[SCSI] ibmvscsi: handle large scatter/gather lists
authorJames Bottomley <jejb@mulgrave.(none)>
Tue, 6 Sep 2005 22:11:54 +0000 (17:11 -0500)
committerJames Bottomley <jejb@mulgrave.(none)>
Tue, 6 Sep 2005 22:11:54 +0000 (17:11 -0500)
The maximum size of a scatter-gather list that the current IBM VSCSI
Client can handle is 10.  This patch adds large scatter-gather support
to the client so that it is capable of handling up to SG_ALL(255)
number of requests in the scatter-gather list.

Signed-off-by: Linda Xie <lxie@us.ibm.com>
Acked by: Dave C Boutcher <sleddog@us.ibm.com>

Rejections fixed up and
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
drivers/scsi/ibmvscsi/ibmvscsi.c
drivers/scsi/ibmvscsi/ibmvscsi.h

index e3e6752c4104ae6a7cd24ddbca2c07ec346a4407..1b911dadf64bbfbc2408191e4a7c4f56b2b96580 100644 (file)
@@ -87,7 +87,7 @@ static int max_channel = 3;
 static int init_timeout = 5;
 static int max_requests = 50;
 
-#define IBMVSCSI_VERSION "1.5.6"
+#define IBMVSCSI_VERSION "1.5.7"
 
 MODULE_DESCRIPTION("IBM Virtual SCSI");
 MODULE_AUTHOR("Dave Boutcher");
@@ -145,6 +145,8 @@ static int initialize_event_pool(struct event_pool *pool,
                        sizeof(*evt->xfer_iu) * i;
                evt->xfer_iu = pool->iu_storage + i;
                evt->hostdata = hostdata;
+               evt->ext_list = NULL;
+               evt->ext_list_token = 0;
        }
 
        return 0;
@@ -161,9 +163,16 @@ static void release_event_pool(struct event_pool *pool,
                               struct ibmvscsi_host_data *hostdata)
 {
        int i, in_use = 0;
-       for (i = 0; i < pool->size; ++i)
+       for (i = 0; i < pool->size; ++i) {
                if (atomic_read(&pool->events[i].free) != 1)
                        ++in_use;
+               if (pool->events[i].ext_list) {
+                       dma_free_coherent(hostdata->dev,
+                                 SG_ALL * sizeof(struct memory_descriptor),
+                                 pool->events[i].ext_list,
+                                 pool->events[i].ext_list_token);
+               }
+       }
        if (in_use)
                printk(KERN_WARNING
                       "ibmvscsi: releasing event pool with %d "
@@ -286,24 +295,41 @@ static void set_srp_direction(struct scsi_cmnd *cmd,
        } else {
                if (cmd->sc_data_direction == DMA_TO_DEVICE) {
                        srp_cmd->data_out_format = SRP_INDIRECT_BUFFER;
-                       srp_cmd->data_out_count = numbuf;
+                       srp_cmd->data_out_count =
+                               numbuf < MAX_INDIRECT_BUFS ?
+                                       numbuf: MAX_INDIRECT_BUFS;
                } else {
                        srp_cmd->data_in_format = SRP_INDIRECT_BUFFER;
-                       srp_cmd->data_in_count = numbuf;
+                       srp_cmd->data_in_count =
+                               numbuf < MAX_INDIRECT_BUFS ?
+                                       numbuf: MAX_INDIRECT_BUFS;
                }
        }
 }
 
