast routing (non member join), no means to identify different multicast styles has yet been determined, so all joins by the driver are currently full member joins. We are looking for guidance in how to solve this.

The IPoIB protocol/encapsulation is described in the Internet-Drafts
http://www.ietf.org/internet-drafts/draft-ietf-ipoib-architecture-04.txt
http://www.ietf.org/internet-drafts/draft-ietf-ipoib-ip-over-infiniband-08.txt

drivers/infiniband/Kconfig +2 -0
@@ -9,4 +9,6 @@
 
 source "drivers/infiniband/hw/mthca/Kconfig"
 
+source "drivers/infiniband/ulp/ipoib/Kconfig"
+
 endmenu
drivers/infiniband/Makefile +1 -0
@@ -1,2 +1,3 @@
 obj-$(CONFIG_INFINIBAND)		+= core/
 obj-$(CONFIG_INFINIBAND_MTHCA)		+= hw/mthca/
+obj-$(CONFIG_INFINIBAND_IPOIB)		+= ulp/ipoib/
drivers/infiniband/ulp/ipoib/Kconfig new file +33 -0
@@ -0,0 +1,33 @@
+config INFINIBAND_IPOIB
+	tristate "IP-over-InfiniBand"
+	depends on INFINIBAND && NETDEVICES && INET
+	---help---
+	  Support for the IP-over-InfiniBand protocol (IPoIB). This
+	  transports IP packets over InfiniBand so you can use your IB
+	  device as a fancy NIC.
+
+	  The IPoIB protocol is defined by the IETF ipoib working
+	  group: <http://www.ietf.org/html.charters/ipoib-charter.html>.
+
+config INFINIBAND_IPOIB_DEBUG
+	bool "IP-over-InfiniBand debugging"
+	depends on INFINIBAND_IPOIB
+	---help---
+	  This option causes debugging code to be compiled into the
+	  IPoIB driver.  The output can be turned on via the
+	  debug_level and mcast_debug_level module parameters (which
+	  can also be set after the driver is loaded through sysfs).
+
+	  This option also creates an "ipoib_debugfs," which can be
+	  mounted to expose debugging information about IB multicast
+	  groups used by the IPoIB driver.
+
+config INFINIBAND_IPOIB_DEBUG_DATA
+	bool "IP-over-InfiniBand data path debugging"
+	depends on INFINIBAND_IPOIB_DEBUG
+	---help---
+	  This option compiles debugging code into the the data path
+	  of the IPoIB driver.  The output can be turned on via the
+	  data_debug_level module parameter; however, even with output
+	  turned off, this debugging code will have some performance
+	  impact.
drivers/infiniband/ulp/ipoib/Makefile new file +11 -0
@@ -0,0 +1,11 @@
+EXTRA_CFLAGS += -Idrivers/infiniband/include
+
+obj-$(CONFIG_INFINIBAND_IPOIB)			+= ib_ipoib.o
+
+ib_ipoib-y					:= ipoib_main.o \
+						   ipoib_ib.o \
+						   ipoib_multicast.o \
+						   ipoib_verbs.o \
+						   ipoib_vlan.o
+ib_ipoib-$(CONFIG_INFINIBAND_IPOIB_DEBUG)	+= ipoib_fs.o
+
drivers/infiniband/ulp/ipoib/ipoib.h new file +350 -0

350 lines not on this page · raw message

drivers/infiniband/ulp/ipoib/ipoib_fs.c new file +287 -0

287 lines not on this page · raw message

drivers/infiniband/ulp/ipoib/ipoib_ib.c new file +678 -0

678 lines not on this page · raw message

drivers/infiniband/ulp/ipoib/ipoib_main.c new file +1079 -0

1,079 lines not on this page · raw message

drivers/infiniband/ulp/ipoib/ipoib_verbs.c new file +254 -0

254 lines not on this page · raw message

#54 lore raw

PATCH v5 20/24 Add IPoIB multicast & partition code

Add functions for handling IPoIB multicast and multiple partitions.

