]> www.infradead.org Git - users/hch/misc.git/commitdiff
block: change blk_mq_add_to_batch() third argument type to bool
authorShin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Tue, 11 Mar 2025 10:43:59 +0000 (19:43 +0900)
committerJens Axboe <axboe@kernel.dk>
Wed, 12 Mar 2025 14:26:36 +0000 (08:26 -0600)
Commit 1f47ed294a2b ("block: cleanup and fix batch completion adding
conditions") modified the evaluation criteria for the third argument,
'ioerror', in the blk_mq_add_to_batch() function. Initially, the
function had checked if 'ioerror' equals zero. Following the commit, it
started checking for negative error values, with the presumption that
such values, for instance -EIO, would be passed in.

However, blk_mq_add_to_batch() callers do not pass negative error
values. Instead, they pass status codes defined in various ways:

- NVMe PCI and Apple drivers pass NVMe status code
- virtio_blk driver passes the virtblk request header status byte
- null_blk driver passes blk_status_t

These codes are either zero or positive, therefore the revised check
fails to function as intended. Specifically, with the NVMe PCI driver,
this modification led to the failure of the blktests test case nvme/039.
In this test scenario, errors are artificially injected to the NVMe
driver, resulting in positive NVMe status codes passed to
blk_mq_add_to_batch(), which unexpectedly processes the failed I/O in a
batch. Hence the failure.

To correct the ioerror check within blk_mq_add_to_batch(), make all
callers to uniformly pass the argument as boolean. Modify the callers to
check their specific status codes and pass the boolean value 'is_error'.
Also describe the arguments of blK_mq_add_to_batch as kerneldoc.

Fixes: 1f47ed294a2b ("block: cleanup and fix batch completion adding conditions")
Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Link: https://lore.kernel.org/r/20250311104359.1767728-3-shinichiro.kawasaki@wdc.com
[axboe: fold in documentation update]
Signed-off-by: Jens Axboe <axboe@kernel.dk>
drivers/block/null_blk/main.c
drivers/block/virtio_blk.c
drivers/nvme/host/apple.c
drivers/nvme/host/pci.c
include/linux/blk-mq.h

index d94ef37480bd457042c6e1ebef64604bc221ae09..fdc7a0b2af1097f217935017c7a236b7bd80bdba 100644 (file)
@@ -1549,8 +1549,8 @@ static int null_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
                cmd = blk_mq_rq_to_pdu(req);
                cmd->error = null_process_cmd(cmd, req_op(req), blk_rq_pos(req),
                                                blk_rq_sectors(req));
-               if (!blk_mq_add_to_batch(req, iob, (__force int) cmd->error,
-                                       blk_mq_end_request_batch))
+               if (!blk_mq_add_to_batch(req, iob, cmd->error != BLK_STS_OK,
+                                        blk_mq_end_request_batch))
                        blk_mq_end_request(req, cmd->error);
                nr++;
        }
index a4af39fc7ea28a83f9f77312b7145d20aeab25f6..286cab5e53684c34df2e66a293a9f13317c56de9 100644 (file)
@@ -1207,11 +1207,12 @@ static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
 
        while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
                struct request *req = blk_mq_rq_from_pdu(vbr);
+               u8 status = virtblk_vbr_status(vbr);
 
                found++;
                if (!blk_mq_complete_request_remote(req) &&
-                   !blk_mq_add_to_batch(req, iob, virtblk_vbr_status(vbr),
-                                               virtblk_complete_batch))
+                   !blk_mq_add_to_batch(req, iob, status != VIRTIO_BLK_S_OK,
+                                        virtblk_complete_batch))
                        virtblk_request_done(req);
        }
 
index a060f69558e76970bfba046cca5127243e8a51b7..8971aca41e63dc3b6889680de9e62128442c2d6e 100644 (file)
@@ -599,7 +599,8 @@ static inline void apple_nvme_handle_cqe(struct apple_nvme_queue *q,
        }
 
        if (!nvme_try_complete_req(req, cqe->status, cqe->result) &&
-           !blk_mq_add_to_batch(req, iob, nvme_req(req)->status,
+           !blk_mq_add_to_batch(req, iob,
+                                nvme_req(req)->status != NVME_SC_SUCCESS,
                                 apple_nvme_complete_batch))
                apple_nvme_complete_rq(req);
 }
index 640590b2172828cf70c05bd35234b44f286d0d16..75de86e235ad7127c55853e32c782540c604798f 100644 (file)
@@ -1130,8 +1130,9 @@ static inline void nvme_handle_cqe(struct nvme_queue *nvmeq,
 
        trace_nvme_sq(req, cqe->sq_head, nvmeq->sq_tail);
        if (!nvme_try_complete_req(req, cqe->status, cqe->result) &&
-           !blk_mq_add_to_batch(req, iob, nvme_req(req)->status,
-                                       nvme_pci_complete_batch))
+           !blk_mq_add_to_batch(req, iob,
+                                nvme_req(req)->status != NVME_SC_SUCCESS,
+                                nvme_pci_complete_batch))
                nvme_pci_complete_rq(req);
 }
 
index 71f4f0cc3dac672d1b118e9877ba0edf47458d3a..aba9c24486aadb8a3c3b26bf245f0faa916cfb9b 100644 (file)
@@ -852,12 +852,20 @@ static inline bool blk_mq_is_reserved_rq(struct request *rq)
        return rq->rq_flags & RQF_RESV;
 }
 
-/*
+/**
+ * blk_mq_add_to_batch() - add a request to the completion batch
+ * @req: The request to add to batch
+ * @iob: The batch to add the request
+ * @is_error: Specify true if the request failed with an error
+ * @complete: The completaion handler for the request
+ *
  * Batched completions only work when there is no I/O error and no special
  * ->end_io handler.
+ *
+ * Return: true when the request was added to the batch, otherwise false
  */
 static inline bool blk_mq_add_to_batch(struct request *req,
-                                      struct io_comp_batch *iob, int ioerror,
+                                      struct io_comp_batch *iob, bool is_error,
                                       void (*complete)(struct io_comp_batch *))
 {
        /*
@@ -865,7 +873,7 @@ static inline bool blk_mq_add_to_batch(struct request *req,
         * 1) No batch container
         * 2) Has scheduler data attached
         * 3) Not a passthrough request and end_io set
-        * 4) Not a passthrough request and an ioerror
+        * 4) Not a passthrough request and failed with an error
         */
        if (!iob)
                return false;
@@ -874,7 +882,7 @@ static inline bool blk_mq_add_to_batch(struct request *req,
        if (!blk_rq_is_passthrough(req)) {
                if (req->end_io)
                        return false;
-               if (ioerror < 0)
+               if (is_error)
                        return false;
        }