Merge remote-tracking branch 'spi/fix/omap2-mcspi' into spi-omap2-mcspi
[linux-2.6-block.git] / drivers / scsi / storvsc_drv.c
index efc6e446b6c83e31677ab6e5ec01424e6bab59cb..d9dad90344d545a18185ecf0736fa79a4b6a212b 100644 (file)
@@ -308,11 +308,16 @@ enum storvsc_request_type {
  * This is the end of Protocol specific defines.
  */
 
-static int storvsc_ringbuffer_size = (20 * PAGE_SIZE);
+static int storvsc_ringbuffer_size = (256 * PAGE_SIZE);
+static u32 max_outstanding_req_per_channel;
+
+static int storvsc_vcpus_per_sub_channel = 4;
 
 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
 
+module_param(storvsc_vcpus_per_sub_channel, int, S_IRUGO);
+MODULE_PARM_DESC(vcpus_per_sub_channel, "Ratio of VCPUs to subchannels");
 /*
  * Timeout in seconds for all devices managed by this driver.
  */
@@ -320,7 +325,6 @@ static int storvsc_timeout = 180;
 
 static int msft_blist_flags = BLIST_TRY_VPD_PAGES;
 
-#define STORVSC_MAX_IO_REQUESTS                                200
 
 static void storvsc_on_channel_callback(void *context);
 
@@ -347,7 +351,10 @@ struct storvsc_cmd_request {
        /* Synchronize the request/response if needed */
        struct completion wait_event;
 
-       struct hv_multipage_buffer data_buffer;
+       struct vmbus_channel_packet_multipage_buffer mpb;
+       struct vmbus_packet_mpb_array *payload;
+       u32 payload_sz;
+
        struct vstor_packet vstor_packet;
 };
 
@@ -373,6 +380,10 @@ struct storvsc_device {
        unsigned char path_id;
        unsigned char target_id;
 
+       /*
+        * Max I/O, the device can support.
+        */
+       u32   max_transfer_bytes;
        /* Used for vsc/vsp channel reset process */
        struct storvsc_cmd_request init_request;
        struct storvsc_cmd_request reset_request;
@@ -618,19 +629,6 @@ cleanup:
        return NULL;
 }
 
-/* Disgusting wrapper functions */
-static inline unsigned long sg_kmap_atomic(struct scatterlist *sgl, int idx)
-{
-       void *addr = kmap_atomic(sg_page(sgl + idx));
-       return (unsigned long)addr;
-}
-
-static inline void sg_kunmap_atomic(unsigned long addr)
-{
-       kunmap_atomic((void *)addr);
-}
-
-
 /* Assume the original sgl has enough room */
 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
                                            struct scatterlist *bounce_sgl,
@@ -645,32 +643,38 @@ static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
        unsigned long bounce_addr = 0;
        unsigned long dest_addr = 0;
        unsigned long flags;
+       struct scatterlist *cur_dest_sgl;
+       struct scatterlist *cur_src_sgl;
 
        local_irq_save(flags);
-
+       cur_dest_sgl = orig_sgl;
+       cur_src_sgl = bounce_sgl;
        for (i = 0; i < orig_sgl_count; i++) {
-               dest_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset;
+               dest_addr = (unsigned long)
+                               kmap_atomic(sg_page(cur_dest_sgl)) +
+                               cur_dest_sgl->offset;
                dest = dest_addr;
-               destlen = orig_sgl[i].length;
+               destlen = cur_dest_sgl->length;
 
                if (bounce_addr == 0)
-                       bounce_addr = sg_kmap_atomic(bounce_sgl,j);
+                       bounce_addr = (unsigned long)kmap_atomic(
+                                                       sg_page(cur_src_sgl));
 
                while (destlen) {
-                       src = bounce_addr + bounce_sgl[j].offset;
-                       srclen = bounce_sgl[j].length - bounce_sgl[j].offset;
+                       src = bounce_addr + cur_src_sgl->offset;
+                       srclen = cur_src_sgl->length - cur_src_sgl->offset;
 
