Re: fix atomic limits check

[Christoph Hellwig <hch@lst.de>]

On Wed, Jun 18, 2025 at 11:30:20AM -0700, alan.adamson@oracle.com wrote:
>> Can you double check this really is your CTLR2?  At least for me
>> the async probing reorders nvme devices quite a lot with qemu.
>>
> CTRL 0 (nvme0) - AWUN=31 AWUPF=15       nvme1n2
> CTRL 1 (nvme1) - AWUN=31 AWUPF=31       nvme1n1
> CTRL 2 (nvme2) - AWUN=15 AWUPF=7         nvme1n3
> CTRL 3 (nvme3) - AWUN=15 AWUPF=15

Do you mean AWUN or AWUPF?  Because AWUN is totally irrelevant and
not even parsed by Linux.

> I rebooted and it probed differently.  Now CTRL2 is nvme1n3.  We know 
> CTRL2 is the same as nvme1n3 because awun and awupf from nvme id-ctrl 
> matches the qemu config (atomic.awun=15,atomic.awupf=7) for CTRL2 (nvme2).
>
> # nvme id-ctrl /dev/nvme1n3 | grep awun
> awun      : 15
> # nvme id-ctrl /dev/nvme1n3 | grep awupf
> awupf     : 7
> # cat /sys/block/nvme1n3/queue/atomic_write_max_bytes
> 8192

The code in nvme_configure_atomic_write calculates the values pretty
much directly from (N)AWUPF, so I don't see how it would get things
wrong.  Can you run your patched qemu setup with this debug printk
patch?

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 3da5ac71a9b0..245397b217f4 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -2033,6 +2033,8 @@ static u32 nvme_configure_atomic_write(struct nvme_ns *ns,
 		atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
 		if (id->nabspf)
 			boundary = (le16_to_cpu(id->nabspf) + 1) * bs;
+		dev_info(ns->ctrl->device, "NAWUPF: %u, atomic_bs: %u\n",
+			le16_to_cpu(id->nabspf), atomic_bs);
 	} else {
 		/*
 		 * Use the controller wide atomic write unit.  This sucks
@@ -2042,6 +2044,8 @@ static u32 nvme_configure_atomic_write(struct nvme_ns *ns,
 		 * values for different controllers in the subsystem.
 		 */
 		atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
+		dev_info(ns->ctrl->device, "AWUPF: %u, atomic_bs: %u\n",
+			ns->ctrl->subsys->awupf, atomic_bs);
 	}
 
 	lim->atomic_write_hw_max = atomic_bs;