drivers/infiniband/ulp/ipoib/ipoib_multicast.c new file +981 -0
@@ -0,0 +1,981 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: ipoib_multicast.c 1362 2004-12-18 15:56:29Z roland $
+ */
+
+#include <linux/skbuff.h>
+#include <linux/rtnetlink.h>
+#include <linux/ip.h>
+#include <linux/in.h>
+#include <linux/igmp.h>
+#include <linux/inetdevice.h>
+#include <linux/delay.h>
+#include <linux/completion.h>
+
+#include "ipoib.h"
+
+#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
+int mcast_debug_level;
+
+module_param(mcast_debug_level, int, 0644);
+MODULE_PARM_DESC(mcast_debug_level,
+		 "Enable multicast debug tracing if > 0");
+#endif
+
+static DECLARE_MUTEX(mcast_mutex);
+
+/* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */
+struct ipoib_mcast {
+	struct ib_sa_mcmember_rec mcmember;
+	struct ipoib_ah          *ah;
+
+	struct rb_node    rb_node;
+	struct list_head  list;
+	struct completion done;
+
+	int                 query_id;
+	struct ib_sa_query *query;
+
+	unsigned long created;
+	unsigned long backoff;
+
+	unsigned long flags;
+	unsigned char logcount;
+
+	struct list_head  neigh_list;
+
+	struct sk_buff_head pkt_queue;
+
+	struct net_device *dev;
+};
+
+struct ipoib_mcast_iter {
+	struct net_device *dev;
+	union ib_gid       mgid;
+	unsigned long      created;
+	unsigned int       queuelen;
+	unsigned int       complete;
+	unsigned int       send_only;
+};
+
+static void ipoib_mcast_free(struct ipoib_mcast *mcast)
+{
+	struct net_device *dev = mcast->dev;
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_neigh *neigh, *tmp;
+	unsigned long flags;
+
+	ipoib_dbg_mcast(netdev_priv(dev),
+			"deleting multicast group " IPOIB_GID_FMT "\n",
+			IPOIB_GID_ARG(mcast->mcmember.mgid));
+
+	spin_lock_irqsave(&priv->lock, flags);
+
+	list_for_each_entry_safe(neigh, tmp, &mcast->neigh_list, list) {
+		ipoib_put_ah(neigh->ah);
+		*to_ipoib_neigh(neigh->neighbour) = NULL;
+		neigh->neighbour->ops->destructor = NULL;
+		kfree(neigh);
+	}
+
+	spin_unlock_irqrestore(&priv->lock, flags);
+
+	if (mcast->ah)
+		ipoib_put_ah(mcast->ah);
+
+	while (!skb_queue_empty(&mcast->pkt_queue)) {
+		struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
+
+		skb->dev = dev;
+		dev_kfree_skb_any(skb);
+	}
+
+	kfree(mcast);
+}
+
+static struct ipoib_mcast *ipoib_mcast_alloc(struct net_device *dev,
+					     int can_sleep)
+{
+	struct ipoib_mcast *mcast;
+
+	mcast = kmalloc(sizeof (*mcast), can_sleep ? GFP_KERNEL : GFP_ATOMIC);
+	if (!mcast)
+		return NULL;
+
+	memset(mcast, 0, sizeof (*mcast));
+
+	init_completion(&mcast->done);
+
+	mcast->dev = dev;
+	mcast->created = jiffies;
+	mcast->backoff = HZ;
+	mcast->logcount = 0;
+
+	INIT_LIST_HEAD(&mcast->list);
+	INIT_LIST_HEAD(&mcast->neigh_list);
+	skb_queue_head_init(&mcast->pkt_queue);
+
+	mcast->ah    = NULL;
+	mcast->query = NULL;
+
+	return mcast;
+}
+
+static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, union ib_gid *mgid)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct rb_node *n = priv->multicast_tree.rb_node;
+
+	while (n) {
+		struct ipoib_mcast *mcast;
+		int ret;
+
+		mcast = rb_entry(n, struct ipoib_mcast, rb_node);
+
+		ret = memcmp(mgid->raw, mcast->mcmember.mgid.raw,
+			     sizeof (union ib_gid));
+		if (ret < 0)
+			n = n->rb_left;
+		else if (ret > 0)
+			n = n->rb_right;
+		else
+			return mcast;
+	}
+
+	return NULL;
+}
+
+static int __ipoib_mcast_add(struct net_device *dev, struct ipoib_mcast *mcast)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL;
+
+	while (*n) {
+		struct ipoib_mcast *tmcast;
+		int ret;
+
+		pn = *n;
+		tmcast = rb_entry(pn, struct ipoib_mcast, rb_node);
+
+		ret = memcmp(mcast->mcmember.mgid.raw, tmcast->mcmember.mgid.raw,
+			     sizeof (union ib_gid));
+		if (ret < 0)
+			n = &pn->rb_left;
+		else if (ret > 0)
+			n = &pn->rb_right;
+		else
+			return -EEXIST;
+	}
+
+	rb_link_node(&mcast->rb_node, pn, n);
+	rb_insert_color(&mcast->rb_node, &priv->multicast_tree);
+
+	return 0;
+}
+
+static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
+				   struct ib_sa_mcmember_rec *mcmember)
+{
+	struct net_device *dev = mcast->dev;
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	int ret;
+
+	mcast->mcmember = *mcmember;
+
+	/* Set the cached Q_Key before we attach if it's the broadcast group */
+	if (!memcmp(mcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
+		    sizeof (union ib_gid)))
+		priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey);
+
+	if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
+		if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
+			ipoib_warn(priv, "multicast group " IPOIB_GID_FMT
+				   " already attached\n",
+				   IPOIB_GID_ARG(mcast->mcmember.mgid));
+
+			return 0;
+		}
+
+		ret = ipoib_mcast_attach(dev, be16_to_cpu(mcast->mcmember.mlid),
+					 &mcast->mcmember.mgid);
+		if (ret < 0) {
+			ipoib_warn(priv, "couldn't attach QP to multicast group "
+				   IPOIB_GID_FMT "\n",
+				   IPOIB_GID_ARG(mcast->mcmember.mgid));
+
+			clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags);
+			return ret;
+		}