+static void unmap_sg_list(int num_entries, 
+               struct device *dev,
+               struct memory_descriptor *md)
+{ 
+       int i;
+
+       for (i = 0; i < num_entries; ++i) {
+               dma_unmap_single(dev,
+                       md[i].virtual_address,
+                       md[i].length, DMA_BIDIRECTIONAL);
+       }
+}
+
 /**
  * unmap_cmd_data: - Unmap data pointed in srp_cmd based on the format
  * @cmd:       srp_cmd whose additional_data member will be unmapped
  * @dev:       device for which the memory is mapped
  *
 */
-static void unmap_cmd_data(struct srp_cmd *cmd, struct device *dev)
+static void unmap_cmd_data(struct srp_cmd *cmd,
+                          struct srp_event_struct *evt_struct,
+                          struct device *dev)
 {
-       int i;
-
        if ((cmd->data_out_format == SRP_NO_BUFFER) &&
            (cmd->data_in_format == SRP_NO_BUFFER))
                return;
@@ -318,15 +344,34 @@ static void unmap_cmd_data(struct srp_cmd *cmd, struct device *dev)
                        (struct indirect_descriptor *)cmd->additional_data;
                int num_mapped = indirect->head.length / 
                        sizeof(indirect->list[0]);
-               for (i = 0; i < num_mapped; ++i) {
-                       struct memory_descriptor *data = &indirect->list[i];
-                       dma_unmap_single(dev,
-                                        data->virtual_address,
-                                        data->length, DMA_BIDIRECTIONAL);
+
+               if (num_mapped <= MAX_INDIRECT_BUFS) {
+                       unmap_sg_list(num_mapped, dev, &indirect->list[0]);
+                       return;
                }
+
+               unmap_sg_list(num_mapped, dev, evt_struct->ext_list);
        }
 }
 
+static int map_sg_list(int num_entries, 
+                      struct scatterlist *sg,
+                      struct memory_descriptor *md)
+{
+       int i;
+       u64 total_length = 0;
+
+       for (i = 0; i < num_entries; ++i) {
+               struct memory_descriptor *descr = md + i;
+               struct scatterlist *sg_entry = &sg[i];
+               descr->virtual_address = sg_dma_address(sg_entry);
+               descr->length = sg_dma_len(sg_entry);
+               descr->memory_handle = 0;
+               total_length += sg_dma_len(sg_entry);
+       }
+       return total_length;
+}
+
 /**
  * map_sg_data: - Maps dma for a scatterlist and initializes decriptor fields
  * @cmd:       Scsi_Cmnd with the scatterlist
@@ -337,10 +382,11 @@ static void unmap_cmd_data(struct srp_cmd *cmd, struct device *dev)
  * Returns 1 on success.
 */
 static int map_sg_data(struct scsi_cmnd *cmd,
+                      struct srp_event_struct *evt_struct,
                       struct srp_cmd *srp_cmd, struct device *dev)
 {
 
-       int i, sg_mapped;
+       int sg_mapped;
        u64 total_length = 0;
        struct scatterlist *sg = cmd->request_buffer;
        struct memory_descriptor *data =
@@ -363,27 +409,46 @@ static int map_sg_data(struct scsi_cmnd *cmd,
                return 1;
        }
 
-       if (sg_mapped > MAX_INDIRECT_BUFS) {
+       if (sg_mapped > SG_ALL) {
                printk(KERN_ERR
                       "ibmvscsi: More than %d mapped sg entries, got %d\n",
-                      MAX_INDIRECT_BUFS, sg_mapped);
+                      SG_ALL, sg_mapped);
                return 0;
        }
 
        indirect->head.virtual_address = 0;
        indirect->head.length = sg_mapped * sizeof(indirect->list[0]);
        indirect->head.memory_handle = 0;
-       for (i = 0; i < sg_mapped; ++i) {
-               struct memory_descriptor *descr = &indirect->list[i];
-               struct scatterlist *sg_entry = &sg[i];
-               descr->virtual_address = sg_dma_address(sg_entry);
-               descr->length = sg_dma_len(sg_entry);
-               descr->memory_handle = 0;
-               total_length += sg_dma_len(sg_entry);
+
+       if (sg_mapped <= MAX_INDIRECT_BUFS) {
+               total_length = map_sg_list(sg_mapped, sg, &indirect->list[0]);
+               indirect->total_length = total_length;
+               return 1;
        }
-       indirect->total_length = total_length;
 
-       return 1;
+       /* get indirect table */
+       if (!evt_struct->ext_list) {
+               evt_struct->ext_list =(struct memory_descriptor*)
+                       dma_alloc_coherent(dev, 
+                               SG_ALL * sizeof(struct memory_descriptor),
+                               &evt_struct->ext_list_token, 0);
+               if (!evt_struct->ext_list) {
+                   printk(KERN_ERR
+                       "ibmvscsi: Can't allocate memory for indirect table\n");
+                       return 0;
+                       
+               }
+       }
+
+       total_length = map_sg_list(sg_mapped, sg, evt_struct->ext_list);        
+
+       indirect->total_length = total_length;
+       indirect->head.virtual_address = evt_struct->ext_list_token;
+       indirect->head.length = sg_mapped * sizeof(indirect->list[0]);
+       memcpy(indirect->list, evt_struct->ext_list,
+               MAX_INDIRECT_BUFS * sizeof(struct memory_descriptor));
+       
+       return 1;
 }
 