                        copylen = min(srclen, destlen);
                        memcpy((void *)dest, (void *)src, copylen);
 
                        total_copied += copylen;
-                       bounce_sgl[j].offset += copylen;
+                       cur_src_sgl->offset += copylen;
                        destlen -= copylen;
                        dest += copylen;
 
-                       if (bounce_sgl[j].offset == bounce_sgl[j].length) {
+                       if (cur_src_sgl->offset == cur_src_sgl->length) {
                                /* full */
-                               sg_kunmap_atomic(bounce_addr);
+                               kunmap_atomic((void *)bounce_addr);
                                j++;
 
                                /*
@@ -684,21 +688,27 @@ static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
                                        /*
                                         * We are done; cleanup and return.
                                         */
-                                       sg_kunmap_atomic(dest_addr - orig_sgl[i].offset);
+                                       kunmap_atomic((void *)(dest_addr -
+                                               cur_dest_sgl->offset));
                                        local_irq_restore(flags);
                                        return total_copied;
                                }
 
                                /* if we need to use another bounce buffer */
-                               if (destlen || i != orig_sgl_count - 1)
-                                       bounce_addr = sg_kmap_atomic(bounce_sgl,j);
+                               if (destlen || i != orig_sgl_count - 1) {
+                                       cur_src_sgl = sg_next(cur_src_sgl);
+                                       bounce_addr = (unsigned long)
+                                                       kmap_atomic(
+                                                       sg_page(cur_src_sgl));
+                               }
                        } else if (destlen == 0 && i == orig_sgl_count - 1) {
                                /* unmap the last bounce that is < PAGE_SIZE */
-                               sg_kunmap_atomic(bounce_addr);
+                               kunmap_atomic((void *)bounce_addr);
                        }
                }
 
-               sg_kunmap_atomic(dest_addr - orig_sgl[i].offset);
+               kunmap_atomic((void *)(dest_addr - cur_dest_sgl->offset));
+               cur_dest_sgl = sg_next(cur_dest_sgl);
        }
 
        local_irq_restore(flags);
@@ -719,48 +729,62 @@ static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl,
        unsigned long bounce_addr = 0;
        unsigned long src_addr = 0;
        unsigned long flags;
+       struct scatterlist *cur_src_sgl;
+       struct scatterlist *cur_dest_sgl;
 
        local_irq_save(flags);
 
+       cur_src_sgl = orig_sgl;
+       cur_dest_sgl = bounce_sgl;
+
        for (i = 0; i < orig_sgl_count; i++) {
-               src_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset;
+               src_addr = (unsigned long)
+                               kmap_atomic(sg_page(cur_src_sgl)) +
+                               cur_src_sgl->offset;
                src = src_addr;
-               srclen = orig_sgl[i].length;
+               srclen = cur_src_sgl->length;
 
                if (bounce_addr == 0)
-                       bounce_addr = sg_kmap_atomic(bounce_sgl,j);
+                       bounce_addr = (unsigned long)
+                                       kmap_atomic(sg_page(cur_dest_sgl));
 
                while (srclen) {
                        /* assume bounce offset always == 0 */
-                       dest = bounce_addr + bounce_sgl[j].length;
-                       destlen = PAGE_SIZE - bounce_sgl[j].length;
+                       dest = bounce_addr + cur_dest_sgl->length;
+                       destlen = PAGE_SIZE - cur_dest_sgl->length;
 
                        copylen = min(srclen, destlen);
                        memcpy((void *)dest, (void *)src, copylen);
 
                        total_copied += copylen;
-                       bounce_sgl[j].length += copylen;
+                       cur_dest_sgl->length += copylen;
                        srclen -= copylen;
                        src += copylen;
 
