[PATCH] RDMA/rtrs: client: Fix clt_path::max_pages_per_mr calculation

[PATCH] RDMA/rtrs: client: Fix clt_path::max_pages_per_mr calculation

[Honggang LI]

If the low two bytes of ib_dev::attrs::max_mr_size are zeros, the `min3`
function will set clt_path::max_pages_per_mr to zero.

`alloc_path_reqs` will pass zero, which is invalid,  as the third parameter
to `ib_alloc_mr`.

Fixes: 6a98d71daea1 ("RDMA/rtrs: client: main functionality")
Signed-off-by: Honggang LI <honggangli@163.com>
---
 drivers/infiniband/ulp/rtrs/rtrs-clt.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/infiniband/ulp/rtrs/rtrs-clt.c b/drivers/infiniband/ulp/rtrs/rtrs-clt.c
index 71387811b281..b9d66e4fab07 100644
--- a/drivers/infiniband/ulp/rtrs/rtrs-clt.c
+++ b/drivers/infiniband/ulp/rtrs/rtrs-clt.c
@@ -1465,7 +1465,7 @@ static void query_fast_reg_mode(struct rtrs_clt_path *clt_path)
 	max_pages_per_mr   = ib_dev->attrs.max_mr_size;
 	do_div(max_pages_per_mr, (1ull << mr_page_shift));
 	clt_path->max_pages_per_mr =
-		min3(clt_path->max_pages_per_mr, (u32)max_pages_per_mr,
+		min(min_not_zero(clt_path->max_pages_per_mr, (u32)max_pages_per_mr),
 		     ib_dev->attrs.max_fast_reg_page_list_len);
 	clt_path->clt->max_segments =
 		min(clt_path->max_pages_per_mr, clt_path->clt->max_segments);
-- 
2.52.0

Re: [PATCH] RDMA/rtrs: client: Fix clt_path::max_pages_per_mr calculation

[Michael Gur]

On 12/23/2025 5:43 AM, Honggang LI wrote:
> If the low two bytes of ib_dev::attrs::max_mr_size are zeros, the `min3`
> function will set clt_path::max_pages_per_mr to zero.

Can't see how if the low two bytes of max_mr_size are zero it would 
cause the local variable max_pages_per_mr to zero.
The more probable cause is that max_mr_size bits in the range 
[mr_page_shift+31:mr_page_shift] are zero. Since that's what's left 
after division and cast to u32.
This means you are working on a device supporting more pages_per_mr than 
can fit in a u32.

> `alloc_path_reqs` will pass zero, which is invalid,  as the third parameter
> to `ib_alloc_mr`.
>
> Fixes: 6a98d71daea1 ("RDMA/rtrs: client: main functionality")
> Signed-off-by: Honggang LI <honggangli@163.com>
> ---
>   drivers/infiniband/ulp/rtrs/rtrs-clt.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/infiniband/ulp/rtrs/rtrs-clt.c b/drivers/infiniband/ulp/rtrs/rtrs-clt.c
> index 71387811b281..b9d66e4fab07 100644
> --- a/drivers/infiniband/ulp/rtrs/rtrs-clt.c
> +++ b/drivers/infiniband/ulp/rtrs/rtrs-clt.c
> @@ -1465,7 +1465,7 @@ static void query_fast_reg_mode(struct rtrs_clt_path *clt_path)
>   	max_pages_per_mr   = ib_dev->attrs.max_mr_size;
>   	do_div(max_pages_per_mr, (1ull << mr_page_shift));
>   	clt_path->max_pages_per_mr =
> -		min3(clt_path->max_pages_per_mr, (u32)max_pages_per_mr,
> +		min(min_not_zero(clt_path->max_pages_per_mr, (u32)max_pages_per_mr),

This still fixes the issue, but for readability, if max_pages_per_mr is 
larger than U32_MAX, I'd set it to be U32_MAX.

Michael

Re: [PATCH] RDMA/rtrs: client: Fix clt_path::max_pages_per_mr calculation

[Honggang LI]

On Tue, Dec 23, 2025 at 04:08:19PM +0200, Michael Gur wrote:
> Subject: Re: [PATCH] RDMA/rtrs: client: Fix clt_path::max_pages_per_mr
>  calculation
> From: Michael Gur <michaelgur@nvidia.com>
> Date: Tue, 23 Dec 2025 16:08:19 +0200
> 
> 
> On 12/23/2025 5:43 AM, Honggang LI wrote:
> > If the low two bytes of ib_dev::attrs::max_mr_size are zeros, the `min3`
> > function will set clt_path::max_pages_per_mr to zero.
> 
> Can't see how if the low two bytes of max_mr_size are zero it would cause
> the local variable max_pages_per_mr to zero.
> The more probable cause is that max_mr_size bits in the range
> [mr_page_shift+31:mr_page_shift] are zero. Since that's what's left after
> division and cast to u32.

Yes, you are right. The irdma support max_mr_size is 0x200000000000.
[mr_page_shift+31:mr_page_shift], a.k.a [43,12] are zeros.

> This means you are working on a device supporting more pages_per_mr than can
> fit in a u32.
> 
> > -		min3(clt_path->max_pages_per_mr, (u32)max_pages_per_mr,
> > +		min(min_not_zero(clt_path->max_pages_per_mr, (u32)max_pages_per_mr),
> 
> This still fixes the issue, but for readability, if max_pages_per_mr is
> larger than U32_MAX, I'd set it to be U32_MAX.

I will set it to U32_MAX as you suggested.

thanks