This reverts upstream 'commit 530436c45ef2 ("nvme: Discard workaround for non-conformant devices")'
Since commit `530436c45ef2` introduced a regression due to which blk_update_request IO error is observed on formatting device, reverting it.
Fixes: 530436c45ef2 ("nvme: Discard workaround for non-conformant devices") Cc: stable stable@vger.kernel.org # 4.19+ Signed-off-by: Dakshaja Uppalapati dakshaja@chelsio.com --- drivers/nvme/host/core.c | 12 +++--------- 1 file changed, 3 insertions(+), 9 deletions(-)
diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c index f3c037f5a9ba..ec598a86f88e 100644 --- a/drivers/nvme/host/core.c +++ b/drivers/nvme/host/core.c @@ -607,14 +607,8 @@ static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req, struct nvme_dsm_range *range; struct bio *bio;
- /* - * Some devices do not consider the DSM 'Number of Ranges' field when - * determining how much data to DMA. Always allocate memory for maximum - * number of segments to prevent device reading beyond end of buffer. - */ - static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES; - - range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN); + range = kmalloc_array(segments, sizeof(*range), + GFP_ATOMIC | __GFP_NOWARN); if (!range) { /* * If we fail allocation our range, fallback to the controller @@ -654,7 +648,7 @@ static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
req->special_vec.bv_page = virt_to_page(range); req->special_vec.bv_offset = offset_in_page(range); - req->special_vec.bv_len = alloc_size; + req->special_vec.bv_len = sizeof(*range) * segments; req->rq_flags |= RQF_SPECIAL_PAYLOAD;
return BLK_STS_OK;
On Wednesday, June 06/03/20, 2020 at 15:07:50 +0200, Christoph Hellwig wrote:
Hi Christoph,
Here is the link describing the issue initially reported for upstream kernel 5.5:
https://lore.kernel.org/linux-nvme/CH2PR12MB40053A64681EFA3E6F63FDFBDD2A0@CH...
Issue is later fixed with upstream commit b716e688.
Now the same issue is observed with stable kernels 4.19 and 5.4 as commit 530436c4 was pulled in to stable. Here is the link describing the same for stable kernels.
https://www.spinics.net/lists/stable/msg387744.html
Please let me know if you need any further info.
Thanks, Dakshaja
On Wed, Jun 03, 2020 at 09:47:18PM +0530, Dakshaja Uppalapati wrote:
We are talking about two different things here. One is the Linux NVMe host code that can be used with lots of different controllers. Many of them are PCIe controller, especially cheap ones.
The other is the Linux NVMe target code. So if a fix for very common PCIe controller trigger a bug in the target code there is no 1:1 relationship as even if you are talking to a Linux fabrics controller it usually runs a different kernel version on a different system.
That being said you can always backport that fix as well, which probably is a good idea as it fixes a real bug.
Nevermind that nothing in your revert patch indicated it wasn't for mainline.
linux-stable-mirror@lists.linaro.org
-
Christoph Hellwig -
Dakshaja Uppalapati -
Greg KH -
Keith Busch -
Sagi Grimberg