Skip to contentSkip to conversation

Add XEN pvSCSI support

PATCH Applied Juergen Gross 3 participants 15 messages Aug 28, 2014 – Sep 1, 2014

Subsystems (3)
lore
Share & tools
lore

Juergen Gross

author maintainer
#0 lore raw

v6 cover Add XEN pvSCSI support

This series adds XEN pvSCSI support. With pvSCSI it is possible to use physical SCSI devices from a XEN domain.

The support consists of a backend in the privileged Domain-0 doing the real I/O and a frontend in the unprivileged domU passing I/O-requests to the backend.

The code is taken (and adapted) from the original pvSCSI implementation done for Linux 2.6 in 2008 by Fujitsu.

Changes in V6:
- changes in scsifront module after comments from Konrad Wilk
- remove double kfree in scsiback
- comments in vscsiif.h adapted after comments from Jan Beulich and David Vrabel in xen-devel
- added [email protected] to MAINTANERS entry

Changes in V5:
- Added patch to support threaded irqs for interdomain event channels
- several changes in xen-scsifront after comments from Christoph Hellwig
- several changes in xen-scsiback after comments from Christoph Hellwig
- several changes in xen-scsiback after comments from James Bottomley
- several changes in xen-scsiback after comments from Nicholas Bellinger

Changes in V4:
- Re-add define for VSCSIIF_ACT_SCSI_SG_PRESET to vscsiif.h to indicate this action value should not be used in future enhancements

Changes in V3:
- added some comments to the protocol header file
- removed the CDB emulation from xen-scsiback, handled by core target infrastructure
- several changes in xen-scsifront after comments from Christoph Hellwig

Changes in V2:
- use core target infrastructure by backend instead of pure SCSI passthrough
- add support for larger SG lists by putting them in grant page(s)
- add command abort capability

Juergen Gross (5):
xen/events: support threaded irqs for interdomain event channels Add XEN pvSCSI protocol description
Introduce xen-scsifront module
Introduce XEN scsiback module
add xen pvscsi maintainer

 MAINTAINERS                        |    9 +
 drivers/scsi/Kconfig               |    9 +
 drivers/scsi/Makefile              |    1 +
 drivers/scsi/xen-scsifront.c       | 1024 +++++++++++++++++
 drivers/xen/Kconfig                |    9 +
 drivers/xen/Makefile               |    1 +
 drivers/xen/events/events_base.c   |    5 +-
 drivers/xen/xen-scsiback.c         | 2124 ++++++++++++++++++++++++++++++++++++
 include/xen/events.h               |    2 +
 include/xen/interface/io/vscsiif.h |  229 ++++
 10 files changed, 3411 insertions(+), 2 deletions(-)
 create mode 100644 drivers/scsi/xen-scsifront.c
 create mode 100644 drivers/xen/xen-scsiback.c
 create mode 100644 include/xen/interface/io/vscsiif.h
Signature · 1 line
1.8.4.5
author maintainer
#1 lore raw

PATCH v6 1/5 xen/events: support threaded irqs for interdomain event channels

Export bind_interdomain_evtchn_to_irq() so drivers can use threaded interrupt handlers with:

irq = bind_interdomain_evtchn_to_irq(remote_dom, remote_port); if (irq < 0)

     /* error */

ret = request_threaded_irq(...);

---
 drivers/xen/events/events_base.c | 5 +++--
 include/xen/events.h             | 2 ++
 2 files changed, 5 insertions(+), 2 deletions(-)
drivers/xen/events/events_base.c +3 -2
index 5b5c5ff..b4bca2d 100644
@@ -900,8 +900,8 @@ static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
 	return irq;
 }
 
-static int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
-					  unsigned int remote_port)
+int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
+				   unsigned int remote_port)
 {
 	struct evtchn_bind_interdomain bind_interdomain;
 	int err;
@@ -914,6 +914,7 @@ static int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
 
 	return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
 }
+EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq);
 
 static int find_virq(unsigned int virq, unsigned int cpu)
 {
include/xen/events.h +2 -0
index 8bee7a7..5321cd9 100644
@@ -28,6 +28,8 @@ int bind_ipi_to_irqhandler(enum ipi_vector ipi,
 			   unsigned long irqflags,
 			   const char *devname,
 			   void *dev_id);
+int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
+				   unsigned int remote_port);
 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
 					  unsigned int remote_port,
 					  irq_handler_t handler,
Signature · 1 line
1.8.4.5
author maintainer
#2 lore raw

PATCH v6 1/5 xen/events: support threaded irqs for interdomain event channels

In reply to Juergen Gross

Export bind_interdomain_evtchn_to_irq() so drivers can use threaded interrupt handlers with:

irq = bind_interdomain_evtchn_to_irq(remote_dom, remote_port); if (irq < 0)

     /* error */

ret = request_threaded_irq(...);

