Skip to contentSkip to conversation

please pull infiniband.git

GIT PULL Roland Dreier 1 participant 1 message Nov 13, 2006 InfiniBand

lore
Share & tools
lore
#0 lore raw

Linus, please pull from

master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband.git for-linus

This tree is also available from kernel.org mirrors at:

git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband.git for-linus

This includes various small fixes for 2.6.19-rc6:

Hoang-Nam Nguyen (3):
IB/ehca: Assure 4K alignment for firmware control blocks IB/ehca: Use named constant for max mtu IB/ehca: Activate scaling code by default

Jean Delvare (1):

      RDMA/amso1100: Fix && typo

Roland Dreier (1):
IB/mad: Fix race between cancel and receive completion

Tom Tucker (1):
RDMA/amso1100: Fix unitialized pseudo_netdev accessed in c2_register_device

 drivers/infiniband/core/mad.c                |    2 +-
 drivers/infiniband/hw/amso1100/c2.c          |    3 +-
 drivers/infiniband/hw/amso1100/c2_provider.c |   39 +++++++++---------
 drivers/infiniband/hw/amso1100/c2_rnic.c     |    4 +-
 drivers/infiniband/hw/ehca/Kconfig           |    1 +
 drivers/infiniband/hw/ehca/ehca_av.c         |    5 +-
 drivers/infiniband/hw/ehca/ehca_hca.c        |   17 ++++----
 drivers/infiniband/hw/ehca/ehca_irq.c        |   17 ++++----
 drivers/infiniband/hw/ehca/ehca_iverbs.h     |    8 ++++
 drivers/infiniband/hw/ehca/ehca_main.c       |   56 +++++++++++++++++++++----
 drivers/infiniband/hw/ehca/ehca_mrmw.c       |    8 ++--
 drivers/infiniband/hw/ehca/ehca_qp.c         |   10 ++--
 drivers/infiniband/hw/ehca/hipz_hw.h         |    2 +
 13 files changed, 111 insertions(+), 61 deletions(-)

The full log and patch is below:

commit 39798695b4bcc7b145f8910ca56195808d3a7637

Author: Roland Dreier <[email protected]>
Date:   Mon Nov 13 09:38:07 2006 -0800

IB/mad: Fix race between cancel and receive completion

When ib_cancel_mad() is called, it puts the canceled send on a list and schedules a "flushed" callback from process context. However, this leaves a window where a receive completion could be processed before the send is fully flushed.

This is fine, except that ib_find_send_mad() will find the MAD and return it to the receive processing, which results in the sender getting both a successful receive and a "flushed" send completion for the same request. Understandably, this confuses the sender, which is expecting only one of these two callbacks, and leads to grief such as a use-after-free in IPoIB.

Fix this by changing ib_find_send_mad() to return a send struct only if the status is still successful (and not "flushed"). The search of the send_list already had this check, so this patch just adds the same check to the search of the wait_list.

commit b26c791e9ca3365616d40836000285931ca033d0

Author: Jean Delvare <[email protected]>
Date:   Thu Nov 9 21:02:26 2006 +0100

RDMA/amso1100: Fix && typo

Fix the AMSO1100 firmware version computation, which was broken due to "&&" being used where "&" should have.

commit 2ffcab6ae44b02679229ca1852526d0a6e062dd2

Author: Tom Tucker <[email protected]>
Date:   Wed Nov 8 14:23:22 2006 -0600

RDMA/amso1100: Fix unitialized pseudo_netdev accessed in c2_register_device

Rework some load-time error handling: c2_register_device() leaked when it failed, and the function that called it didn't check the return code.

commit f2c238a0c5e155acd49752c5fb93fb8d8534232b

Author: Hoang-Nam Nguyen <[email protected]>
Date:   Sun Nov 5 21:42:20 2006 +0100

IB/ehca: Activate scaling code by default

Change ehca's Kconfig to activates scaling code as default. After several measurements we saw that this feature prevents dropped packets (UD) in stress situation. Thus, enabling it helps to improve ehca's bandwidth through IPoIB.