-                       if (bounce_sgl[j].length == PAGE_SIZE) {
+                       if (cur_dest_sgl->length == PAGE_SIZE) {
                                /* full..move to next entry */
-                               sg_kunmap_atomic(bounce_addr);
+                               kunmap_atomic((void *)bounce_addr);
+                               bounce_addr = 0;
                                j++;
+                       }
 
-                               /* if we need to use another bounce buffer */
-                               if (srclen || i != orig_sgl_count - 1)
-                                       bounce_addr = sg_kmap_atomic(bounce_sgl,j);
-
-                       } else if (srclen == 0 && i == orig_sgl_count - 1) {
-                               /* unmap the last bounce that is < PAGE_SIZE */
-                               sg_kunmap_atomic(bounce_addr);
+                       /* if we need to use another bounce buffer */
+                       if (srclen && bounce_addr == 0) {
+                               cur_dest_sgl = sg_next(cur_dest_sgl);
+                               bounce_addr = (unsigned long)
+                                               kmap_atomic(
+                                               sg_page(cur_dest_sgl));
                        }
+
                }
 
-               sg_kunmap_atomic(src_addr - orig_sgl[i].offset);
+               kunmap_atomic((void *)(src_addr - cur_src_sgl->offset));
+               cur_src_sgl = sg_next(cur_src_sgl);
        }
 
+       if (bounce_addr)
+               kunmap_atomic((void *)bounce_addr);
+
        local_irq_restore(flags);
 
        return total_copied;
@@ -970,6 +994,8 @@ static int storvsc_channel_init(struct hv_device *device)
                    STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL)
                        process_sub_channels = true;
        }
+       stor_device->max_transfer_bytes =
+               vstor_packet->storage_channel_properties.max_transfer_bytes;
 
        memset(vstor_packet, 0, sizeof(struct vstor_packet));
        vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
@@ -1080,6 +1106,8 @@ static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request)
        struct Scsi_Host *host;
        struct storvsc_device *stor_dev;
        struct hv_device *dev = host_dev->dev;
+       u32 payload_sz = cmd_request->payload_sz;
+       void *payload = cmd_request->payload;
 
        stor_dev = get_in_stor_device(dev);
        host = stor_dev->host;
@@ -1109,10 +1137,14 @@ static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request)
                                         sense_hdr.ascq);
 
        scsi_set_resid(scmnd,
-               cmd_request->data_buffer.len -
+               cmd_request->payload->range.len -
                vm_srb->data_transfer_length);
 
        scmnd->scsi_done(scmnd);
+
+       if (payload_sz >
+               sizeof(struct vmbus_channel_packet_multipage_buffer))
+               kfree(payload);
 }
 
 static void storvsc_on_io_completion(struct hv_device *device,
@@ -1314,7 +1346,7 @@ static int storvsc_dev_remove(struct hv_device *device)
 }
 
 static int storvsc_do_io(struct hv_device *device,
-                             struct storvsc_cmd_request *request)
+                        struct storvsc_cmd_request *request)
 {
        struct storvsc_device *stor_device;
        struct vstor_packet *vstor_packet;
@@ -1346,19 +1378,20 @@ static int storvsc_do_io(struct hv_device *device,
 
 
        vstor_packet->vm_srb.data_transfer_length =
-       request->data_buffer.len;
+       request->payload->range.len;
 
        vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
 
-       if (request->data_buffer.len) {
-               ret = vmbus_sendpacket_multipagebuffer(outgoing_channel,
-                               &request->data_buffer,
+       if (request->payload->range.len) {
+
+               ret = vmbus_sendpacket_mpb_desc(outgoing_channel,
+                               request->payload, request->payload_sz,
                                vstor_packet,
                                (sizeof(struct vstor_packet) -
                                vmscsi_size_delta),
                                (unsigned long)request);
        } else {
-               ret = vmbus_sendpacket(device->channel, vstor_packet,
+               ret = vmbus_sendpacket(outgoing_channel, vstor_packet,
                               (sizeof(struct vstor_packet) -
                                vmscsi_size_delta),
                               (unsigned long)request,
@@ -1376,7 +1409,6 @@ static int storvsc_do_io(struct hv_device *device,
 
 static int storvsc_device_configure(struct scsi_device *sdevice)
 {
-       scsi_change_queue_depth(sdevice, STORVSC_MAX_IO_REQUESTS);
 
        blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE);
 