+	}
+
+	{
+		/*
+		 * For now we set static_rate to 0.  This is not
+		 * really correct: we should look at the rate
+		 * component of the MC member record, compare it with
+		 * the rate of our local port (calculated from the
+		 * active link speed and link width) and set an
+		 * inter-packet delay appropriately.
+		 */
+		struct ib_ah_attr av = {
+			.dlid	       = be16_to_cpu(mcast->mcmember.mlid),
+			.port_num      = priv->port,
+			.sl	       = mcast->mcmember.sl,
+			.static_rate   = 0,
+			.ah_flags      = IB_AH_GRH,
+			.grh	       = {
+				.flow_label    = be32_to_cpu(mcast->mcmember.flow_label),
+				.hop_limit     = mcast->mcmember.hop_limit,
+				.sgid_index    = 0,
+				.traffic_class = mcast->mcmember.traffic_class
+			}
+		};
+
+		av.grh.dgid = mcast->mcmember.mgid;
+
+		mcast->ah = ipoib_create_ah(dev, priv->pd, &av);
+		if (!mcast->ah) {
+			ipoib_warn(priv, "ib_address_create failed\n");
+		} else {
+			ipoib_dbg_mcast(priv, "MGID " IPOIB_GID_FMT
+					" AV %p, LID 0x%04x, SL %d\n",
+					IPOIB_GID_ARG(mcast->mcmember.mgid),
+					mcast->ah->ah,
+					be16_to_cpu(mcast->mcmember.mlid),
+					mcast->mcmember.sl);
+		}
+	}
+
+	/* actually send any queued packets */
+	while (!skb_queue_empty(&mcast->pkt_queue)) {
+		struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
+
+		skb->dev = dev;
+
+		if (!skb->dst || !skb->dst->neighbour) {
+			/* put pseudoheader back on for next time */
+			skb_push(skb, sizeof (struct ipoib_pseudoheader));
+		}
+
+		if (dev_queue_xmit(skb))
+			ipoib_warn(priv, "dev_queue_xmit failed to requeue packet\n");
+	}
+
+	return 0;
+}
+
+static void
+ipoib_mcast_sendonly_join_complete(int status,
+				   struct ib_sa_mcmember_rec *mcmember,
+				   void *mcast_ptr)
+{
+	struct ipoib_mcast *mcast = mcast_ptr;
+	struct net_device *dev = mcast->dev;
+
+	if (!status)
+		ipoib_mcast_join_finish(mcast, mcmember);
+	else {
+		if (mcast->logcount++ < 20)
+			ipoib_dbg_mcast(netdev_priv(dev), "multicast join failed for "
+					IPOIB_GID_FMT ", status %d\n",
+					IPOIB_GID_ARG(mcast->mcmember.mgid), status);
+
+		/* Flush out any queued packets */
+		while (!skb_queue_empty(&mcast->pkt_queue)) {
+			struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
+
+			skb->dev = dev;
+
+			dev_kfree_skb_any(skb);
+		}
+
+		/* Clear the busy flag so we try again */
+		clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
+	}
+
+	complete(&mcast->done);
+}
+
+static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast)
+{
+	struct net_device *dev = mcast->dev;
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ib_sa_mcmember_rec rec = {
+#if 0				/* Some SMs don't support send-only yet */
+		.join_state = 4
+#else
+		.join_state = 1
+#endif
+	};
+	int ret = 0;
+
+	if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
+		ipoib_dbg_mcast(priv, "device shutting down, no multicast joins\n");
+		return -ENODEV;
+	}
+
+	if (test_and_set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) {
+		ipoib_dbg_mcast(priv, "multicast entry busy, skipping\n");
+		return -EBUSY;
+	}
+
+	rec.mgid     = mcast->mcmember.mgid;
+	rec.port_gid = priv->local_gid;
+	rec.pkey     = be16_to_cpu(priv->pkey);
+
+	ret = ib_sa_mcmember_rec_set(priv->ca, priv->port, &rec,
+				     IB_SA_MCMEMBER_REC_MGID		|
+				     IB_SA_MCMEMBER_REC_PORT_GID	|
+				     IB_SA_MCMEMBER_REC_PKEY		|
+				     IB_SA_MCMEMBER_REC_JOIN_STATE,
+				     1000, GFP_ATOMIC,
+				     ipoib_mcast_sendonly_join_complete,
+				     mcast, &mcast->query);
+	if (ret < 0) {
+		ipoib_warn(priv, "ib_sa_mcmember_rec_set failed (ret = %d)\n",
+			   ret);
+	} else {
+		ipoib_dbg_mcast(priv, "no multicast record for " IPOIB_GID_FMT
+				", starting join\n",
+				IPOIB_GID_ARG(mcast->mcmember.mgid));
+
+		mcast->query_id = ret;
+	}
+
+	return ret;
+}
+
+static void ipoib_mcast_join_complete(int status,
+				      struct ib_sa_mcmember_rec *mcmember,
+				      void *mcast_ptr)
+{
+	struct ipoib_mcast *mcast = mcast_ptr;
+	struct net_device *dev = mcast->dev;
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+
+	ipoib_dbg_mcast(priv, "join completion for " IPOIB_GID_FMT
+			" (status %d)\n",
+			IPOIB_GID_ARG(mcast->mcmember.mgid), status);
+
+	if (!status && !ipoib_mcast_join_finish(mcast, mcmember)) {
+		mcast->backoff = HZ;
+		down(&mcast_mutex);
+		if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
+			queue_work(ipoib_workqueue, &priv->mcast_task);
+		up(&mcast_mutex);
+		complete(&mcast->done);
+		return;
+	}
+
+	if (status == -EINTR) {
+		complete(&mcast->done);
+		return;
+	}
+
+	if (status && mcast->logcount++ < 20) {
+		if (status == -ETIMEDOUT || status == -EINTR) {
+			ipoib_dbg_mcast(priv, "multicast join failed for " IPOIB_GID_FMT
+					", status %d\n",
+					IPOIB_GID_ARG(mcast->mcmember.mgid),
+					status);
+		} else {
+			ipoib_warn(priv, "multicast join failed for "
+				   IPOIB_GID_FMT ", status %d\n",
+				   IPOIB_GID_ARG(mcast->mcmember.mgid),
+				   status);
+		}
+	}
+
+	mcast->backoff *= 2;
+	if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