---
 drivers/xen/events/events_base.c | 5 +++--
 include/xen/events.h             | 2 ++
 2 files changed, 5 insertions(+), 2 deletions(-)
drivers/xen/events/events_base.c +3 -2
index 5b5c5ff..b4bca2d 100644
@@ -900,8 +900,8 @@ static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
 	return irq;
 }
 
-static int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
-					  unsigned int remote_port)
+int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
+				   unsigned int remote_port)
 {
 	struct evtchn_bind_interdomain bind_interdomain;
 	int err;
@@ -914,6 +914,7 @@ static int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
 
 	return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
 }
+EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq);
 
 static int find_virq(unsigned int virq, unsigned int cpu)
 {
include/xen/events.h +2 -0
index 8bee7a7..5321cd9 100644
@@ -28,6 +28,8 @@ int bind_ipi_to_irqhandler(enum ipi_vector ipi,
 			   unsigned long irqflags,
 			   const char *devname,
 			   void *dev_id);
+int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
+				   unsigned int remote_port);
 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
 					  unsigned int remote_port,
 					  irq_handler_t handler,
Signature · 1 line
1.8.4.5
author maintainer
#3 lore raw

PATCH v6 2/5 Add XEN pvSCSI protocol description

In reply to Juergen Gross

Add the definition of pvSCSI protocol used between the pvSCSI frontend in a XEN domU and the pvSCSI backend in a XEN driver domain (usually Dom0).

This header was originally provided by Fujitsu for XEN based on Linux 2.6.18. Changes are:
- added comment
- adapt to Linux style guide
- add support for larger SG-lists by putting them in an own granted page
- remove stale definitions