commit c58121143f87930621c1a6fa9683b6862f2b42c9

Author: Hoang-Nam Nguyen <[email protected]>
Date:   Sun Nov 5 21:42:56 2006 +0100

IB/ehca: Use named constant for max mtu

Define and use a constant EHCA_MAX_MTU instead hardcoded value.

commit 7e28db5d8ff63b1cabc221c5cb84a5f45752f1c2

Author: Hoang-Nam Nguyen <[email protected]>
Date:   Tue Nov 7 00:56:39 2006 +0100

IB/ehca: Assure 4K alignment for firmware control blocks

Assure 4K alignment for firmware control blocks in 64K page mode, because kzalloc()'s result address might not be 4K aligned if 64K pages are enabled. Thus, we introduce wrappers called ehca_{alloc,free}_fw_ctrlblock(), which use a slab cache for objects with 4K length and 4K alignment in order to alloc/free firmware control blocks in 64K page mode. In 4K page mode those wrappers just are defines of get_zeroed_page() and free_page().

commit 39798695b4bcc7b145f8910ca56195808d3a7637

Author: Roland Dreier <[email protected]>
Date:   Mon Nov 13 09:38:07 2006 -0800

IB/mad: Fix race between cancel and receive completion

When ib_cancel_mad() is called, it puts the canceled send on a list and schedules a "flushed" callback from process context. However, this leaves a window where a receive completion could be processed before the send is fully flushed.

This is fine, except that ib_find_send_mad() will find the MAD and return it to the receive processing, which results in the sender getting both a successful receive and a "flushed" send completion for the same request. Understandably, this confuses the sender, which is expecting only one of these two callbacks, and leads to grief such as a use-after-free in IPoIB.

Fix this by changing ib_find_send_mad() to return a send struct only if the status is still successful (and not "flushed"). The search of the send_list already had this check, so this patch just adds the same check to the search of the wait_list.

commit b26c791e9ca3365616d40836000285931ca033d0

Author: Jean Delvare <[email protected]>
Date:   Thu Nov 9 21:02:26 2006 +0100

RDMA/amso1100: Fix && typo

Fix the AMSO1100 firmware version computation, which was broken due to "&&" being used where "&" should have.

commit 2ffcab6ae44b02679229ca1852526d0a6e062dd2

Author: Tom Tucker <[email protected]>
Date:   Wed Nov 8 14:23:22 2006 -0600

RDMA/amso1100: Fix unitialized pseudo_netdev accessed in c2_register_device

Rework some load-time error handling: c2_register_device() leaked when it failed, and the function that called it didn't check the return code.

commit f2c238a0c5e155acd49752c5fb93fb8d8534232b

Author: Hoang-Nam Nguyen <[email protected]>
Date:   Sun Nov 5 21:42:20 2006 +0100

IB/ehca: Activate scaling code by default

Change ehca's Kconfig to activates scaling code as default. After several measurements we saw that this feature prevents dropped packets (UD) in stress situation. Thus, enabling it helps to improve ehca's bandwidth through IPoIB.

commit c58121143f87930621c1a6fa9683b6862f2b42c9

Author: Hoang-Nam Nguyen <[email protected]>
Date:   Sun Nov 5 21:42:56 2006 +0100

IB/ehca: Use named constant for max mtu

Define and use a constant EHCA_MAX_MTU instead hardcoded value.

commit 7e28db5d8ff63b1cabc221c5cb84a5f45752f1c2

Author: Hoang-Nam Nguyen <[email protected]>
Date:   Tue Nov 7 00:56:39 2006 +0100

IB/ehca: Assure 4K alignment for firmware control blocks

Assure 4K alignment for firmware control blocks in 64K page mode, because kzalloc()'s result address might not be 4K aligned if 64K pages are enabled. Thus, we introduce wrappers called ehca_{alloc,free}_fw_ctrlblock(), which use a slab cache for objects with 4K length and 4K alignment in order to alloc/free firmware control blocks in 64K page mode. In 4K page mode those wrappers just are defines of get_zeroed_page() and free_page().