+		mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
+
+	mcast->query = NULL;
+
+	down(&mcast_mutex);
+	if (test_bit(IPOIB_MCAST_RUN, &priv->flags)) {
+		if (status == -ETIMEDOUT)
+			queue_work(ipoib_workqueue, &priv->mcast_task);
+		else
+			queue_delayed_work(ipoib_workqueue, &priv->mcast_task,
+					   mcast->backoff * HZ);
+	} else
+		complete(&mcast->done);
+	up(&mcast_mutex);
+
+	return;
+}
+
+static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast,
+			     int create)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ib_sa_mcmember_rec rec = {
+		.join_state = 1
+	};
+	ib_sa_comp_mask comp_mask;
+	int ret = 0;
+
+	ipoib_dbg_mcast(priv, "joining MGID " IPOIB_GID_FMT "\n",
+			IPOIB_GID_ARG(mcast->mcmember.mgid));
+
+	rec.mgid     = mcast->mcmember.mgid;
+	rec.port_gid = priv->local_gid;
+	rec.pkey     = be16_to_cpu(priv->pkey);
+
+	comp_mask =
+		IB_SA_MCMEMBER_REC_MGID		|
+		IB_SA_MCMEMBER_REC_PORT_GID	|
+		IB_SA_MCMEMBER_REC_PKEY		|
+		IB_SA_MCMEMBER_REC_JOIN_STATE;
+
+	if (create) {
+		comp_mask |=
+			IB_SA_MCMEMBER_REC_QKEY		|
+			IB_SA_MCMEMBER_REC_SL		|
+			IB_SA_MCMEMBER_REC_FLOW_LABEL	|
+			IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
+
+		rec.qkey	  = priv->broadcast->mcmember.qkey;
+		rec.sl		  = priv->broadcast->mcmember.sl;
+		rec.flow_label	  = priv->broadcast->mcmember.flow_label;
+		rec.traffic_class = priv->broadcast->mcmember.traffic_class;
+	}
+
+	ret = ib_sa_mcmember_rec_set(priv->ca, priv->port, &rec, comp_mask,
+				     mcast->backoff * 1000, GFP_ATOMIC,
+				     ipoib_mcast_join_complete,
+				     mcast, &mcast->query);
+
+	if (ret < 0) {
+		ipoib_warn(priv, "ib_sa_mcmember_rec_set failed, status %d\n", ret);
+
+		mcast->backoff *= 2;
+		if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
+			mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
+
+		down(&mcast_mutex);
+		if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
+			queue_delayed_work(ipoib_workqueue,
+					   &priv->mcast_task,
+					   mcast->backoff);
+		up(&mcast_mutex);
+	} else
+		mcast->query_id = ret;
+}
+
+void ipoib_mcast_join_task(void *dev_ptr)
+{
+	struct net_device *dev = dev_ptr;
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+
+	if (!test_bit(IPOIB_MCAST_RUN, &priv->flags))
+		return;
+
+	if (ib_query_gid(priv->ca, priv->port, 0, &priv->local_gid))
+		ipoib_warn(priv, "ib_gid_entry_get() failed\n");
+	else
+		memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid));
+
+	if (!priv->broadcast) {
+		priv->broadcast = ipoib_mcast_alloc(dev, 1);
+		if (!priv->broadcast) {
+			ipoib_warn(priv, "failed to allocate broadcast group\n");
+			down(&mcast_mutex);
+			if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
+				queue_delayed_work(ipoib_workqueue,
+						   &priv->mcast_task, HZ);
+			up(&mcast_mutex);
+			return;
+		}
+
+		memcpy(priv->broadcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
+		       sizeof (union ib_gid));
+
+		spin_lock_irq(&priv->lock);
+		__ipoib_mcast_add(dev, priv->broadcast);
+		spin_unlock_irq(&priv->lock);
+	}
+
+	if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
+		ipoib_mcast_join(dev, priv->broadcast, 0);
+		return;
+	}
+
+	while (1) {
+		struct ipoib_mcast *mcast = NULL;
+
+		spin_lock_irq(&priv->lock);
+		list_for_each_entry(mcast, &priv->multicast_list, list) {
+			if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)
+			    && !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)
+			    && !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
+				/* Found the next unjoined group */
+				break;
+			}
+		}
+		spin_unlock_irq(&priv->lock);
+
+		if (&mcast->list == &priv->multicast_list) {
+			/* All done */
+			break;
+		}
+
+		ipoib_mcast_join(dev, mcast, 1);
+		return;
+	}
+
+	{
+		struct ib_port_attr attr;
+
+		if (!ib_query_port(priv->ca, priv->port, &attr))
+			priv->local_lid = attr.lid;
+		else
+			ipoib_warn(priv, "ib_query_port failed\n");
+	}
+
+	priv->mcast_mtu = ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu) -
+		IPOIB_ENCAP_LEN;
+	dev->mtu = min(priv->mcast_mtu, priv->admin_mtu);
+
+	ipoib_dbg_mcast(priv, "successfully joined all multicast groups\n");
+
+	clear_bit(IPOIB_MCAST_RUN, &priv->flags);
+	netif_carrier_on(dev);
+}
+
+int ipoib_mcast_start_thread(struct net_device *dev)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+
+	ipoib_dbg_mcast(priv, "starting multicast thread\n");
+
+	down(&mcast_mutex);
+	if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags))
+		queue_work(ipoib_workqueue, &priv->mcast_task);
+	up(&mcast_mutex);
+
+	return 0;
+}
+
+int ipoib_mcast_stop_thread(struct net_device *dev)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_mcast *mcast;
+
+	ipoib_dbg_mcast(priv, "stopping multicast thread\n");
+
+	down(&mcast_mutex);
+	clear_bit(IPOIB_MCAST_RUN, &priv->flags);
+	cancel_delayed_work(&priv->mcast_task);
+	up(&mcast_mutex);
+
+	flush_workqueue(ipoib_workqueue);
+
+	if (priv->broadcast && priv->broadcast->query) {
+		ib_sa_cancel_query(priv->broadcast->query_id, priv->broadcast->query);
+		priv->broadcast->query = NULL;
