[PATCH] nvme-pci: prevent SK Hynix PE8000 from using Write Zeroes command

[Chaitanya Kulkarni]

On 2/10/21 5:20 AM, Christoph Hellwig wrote:
> MDTS vs Write Zeroes is a common misconception.  We had that discussion
> in the NVMe working group, and while the text should pretty clear that
> MDTS only applies to data transfers many implementators did not understand
> that, which is why we added the clarification you quoted.
>
> Also the next NVMe spec will allow devices to advertise an explicit
> Write Zeroes limit, take a look at TP4040 from the "NVM Express 1.4 Ratified
> TPs" at https://nvmexpress.org/developers/nvme-specification/.
>
> So I think limiting to MDTS unless the new WZSL is set might be a fail
> safe option, even if it is more pessimistic than what the spec says.
>
> I'd also love to retest most Write Zeroes quirks with that in place.

Do you prefer some variant of the following patch (totally untested)? OR
something else ?

Also, I think we should gradually roll out the MDTS replacement and
still keep the
write-zeroes quirk unless you are okay to just replace at one go.

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index d77f3f26d8d3..d5ca27608881 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -1167,7 +1167,8 @@ static int nvme_submit_user_cmd(struct
request_queue *q,
                if (ret)
                        goto out;
                bio = req->bio;
-               bio_set_dev(bio, bdev);
+               if(bdev)
+                       bio_set_dev(bio, bdev);
                if (bdev && meta_buffer && meta_len) {
                        meta = nvme_add_user_metadata(bio, meta_buffer,
meta_len,
                                        meta_seed, write);
@@ -1967,6 +1968,7 @@ static void nvme_config_write_zeroes(struct
gendisk *disk, struct nvme_ns *ns)
        u64 max_blocks;
 
        if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
+           !(ns->ctrl->quirks & NVME_QUIRK_USE_MDTS_WRITE_ZEROES) ||
            (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
                return;
        /*
@@ -1979,10 +1981,15 @@ static void nvme_config_write_zeroes(struct
gendisk *disk, struct nvme_ns *ns)
         * configured based on the controller's MDTS field in the
         * nvme_init_identify() if available.
         */
-       if (ns->ctrl->max_hw_sectors == UINT_MAX)
-               max_blocks = (u64)USHRT_MAX + 1;
-       else
-               max_blocks = ns->ctrl->max_hw_sectors + 1;
+
+       if (!(ns->ctrl->quirks & NVME_QUIRK_USE_MDTS_WRITE_ZEROES)) {
+               if (ns->ctrl->max_hw_sectors == UINT_MAX)
+                       max_blocks = (u64)USHRT_MAX + 1;
+               else
+                       max_blocks = ns->ctrl->max_hw_sectors + 1;
+       } else {
+               max_blocks = ns->ctrl->max_mdts_hw_sectors;
+       }
 
        blk_queue_max_write_zeroes_sectors(disk->queue,
                                           nvme_lba_to_sect(ns,
max_blocks));
@@ -3141,6 +3148,11 @@ int nvme_init_identify(struct nvme_ctrl *ctrl)
        ctrl->max_hw_sectors =
                min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
 
+       /* explicitely set the mdts value to 0 */
+       ctrl->max_mdts_hw_sectors = 0;
+       if (id->mdts)
+               ctrl->max_mdts_hw_sectors =  1 << (id->mdts + page_shift
- 9);
+
        nvme_set_queue_limits(ctrl, ctrl->admin_q);
        ctrl->sgls = le32_to_cpu(id->sgls);
        ctrl->kas = le16_to_cpu(id->kas);
diff --git a/drivers/nvme/host/nvme.h b/drivers/nvme/host/nvme.h
index 0a73e78f09c8..7b60b2196d23 100644
--- a/drivers/nvme/host/nvme.h
+++ b/drivers/nvme/host/nvme.h
@@ -144,6 +144,11 @@ enum nvme_quirks {
         * NVMe 1.3 compliance.
         */
        NVME_QUIRK_NO_NS_DESC_LIST              = (1 << 15),
+       /*
+        * The controller doesn't allow the write zeores command of size
+        * more than reported by the MDTS
+        */
+       NVME_QUIRK_USE_MDTS_WRITE_ZEROES        = (1 << 16)
 };
 
 /*
@@ -268,6 +273,7 @@ struct nvme_ctrl {
 
        u64 cap;
        u32 max_hw_sectors;
+       u32 max_mdts_hw_sectors;
        u32 max_segments;
        u32 max_integrity_segments;
 #ifdef CONFIG_BLK_DEV_ZONED