drivers/infiniband/core/mad.c +1 -1
index 493f4c6..a72bcea 100644
@@ -1750,7 +1750,7 @@ ib_find_send_mad(struct ib_mad_agent_pri
 		     */
 		    (is_direct(wc->recv_buf.mad->mad_hdr.mgmt_class) ||
 		     rcv_has_same_gid(mad_agent_priv, wr, wc)))
-			return wr;
+			return (wr->status == IB_WC_SUCCESS) ? wr : NULL;
 	}
 
 	/*
drivers/infiniband/hw/amso1100/c2.c +2 -1
index 9e7bd94..27fe242 100644
@@ -1155,7 +1155,8 @@ static int __devinit c2_probe(struct pci
 		goto bail10;
 	}
 
-	c2_register_device(c2dev);
+	if (c2_register_device(c2dev))
+		goto bail10;
 
 	return 0;
 
drivers/infiniband/hw/amso1100/c2_provider.c +20 -19
index da98d9f..fef9727 100644
@@ -757,20 +757,17 @@ #endif
 
 int c2_register_device(struct c2_dev *dev)
 {
-	int ret;
+	int ret = -ENOMEM;
 	int i;
 
 	/* Register pseudo network device */
 	dev->pseudo_netdev = c2_pseudo_netdev_init(dev);
-	if (dev->pseudo_netdev) {
-		ret = register_netdev(dev->pseudo_netdev);
-		if (ret) {
-			printk(KERN_ERR PFX
-				"Unable to register netdev, ret = %d\n", ret);
-			free_netdev(dev->pseudo_netdev);
-			return ret;
-		}
-	}
+	if (!dev->pseudo_netdev)
+		goto out3;
+
+	ret = register_netdev(dev->pseudo_netdev);
+	if (ret)
+		goto out2;
 
 	pr_debug("%s:%u\n", __FUNCTION__, __LINE__);
 	strlcpy(dev->ibdev.name, "amso%d", IB_DEVICE_NAME_MAX);
@@ -848,21 +845,25 @@ int c2_register_device(struct c2_dev *de
 
 	ret = ib_register_device(&dev->ibdev);
 	if (ret)
-		return ret;
+		goto out1;
 
 	for (i = 0; i < ARRAY_SIZE(c2_class_attributes); ++i) {
 		ret = class_device_create_file(&dev->ibdev.class_dev,
 					       c2_class_attributes[i]);
-		if (ret) {
-			unregister_netdev(dev->pseudo_netdev);
-			free_netdev(dev->pseudo_netdev);
-			ib_unregister_device(&dev->ibdev);
-			return ret;
-		}
+		if (ret)
+			goto out0;
 	}
+	goto out3;
 
-	pr_debug("%s:%u\n", __FUNCTION__, __LINE__);
-	return 0;
+out0:
+	ib_unregister_device(&dev->ibdev);
+out1:
+	unregister_netdev(dev->pseudo_netdev);
+out2:
+	free_netdev(dev->pseudo_netdev);
+out3:
+	pr_debug("%s:%u ret=%d\n", __FUNCTION__, __LINE__, ret);
+	return ret;
 }
 
 void c2_unregister_device(struct c2_dev *dev)
drivers/infiniband/hw/amso1100/c2_rnic.c +2 -2
index 21d9612..623dc95 100644
@@ -157,8 +157,8 @@ static int c2_rnic_query(struct c2_dev *
 