+		ipoib_dbg_mcast(priv, "waiting for bcast\n");
+		wait_for_completion(&priv->broadcast->done);
+	}
+
+	list_for_each_entry(mcast, &priv->multicast_list, list) {
+		if (mcast->query) {
+			ib_sa_cancel_query(mcast->query_id, mcast->query);
+			mcast->query = NULL;
+			ipoib_dbg_mcast(priv, "waiting for MGID " IPOIB_GID_FMT "\n",
+					IPOIB_GID_ARG(mcast->mcmember.mgid));
+			wait_for_completion(&mcast->done);
+		}
+	}
+
+	return 0;
+}
+
+int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ib_sa_mcmember_rec rec = {
+		.join_state = 1
+	};
+	int ret = 0;
+
+	if (!test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags))
+		return 0;
+
+	ipoib_dbg_mcast(priv, "leaving MGID " IPOIB_GID_FMT "\n",
+			IPOIB_GID_ARG(mcast->mcmember.mgid));
+
+	rec.mgid     = mcast->mcmember.mgid;
+	rec.port_gid = priv->local_gid;
+	rec.pkey     = be16_to_cpu(priv->pkey);
+
+	/* Remove ourselves from the multicast group */
+	ret = ipoib_mcast_detach(dev, be16_to_cpu(mcast->mcmember.mlid),
+				 &mcast->mcmember.mgid);
+	if (ret)
+		ipoib_warn(priv, "ipoib_mcast_detach failed (result = %d)\n", ret);
+
+	/*
+	 * Just make one shot at leaving and don't wait for a reply;
+	 * if we fail, too bad.
+	 */
+	ret = ib_sa_mcmember_rec_delete(priv->ca, priv->port, &rec,
+					IB_SA_MCMEMBER_REC_MGID		|
+					IB_SA_MCMEMBER_REC_PORT_GID	|
+					IB_SA_MCMEMBER_REC_PKEY		|
+					IB_SA_MCMEMBER_REC_JOIN_STATE,
+					0, GFP_ATOMIC, NULL,
+					mcast, &mcast->query);
+	if (ret < 0)
+		ipoib_warn(priv, "ib_sa_mcmember_rec_delete failed "
+			   "for leave (result = %d)\n", ret);
+
+	return 0;
+}
+
+void ipoib_mcast_send(struct net_device *dev, union ib_gid *mgid,
+		      struct sk_buff *skb)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_mcast *mcast;
+
+	/*
+	 * We can only be called from ipoib_start_xmit, so we're
+	 * inside tx_lock -- no need to save/restore flags.
+	 */
+	spin_lock(&priv->lock);
+
+	mcast = __ipoib_mcast_find(dev, mgid);
+	if (!mcast) {
+		/* Let's create a new send only group now */
+		ipoib_dbg_mcast(priv, "setting up send only multicast group for "
+				IPOIB_GID_FMT "\n", IPOIB_GID_ARG(*mgid));
+
+		mcast = ipoib_mcast_alloc(dev, 0);
+		if (!mcast) {
+			ipoib_warn(priv, "unable to allocate memory for "
+				   "multicast structure\n");
+			dev_kfree_skb_any(skb);
+			goto out;
+		}
+
+		set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags);
+		mcast->mcmember.mgid = *mgid;
+		__ipoib_mcast_add(dev, mcast);
+		list_add_tail(&mcast->list, &priv->multicast_list);
+	}
+
+	if (!mcast->ah) {
+		if (skb_queue_len(&mcast->pkt_queue) < IPOIB_MAX_MCAST_QUEUE)
+			skb_queue_tail(&mcast->pkt_queue, skb);
+		else
+			dev_kfree_skb_any(skb);
+
+		if (mcast->query)
+			ipoib_dbg_mcast(priv, "no address vector, "
+					"but multicast join already started\n");
+		else if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
+			ipoib_mcast_sendonly_join(mcast);
+
+		/*
+		 * If lookup completes between here and out:, don't
+		 * want to send packet twice.
+		 */
+		mcast = NULL;
+	}
+
+out:
+	if (mcast && mcast->ah) {
+		if (skb->dst            &&
+		    skb->dst->neighbour &&
+		    !*to_ipoib_neigh(skb->dst->neighbour)) {
+			struct ipoib_neigh *neigh = kmalloc(sizeof *neigh, GFP_ATOMIC);
+
+			if (neigh) {
+				kref_get(&mcast->ah->ref);
+				neigh->ah  	= mcast->ah;
+				neigh->neighbour = skb->dst->neighbour;
+				*to_ipoib_neigh(skb->dst->neighbour) = neigh;
+				list_add_tail(&neigh->list, &mcast->neigh_list);
+			}
+		}
+
+		ipoib_send(dev, skb, mcast->ah, IB_MULTICAST_QPN);
+	}
+
+	spin_unlock(&priv->lock);
+}
+
+void ipoib_mcast_dev_flush(struct net_device *dev)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	LIST_HEAD(remove_list);
+	struct ipoib_mcast *mcast, *tmcast, *nmcast;
+	unsigned long flags;
+
+	ipoib_dbg_mcast(priv, "flushing multicast list\n");
+
+	spin_lock_irqsave(&priv->lock, flags);
+	list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
+		nmcast = ipoib_mcast_alloc(dev, 0);
+		if (nmcast) {
+			nmcast->flags =
+				mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY);
+
+			nmcast->mcmember.mgid = mcast->mcmember.mgid;
+
+			/* Add the new group in before the to-be-destroyed group */
+			list_add_tail(&nmcast->list, &mcast->list);
+			list_del_init(&mcast->list);
+
+			rb_replace_node(&mcast->rb_node, &nmcast->rb_node,
+					&priv->multicast_tree);
+
+			list_add_tail(&mcast->list, &remove_list);
+		} else {
+			ipoib_warn(priv, "could not reallocate multicast group "
+				   IPOIB_GID_FMT "\n",
+				   IPOIB_GID_ARG(mcast->mcmember.mgid));
+		}
+	}
+
+	if (priv->broadcast) {
+		nmcast = ipoib_mcast_alloc(dev, 0);
+		if (nmcast) {
+			nmcast->mcmember.mgid = priv->broadcast->mcmember.mgid;
+
+			rb_replace_node(&priv->broadcast->rb_node,
+					&nmcast->rb_node,
+					&priv->multicast_tree);
+
+			list_add_tail(&priv->broadcast->list, &remove_list);
+		}
+
+		priv->broadcast = nmcast;
+	}
+
+	spin_unlock_irqrestore(&priv->lock, flags);
+
+	list_for_each_entry(mcast, &remove_list, list) {
+		ipoib_mcast_leave(dev, mcast);
+		ipoib_mcast_free(mcast);