@@ -1526,6 +1558,10 @@ static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
        struct scatterlist *sgl;
        unsigned int sg_count = 0;
        struct vmscsi_request *vm_srb;
+       struct scatterlist *cur_sgl;
+       struct vmbus_packet_mpb_array  *payload;
+       u32 payload_sz;
+       u32 length;
 
        if (vmstor_current_major <= VMSTOR_WIN8_MAJOR) {
                /*
@@ -1579,46 +1615,71 @@ static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
 
        memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
 
-       cmd_request->data_buffer.len = scsi_bufflen(scmnd);
-       if (scsi_sg_count(scmnd)) {
-               sgl = (struct scatterlist *)scsi_sglist(scmnd);
-               sg_count = scsi_sg_count(scmnd);
+       sgl = (struct scatterlist *)scsi_sglist(scmnd);
+       sg_count = scsi_sg_count(scmnd);
+
+       length = scsi_bufflen(scmnd);
+       payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb;
+       payload_sz = sizeof(cmd_request->mpb);
 
+       if (sg_count) {
                /* check if we need to bounce the sgl */
                if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) {
                        cmd_request->bounce_sgl =
-                               create_bounce_buffer(sgl, scsi_sg_count(scmnd),
-                                                    scsi_bufflen(scmnd),
+                               create_bounce_buffer(sgl, sg_count,
+                                                    length,
                                                     vm_srb->data_in);
                        if (!cmd_request->bounce_sgl)
                                return SCSI_MLQUEUE_HOST_BUSY;
 
                        cmd_request->bounce_sgl_count =
-                               ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >>
-                                       PAGE_SHIFT;
+                               ALIGN(length, PAGE_SIZE) >> PAGE_SHIFT;
 
                        if (vm_srb->data_in == WRITE_TYPE)
                                copy_to_bounce_buffer(sgl,
-                                       cmd_request->bounce_sgl,
-                                       scsi_sg_count(scmnd));
+                                       cmd_request->bounce_sgl, sg_count);
 
                        sgl = cmd_request->bounce_sgl;
                        sg_count = cmd_request->bounce_sgl_count;
                }
 
-               cmd_request->data_buffer.offset = sgl[0].offset;
 
-               for (i = 0; i < sg_count; i++)
-                       cmd_request->data_buffer.pfn_array[i] =
-                               page_to_pfn(sg_page((&sgl[i])));
+               if (sg_count > MAX_PAGE_BUFFER_COUNT) {
+
+                       payload_sz = (sg_count * sizeof(void *) +
+                                     sizeof(struct vmbus_packet_mpb_array));
+                       payload = kmalloc(payload_sz, GFP_ATOMIC);
+                       if (!payload) {
+                               if (cmd_request->bounce_sgl_count)
+                                       destroy_bounce_buffer(
+                                       cmd_request->bounce_sgl,
+                                       cmd_request->bounce_sgl_count);
+
+                                       return SCSI_MLQUEUE_DEVICE_BUSY;
+                       }
+               }
+
+               payload->range.len = length;
+               payload->range.offset = sgl[0].offset;
+
+               cur_sgl = sgl;
+               for (i = 0; i < sg_count; i++) {
+                       payload->range.pfn_array[i] =
+                               page_to_pfn(sg_page((cur_sgl)));
+                       cur_sgl = sg_next(cur_sgl);
+               }
 