 	props->fw_ver =
 		((u64)be32_to_cpu(reply->fw_ver_major) << 32) |
-		((be32_to_cpu(reply->fw_ver_minor) && 0xFFFF) << 16) |
-		(be32_to_cpu(reply->fw_ver_patch) && 0xFFFF);
+		((be32_to_cpu(reply->fw_ver_minor) & 0xFFFF) << 16) |
+		(be32_to_cpu(reply->fw_ver_patch) & 0xFFFF);
 	memcpy(&props->sys_image_guid, c2dev->netdev->dev_addr, 6);
 	props->max_mr_size         = 0xFFFFFFFF;
 	props->page_size_cap       = ~(C2_MIN_PAGESIZE-1);
drivers/infiniband/hw/ehca/Kconfig +1 -0
index 922389b..727b10d 100644
@@ -10,6 +10,7 @@ config INFINIBAND_EHCA
 config INFINIBAND_EHCA_SCALING
 	bool "Scaling support (EXPERIMENTAL)"
 	depends on IBMEBUS && INFINIBAND_EHCA && HOTPLUG_CPU && EXPERIMENTAL
+	default y
 	---help---
 	eHCA scaling support schedules the CQ callbacks to different CPUs.
 
drivers/infiniband/hw/ehca/ehca_av.c +2 -3
index 3bac197..214e2fd 100644
@@ -118,8 +118,7 @@ struct ib_ah *ehca_create_ah(struct ib_p
 		}
 		memcpy(&av->av.grh.word_1, &gid, sizeof(gid));
 	}
-	/* for the time being we use a hard coded PMTU of 2048 Bytes */
-	av->av.pmtu = 4;
+	av->av.pmtu = EHCA_MAX_MTU;
 
 	/* dgid comes in grh.word_3 */
 	memcpy(&av->av.grh.word_3, &ah_attr->grh.dgid,
@@ -193,7 +192,7 @@ int ehca_modify_ah(struct ib_ah *ah, str
 		memcpy(&new_ehca_av.grh.word_1, &gid, sizeof(gid));
 	}
 
-	new_ehca_av.pmtu = 4; /* see also comment in create_ah() */
+	new_ehca_av.pmtu = EHCA_MAX_MTU;
 
 	memcpy(&new_ehca_av.grh.word_3, &ah_attr->grh.dgid,
 	       sizeof(ah_attr->grh.dgid));
drivers/infiniband/hw/ehca/ehca_hca.c +9 -8
index 5eae6ac..e1b618c 100644
@@ -40,6 +40,7 @@
  */
 
 #include "ehca_tools.h"
+#include "ehca_iverbs.h"
 #include "hcp_if.h"
 
 int ehca_query_device(struct ib_device *ibdev, struct ib_device_attr *props)
@@ -49,7 +50,7 @@ int ehca_query_device(struct ib_device *
 					      ib_device);
 	struct hipz_query_hca *rblock;
 
-	rblock = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
+	rblock = ehca_alloc_fw_ctrlblock();
 	if (!rblock) {
 		ehca_err(&shca->ib_device, "Can't allocate rblock memory.");
 		return -ENOMEM;
@@ -96,7 +97,7 @@ int ehca_query_device(struct ib_device *
 		= min_t(int, rblock->max_total_mcast_qp_attach, INT_MAX);
 
 query_device1:
-	kfree(rblock);
+	ehca_free_fw_ctrlblock(rblock);
 
 	return ret;
 }
@@ -109,7 +110,7 @@ int ehca_query_port(struct ib_device *ib
 					      ib_device);
 	struct hipz_query_port *rblock;
 
-	rblock = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
+	rblock = ehca_alloc_fw_ctrlblock();
 	if (!rblock) {
 		ehca_err(&shca->ib_device, "Can't allocate rblock memory.");
 		return -ENOMEM;
@@ -162,7 +163,7 @@ int ehca_query_port(struct ib_device *ib
 	props->active_speed    = 0x1;
 
 query_port1:
-	kfree(rblock);
+	ehca_free_fw_ctrlblock(rblock);
 
 	return ret;
 }
@@ -178,7 +179,7 @@ int ehca_query_pkey(struct ib_device *ib
 		return -EINVAL;
 	}
 
