Re: [PATCH] drm/i915/irq: crtc guard inside interrupt handlers

[Jani Nikula]

On Fri, 25 Sep 2026, Marcel Hamer <marcel@xxxxxxxxxx> wrote:
> Interrupts are enabled before the crtc datastructures are initialized.
> As intel_crtc_for_pipe() is called inside interrupt handlers and can
> return a NULL value, this can in rare cases lead to the following NULL
> pointer dereference and panic:
>
> RIP: 0010:handle_plane_fault+0x19/0xc0 [i915]
> Code: 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 f3 0f 1e fa 0f
> 1f 44 00 00 55 48 89 e5 41 55 41 54 49 89 fc 53 48 83 ec 18 <48> 8b 1f
> 65 4c 8b 2d 24 90 31 ef 4c 89 6d e0 41 89 f5 48 85 db 74
> RSP:
> 0018:ffff92b180003e50 EFLAGS: 00010086
> RAX: ffffffffc0911fc0 RBX: ffffffffc06e0b90 RCX: 0000000000000005
> RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000000
> RBP: ffff92b180003e80 R08: 0000000000000005 R09: 000000010517d5f6
> R10: 0000008000000041 R11: 000000010517d5f6 R12: 0000000000000000
> R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000
> FS:  00007f72f5b72780(0000) GS:ffff8bbdcbfe9000(0000) knlGS:0000000000000000
> CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> CR2: 0000000000000000 CR3: 0000000102f2c000 CR4: 00000000003506f0
> Kernel panic - not syncing: Fatal exception in interrupt
> Kernel Offset: 0x2c800000 from 0xffffffff81000000 (relocation range:
> 0xffffffff80000000-0xffffffffbfffffff)
> Call Trace:
>  <IRQ>
>  intel_pipe_fault_irq_handler+0x42/0xc0 [i915]
>  gen8_de_irq_handler+0x1ed/0xdf0 [i915]
>  gen8_irq_handler+0xb5/0xd0 [i915]
>  __handle_irq_event_percpu+0x6e/0x230
>  handle_irq_event+0x3a/0x90
>  handle_edge_irq+0xd1/0x1a0
>  __common_interrupt+0x48/0xe0
>  ? generic_smp_call_function_single_interrupt+0x17/0x20
>  common_interrupt+0x84/0xa0
>
> Fixes: f13011a79999 ("drm/i915: Pimp display fault reporting")
> Fixes: 138d2bda4eaf ("drm/i915/display: pass display to intel_crtc_for_pipe()")
> Cc: stable@xxxxxxxxxxxxxxx # 6.18.x
> Signed-off-by: Marcel Hamer <marcel@xxxxxxxxxx>

Please file a bug as described at [1], attaching full dmesg reproducing
the problem. I don't think we're seeing this in our CI, so would be good
to understand the root cause better.

BR,
Jani.


[1] https://drm.pages.freedesktop.org/intel-docs/how-to-file-i915-bugs.html


> ---
>  drivers/gpu/drm/i915/display/intel_display_irq.c | 16 ++++++++++++++--
>  1 file changed, 14 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/display/intel_display_irq.c b/drivers/gpu/drm/i915/display/intel_display_irq.c
> index a59b75830bd1..4ffc8789c9f3 100644
> --- a/drivers/gpu/drm/i915/display/intel_display_irq.c
> +++ b/drivers/gpu/drm/i915/display/intel_display_irq.c
> @@ -131,6 +131,9 @@ static void intel_pipe_fault_irq_handler(struct intel_display *display,
>  	struct intel_crtc *crtc = intel_crtc_for_pipe(display, pipe);
>  	const struct pipe_fault_handler *handler;
>  
> +	if (!crtc)
> +		return;
> +
>  	for (handler = handlers; handler && handler->fault; handler++) {
>  		if ((fault_errors & handler->fault) == 0)
>  			continue;
> @@ -148,7 +151,8 @@ intel_handle_vblank(struct intel_display *display, enum pipe pipe)
>  {
>  	struct intel_crtc *crtc = intel_crtc_for_pipe(display, pipe);
>  
> -	drm_crtc_handle_vblank(&crtc->base);
> +	if (crtc)
> +		drm_crtc_handle_vblank(&crtc->base);
>  }
>  
>  /**
> @@ -425,9 +429,14 @@ static void display_pipe_crc_irq_handler(struct intel_display *display,
>  					 u32 crc4)
>  {
>  	struct intel_crtc *crtc = intel_crtc_for_pipe(display, pipe);
> -	struct intel_pipe_crc *pipe_crc = &crtc->pipe_crc;
> +	struct intel_pipe_crc *pipe_crc;
>  	u32 crcs[5] = { crc0, crc1, crc2, crc3, crc4 };
>  
> +	if (!crtc)
> +		return;
> +
> +	pipe_crc = &crtc->pipe_crc;
> +
>  	trace_intel_pipe_crc(crtc, crcs);
>  
>  	spin_lock(&pipe_crc->lock);
> @@ -465,6 +474,9 @@ static void flip_done_handler(struct intel_display *display,
>  {
>  	struct intel_crtc *crtc = intel_crtc_for_pipe(display, pipe);
>  
> +	if (!crtc)
> +		return;
> +
>  	spin_lock(&display->drm->event_lock);
>  
>  	if (crtc->flip_done_event) {
>
> base-commit: f49a343b305c0b6c19a3b50c0bbf10bcd0e2e8fd

-- 
Jani Nikula, Intel