 /**
@@ -428,6 +493,7 @@ static int map_single_data(struct scsi_cmnd *cmd,
  * Returns 1 on success.
 */
 static int map_data_for_srp_cmd(struct scsi_cmnd *cmd,
+                               struct srp_event_struct *evt_struct,
                                struct srp_cmd *srp_cmd, struct device *dev)
 {
        switch (cmd->sc_data_direction) {
@@ -450,7 +516,7 @@ static int map_data_for_srp_cmd(struct scsi_cmnd *cmd,
        if (!cmd->request_buffer)
                return 1;
        if (cmd->use_sg)
-               return map_sg_data(cmd, srp_cmd, dev);
+               return map_sg_data(cmd, evt_struct, srp_cmd, dev);
        return map_single_data(cmd, srp_cmd, dev);
 }
 
@@ -486,6 +552,7 @@ static int ibmvscsi_send_srp_event(struct srp_event_struct *evt_struct,
                printk(KERN_WARNING 
                       "ibmvscsi: Warning, request_limit exceeded\n");
                unmap_cmd_data(&evt_struct->iu.srp.cmd,
+                              evt_struct,
                               hostdata->dev);
                free_event_struct(&hostdata->pool, evt_struct);
                return SCSI_MLQUEUE_HOST_BUSY;
@@ -513,7 +580,7 @@ static int ibmvscsi_send_srp_event(struct srp_event_struct *evt_struct,
        return 0;
 
  send_error:
-       unmap_cmd_data(&evt_struct->iu.srp.cmd, hostdata->dev);
+       unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
 
        if ((cmnd = evt_struct->cmnd) != NULL) {
                cmnd->result = DID_ERROR << 16;
@@ -551,6 +618,7 @@ static void handle_cmd_rsp(struct srp_event_struct *evt_struct)
                               rsp->sense_and_response_data,
                               rsp->sense_data_list_length);
                unmap_cmd_data(&evt_struct->iu.srp.cmd, 
+                              evt_struct, 
                               evt_struct->hostdata->dev);
 
                if (rsp->doover)
@@ -583,6 +651,7 @@ static int ibmvscsi_queuecommand(struct scsi_cmnd *cmnd,
 {
        struct srp_cmd *srp_cmd;
        struct srp_event_struct *evt_struct;
+       struct indirect_descriptor *indirect;
        struct ibmvscsi_host_data *hostdata =
                (struct ibmvscsi_host_data *)&cmnd->device->host->hostdata;
        u16 lun = lun_from_dev(cmnd->device);
@@ -591,14 +660,6 @@ static int ibmvscsi_queuecommand(struct scsi_cmnd *cmnd,
        if (!evt_struct)
                return SCSI_MLQUEUE_HOST_BUSY;
 
-       init_event_struct(evt_struct,
-                         handle_cmd_rsp,
-                         VIOSRP_SRP_FORMAT,
-                         cmnd->timeout_per_command/HZ);
-
-       evt_struct->cmnd = cmnd;
-       evt_struct->cmnd_done = done;
-
        /* Set up the actual SRP IU */
        srp_cmd = &evt_struct->iu.srp.cmd;
        memset(srp_cmd, 0x00, sizeof(*srp_cmd));
@@ -606,17 +667,25 @@ static int ibmvscsi_queuecommand(struct scsi_cmnd *cmnd,
        memcpy(srp_cmd->cdb, cmnd->cmnd, sizeof(cmnd->cmnd));
        srp_cmd->lun = ((u64) lun) << 48;
 
-       if (!map_data_for_srp_cmd(cmnd, srp_cmd, hostdata->dev)) {
+       if (!map_data_for_srp_cmd(cmnd, evt_struct, srp_cmd, hostdata->dev)) {
                printk(KERN_ERR "ibmvscsi: couldn't convert cmd to srp_cmd\n");
                free_event_struct(&hostdata->pool, evt_struct);
                return SCSI_MLQUEUE_HOST_BUSY;
        }
 