-	rblock = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
+	rblock = ehca_alloc_fw_ctrlblock();
 	if (!rblock) {
 		ehca_err(&shca->ib_device,  "Can't allocate rblock memory.");
 		return -ENOMEM;
@@ -193,7 +194,7 @@ int ehca_query_pkey(struct ib_device *ib
 	memcpy(pkey, &rblock->pkey_entries + index, sizeof(u16));
 
 query_pkey1:
-	kfree(rblock);
+	ehca_free_fw_ctrlblock(rblock);
 
 	return ret;
 }
@@ -211,7 +212,7 @@ int ehca_query_gid(struct ib_device *ibd
 		return -EINVAL;
 	}
 
-	rblock = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
+	rblock = ehca_alloc_fw_ctrlblock();
 	if (!rblock) {
 		ehca_err(&shca->ib_device, "Can't allocate rblock memory.");
 		return -ENOMEM;
@@ -227,7 +228,7 @@ int ehca_query_gid(struct ib_device *ibd
 	memcpy(&gid->raw[8], &rblock->guid_entries[index], sizeof(u64));
 
 query_gid1:
-	kfree(rblock);
+	ehca_free_fw_ctrlblock(rblock);
 
 	return ret;
 }
drivers/infiniband/hw/ehca/ehca_irq.c +8 -9
index 048cc44..c3ea746 100644
@@ -45,6 +45,7 @@ #include "ehca_iverbs.h"
 #include "ehca_tools.h"
 #include "hcp_if.h"
 #include "hipz_fns.h"
+#include "ipz_pt_fn.h"
 
 #define EQE_COMPLETION_EVENT   EHCA_BMASK_IBM(1,1)
 #define EQE_CQ_QP_NUMBER       EHCA_BMASK_IBM(8,31)
@@ -137,38 +138,36 @@ int ehca_error_data(struct ehca_shca *sh
 	u64 *rblock;
 	unsigned long block_count;
 
-	rblock = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
+	rblock = ehca_alloc_fw_ctrlblock();
 	if (!rblock) {
 		ehca_err(&shca->ib_device, "Cannot allocate rblock memory.");
 		ret = -ENOMEM;
 		goto error_data1;
 	}
 
+	/* rblock must be 4K aligned and should be 4K large */
 	ret = hipz_h_error_data(shca->ipz_hca_handle,
 				resource,
 				rblock,
 				&block_count);
 
-	if (ret == H_R_STATE) {
+	if (ret == H_R_STATE)
 		ehca_err(&shca->ib_device,
 			 "No error data is available: %lx.", resource);
-	}
 	else if (ret == H_SUCCESS) {
 		int length;
 
 		length = EHCA_BMASK_GET(ERROR_DATA_LENGTH, rblock[0]);
 
-		if (length > PAGE_SIZE)
-			length = PAGE_SIZE;
+		if (length > EHCA_PAGESIZE)
+			length = EHCA_PAGESIZE;
 
 		print_error_data(shca, data, rblock, length);
-	}
-	else {
+	} else
 		ehca_err(&shca->ib_device,
 			 "Error data could not be fetched: %lx", resource);
-	}
 
-	kfree(rblock);
+	ehca_free_fw_ctrlblock(rblock);
 
 error_data1:
 	return ret;
drivers/infiniband/hw/ehca/ehca_iverbs.h +8 -0
index 319c39d..3720e30 100644
@@ -179,4 +179,12 @@ int ehca_mmap_register(u64 physical,void
 
 int ehca_munmap(unsigned long addr, size_t len);
 
+#ifdef CONFIG_PPC_64K_PAGES
+void *ehca_alloc_fw_ctrlblock(void);
+void ehca_free_fw_ctrlblock(void *ptr);
+#else
+#define ehca_alloc_fw_ctrlblock() ((void *) get_zeroed_page(GFP_KERNEL))
+#define ehca_free_fw_ctrlblock(ptr) free_page((unsigned long)(ptr))
+#endif
+
 #endif
drivers/infiniband/hw/ehca/ehca_main.c +47 -9