+	}
+}
+
+void ipoib_mcast_dev_down(struct net_device *dev)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	unsigned long flags;
+
+	/* Delete broadcast since it will be recreated */
+	if (priv->broadcast) {
+		ipoib_dbg_mcast(priv, "deleting broadcast group\n");
+
+		spin_lock_irqsave(&priv->lock, flags);
+		rb_erase(&priv->broadcast->rb_node, &priv->multicast_tree);
+		spin_unlock_irqrestore(&priv->lock, flags);
+		ipoib_mcast_leave(dev, priv->broadcast);
+		ipoib_mcast_free(priv->broadcast);
+		priv->broadcast = NULL;
+	}
+}
+
+void ipoib_mcast_restart_task(void *dev_ptr)
+{
+	struct net_device *dev = dev_ptr;
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct dev_mc_list *mclist;
+	struct ipoib_mcast *mcast, *tmcast;
+	LIST_HEAD(remove_list);
+	unsigned long flags;
+
+	ipoib_dbg_mcast(priv, "restarting multicast task\n");
+
+	ipoib_mcast_stop_thread(dev);
+
+	spin_lock_irqsave(&priv->lock, flags);
+
+	/*
+	 * Unfortunately, the networking core only gives us a list of all of
+	 * the multicast hardware addresses. We need to figure out which ones
+	 * are new and which ones have been removed
+	 */
+
+	/* Clear out the found flag */
+	list_for_each_entry(mcast, &priv->multicast_list, list)
+		clear_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
+
+	/* Mark all of the entries that are found or don't exist */
+	for (mclist = dev->mc_list; mclist; mclist = mclist->next) {
+		union ib_gid mgid;
+
+		memcpy(mgid.raw, mclist->dmi_addr + 4, sizeof mgid);
+
+		/* Add in the P_Key */
+		mgid.raw[4] = (priv->pkey >> 8) & 0xff;
+		mgid.raw[5] = priv->pkey & 0xff;
+
+		mcast = __ipoib_mcast_find(dev, &mgid);
+		if (!mcast || test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
+			struct ipoib_mcast *nmcast;
+
+			/* Not found or send-only group, let's add a new entry */
+			ipoib_dbg_mcast(priv, "adding multicast entry for mgid "
+					IPOIB_GID_FMT "\n", IPOIB_GID_ARG(mgid));
+
+			nmcast = ipoib_mcast_alloc(dev, 0);
+			if (!nmcast) {
+				ipoib_warn(priv, "unable to allocate memory for multicast structure\n");
+				continue;
+			}
+
+			set_bit(IPOIB_MCAST_FLAG_FOUND, &nmcast->flags);
+
+			nmcast->mcmember.mgid = mgid;
+
+			if (mcast) {
+				/* Destroy the send only entry */
+				list_del(&mcast->list);
+				list_add_tail(&mcast->list, &remove_list);
+
+				rb_replace_node(&mcast->rb_node,
+						&nmcast->rb_node,
+						&priv->multicast_tree);
+			} else
+				__ipoib_mcast_add(dev, nmcast);
+
+			list_add_tail(&nmcast->list, &priv->multicast_list);
+		}
+
+		if (mcast)
+			set_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
+	}
+
+	/* Remove all of the entries don't exist anymore */
+	list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
+		if (!test_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags) &&
+		    !test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
+			ipoib_dbg_mcast(priv, "deleting multicast group " IPOIB_GID_FMT "\n",
+					IPOIB_GID_ARG(mcast->mcmember.mgid));
+
+			rb_erase(&mcast->rb_node, &priv->multicast_tree);
+
+			/* Move to the remove list */
+			list_del(&mcast->list);
+			list_add_tail(&mcast->list, &remove_list);
+		}
+	}
+	spin_unlock_irqrestore(&priv->lock, flags);
+
+	/* We have to cancel outside of the spinlock */
+	list_for_each_entry(mcast, &remove_list, list) {
+		ipoib_mcast_leave(mcast->dev, mcast);
+		ipoib_mcast_free(mcast);
+	}
+
+	if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
+		ipoib_mcast_start_thread(dev);
+}
+
+struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev)
+{
+	struct ipoib_mcast_iter *iter;
+
+	iter = kmalloc(sizeof *iter, GFP_KERNEL);
+	if (!iter)
+		return NULL;
+
+	iter->dev = dev;
+	memset(iter->mgid.raw, 0, sizeof iter->mgid);
+
+	if (ipoib_mcast_iter_next(iter)) {
+		ipoib_mcast_iter_free(iter);
+		return NULL;
+	}
+
+	return iter;
+}
+
+void ipoib_mcast_iter_free(struct ipoib_mcast_iter *iter)
+{
+	kfree(iter);
+}
+
+int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
+	struct rb_node *n;
+	struct ipoib_mcast *mcast;
+	int ret = 1;
+
+	spin_lock_irq(&priv->lock);
+
+	n = rb_first(&priv->multicast_tree);
+
+	while (n) {
+		mcast = rb_entry(n, struct ipoib_mcast, rb_node);
+
+		if (memcmp(iter->mgid.raw, mcast->mcmember.mgid.raw,
+			   sizeof (union ib_gid)) < 0) {
+			iter->mgid      = mcast->mcmember.mgid;
+			iter->created   = mcast->created;
+			iter->queuelen  = skb_queue_len(&mcast->pkt_queue);
+			iter->complete  = !!mcast->ah;
+			iter->send_only = !!(mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY));
+
+			ret = 0;
+
+			break;
+		}
+
+		n = rb_next(n);
+	}
+
+	spin_unlock_irq(&priv->lock);
+
+	return ret;
+}
+
+void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
+			   union ib_gid *mgid,
+			   unsigned long *created,
+			   unsigned int *queuelen,
+			   unsigned int *complete,
+			   unsigned int *send_only)
+{
+	*mgid      = iter->mgid;
+	*created   = iter->created;
+	*queuelen  = iter->queuelen;
+	*complete  = iter->complete;
+	*send_only = iter->send_only;
+}
drivers/infiniband/ulp/ipoib/ipoib_vlan.c new file +177 -0

