[SCSI] ipr: Faster sg list fetch

Improve overall command performance by embedding the scatterlist
in the command block used by the adapter. This decreases
the overall number of DMAs required for a single command.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

2 files changed, 39 insertions, 20 deletions

diff --git a/drivers/scsi/ipr.c b/drivers/scsi/ipr.c
index aa1fb72ca235..5cf1002283b4 100644
--- a/drivers/scsi/ipr.c
+++ b/drivers/scsi/ipr.c
@@ -480,12 +480,16 @@ static void ipr_reinit_ipr_cmnd(struct ipr_cmnd *ipr_cmd)
 {
         struct ipr_ioarcb *ioarcb = &ipr_cmd->ioarcb;
         struct ipr_ioasa *ioasa = &ipr_cmd->ioasa;
+        dma_addr_t dma_addr = be32_to_cpu(ioarcb->ioarcb_host_pci_addr);
 
         memset(&ioarcb->cmd_pkt, 0, sizeof(struct ipr_cmd_pkt));
         ioarcb->write_data_transfer_length = 0;
         ioarcb->read_data_transfer_length = 0;
         ioarcb->write_ioadl_len = 0;
         ioarcb->read_ioadl_len = 0;
+        ioarcb->write_ioadl_addr =
+                cpu_to_be32(dma_addr + offsetof(struct ipr_cmnd, ioadl));
+        ioarcb->read_ioadl_addr = ioarcb->write_ioadl_addr;
         ioasa->ioasc = 0;
         ioasa->residual_data_len = 0;
         ioasa->u.gata.status = 0;
@@ -4230,6 +4234,14 @@ static int ipr_build_ioadl(struct ipr_ioa_cfg *ioa_cfg,
 
                 sglist = scsi_cmd->request_buffer;
 
+                if (ipr_cmd->dma_use_sg <= ARRAY_SIZE(ioarcb->add_data.u.ioadl)) {
+                        ioadl = ioarcb->add_data.u.ioadl;
+                        ioarcb->write_ioadl_addr =
+                                cpu_to_be32(be32_to_cpu(ioarcb->ioarcb_host_pci_addr) +
+                                            offsetof(struct ipr_ioarcb, add_data));
+                        ioarcb->read_ioadl_addr = ioarcb->write_ioadl_addr;
+                }
+
                 for (i = 0; i < ipr_cmd->dma_use_sg; i++) {
                         ioadl[i].flags_and_data_len =
                                 cpu_to_be32(ioadl_flags | sg_dma_len(&sglist[i]));
@@ -4260,6 +4272,11 @@ static int ipr_build_ioadl(struct ipr_ioa_cfg *ioa_cfg,
                                                      scsi_cmd->sc_data_direction);
 
                 if (likely(!pci_dma_mapping_error(ipr_cmd->dma_handle))) {
+                        ioadl = ioarcb->add_data.u.ioadl;
+                        ioarcb->write_ioadl_addr =
+                                cpu_to_be32(be32_to_cpu(ioarcb->ioarcb_host_pci_addr) +
+                                            offsetof(struct ipr_ioarcb, add_data));
+                        ioarcb->read_ioadl_addr = ioarcb->write_ioadl_addr;
                         ipr_cmd->dma_use_sg = 1;
                         ioadl[0].flags_and_data_len =
                                 cpu_to_be32(ioadl_flags | length | IPR_IOADL_FLAGS_LAST);
@@ -4346,11 +4363,9 @@ static void ipr_erp_done(struct ipr_cmnd *ipr_cmd)
  **/
 static void ipr_reinit_ipr_cmnd_for_erp(struct ipr_cmnd *ipr_cmd)
 {
-        struct ipr_ioarcb *ioarcb;
-        struct ipr_ioasa *ioasa;
-
-        ioarcb = &ipr_cmd->ioarcb;
-        ioasa = &ipr_cmd->ioasa;
+        struct ipr_ioarcb *ioarcb = &ipr_cmd->ioarcb;
+        struct ipr_ioasa *ioasa = &ipr_cmd->ioasa;
+        dma_addr_t dma_addr = be32_to_cpu(ioarcb->ioarcb_host_pci_addr);
 