---
 include/xen/interface/io/vscsiif.h | 229 +++++++++++++++++++++++++++++++++++++
 1 file changed, 229 insertions(+)
 create mode 100644 include/xen/interface/io/vscsiif.h
include/xen/interface/io/vscsiif.h new file +229 -0
new file mode 100644
index 0000000..d07d7ac
@@ -0,0 +1,229 @@
+/******************************************************************************
+ * vscsiif.h
+ *
+ * Based on the blkif.h code.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * 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) FUJITSU Limited 2008.
+ */
+
+#ifndef __XEN__PUBLIC_IO_SCSI_H__
+#define __XEN__PUBLIC_IO_SCSI_H__
+
+#include "ring.h"
+#include "../grant_table.h"
+
+/*
+ * Feature and Parameter Negotiation
+ * =================================
+ * The two halves of a Xen pvSCSI driver utilize nodes within the XenStore to
+ * communicate capabilities and to negotiate operating parameters.  This
+ * section enumerates these nodes which reside in the respective front and
+ * backend portions of the XenStore, following the XenBus convention.
+ *
+ * Any specified default value is in effect if the corresponding XenBus node
+ * is not present in the XenStore.
+ *
+ * XenStore nodes in sections marked "PRIVATE" are solely for use by the
+ * driver side whose XenBus tree contains them.
+ *
+ *****************************************************************************
+ *                            Backend XenBus Nodes
+ *****************************************************************************
+ *
+ *------------------ Backend Device Identification (PRIVATE) ------------------
+ *
+ * p-devname
+ *      Values:         string
+ *
+ *      A free string used to identify the physical device (e.g. a disk name).
+ *
+ * p-dev
+ *      Values:         string
+ *
+ *      A string specifying the backend device: either a 4-tuple "h:c:t:l"
+ *      (host, controller, target, lun, all integers), or a WWN (e.g.
+ *      "naa.60014054ac780582").
+ *
+ * v-dev
+ *      Values:         string
+ *
+ *      A string specifying the frontend device in form of a 4-tuple "h:c:t:l"
+ *      (host, controller, target, lun, all integers).
+ *
+ *--------------------------------- Features ---------------------------------
+ *
+ * feature-sg-grant
+ *      Values:         unsigned [VSCSIIF_SG_TABLESIZE...65535]
+ *      Default Value:  0
+ *
+ *      Specifies the maximum number of scatter/gather elements in grant pages
+ *      supported. If not set, the backend supports up to VSCSIIF_SG_TABLESIZE
+ *      SG elements specified directly in the request.
+ *
+ *****************************************************************************
+ *                            Frontend XenBus Nodes
+ *****************************************************************************
+ *
+ *----------------------- Request Transport Parameters -----------------------
+ *
+ * event-channel
+ *      Values:         unsigned
+ *
+ *      The identifier of the Xen event channel used to signal activity
+ *      in the ring buffer.
+ *
+ * ring-ref
+ *      Values:         unsigned
+ *
+ *      The Xen grant reference granting permission for the backend to map
+ *      the sole page in a single page sized ring buffer.
+ *
+ * protocol
+ *      Values:         string (XEN_IO_PROTO_ABI_*)
+ *      Default Value:  XEN_IO_PROTO_ABI_NATIVE
+ *
+ *      The machine ABI rules governing the format of all ring request and
+ *      response structures.
+ */
+
+/* Requests from the frontend to the backend */
+
+/*
+ * Request a SCSI operation specified via a CDB in vscsiif_request.cmnd.
+ * The target is specified via channel, id and lun.
+ *
+ * The operation to be performed is specified via a CDB in cmnd[], the length
+ * of the CDB is in cmd_len. sc_data_direction specifies the direction of data
+ * (to the device, from the device, or none at all).
+ *
+ * If data is to be transferred to or from the device the buffer(s) in the
+ * guest memory is/are specified via one or multiple scsiif_request_segment
+ * descriptors each specifying a memory page via a grant_ref_t, a offset into
+ * the page and the length of the area in that page. All scsiif_request_segment
+ * areas concatenated form the resulting data buffer used by the operation.
+ * If the number of scsiif_request_segment areas is not too large (less than
+ * or equal VSCSIIF_SG_TABLESIZE) the areas can be specified directly in the
+ * seg[] array and the number of valid scsiif_request_segment elements is to be
+ * set in nr_segments.
+ *
+ * If "feature-sg-grant" in the Xenstore is set it is possible to specify more
+ * than VSCSIIF_SG_TABLESIZE scsiif_request_segment elements via indirection.
+ * The maximum number of allowed scsiif_request_segment elements is the value
+ * of the "feature-sg-grant" entry from Xenstore. When using indirection the
+ * seg[] array doesn't contain specifications of the data buffers, but
+ * references to scsiif_request_segment arrays, which in turn reference the
+ * data buffers. While nr_segments holds the number of populated seg[] entries
+ * (plus the set VSCSIIF_SG_GRANT bit), the number of scsiif_request_segment
+ * elements referencing the target data buffers is calculated from the lengths
+ * of the seg[] elements (the sum of all valid seg[].length divided by the
+ * size of one scsiif_request_segment structure).
+ */
+#define VSCSIIF_ACT_SCSI_CDB		1
+
+/*
+ * Request abort of a running operation for the specified target given by
+ * channel, id, lun and the operation's rqid in ref_rqid.
+ */
+#define VSCSIIF_ACT_SCSI_ABORT		2
+
+/*
+ * Request a device reset of the specified target (channel and id).
+ */
+#define VSCSIIF_ACT_SCSI_RESET		3
+
+/*
+ * Preset scatter/gather elements for a following request. Deprecated.
+ * Keeping the define only to avoid usage of the value "4" for other actions.
+ */
+#define VSCSIIF_ACT_SCSI_SG_PRESET	4
+
+/*
+ * Maximum scatter/gather segments per request.
+ *
+ * Considering balance between allocating at least 16 "vscsiif_request"
+ * structures on one page (4096 bytes) and the number of scatter/gather
+ * elements needed, we decided to use 26 as a magic number.
+ *
+ * If "feature-sg-grant" is set, more scatter/gather elements can be specified
+ * by placing them in one or more (up to VSCSIIF_SG_TABLESIZE) granted pages.
+ * In this case the vscsiif_request seg elements don't contain references to
+ * the user data, but to the SG elements referencing the user data.
+ */
+#define VSCSIIF_SG_TABLESIZE		26
+
+/*
+ * based on Linux kernel 2.6.18, still valid
+ * Changing these values requires support of multiple protocols via the rings
+ * as "old clients" will blindly use these values and the resulting structure
+ * sizes.
+ */
+#define VSCSIIF_MAX_COMMAND_SIZE	16
+#define VSCSIIF_SENSE_BUFFERSIZE	96
+
+struct scsiif_request_segment {
+	grant_ref_t gref;
+	uint16_t offset;
+	uint16_t length;
+};
+
+#define VSCSIIF_SG_PER_PAGE (PAGE_SIZE / sizeof(struct scsiif_request_segment))
+
+/* Size of one request is 252 bytes */
+struct vscsiif_request {
+	uint16_t rqid;		/* private guest value, echoed in resp  */
+	uint8_t act;		/* command between backend and frontend */
+	uint8_t cmd_len;	/* valid CDB bytes */
+
+	uint8_t cmnd[VSCSIIF_MAX_COMMAND_SIZE];	/* the CDB */
+	uint16_t timeout_per_command;	/* deprecated */
+	uint16_t channel, id, lun;	/* (virtual) device specification */
+	uint16_t ref_rqid;		/* command abort reference */
+	uint8_t sc_data_direction;	/* for DMA_TO_DEVICE(1)
+					   DMA_FROM_DEVICE(2)
+					   DMA_NONE(3) requests */
+	uint8_t nr_segments;		/* Number of pieces of scatter-gather */
+/*
+ * flag in nr_segments: SG elements via grant page
+ *
+ * If VSCSIIF_SG_GRANT is set, the low 7 bits of nr_segments specify the number
+ * of grant pages containing SG elements. Usable if "feature-sg-grant" set.
+ */
+#define VSCSIIF_SG_GRANT	0x80
+
+	struct scsiif_request_segment seg[VSCSIIF_SG_TABLESIZE];
+	uint32_t reserved[3];
+};
+
+/* Size of one response is 252 bytes */
+struct vscsiif_response {
+	uint16_t rqid;		/* identifies request */
+	uint8_t padding;
+	uint8_t sense_len;
+	uint8_t sense_buffer[VSCSIIF_SENSE_BUFFERSIZE];
+	int32_t rslt;
+	uint32_t residual_len;	/* request bufflen -
+				   return the value from physical device */
+	uint32_t reserved[36];
+};
+
+DEFINE_RING_TYPES(vscsiif, struct vscsiif_request, struct vscsiif_response);
+
+#endif /*__XEN__PUBLIC_IO_SCSI_H__*/
Signature · 1 line
1.8.4.5
author maintainer
#4 lore raw