+       init_event_struct(evt_struct,
+                         handle_cmd_rsp,
+                         VIOSRP_SRP_FORMAT,
+                         cmnd->timeout_per_command/HZ);
+
+       evt_struct->cmnd = cmnd;
+       evt_struct->cmnd_done = done;
+
        /* Fix up dma address of the buffer itself */
-       if ((srp_cmd->data_out_format == SRP_INDIRECT_BUFFER) ||
-           (srp_cmd->data_in_format == SRP_INDIRECT_BUFFER)) {
-               struct indirect_descriptor *indirect =
-                   (struct indirect_descriptor *)srp_cmd->additional_data;
+       indirect = (struct indirect_descriptor *)srp_cmd->additional_data;
+       if (((srp_cmd->data_out_format == SRP_INDIRECT_BUFFER) ||
+           (srp_cmd->data_in_format == SRP_INDIRECT_BUFFER)) &&
+           (indirect->head.virtual_address == 0)) {
                indirect->head.virtual_address = evt_struct->crq.IU_data_ptr +
                    offsetof(struct srp_cmd, additional_data) +
                    offsetof(struct indirect_descriptor, list);
@@ -931,7 +1000,8 @@ static int ibmvscsi_eh_abort_handler(struct scsi_cmnd *cmd)
 
        cmd->result = (DID_ABORT << 16);
        list_del(&found_evt->list);
-       unmap_cmd_data(&found_evt->iu.srp.cmd, found_evt->hostdata->dev);
+       unmap_cmd_data(&found_evt->iu.srp.cmd, found_evt,
+                      found_evt->hostdata->dev);
        free_event_struct(&found_evt->hostdata->pool, found_evt);
        spin_unlock_irqrestore(hostdata->host->host_lock, flags);
        atomic_inc(&hostdata->request_limit);
@@ -1023,7 +1093,8 @@ static int ibmvscsi_eh_device_reset_handler(struct scsi_cmnd *cmd)
                        if (tmp_evt->cmnd)
                                tmp_evt->cmnd->result = (DID_RESET << 16);
                        list_del(&tmp_evt->list);
-                       unmap_cmd_data(&tmp_evt->iu.srp.cmd, tmp_evt->hostdata->dev);
+                       unmap_cmd_data(&tmp_evt->iu.srp.cmd, tmp_evt,
+                                      tmp_evt->hostdata->dev);
                        free_event_struct(&tmp_evt->hostdata->pool,
                                                   tmp_evt);
                        atomic_inc(&hostdata->request_limit);
@@ -1052,6 +1123,7 @@ static void purge_requests(struct ibmvscsi_host_data *hostdata)
                if (tmp_evt->cmnd) {
                        tmp_evt->cmnd->result = (DID_ERROR << 16);
                        unmap_cmd_data(&tmp_evt->iu.srp.cmd, 
+                                      tmp_evt, 
                                       tmp_evt->hostdata->dev);
                        if (tmp_evt->cmnd_done)
                                tmp_evt->cmnd_done(tmp_evt->cmnd);
@@ -1356,7 +1428,7 @@ static struct scsi_host_template driver_template = {
        .cmd_per_lun = 16,
        .can_queue = 1,         /* Updated after SRP_LOGIN */
        .this_id = -1,
-       .sg_tablesize = MAX_INDIRECT_BUFS,
+       .sg_tablesize = SG_ALL,
        .use_clustering = ENABLE_CLUSTERING,
        .shost_attrs = ibmvscsi_attrs,
 };
index 1030b703c30e93ba415f2790aa7123712530dac7..8bec0438dc8a9151f3aed2b04a717a0ee6a65be2 100644 (file)
@@ -68,6 +68,8 @@ struct srp_event_struct {
        void (*cmnd_done) (struct scsi_cmnd *);
        struct completion comp;
        union viosrp_iu *sync_srp;
+       struct memory_descriptor *ext_list;
+       dma_addr_t ext_list_token;
 };
 
 /* a pool of event structs for use */