177 lines not on this page · raw message

#55 lore raw

PATCH v5 21/24 Add InfiniBand userspace MAD support

Add a driver that provides a character special device for each InfiniBand port. This device allows userspace to send and receive MADs via write() and read() (with some control operations implemented as ioctls).

All operations are 32/64 clean and have been tested with 32-bit userspace running on a ppc64 kernel.

drivers/infiniband/core/Makefile +3 -1
@@ -1,6 +1,6 @@
 EXTRA_CFLAGS += -Idrivers/infiniband/include
 
-obj-$(CONFIG_INFINIBAND) +=	ib_core.o ib_mad.o ib_sa.o
+obj-$(CONFIG_INFINIBAND) +=	ib_core.o ib_mad.o ib_sa.o ib_umad.o
 
 ib_core-y :=			packer.o ud_header.o verbs.o sysfs.o \
 				device.o fmr_pool.o cache.o
@@ -8,3 +8,5 @@
 ib_mad-y :=			mad.o smi.o agent.o
 
 ib_sa-y :=			sa_query.o
+
+ib_umad-y :=			user_mad.o
drivers/infiniband/core/user_mad.c new file +738 -0

738 lines not on this page · raw message

drivers/infiniband/include/ib_user_mad.h new file +123 -0

123 lines not on this page · raw message

#56 lore raw

PATCH v5 22/24 Document InfiniBand ioctl use

Add the 0x1b ioctl magic number used by ib_umad module to Documentation/ioctl-number.txt.

Documentation/ioctl-number.txt +1 -0
@@ -72,6 +72,7 @@
 0x09	all	linux/md.h
 0x12	all	linux/fs.h
 		linux/blkpg.h
+0x1b	all	InfiniBand Subsystem	<http://www.openib.org/>
 0x20	all	drivers/cdrom/cm206.h
 0x22	all	scsi/sg.h
 '#'	00-3F	IEEE 1394 Subsystem	Block for the entire subsystem
#57 lore raw

PATCH v5 23/24 Add InfiniBand Documentation files

Add files to Documentation/infiniband that describe the tree under /sys/class/infiniband, the IPoIB driver and the userspace MAD access driver.

Documentation/infiniband/ipoib.txt new file +56 -0
@@ -0,0 +1,56 @@
+IP OVER INFINIBAND
+
+  The ib_ipoib driver is an implementation of the IP over InfiniBand
+  protocol as specified by the latest Internet-Drafts issued by the
+  IETF ipoib working group.  It is a "native" implementation in the
+  sense of setting the interface type to ARPHRD_INFINIBAND and the
+  hardware address length to 20 (earlier proprietary implementations
+  masqueraded to the kernel as ethernet interfaces).
+
+Partitions and P_Keys
+
+  When the IPoIB driver is loaded, it creates one interface for each
+  port using the P_Key at index 0.  To create an interface with a
+  different P_Key, write the desired P_Key into the main interface's
+  /sys/class/net/<intf name>/create_child file.  For example:
+
+    echo 0x8001 > /sys/class/net/ib0/create_child
+
+  This will create an interface named ib0.8001 with P_Key 0x8001.  To
+  remove a subinterface, use the "delete_child" file:
+
+    echo 0x8001 > /sys/class/net/ib0/delete_child
+
+  The P_Key for any interface is given by the "pkey" file, and the
+  main interface for a subinterface is in "parent."
+
+Debugging Information
+
+  By compiling the IPoIB driver with CONFIG_INFINIBAND_IPOIB_DEBUG set
+  to 'y', tracing messages are compiled into the driver.  They are
+  turned on by setting the module parameters debug_level and
+  mcast_debug_level to 1.  These parameters can be controlled at
+  runtime through files in /sys/module/ib_ipoib/.
+
+  CONFIG_INFINIBAND_IPOIB_DEBUG also enables the "ipoib_debugfs"
+  virtual filesystem.  By mounting this filesystem, for example with
+
+    mkdir -p /ipoib_debugfs
+    mount -t ipoib_debugfs none /ipoib_debufs
+
+  it is possible to get statistics about multicast groups from the
+  files /ipoib_debugfs/ib0_mcg and so on.
+
+  The performance impact of this option is negligible, so it
+  is safe to enable this option with debug_level set to 0 for normal
+  operation.
+
+  CONFIG_INFINIBAND_IPOIB_DEBUG_DATA enables even more debug output in
+  the data path when data_debug_level is set to 1.  However, even with
+  the output disabled, enabling this configuration option will affect
+  performance, because it adds tests to the fast path.
+
+References
+
+  IETF IP over InfiniBand (ipoib) Working Group
+    http://ietf.org/html.charters/ipoib-charter.html
Documentation/infiniband/sysfs.txt new file +64 -0

64 lines not on this page · raw message

Documentation/infiniband/user_mad.txt new file +81 -0

81 lines not on this page · raw message

#58 lore raw

PATCH v5 24/24 InfiniBand MAINTAINERS entry

Add OpenIB maintainers information to MAINTAINERS.

MAINTAINERS +11 -0
@@ -1081,6 +1081,17 @@
 L:	[email protected]
 S:	Maintained
 
+INFINIBAND SUBSYSTEM
+P:	Roland Dreier
+M:	[email protected]
+P:	Sean Hefty
+M:	[email protected]
+P:	Hal Rosenstock
+M:	[email protected]
+L:	[email protected]
+W:	http://www.openib.org/
+S:	Supported
+
 INPUT (KEYBOARD, MOUSE, JOYSTICK) DRIVERS
 P:	Vojtech Pavlik
 M:	[email protected]
maintainer
#59 lore raw

Re: [PATCH][v5][0/24] Latest IB patch queue

In reply to Roland Dreier
Quoted 6 lines · On Mon, 27 Dec 2004 21:50:47 -0800 Roland Dreier <[email protected]> wrote:
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"David" == David S Miller <[email protected]> writes:

David> Send it all over.

OK, you asked for it... here's our latest tree, which should incorporate all the feedback I've seen.

W00t :-) All applied, thanks Roland.

I'll run it through some build tests then toss it upstream.