137 lines not on this page · raw message

drivers/infiniband/hw/ehca/ehca_mrmw.c +4 -4
index 5ca6544..abce676 100644
@@ -1013,7 +1013,7 @@ int ehca_reg_mr_rpages(struct ehca_shca
 	u32 i;
 	u64 *kpage;
 
-	kpage = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
+	kpage = ehca_alloc_fw_ctrlblock();
 	if (!kpage) {
 		ehca_err(&shca->ib_device, "kpage alloc failed");
 		ret = -ENOMEM;
@@ -1092,7 +1092,7 @@ int ehca_reg_mr_rpages(struct ehca_shca
 
 
 ehca_reg_mr_rpages_exit1:
-	kfree(kpage);
+	ehca_free_fw_ctrlblock(kpage);
 ehca_reg_mr_rpages_exit0:
 	if (ret)
 		ehca_err(&shca->ib_device, "ret=%x shca=%p e_mr=%p pginfo=%p "
@@ -1124,7 +1124,7 @@ inline int ehca_rereg_mr_rereg1(struct e
 	ehca_mrmw_map_acl(acl, &hipz_acl);
 	ehca_mrmw_set_pgsize_hipz_acl(&hipz_acl);
 
-	kpage = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
+	kpage = ehca_alloc_fw_ctrlblock();
 	if (!kpage) {
 		ehca_err(&shca->ib_device, "kpage alloc failed");
 		ret = -ENOMEM;
@@ -1181,7 +1181,7 @@ inline int ehca_rereg_mr_rereg1(struct e
 	}
 
 ehca_rereg_mr_rereg1_exit1:
-	kfree(kpage);
+	ehca_free_fw_ctrlblock(kpage);
 ehca_rereg_mr_rereg1_exit0:
 	if ( ret && (ret != -EAGAIN) )
 		ehca_err(&shca->ib_device, "ret=%x lkey=%x rkey=%x "
drivers/infiniband/hw/ehca/ehca_qp.c +5 -5
index 4394123..cf3e50e 100644
@@ -811,8 +811,8 @@ static int internal_modify_qp(struct ib_
 	unsigned long spl_flags = 0;
 
 	/* do query_qp to obtain current attr values */
-	mqpcb = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
-	if (mqpcb == NULL) {
+	mqpcb = ehca_alloc_fw_ctrlblock();
+	if (!mqpcb) {
 		ehca_err(ibqp->device, "Could not get zeroed page for mqpcb "
 			 "ehca_qp=%p qp_num=%x ", my_qp, ibqp->qp_num);
 		return -ENOMEM;
@@ -1225,7 +1225,7 @@ modify_qp_exit2:
 	}
 
 modify_qp_exit1:
-	kfree(mqpcb);
+	ehca_free_fw_ctrlblock(mqpcb);
 
 	return ret;
 }
@@ -1277,7 +1277,7 @@ int ehca_query_qp(struct ib_qp *qp,
 		return -EINVAL;
 	}
 
-	qpcb = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL );
+	qpcb = ehca_alloc_fw_ctrlblock();
 	if (!qpcb) {
 		ehca_err(qp->device,"Out of memory for qpcb "
 			 "ehca_qp=%p qp_num=%x", my_qp, qp->qp_num);
@@ -1401,7 +1401,7 @@ int ehca_query_qp(struct ib_qp *qp,
 		ehca_dmp(qpcb, 4*70, "qp_num=%x", qp->qp_num);
 
 query_qp_exit1:
-	kfree(qpcb);
+	ehca_free_fw_ctrlblock(qpcb);
 
 	return ret;
 }
drivers/infiniband/hw/ehca/hipz_hw.h +2 -0
index 3fc92b0..fad9136 100644
@@ -45,6 +45,8 @@ #define __HIPZ_HW_H__
 
 #include "ehca_tools.h"
 
+#define EHCA_MAX_MTU 4
+
 /* QP Table Entry Memory Map */
 struct hipz_qptemm {
 	u64 qpx_hcr;