skd: remove broken discard support
authorJeff Moyer <jmoyer@redhat.com>
Tue, 26 Apr 2016 01:12:38 +0000 (19:12 -0600)
committerJens Axboe <axboe@fb.com>
Tue, 26 Apr 2016 01:12:38 +0000 (19:12 -0600)
Simply creating a file system on an skd device, followed by mount and
fstrim will result in errors in the logs and then a BUG().  Let's remove
discard support from that driver.  As far as I can tell, it hasn't
worked right since it was merged.  This patch also has a side-effect of
cleaning up an unintentional shadowed declaration inside of
skd_end_request.

I tested to ensure that I can still do I/O to the device using xfstests
./check -g quick.  I didn't do anything more extensive than that,
though.

Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
drivers/block/skd_main.c

index 41aaae30c00567cd75e3a97edfca66aa1e8753ea..910e065918af13d2536f6fe6736a4fad3303e243 100644 (file)
@@ -133,7 +133,6 @@ MODULE_VERSION(DRV_VERSION "-" DRV_BUILD_ID);
 #define SKD_TIMER_MINUTES(minutes) ((minutes) * (60))
 
 #define INQ_STD_NBYTES 36
-#define SKD_DISCARD_CDB_LENGTH 24
 
 enum skd_drvr_state {
        SKD_DRVR_STATE_LOAD,
@@ -212,7 +211,6 @@ struct skd_request_context {
 
        struct request *req;
        u8 flush_cmd;
-       u8 discard_page;
 
        u32 timeout_stamp;
        u8 sg_data_dir;
@@ -230,7 +228,6 @@ struct skd_request_context {
 };
 #define SKD_DATA_DIR_HOST_TO_CARD       1
 #define SKD_DATA_DIR_CARD_TO_HOST       2
-#define SKD_DATA_DIR_NONE              3       /* especially for DISCARD requests. */
 
 struct skd_special_context {
        struct skd_request_context req;
@@ -540,31 +537,6 @@ skd_prep_zerosize_flush_cdb(struct skd_scsi_request *scsi_req,
        scsi_req->cdb[9] = 0;
 }
 
-static void
-skd_prep_discard_cdb(struct skd_scsi_request *scsi_req,
-                    struct skd_request_context *skreq,
-                    struct page *page,
-                    u32 lba, u32 count)
-{
-       char *buf;
-       unsigned long len;
-       struct request *req;
-
-       buf = page_address(page);
-       len = SKD_DISCARD_CDB_LENGTH;
-
-       scsi_req->cdb[0] = UNMAP;
-       scsi_req->cdb[8] = len;
-
-       put_unaligned_be16(6 + 16, &buf[0]);
-       put_unaligned_be16(16, &buf[2]);
-       put_unaligned_be64(lba, &buf[8]);
-       put_unaligned_be32(count, &buf[16]);
-
-       req = skreq->req;
-       blk_add_request_payload(req, page, 0, len);
-}
-
 static void skd_request_fn_not_online(struct request_queue *q);
 
 static void skd_request_fn(struct request_queue *q)
@@ -575,7 +547,6 @@ static void skd_request_fn(struct request_queue *q)
        struct skd_request_context *skreq;
        struct request *req = NULL;
        struct skd_scsi_request *scsi_req;
-       struct page *page;
        unsigned long io_flags;
        int error;
        u32 lba;
@@ -669,7 +640,6 @@ static void skd_request_fn(struct request_queue *q)
                skreq->flush_cmd = 0;
                skreq->n_sg = 0;
                skreq->sg_byte_count = 0;
-               skreq->discard_page = 0;
 
                /*
                 * OK to now dequeue request from q.
@@ -735,18 +705,7 @@ static void skd_request_fn(struct request_queue *q)
                else
                        skreq->sg_data_dir = SKD_DATA_DIR_HOST_TO_CARD;
 
-               if (io_flags & REQ_DISCARD) {
-                       page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
-                       if (!page) {
-                               pr_err("request_fn:Page allocation failed.\n");
-                               skd_end_request(skdev, skreq, -ENOMEM);
-                               break;
-                       }
-                       skreq->discard_page = 1;
-                       req->completion_data = page;
-                       skd_prep_discard_cdb(scsi_req, skreq, page, lba, count);
-
-               } else if (flush == SKD_FLUSH_ZERO_SIZE_FIRST) {
+               if (flush == SKD_FLUSH_ZERO_SIZE_FIRST) {
                        skd_prep_zerosize_flush_cdb(scsi_req, skreq);
                        SKD_ASSERT(skreq->flush_cmd == 1);
 
@@ -851,16 +810,6 @@ skip_sg:
 static void skd_end_request(struct skd_device *skdev,
                            struct skd_request_context *skreq, int error)
 {
-       struct request *req = skreq->req;
-       unsigned int io_flags = req->cmd_flags;
-
-       if ((io_flags & REQ_DISCARD) &&
-               (skreq->discard_page == 1)) {
-               pr_debug("%s:%s:%d, free the page!",
-                        skdev->name, __func__, __LINE__);
-               __free_page(req->completion_data);
-       }
-
        if (unlikely(error)) {
                struct request *req = skreq->req;
                char *cmd = (rq_data_dir(req) == READ) ? "read" : "write";
@@ -4419,12 +4368,6 @@ static int skd_cons_disk(struct skd_device *skdev)
        /* set sysfs ptimal_io_size to 8K */
        blk_queue_io_opt(q, 8192);
 
-       /* DISCARD Flag initialization. */
-       q->limits.discard_granularity = 8192;
-       q->limits.discard_alignment = 0;
-       blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
-       q->limits.discard_zeroes_data = 1;
-       queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
        queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
        queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);