[PATCH] drm/xe/relay: Don't use GFP_KERNEL for new transactions

[Bernatowicz, Marcin]

On 1/31/2025 4:37 PM, Michal Wajdeczko wrote:
> VFs use a relay transaction during the resume/reset flow and use
> of the GFP_KERNEL flag may conflict with the reclaim:
> 
>       -> #0 (fs_reclaim){+.+.}-{0:0}:
>   [ ]        __lock_acquire+0x1874/0x2bc0
>   [ ]        lock_acquire+0xd2/0x310
>   [ ]        fs_reclaim_acquire+0xc5/0x100
>   [ ]        mempool_alloc_noprof+0x5c/0x1b0
>   [ ]        __relay_get_transaction+0xdc/0xa10 [xe]
>   [ ]        relay_send_to+0x251/0xe50 [xe]
>   [ ]        xe_guc_relay_send_to_pf+0x79/0x3a0 [xe]
>   [ ]        xe_gt_sriov_vf_connect+0x90/0x4d0 [xe]
>   [ ]        xe_uc_init_hw+0x157/0x3b0 [xe]
>   [ ]        do_gt_restart+0x1ae/0x650 [xe]
>   [ ]        xe_gt_resume+0xb6/0x120 [xe]
>   [ ]        xe_pm_runtime_resume+0x15b/0x370 [xe]
>   [ ]        xe_pci_runtime_resume+0x73/0x90 [xe]
>   [ ]        pci_pm_runtime_resume+0xa0/0x100
>   [ ]        __rpm_callback+0x4d/0x170
>   [ ]        rpm_callback+0x64/0x70
>   [ ]        rpm_resume+0x594/0x790
>   [ ]        __pm_runtime_resume+0x4e/0x90
>   [ ]        xe_pm_runtime_get_ioctl+0x9c/0x160 [xe]
> 
> Since we have a preallocated pool of relay transactions, which
> should cover all our normal relay use cases, we may use the
> GFP_NOWAIT flag when allocating new outgoing transactions.
> 
> Signed-off-by: Michal Wajdeczko <michal.wajdeczko at intel.com>
> ---
> Cc: Michał Winiarski <michal.winiarski at intel.com>
> Cc: Marcin Bernatowicz <marcin.bernatowicz at linux.intel.com>
> ---
>   drivers/gpu/drm/xe/xe_guc_relay.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_guc_relay.c b/drivers/gpu/drm/xe/xe_guc_relay.c
> index 8f62de026724..e5dc94f3e618 100644
> --- a/drivers/gpu/drm/xe/xe_guc_relay.c
> +++ b/drivers/gpu/drm/xe/xe_guc_relay.c
> @@ -225,7 +225,7 @@ __relay_get_transaction(struct xe_guc_relay *relay, bool incoming, u32 remote, u
>   	 * with CTB lock held which is marked as used in the reclaim path.
>   	 * Btw, that's one of the reason why we use mempool here!
>   	 */
> -	txn = mempool_alloc(&relay->pool, incoming ? GFP_ATOMIC : GFP_KERNEL);
> +	txn = mempool_alloc(&relay->pool, incoming ? GFP_ATOMIC : GFP_NOWAIT);
>   	if (!txn)
>   		return ERR_PTR(-ENOMEM);
>   

LGTM. I tested this patch and no longer see the lockdep warning.

Tested-by: Marcin Bernatowicz marcin.bernatowicz at linux.intel.com
Reviewed-by: Marcin Bernatowicz marcin.bernatowicz at linux.intel.com