Re: [PATCH] ntb: fix peer_spad_read() returning -EINVAL instead of 0
From: Koichiro Den
Date: Fri Nov 20 2026 - 11:34:43 EST
On Fri, Nov 20, 2026 at 05:45:45PM +0900, zhangjiao2 wrote:
> From: zhang jiao <zhangjiao2@xxxxxxxxxxxxxxxxxxxx>
>
> The amd_ntb_peer_spad_read(), ntb_epf_peer_spad_read() and
> switchtec_ntb_peer_spad_read() functions have a return type of u32
> but return -EINVAL on error, which results in a garbage value
> (0xFFFFFFEA) being returned to callers. Other NTB hardware drivers
> (e.g. Intel's ndev_spad_read) correctly return 0 for out-of-range
> scratchpad indices.
>
> Fix all three functions to return 0 on error, matching the
> established convention in the NTB subsystem.
>
> Signed-off-by: zhang jiao <zhangjiao2@xxxxxxxxxxxxxxxxxxxx>
I wonder if it would make more sense for ntb_peer_spad_read() and its callback
to take a u32 *val for the result and return 0 on success or -errno on failure.
(The wrapper could then return -EOPNOTSUPP if the callback is missing.)
It looks like ntb_msi_test can legitimately publish an irq_count of 0 to its
SPAD, and ntb_perf uses all 1's for an empty command slot (ie. PERF_CMD_INVAL).
So I'm not sure overloading the return value to signal a failed read is a good
choice from the beginning.
That said, I agree that returning -EINVAL as a u32 is misleading.
One thing I'm not sure about is whether switching to 0 is really harmless.
In ntb_msit_db_event(), what if the doorbell bit isn't a valid peer index?
>From what I can see, on ntb_hw_epf or switchtec, the old value (-EINVAL) makes
the allocation fail, while with 0, kzalloc_objs() returns ZERO_SIZE_PTR rather
than NULL. Wouldn't the code then use that invalid peer index in nm->peers[]?
Best regards,
Koichiro Den
> ---
> drivers/ntb/hw/amd/ntb_hw_amd.c | 2 +-
> drivers/ntb/hw/epf/ntb_hw_epf.c | 4 ++--
> drivers/ntb/hw/mscc/ntb_hw_switchtec.c | 2 +-
> 3 files changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/ntb/hw/amd/ntb_hw_amd.c b/drivers/ntb/hw/amd/ntb_hw_amd.c
> index 36bee0d8ac6c..40ced4043340 100644
> --- a/drivers/ntb/hw/amd/ntb_hw_amd.c
> +++ b/drivers/ntb/hw/amd/ntb_hw_amd.c
> @@ -529,7 +529,7 @@ static u32 amd_ntb_peer_spad_read(struct ntb_dev *ntb, int pidx, int sidx)
> u32 offset;
>
> if (sidx < 0 || sidx >= ndev->spad_count)
> - return -EINVAL;
> + return 0;
>
> offset = ndev->peer_spad + (sidx << 2);
> return readl(mmio + AMD_SPAD_OFFSET + offset);
> diff --git a/drivers/ntb/hw/epf/ntb_hw_epf.c b/drivers/ntb/hw/epf/ntb_hw_epf.c
> index c47607d4f8a7..dcc5eb4bea74 100644
> --- a/drivers/ntb/hw/epf/ntb_hw_epf.c
> +++ b/drivers/ntb/hw/epf/ntb_hw_epf.c
> @@ -273,12 +273,12 @@ static u32 ntb_epf_peer_spad_read(struct ntb_dev *ntb, int pidx, int idx)
>
> if (pidx != NTB_DEF_PEER_IDX) {
> dev_err(dev, "Unsupported Peer ID %d\n", pidx);
> - return -EINVAL;
> + return 0;
> }
>
> if (idx < 0 || idx >= ndev->spad_count) {
> dev_err(dev, "WRITE: Invalid Peer ScratchPad Index %d\n", idx);
> - return -EINVAL;
> + return 0;
> }
>
> offset = (idx << 2);
> diff --git a/drivers/ntb/hw/mscc/ntb_hw_switchtec.c b/drivers/ntb/hw/mscc/ntb_hw_switchtec.c
> index e38540b92716..a9fc94869aeb 100644
> --- a/drivers/ntb/hw/mscc/ntb_hw_switchtec.c
> +++ b/drivers/ntb/hw/mscc/ntb_hw_switchtec.c
> @@ -762,7 +762,7 @@ static u32 switchtec_ntb_peer_spad_read(struct ntb_dev *ntb, int pidx,
> struct switchtec_ntb *sndev = ntb_sndev(ntb);
>
> if (pidx != NTB_DEF_PEER_IDX)
> - return -EINVAL;
> + return 0;
>
> if (sidx < 0 || sidx >= ARRAY_SIZE(sndev->peer_shared->spad))
> return 0;
> --
> 2.33.0
>
>
>