PATCH v6 2/5 Add XEN pvSCSI protocol description

In reply to Juergen Gross

Add the definition of pvSCSI protocol used between the pvSCSI frontend in a XEN domU and the pvSCSI backend in a XEN driver domain (usually Dom0).

This header was originally provided by Fujitsu for XEN based on Linux 2.6.18. Changes are:
- added comment
- adapt to Linux style guide
- add support for larger SG-lists by putting them in an own granted page
- remove stale definitions

---
 include/xen/interface/io/vscsiif.h | 229 +++++++++++++++++++++++++++++++++++++
 1 file changed, 229 insertions(+)
 create mode 100644 include/xen/interface/io/vscsiif.h
include/xen/interface/io/vscsiif.h new file +229 -0
new file mode 100644
index 0000000..d07d7ac
@@ -0,0 +1,229 @@
+/******************************************************************************
+ * vscsiif.h
+ *
+ * Based on the blkif.h code.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * 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) FUJITSU Limited 2008.
+ */
+
+#ifndef __XEN__PUBLIC_IO_SCSI_H__
+#define __XEN__PUBLIC_IO_SCSI_H__
+
+#include "ring.h"
+#include "../grant_table.h"
+
+/*
+ * Feature and Parameter Negotiation
+ * =================================
+ * The two halves of a Xen pvSCSI driver utilize nodes within the XenStore to
+ * communicate capabilities and to negotiate operating parameters.  This
+ * section enumerates these nodes which reside in the respective front and
+ * backend portions of the XenStore, following the XenBus convention.
+ *
+ * Any specified default value is in effect if the corresponding XenBus node
+ * is not present in the XenStore.
+ *
+ * XenStore nodes in sections marked "PRIVATE" are solely for use by the
+ * driver side whose XenBus tree contains them.
+ *
+ *****************************************************************************
+ *                            Backend XenBus Nodes
+ *****************************************************************************
+ *
+ *------------------ Backend Device Identification (PRIVATE) ------------------
+ *
+ * p-devname
+ *      Values:         string
+ *
+ *      A free string used to identify the physical device (e.g. a disk name).
+ *
+ * p-dev
+ *      Values:         string
+ *
+ *      A string specifying the backend device: either a 4-tuple "h:c:t:l"
+ *      (host, controller, target, lun, all integers), or a WWN (e.g.
+ *      "naa.60014054ac780582").
+ *
+ * v-dev
+ *      Values:         string
+ *
+ *      A string specifying the frontend device in form of a 4-tuple "h:c:t:l"
+ *      (host, controller, target, lun, all integers).
+ *
+ *--------------------------------- Features ---------------------------------
+ *
+ * feature-sg-grant
+ *      Values:         unsigned [VSCSIIF_SG_TABLESIZE...65535]
+ *      Default Value:  0
+ *
+ *      Specifies the maximum number of scatter/gather elements in grant pages
+ *      supported. If not set, the backend supports up to VSCSIIF_SG_TABLESIZE
+ *      SG elements specified directly in the request.
+ *
+ *****************************************************************************
+ *                            Frontend XenBus Nodes
+ *****************************************************************************
+ *
+ *----------------------- Request Transport Parameters -----------------------
+ *
+ * event-channel
+ *      Values:         unsigned
+ *
+ *      The identifier of the Xen event channel used to signal activity
+ *      in the ring buffer.
+ *
+ * ring-ref
+ *      Values:         unsigned
+ *
+ *      The Xen grant reference granting permission for the backend to map
+ *      the sole page in a single page sized ring buffer.
+ *
+ * protocol
+ *      Values:         string (XEN_IO_PROTO_ABI_*)
+ *      Default Value:  XEN_IO_PROTO_ABI_NATIVE
+ *
+ *      The machine ABI rules governing the format of all ring request and
+ *      response structures.
+ */
+
+/* Requests from the frontend to the backend */
+
+/*
+ * Request a SCSI operation specified via a CDB in vscsiif_request.cmnd.
+ * The target is specified via channel, id and lun.
+ *
+ * The operation to be performed is specified via a CDB in cmnd[], the length
+ * of the CDB is in cmd_len. sc_data_direction specifies the direction of data
+ * (to the device, from the device, or none at all).
+ *
+ * If data is to be transferred to or from the device the buffer(s) in the
+ * guest memory is/are specified via one or multiple scsiif_request_segment
+ * descriptors each specifying a memory page via a grant_ref_t, a offset into
+ * the page and the length of the area in that page. All scsiif_request_segment