         memset(&ioarcb->cmd_pkt, 0, sizeof(struct ipr_cmd_pkt));
         ioarcb->write_data_transfer_length = 0;
@@ -4359,6 +4374,9 @@ static void ipr_reinit_ipr_cmnd_for_erp(struct ipr_cmnd *ipr_cmd)
         ioarcb->read_ioadl_len = 0;
         ioasa->ioasc = 0;
         ioasa->residual_data_len = 0;
+        ioarcb->write_ioadl_addr =
+                cpu_to_be32(dma_addr + offsetof(struct ipr_cmnd, ioadl));
+        ioarcb->read_ioadl_addr = ioarcb->write_ioadl_addr;
 }
 
 /**
diff --git a/drivers/scsi/ipr.h b/drivers/scsi/ipr.h
index 4d068887608c..50e14f59f2fa 100644
--- a/drivers/scsi/ipr.h
+++ b/drivers/scsi/ipr.h
@@ -422,9 +422,25 @@ struct ipr_ioarcb_ata_regs {
         u8 ctl;
 }__attribute__ ((packed, aligned(4)));
 
+struct ipr_ioadl_desc {
+        __be32 flags_and_data_len;
+#define IPR_IOADL_FLAGS_MASK            0xff000000
+#define IPR_IOADL_GET_FLAGS(x) (be32_to_cpu(x) & IPR_IOADL_FLAGS_MASK)
+#define IPR_IOADL_DATA_LEN_MASK         0x00ffffff
+#define IPR_IOADL_GET_DATA_LEN(x) (be32_to_cpu(x) & IPR_IOADL_DATA_LEN_MASK)
+#define IPR_IOADL_FLAGS_READ            0x48000000
+#define IPR_IOADL_FLAGS_READ_LAST       0x49000000
+#define IPR_IOADL_FLAGS_WRITE           0x68000000
+#define IPR_IOADL_FLAGS_WRITE_LAST      0x69000000
+#define IPR_IOADL_FLAGS_LAST            0x01000000
+
+        __be32 address;
+}__attribute__((packed, aligned (8)));
+
 struct ipr_ioarcb_add_data {
         union {
                 struct ipr_ioarcb_ata_regs regs;
+                struct ipr_ioadl_desc ioadl[5];
                 __be32 add_cmd_parms[10];
         }u;
 }__attribute__ ((packed, aligned(4)));
@@ -456,21 +472,6 @@ struct ipr_ioarcb {
         struct ipr_ioarcb_add_data add_data;
 }__attribute__((packed, aligned (4)));
 
-struct ipr_ioadl_desc {
-        __be32 flags_and_data_len;
-#define IPR_IOADL_FLAGS_MASK            0xff000000
-#define IPR_IOADL_GET_FLAGS(x) (be32_to_cpu(x) & IPR_IOADL_FLAGS_MASK)
-#define IPR_IOADL_DATA_LEN_MASK         0x00ffffff
-#define IPR_IOADL_GET_DATA_LEN(x) (be32_to_cpu(x) & IPR_IOADL_DATA_LEN_MASK)
-#define IPR_IOADL_FLAGS_READ            0x48000000
-#define IPR_IOADL_FLAGS_READ_LAST       0x49000000
-#define IPR_IOADL_FLAGS_WRITE           0x68000000
-#define IPR_IOADL_FLAGS_WRITE_LAST      0x69000000
-#define IPR_IOADL_FLAGS_LAST            0x01000000
-
-        __be32 address;
-}__attribute__((packed, aligned (8)));
-
 struct ipr_ioasa_vset {
         __be32 failing_lba_hi;
         __be32 failing_lba_lo;