#60 lore raw

Re: [PATCH][v5][0/24] Latest IB patch queue

David> W00t :-) All applied, thanks Roland.

David> I'll run it through some build tests then toss it upstream.

Very cool, thanks a lot. Let me know if you see any build failures -- I test on about 6 or 7 different archs/configs but the bug gods always seem to hide problems from me.

Speaking of build failures, one of my test builds is cross-compiling for sparc64 with gcc 3.4.2, which adds __attribute__((warn_unused_result)) to copy_to_user() et al. The -Werror in the arch/sparc64 means the build fails with

linux-2.6.10/arch/sparc64/kernel/sys_sparc32.c:1686: warning: ignoring return value of `copy_to_user', declared with attribute warn_unused_result

Of course binfmt_elf.c and compat_ioctl.c still have issues but those probably get more visibility...

Thanks,
Roland

Check copy_to_user() return value in sys_sparc32.c and sys_sunos32.c.

arch/sparc64/kernel/sys_sparc32.c +2 -1
Index: linux-2.6.10/arch/sparc64/kernel/sys_sparc32.c
===================================================================
@@ -1683,7 +1683,8 @@
 			    put_user(oldlen, (u32 __user *)(unsigned long) tmp.oldlenp))
 				error = -EFAULT;
 		}
-		copy_to_user(args->__unused, tmp.__unused, sizeof(tmp.__unused));
+		if (copy_to_user(args->__unused, tmp.__unused, sizeof(tmp.__unused)))
+			error = -EFAULT;
 	}
 	return error;
 #endif
arch/sparc64/kernel/sys_sunos32.c +4 -2
Index: linux-2.6.10/arch/sparc64/kernel/sys_sunos32.c
===================================================================
@@ -291,7 +291,8 @@
 	put_user(ino, &dirent->d_ino);
 	put_user(namlen, &dirent->d_namlen);
 	put_user(reclen, &dirent->d_reclen);
-	copy_to_user(dirent->d_name, name, namlen);
+	if (copy_to_user(dirent->d_name, name, namlen))
+		return -EFAULT;
 	put_user(0, dirent->d_name + namlen);
 	dirent = (void __user *) dirent + reclen;
 	buf->curr = dirent;
@@ -371,7 +372,8 @@
 	put_user(ino, &dirent->d_ino);
 	put_user(namlen, &dirent->d_namlen);
 	put_user(reclen, &dirent->d_reclen);
-	copy_to_user(dirent->d_name, name, namlen);
+	if (copy_to_user(dirent->d_name, name, namlen))
+		return -EFAULT;
 	put_user(0, dirent->d_name + namlen);
 	dirent = (void __user *) dirent + reclen;
 	buf->curr = dirent;
maintainer
#61 lore raw

Re: [PATCH][v5][0/24] Latest IB patch queue

Quoted 4 lines · On Tue, 28 Dec 2004 11:48:13 -0800 Roland Dreier <[email protected]> wrote:

Speaking of build failures, one of my test builds is cross-compiling for sparc64 with gcc 3.4.2, which adds __attribute__((warn_unused_result)) to copy_to_user() et al. The -Werror in the arch/sparc64 means the build fails with

Thanks, I'll check that out.

I believe that you didn't test the sparc64 build of the infiniband stuff because arch/sparc64/Kconfig needs to explicitly include the infiniband Kconfig since it does not use drivers/Kconfig. You didn't send me any such changes.

There are a few platforms which also are in this situation. I added the sparc64 one to my tree while integrating your changes, but the others need to be attended to if you wish infiniband to be configurable on them.

#62 lore raw

Re: [PATCH][v5][0/24] Latest IB patch queue

David> I believe that you didn't test the sparc64 build of the David> infiniband stuff because arch/sparc64/Kconfig needs to David> explicitly include the infiniband Kconfig since it does not David> use drivers/Kconfig. You didn't send me any such changes.

Actually I did test the build (and Tom Duffy at Sun has actually run the drivers on his system), but I forgot to include the required Kconfig change -- I just have it in my local test tree.

David> There are a few platforms which also are in this situation. David> I added the sparc64 one to my tree while integrating your David> changes, but the others need to be attended to if you wish David> infiniband to be configurable on them.

I think sparc64 is the only such platform where InfiniBand is likely to be of much interest. However I'll check out all of arch/ and send patches to hook up drivers/infiniband/ to the relevant maintainers once IB makes it upstream.

Thanks,
Roland

#63 lore raw

Re: [PATCH][v5][0/24] Latest IB patch queue

Quoted 6 lines · Roland Dreier

I think sparc64 is the only such platform where InfiniBand is likely to be of much interest. However I'll check out all of arch/ and send patches to hook up drivers/infiniband/ to the relevant maintainers once IB makes it upstream.

Hi Roland,

I am interested in Infiniband with x86_64 Opterons.

Thanks,
Karen

Signature · 3 lines
 Karen Shaeffer
 Neuralscape, Palo Alto, Ca. 94306
 [email protected]  http://www.neuralscape.com
#64 lore raw

Re: [PATCH][v5][0/24] Latest IB patch queue

Roland> I think sparc64 is the only such platform where InfiniBand Roland> is likely to be of much interest. However I'll check out Roland> all of arch/ and send patches to hook up Roland> drivers/infiniband/ to the relevant maintainers once IB Roland> makes it upstream.

Karen> I am interested in Infiniband with x86_64 Opterons.

OK, the current code should work well for you -- x86_64 is probably the most-tested architecture.

"such platform[s]" in my comment above referred to architectures where arch/xxx/Kconfig does _not_ include drivers/Kconfig; arch/x86_64/Kconfig does include that file. So no change is required to use the current IB patches on x86_64. I believe the only architectures that both support PCI and do not include drivers/Kconfig in their arch Kconfig are arm, sparc, sparc64 and v850. Perhaps I'm wrong, but of those four architectures, sparc64 seems to be the only one where there would be any interest in using IB.

- Roland