+ * areas concatenated form the resulting data buffer used by the operation.
+ * If the number of scsiif_request_segment areas is not too large (less than
+ * or equal VSCSIIF_SG_TABLESIZE) the areas can be specified directly in the
+ * seg[] array and the number of valid scsiif_request_segment elements is to be
+ * set in nr_segments.
+ *
+ * If "feature-sg-grant" in the Xenstore is set it is possible to specify more
+ * than VSCSIIF_SG_TABLESIZE scsiif_request_segment elements via indirection.
+ * The maximum number of allowed scsiif_request_segment elements is the value
+ * of the "feature-sg-grant" entry from Xenstore. When using indirection the
+ * seg[] array doesn't contain specifications of the data buffers, but
+ * references to scsiif_request_segment arrays, which in turn reference the
+ * data buffers. While nr_segments holds the number of populated seg[] entries
+ * (plus the set VSCSIIF_SG_GRANT bit), the number of scsiif_request_segment
+ * elements referencing the target data buffers is calculated from the lengths
+ * of the seg[] elements (the sum of all valid seg[].length divided by the
+ * size of one scsiif_request_segment structure).
+ */
+#define VSCSIIF_ACT_SCSI_CDB		1
+
+/*
+ * Request abort of a running operation for the specified target given by
+ * channel, id, lun and the operation's rqid in ref_rqid.
+ */
+#define VSCSIIF_ACT_SCSI_ABORT		2
+
+/*
+ * Request a device reset of the specified target (channel and id).
+ */
+#define VSCSIIF_ACT_SCSI_RESET		3
+
+/*
+ * Preset scatter/gather elements for a following request. Deprecated.
+ * Keeping the define only to avoid usage of the value "4" for other actions.
+ */
+#define VSCSIIF_ACT_SCSI_SG_PRESET	4
+
+/*
+ * Maximum scatter/gather segments per request.
+ *
+ * Considering balance between allocating at least 16 "vscsiif_request"
+ * structures on one page (4096 bytes) and the number of scatter/gather
+ * elements needed, we decided to use 26 as a magic number.
+ *
+ * If "feature-sg-grant" is set, more scatter/gather elements can be specified
+ * by placing them in one or more (up to VSCSIIF_SG_TABLESIZE) granted pages.
+ * In this case the vscsiif_request seg elements don't contain references to
+ * the user data, but to the SG elements referencing the user data.
+ */
+#define VSCSIIF_SG_TABLESIZE		26
+
+/*
+ * based on Linux kernel 2.6.18, still valid
+ * Changing these values requires support of multiple protocols via the rings
+ * as "old clients" will blindly use these values and the resulting structure
+ * sizes.
+ */
+#define VSCSIIF_MAX_COMMAND_SIZE	16
+#define VSCSIIF_SENSE_BUFFERSIZE	96
+
+struct scsiif_request_segment {
+	grant_ref_t gref;
+	uint16_t offset;
+	uint16_t length;
+};
+
+#define VSCSIIF_SG_PER_PAGE (PAGE_SIZE / sizeof(struct scsiif_request_segment))
+
+/* Size of one request is 252 bytes */
+struct vscsiif_request {
+	uint16_t rqid;		/* private guest value, echoed in resp  */
+	uint8_t act;		/* command between backend and frontend */
+	uint8_t cmd_len;	/* valid CDB bytes */
+
+	uint8_t cmnd[VSCSIIF_MAX_COMMAND_SIZE];	/* the CDB */
+	uint16_t timeout_per_command;	/* deprecated */
+	uint16_t channel, id, lun;	/* (virtual) device specification */
+	uint16_t ref_rqid;		/* command abort reference */
+	uint8_t sc_data_direction;	/* for DMA_TO_DEVICE(1)
+					   DMA_FROM_DEVICE(2)
+					   DMA_NONE(3) requests */
+	uint8_t nr_segments;		/* Number of pieces of scatter-gather */
+/*
+ * flag in nr_segments: SG elements via grant page
+ *
+ * If VSCSIIF_SG_GRANT is set, the low 7 bits of nr_segments specify the number
+ * of grant pages containing SG elements. Usable if "feature-sg-grant" set.
+ */
+#define VSCSIIF_SG_GRANT	0x80
+
+	struct scsiif_request_segment seg[VSCSIIF_SG_TABLESIZE];
+	uint32_t reserved[3];
+};
+
+/* Size of one response is 252 bytes */
+struct vscsiif_response {
+	uint16_t rqid;		/* identifies request */
+	uint8_t padding;
+	uint8_t sense_len;
+	uint8_t sense_buffer[VSCSIIF_SENSE_BUFFERSIZE];
+	int32_t rslt;
+	uint32_t residual_len;	/* request bufflen -
+				   return the value from physical device */
+	uint32_t reserved[36];
+};
+
+DEFINE_RING_TYPES(vscsiif, struct vscsiif_request, struct vscsiif_response);
+
+#endif /*__XEN__PUBLIC_IO_SCSI_H__*/
Signature · 1 line
1.8.4.5
author maintainer
#5 lore raw

PATCH v6 3/5 Introduce xen-scsifront module

In reply to Juergen Gross

Introduces the XEN pvSCSI frontend. With pvSCSI it is possible for a XEN domU to issue SCSI commands to a SCSI LUN assigned to that domU. The SCSI commands are passed to the pvSCSI backend in a driver domain (usually Dom0) which is owner of the physical device. This allows e.g. to use SCSI tape drives in a XEN domU.

