Re: [PATCH 2/2] drm/nouveau/fifo: use global nonstall event on legacy FIFO

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [Critical] If channel initialization fails early in `nvkm_chan_new_()`, the cleanup path dereferences a NULL `chan->cgrp`.
--

commit 6a368e9f19db8e6f77b9767f67dd0d51b8625cd8
Author: Risto Pajula <or.pajula@gmail.com>

drm/nouveau/fifo: use global nonstall event on legacy FIFO

This commit changes the nonstall event registration to use the runlist ID
only when per-runlist nonstall interrupts are supported, falling back to
event index 0 for legacy FIFO implementations. This fixes an issue where
fence completion waits for an event that is never signalled.

> diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/uchan.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/uchan.c
> index d6a87cec21503..b96e841920320 100644
> --- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/uchan.c
> +++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/uchan.c
> @@ -52,7 +52,8 @@ nvkm_uchan_uevent(struct nvkm_object *object, void *argv, u32 argc, struct nvkm_
>  
>  	switch (args->v0.type) {
>  	case NVIF_CHAN_EVENT_V0_NON_STALL_INTR:
> -		return nvkm_uevent_add(uevent, &runl->fifo->nonstall.event, runl->id,
> +		return nvkm_uevent_add(uevent, &runl->fifo->nonstall.event,
> +				runl->fifo->func->nonstall_ctor ? runl->id : 0,
>  				       NVKM_FIFO_NONSTALL_EVENT, NULL);
>  	case NVIF_CHAN_EVENT_V0_KILLED:
>  		return nvkm_uevent_add(uevent, &runl->chid->event, chan->id,

[Severity: Critical]
This is a pre-existing issue, but while reviewing the channel handling code, I
noticed a potential NULL pointer dereference if channel initialization fails
early.

In drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.c:nvkm_chan_new_():

    ret = nvkm_cgrp_new(runl, chan->name, vmm, fifo->func->cgrp.force,
                        &chan->cgrp);
    if (ret) {
        RUNL_DEBUG(runl, "cgrp %d", ret);
        return ret;
    }

If nvkm_cgrp_new() fails, chan->cgrp remains NULL but the error is returned
to the caller while leaving the partially initialized channel allocated.

The cleanup path in nvkm_chan_del() unconditionally executes the RAMFC clear
callback on the partially initialized channel object:

drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.c:nvkm_chan_del() {
    ...
    if (chan->func->ramfc->clear)
        chan->func->ramfc->clear(chan);
    ...
}

For hardware-specific implementations like nv04_chan_ramfc_clear() in
drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c, this directly dereferences
chan->cgrp:

    struct nvkm_memory *ramfc =
        chan->cgrp->runl->fifo->engine.subdev.device->imem->ramfc;

Could this lead to a kernel panic when a userspace program invokes the
NVIF_IOCTL_NEW ioctl to create a new channel and nvkm_cgrp_new() fails? This
seems reachable if a user provides an invalid channel group or during memory
pressure.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260913201225.5975-1-or.pajula@gmail.com?part=2