        } else if (scsi_sglist(scmnd)) {
-               cmd_request->data_buffer.offset =
+               payload->range.len = length;
+               payload->range.offset =
                        virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1);
-               cmd_request->data_buffer.pfn_array[0] =
+               payload->range.pfn_array[0] =
                        virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT;
        }
 
+       cmd_request->payload = payload;
+       cmd_request->payload_sz = payload_sz;
+
        /* Invokes the vsc to start an IO */
        ret = storvsc_do_io(dev, cmd_request);
 
@@ -1646,12 +1707,8 @@ static struct scsi_host_template scsi_driver = {
        .eh_timed_out =         storvsc_eh_timed_out,
        .slave_configure =      storvsc_device_configure,
        .cmd_per_lun =          255,
-       .can_queue =            STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS,
        .this_id =              -1,
-       /* no use setting to 0 since ll_blk_rw reset it to 1 */
-       /* currently 32 */
-       .sg_tablesize =         MAX_MULTIPAGE_BUFFER_COUNT,
-       .use_clustering =       DISABLE_CLUSTERING,
+       .use_clustering =       ENABLE_CLUSTERING,
        /* Make sure we dont get a sg segment crosses a page boundary */
        .dma_boundary =         PAGE_SIZE-1,
        .no_write_same =        1,
@@ -1686,6 +1743,7 @@ static int storvsc_probe(struct hv_device *device,
                        const struct hv_vmbus_device_id *dev_id)
 {
        int ret;
+       int num_cpus = num_online_cpus();
        struct Scsi_Host *host;
        struct hv_host_device *host_dev;
        bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
@@ -1694,6 +1752,7 @@ static int storvsc_probe(struct hv_device *device,
        int max_luns_per_target;
        int max_targets;
        int max_channels;
+       int max_sub_channels = 0;
 
        /*
         * Based on the windows host we are running on,
@@ -1719,12 +1778,18 @@ static int storvsc_probe(struct hv_device *device,
                max_luns_per_target = STORVSC_MAX_LUNS_PER_TARGET;
                max_targets = STORVSC_MAX_TARGETS;
                max_channels = STORVSC_MAX_CHANNELS;
+               /*
+                * On Windows8 and above, we support sub-channels for storage.
+                * The number of sub-channels offerred is based on the number of
+                * VCPUs in the guest.
+                */
+               max_sub_channels = (num_cpus / storvsc_vcpus_per_sub_channel);
                break;
        }
 
-       if (dev_id->driver_data == SFC_GUID)
-               scsi_driver.can_queue = (STORVSC_MAX_IO_REQUESTS *
-                                        STORVSC_FC_MAX_TARGETS);
+       scsi_driver.can_queue = (max_outstanding_req_per_channel *
+                                (max_sub_channels + 1));
+
        host = scsi_host_alloc(&scsi_driver,
                               sizeof(struct hv_host_device));
        if (!host)
@@ -1780,6 +1845,12 @@ static int storvsc_probe(struct hv_device *device,
        /* max cmd length */
        host->max_cmd_len = STORVSC_MAX_CMD_LEN;
 
+       /*
+        * set the table size based on the info we got
+        * from the host.
+        */
+       host->sg_tablesize = (stor_device->max_transfer_bytes >> PAGE_SHIFT);
+
        /* Register the HBA and start the scsi bus scan */
        ret = scsi_add_host(host, &device->device);
        if (ret != 0)
@@ -1837,7 +1908,6 @@ static struct hv_driver storvsc_drv = {
 
 static int __init storvsc_drv_init(void)
 {
-       u32 max_outstanding_req_per_channel;
 
        /*
         * Divide the ring buffer data size (which is 1 page less
@@ -1852,10 +1922,6 @@ static int __init storvsc_drv_init(void)
                vmscsi_size_delta,
                sizeof(u64)));
 
-       if (max_outstanding_req_per_channel <
-           STORVSC_MAX_IO_REQUESTS)
-               return -EINVAL;
-
        return vmbus_driver_register(&storvsc_drv);
 }