The code is taken from the pvSCSI implementation in XEN done by Fujitsu based on Linux kernel 2.6.18.

Changes from the original version are:
- port to upstream kernel
- put all code in just one source file
- move module to appropriate location in kernel tree
- adapt to Linux style guide
- some minor code simplifications
- replace constants with defines
- remove not used defines
- add support for larger SG lists by putting them in a granted page

Xen related parts

---
 drivers/scsi/Kconfig         |    9 +
 drivers/scsi/Makefile        |    1 +
 drivers/scsi/xen-scsifront.c | 1024 ++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 1034 insertions(+)
 create mode 100644 drivers/scsi/xen-scsifront.c
drivers/scsi/Kconfig +9 -0
index 18a3358..9130df1 100644
@@ -576,6 +576,15 @@ config VMWARE_PVSCSI
 	  To compile this driver as a module, choose M here: the
 	  module will be called vmw_pvscsi.
 
+config XEN_SCSI_FRONTEND
+	tristate "XEN SCSI frontend driver"
+	depends on SCSI && XEN
+	help
+	  The XEN SCSI frontend driver allows the kernel to access SCSI Devices
+	  within another guest OS (usually Dom0).
+	  Only needed if the kernel is running in a XEN guest and generic
+	  SCSI access to a device is needed.
+
 config HYPERV_STORAGE
 	tristate "Microsoft Hyper-V virtual storage driver"
 	depends on SCSI && HYPERV
drivers/scsi/Makefile +1 -0
index 5f0d299..59f1ce6 100644
@@ -141,6 +141,7 @@ obj-$(CONFIG_SCSI_ESAS2R)	+= esas2r/
 obj-$(CONFIG_SCSI_PMCRAID)	+= pmcraid.o
 obj-$(CONFIG_SCSI_VIRTIO)	+= virtio_scsi.o
 obj-$(CONFIG_VMWARE_PVSCSI)	+= vmw_pvscsi.o
+obj-$(CONFIG_XEN_SCSI_FRONTEND)	+= xen-scsifront.o
 obj-$(CONFIG_HYPERV_STORAGE)	+= hv_storvsc.o
 
 obj-$(CONFIG_ARM)		+= arm/
drivers/scsi/xen-scsifront.c new file +1024 -0

1,024 lines not on this page · raw message

Signature · 1 line
1.8.4.5
Acked-by on v6 3/5
#6 lore raw

Re: [PATCH V6 3/5] Introduce xen-scsifront module

Looks good to me from the SCSI POV. As discussed earlier I'm fine with picking this up via the Xen tree. If you really want it to go through the SCSI tree and I'd be happy to pick it up once I get another full review.

author maintainer
#8 lore raw

PATCH v6 3/5 Introduce xen-scsifront module

In reply to Juergen Gross

Introduces the XEN pvSCSI frontend. With pvSCSI it is possible for a XEN domU to issue SCSI commands to a SCSI LUN assigned to that domU. The SCSI commands are passed to the pvSCSI backend in a driver domain (usually Dom0) which is owner of the physical device. This allows e.g. to use SCSI tape drives in a XEN domU.

The code is taken from the pvSCSI implementation in XEN done by Fujitsu based on Linux kernel 2.6.18.

Changes from the original version are:
- port to upstream kernel
- put all code in just one source file
- move module to appropriate location in kernel tree
- adapt to Linux style guide
- some minor code simplifications
- replace constants with defines
- remove not used defines
- add support for larger SG lists by putting them in a granted page

Xen related parts

---
 drivers/scsi/Kconfig         |    9 +
 drivers/scsi/Makefile        |    1 +
 drivers/scsi/xen-scsifront.c | 1024 ++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 1034 insertions(+)
 create mode 100644 drivers/scsi/xen-scsifront.c
drivers/scsi/Kconfig +9 -0
index 18a3358..9130df1 100644
@@ -576,6 +576,15 @@ config VMWARE_PVSCSI
 	  To compile this driver as a module, choose M here: the
 	  module will be called vmw_pvscsi.
 
