b=23193 Land ISER performance patch for Sundance

The purpose of this patch is to make ISER support 1MB RDMAs instead of splitting 2 512K RDMAs and increase the maximum number of ISCSI commands.

i=alexg

diff --git a/lustre/kernel_patches/patches/lustre_iser_max_sectors_tuning_lustre2.0.patch b/lustre/kernel_patches/patches/lustre_iser_max_sectors_tuning_lustre2.0.patch
new file mode 100644 (file)
index 0000000..9f3c500
--- /dev/null
+++ b/lustre/kernel_patches/patches/lustre_iser_max_sectors_tuning_lustre2.0.patch
@@ -0,0 +1,78 @@
+diff -Naur base.linux-2.6.18.x86_64/drivers/infiniband/ulp/iser/iscsi_iser.c linux-2.6.18.x86_64/drivers/infiniband/ulp/iser/iscsi_iser.c
+--- base.linux-2.6.18.x86_64/drivers/infiniband/ulp/iser/iscsi_iser.c  2010-09-09 16:57:15.000000000 -0400
++++ linux-2.6.18.x86_64/drivers/infiniband/ulp/iser/iscsi_iser.c       2010-09-09 17:02:17.000000000 -0400
+@@ -586,18 +586,25 @@
+       iser_conn_terminate(ib_conn);
+ }
+ 
++static int iscsi_iser_slave_configure(struct scsi_device *sdev)
++{
++      blk_queue_dma_alignment(sdev->request_queue, 0);
++      return 0;
++}
++
+ static struct scsi_host_template iscsi_iser_sht = {
+       .module                 = THIS_MODULE,
+       .name                   = "iSCSI Initiator over iSER, v." DRV_VER,
+       .queuecommand           = iscsi2_queuecommand,
+       .change_queue_depth     = iscsi2_change_queue_depth,
+       .sg_tablesize           = ISCSI_ISER_SG_TABLESIZE,
+-      .max_sectors            = 1024,
++      .max_sectors            = 0xffff,
+       .cmd_per_lun            = ISER_DEF_CMD_PER_LUN,
+       .eh_abort_handler       = iscsi2_eh_abort,
+       .eh_device_reset_handler= iscsi2_eh_device_reset,
+       .eh_host_reset_handler= iscsi2_eh_target_reset,
+       .use_clustering         = DISABLE_CLUSTERING,
++      .slave_configure        = iscsi_iser_slave_configure,
+       .proc_name              = "iscsi_iser",
+       .this_id                = -1,
+ };
+diff -Naur base.linux-2.6.18.x86_64/drivers/infiniband/ulp/iser/iscsi_iser.h linux-2.6.18.x86_64/drivers/infiniband/ulp/iser/iscsi_iser.h
+--- base.linux-2.6.18.x86_64/drivers/infiniband/ulp/iser/iscsi_iser.h  2010-09-09 16:57:15.000000000 -0400
++++ linux-2.6.18.x86_64/drivers/infiniband/ulp/iser/iscsi_iser.h       2010-09-09 17:03:17.000000000 -0400
+@@ -92,7 +92,8 @@
+ #define MASK_4K       (~(SIZE_4K-1))
+ 
+                                       /* support upto 512KB in one RDMA */
+-#define ISCSI_ISER_SG_TABLESIZE         (0x80000 >> SHIFT_4K)
++/* FMR space for 1 MB of 4k-page transfers, plus 1 if not page aligned */
++#define ISCSI_ISER_SG_TABLESIZE                (((1<<20) >> SHIFT_4K) + 1)
+ #define ISER_DEF_CMD_PER_LUN          128
+ 
+ /* QP settings */
+diff -Naur base.linux-2.6.18.x86_64/drivers/infiniband/ulp/iser/iser_verbs.c linux-2.6.18.x86_64/drivers/infiniband/ulp/iser/iser_verbs.c
+--- base.linux-2.6.18.x86_64/drivers/infiniband/ulp/iser/iser_verbs.c  2010-09-09 16:57:15.000000000 -0400
++++ linux-2.6.18.x86_64/drivers/infiniband/ulp/iser/iser_verbs.c       2010-09-09 17:04:44.000000000 -0400
+@@ -137,7 +137,7 @@
+       device = ib_conn->device;
+ 
+       ib_conn->page_vec = kmalloc(sizeof(struct iser_page_vec) +
+-                                  (sizeof(u64) * (ISCSI_ISER_SG_TABLESIZE +1)),
++                                  sizeof(u64) * ISCSI_ISER_SG_TABLESIZE,
+                                   GFP_KERNEL);
+       if (!ib_conn->page_vec) {
+               ret = -ENOMEM;
+@@ -146,9 +146,7 @@
+       ib_conn->page_vec->pages = (u64 *) (ib_conn->page_vec + 1);
+ 
+       params.page_shift        = SHIFT_4K;
+-      /* when the first/last SG element are not start/end *
+-       * page aligned, the map whould be of N+1 pages     */
+-      params.max_pages_per_fmr = ISCSI_ISER_SG_TABLESIZE + 1;
++      params.max_pages_per_fmr = ISCSI_ISER_SG_TABLESIZE;
+       /* make the pool size twice the max number of SCSI commands *
+        * the ML is expected to queue, watermark for unmap at 50%  */
+       params.pool_size         = ISCSI_DEF_XMIT_CMDS_MAX * 2;
+diff -Naur base.linux-2.6.18.x86_64/include/scsi/libiscsi2.h linux-2.6.18.x86_64/include/scsi/libiscsi2.h
+--- base.linux-2.6.18.x86_64/include/scsi/libiscsi2.h  2010-09-09 16:57:35.000000000 -0400
++++ linux-2.6.18.x86_64/include/scsi/libiscsi2.h       2010-09-09 17:05:34.000000000 -0400
+@@ -43,7 +43,7 @@
+ struct iscsi_nopin;
+ struct device;
+ 
+-#define ISCSI_DEF_XMIT_CMDS_MAX       128     /* must be power of 2 */
++#define ISCSI_DEF_XMIT_CMDS_MAX       256     /* must be power of 2 */
+ #define ISCSI_MGMT_CMDS_MAX   15
+ 
+ #define ISCSI_DEF_CMD_PER_LUN 32

diff --git a/lustre/kernel_patches/series/2.6-rhel5.series b/lustre/kernel_patches/series/2.6-rhel5.series
index 240695b..db50c09 100644 (file)
--- a/lustre/kernel_patches/series/2.6-rhel5.series
+++ b/lustre/kernel_patches/series/2.6-rhel5.series
@@ -28,3 +28,4 @@ jbd2-commit-timer-no-jiffies-rounding.diff
 md-avoid-bug_on-when-bmc-overflow.patch
 export-symbols-for-dmu.patch
 jbd2_stats_proc_init-wrong-place.patch
+lustre_iser_max_sectors_tuning_lustre2.0.patch