+config XEN_SCSI_FRONTEND
+	tristate "XEN SCSI frontend driver"
+	depends on SCSI && XEN
+	help
+	  The XEN SCSI frontend driver allows the kernel to access SCSI Devices
+	  within another guest OS (usually Dom0).
+	  Only needed if the kernel is running in a XEN guest and generic
+	  SCSI access to a device is needed.
+
 config HYPERV_STORAGE
 	tristate "Microsoft Hyper-V virtual storage driver"
 	depends on SCSI && HYPERV
drivers/scsi/Makefile +1 -0
index 5f0d299..59f1ce6 100644
@@ -141,6 +141,7 @@ obj-$(CONFIG_SCSI_ESAS2R)	+= esas2r/
 obj-$(CONFIG_SCSI_PMCRAID)	+= pmcraid.o
 obj-$(CONFIG_SCSI_VIRTIO)	+= virtio_scsi.o
 obj-$(CONFIG_VMWARE_PVSCSI)	+= vmw_pvscsi.o
+obj-$(CONFIG_XEN_SCSI_FRONTEND)	+= xen-scsifront.o
 obj-$(CONFIG_HYPERV_STORAGE)	+= hv_storvsc.o
 
 obj-$(CONFIG_ARM)		+= arm/
drivers/scsi/xen-scsifront.c new file +1024 -0

1,024 lines not on this page · raw message

Signature · 1 line
1.8.4.5
author maintainer
#9 lore raw

PATCH v6 4/5 Introduce XEN scsiback module

In reply to Juergen Gross

Introduces the XEN pvSCSI backend. With pvSCSI it is possible for a XEN domU to issue SCSI commands to a SCSI LUN assigned to that domU. The SCSI commands are passed to the pvSCSI backend in a driver domain (usually Dom0) which is owner of the physical device. This allows e.g. to use SCSI tape drives in a XEN domU.

The code is taken from the pvSCSI implementation in XEN done by Fujitsu based on Linux kernel 2.6.18.

Changes from the original version are:
- port to upstream kernel
- put all code in just one source file
- adapt to Linux style guide
- use target core infrastructure instead doing pure pass-through
- enable module unloading
- support SG-list in grant page(s)
- support task abort
- remove redundant struct backend
- allocate resources dynamically
- correct minor error in scsiback_fast_flush_area
- free allocated resources in case of error during I/O preparation
- remove CDB emulation, now handled by target core infrastructure

Xen related parts

---
 drivers/xen/Kconfig        |    9 +
 drivers/xen/Makefile       |    1 +
 drivers/xen/xen-scsiback.c | 2124 ++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 2134 insertions(+)
 create mode 100644 drivers/xen/xen-scsiback.c
drivers/xen/Kconfig +9 -0
index 8bc0183..b812462 100644
@@ -172,6 +172,15 @@ config XEN_PCIDEV_BACKEND
 
 	  If in doubt, say m.
 
+config XEN_SCSI_BACKEND
+	tristate "XEN SCSI backend driver"
+	depends on XEN && XEN_BACKEND && TARGET_CORE
+	help
+	  The SCSI backend driver allows the kernel to export its SCSI Devices
+	  to other guests via a high-performance shared-memory interface.
+	  Only needed for systems running as XEN driver domains (e.g. Dom0) and
+	  if guests need generic access to SCSI devices.
+
 config XEN_PRIVCMD
 	tristate
 	depends on XEN
drivers/xen/Makefile +1 -0
index 84044b5..2140398 100644
@@ -36,6 +36,7 @@ obj-$(CONFIG_XEN_ACPI_HOTPLUG_MEMORY)	+= xen-acpi-memhotplug.o
 obj-$(CONFIG_XEN_ACPI_HOTPLUG_CPU)	+= xen-acpi-cpuhotplug.o
 obj-$(CONFIG_XEN_ACPI_PROCESSOR)	+= xen-acpi-processor.o
 obj-$(CONFIG_XEN_EFI)			+= efi.o
+obj-$(CONFIG_XEN_SCSI_BACKEND)		+= xen-scsiback.o
 xen-evtchn-y				:= evtchn.o
 xen-gntdev-y				:= gntdev.o
 xen-gntalloc-y				:= gntalloc.o
drivers/xen/xen-scsiback.c new file +2124 -0

2,124 lines not on this page · raw message

Signature · 1 line
1.8.4.5
author maintainer
#10 lore raw

PATCH v6 4/5 Introduce XEN scsiback module

In reply to Juergen Gross

Introduces the XEN pvSCSI backend. With pvSCSI it is possible for a XEN domU to issue SCSI commands to a SCSI LUN assigned to that domU. The SCSI commands are passed to the pvSCSI backend in a driver domain (usually Dom0) which is owner of the physical device. This allows e.g. to use SCSI tape drives in a XEN domU.

The code is taken from the pvSCSI implementation in XEN done by Fujitsu based on Linux kernel 2.6.18.

Changes from the original version are:
- port to upstream kernel
- put all code in just one source file
- adapt to Linux style guide
- use target core infrastructure instead doing pure pass-through
- enable module unloading
- support SG-list in grant page(s)
- support task abort
- remove redundant struct backend
- allocate resources dynamically
- correct minor error in scsiback_fast_flush_area
- free allocated resources in case of error during I/O preparation
- remove CDB emulation, now handled by target core infrastructure

Xen related parts

---
 drivers/xen/Kconfig        |    9 +
 drivers/xen/Makefile       |    1 +
 drivers/xen/xen-scsiback.c | 2124 ++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 2134 insertions(+)
 create mode 100644 drivers/xen/xen-scsiback.c
drivers/xen/Kconfig +9 -0
index 8bc0183..b812462 100644
@@ -172,6 +172,15 @@ config XEN_PCIDEV_BACKEND
 
 	  If in doubt, say m.
 
+config XEN_SCSI_BACKEND
+	tristate "XEN SCSI backend driver"
+	depends on XEN && XEN_BACKEND && TARGET_CORE
+	help
+	  The SCSI backend driver allows the kernel to export its SCSI Devices
+	  to other guests via a high-performance shared-memory interface.
+	  Only needed for systems running as XEN driver domains (e.g. Dom0) and
+	  if guests need generic access to SCSI devices.
+
 config XEN_PRIVCMD
 	tristate
 	depends on XEN
drivers/xen/Makefile +1 -0
index 84044b5..2140398 100644
@@ -36,6 +36,7 @@ obj-$(CONFIG_XEN_ACPI_HOTPLUG_MEMORY)	+= xen-acpi-memhotplug.o
 obj-$(CONFIG_XEN_ACPI_HOTPLUG_CPU)	+= xen-acpi-cpuhotplug.o
 obj-$(CONFIG_XEN_ACPI_PROCESSOR)	+= xen-acpi-processor.o
 obj-$(CONFIG_XEN_EFI)			+= efi.o
+obj-$(CONFIG_XEN_SCSI_BACKEND)		+= xen-scsiback.o
 xen-evtchn-y				:= evtchn.o
 xen-gntdev-y				:= gntdev.o
 xen-gntalloc-y				:= gntalloc.o
drivers/xen/xen-scsiback.c new file +2124 -0

2,124 lines not on this page · raw message

Signature · 1 line
1.8.4.5
author maintainer
#11 lore raw

PATCH v6 5/5 add xen pvscsi maintainer

In reply to Juergen Gross

Add myself as maintainer for the Xen pvSCSI stuff.

---
 MAINTAINERS | 9 +++++++++
 1 file changed, 9 insertions(+)
MAINTAINERS +9 -0
index 1ff06de..0ae6238 100644
@@ -10128,6 +10128,15 @@ S:	Supported
 F:	drivers/block/xen-blkback/*
 F:	drivers/block/xen*
 
+XEN PVSCSI DRIVERS
+M:	Juergen Gross <[email protected]>
+L:	[email protected] (moderated for non-subscribers)
+L:	[email protected]
+S:	Supported
+F:	drivers/scsi/xen-scsifront.c
+F:	drivers/xen/xen-scsiback.c
+F:	include/xen/interface/io/vscsiif.h
+
 XEN SWIOTLB SUBSYSTEM
 M:	Konrad Rzeszutek Wilk <[email protected]>
 L:	[email protected] (moderated for non-subscribers)
Signature · 1 line
1.8.4.5
author maintainer
#12 lore raw

PATCH v6 5/5 add xen pvscsi maintainer

In reply to Juergen Gross

Add myself as maintainer for the Xen pvSCSI stuff.

---
 MAINTAINERS | 9 +++++++++
 1 file changed, 9 insertions(+)
MAINTAINERS +9 -0
index 1ff06de..0ae6238 100644
@@ -10128,6 +10128,15 @@ S:	Supported
 F:	drivers/block/xen-blkback/*
 F:	drivers/block/xen*
 
+XEN PVSCSI DRIVERS
+M:	Juergen Gross <[email protected]>
+L:	[email protected] (moderated for non-subscribers)
+L:	[email protected]
+S:	Supported
+F:	drivers/scsi/xen-scsifront.c
+F:	drivers/xen/xen-scsiback.c
+F:	include/xen/interface/io/vscsiif.h
+
 XEN SWIOTLB SUBSYSTEM
 M:	Konrad Rzeszutek Wilk <[email protected]>
 L:	[email protected] (moderated for non-subscribers)
Signature · 1 line
1.8.4.5
Applied
#13 lore raw
In reply to Juergen Gross
Quoted 2 lines · Juergen Gross

This series adds XEN pvSCSI support. With pvSCSI it is possible to use physical SCSI devices from a XEN domain.

Applied to devel/for-linus-3.18.

Thanks.

Your lines in your descriptions were too long for the default git log format. I fixed this up but can you ensure they're wrapped at 72 characters next time. Also Xen isn't an abbreviation and should not be capitalized.

David

Applied
#14 lore raw
In reply to Juergen Gross
Quoted 2 lines · Juergen Gross

This series adds XEN pvSCSI support. With pvSCSI it is possible to use physical SCSI devices from a XEN domain.

Applied to devel/for-linus-3.18.

Thanks.

Your lines in your descriptions were too long for the default git log format. I fixed this up but can you ensure they're wrapped at 72 characters next time. Also Xen isn't an abbreviation and should not be capitalized.

David