[Intel-gfx] [PATCH v6 1/4] drm/i915/hpd: postpone HPD cancel work after last user suspension

[Imre Deak]

On Fri, Jul 22, 2022 at 02:51:40PM +0200, Andrzej Hajda wrote:
> i915->hotplug.dig_port_work can be queued from intel_hpd_irq_handler
> called by IRQ handler or by intel_hpd_trigger_irq called from dp_mst.
> Since dp_mst is suspended after irq handler uninstall, a cleaner approach
> is to cancel hpd work after intel_dp_mst_suspend, otherwise we risk
> use-after-free.
> 
> It should fix following WARNINGS:
> [283.405824] cpu_latency_qos_update_request called for unknown object
> [283.405866] WARNING: CPU: 2 PID: 240 at kernel/power/qos.c:296 cpu_latency_qos_update_request+0x2d/0x100
> [283.405912] CPU: 2 PID: 240 Comm: kworker/u64:9 Not tainted 5.18.0-rc6-Patchwork_103738v3-g1672d1c43e43+ #1
> [283.405915] Hardware name: Intel Corporation Raptor Lake Client Platform/RPL-S ADP-S DDR5 UDIMM CRB, BIOS RPLSFWI1.R00.2397.A01.2109300731 09/30/2021
> [283.405916] Workqueue: i915-dp i915_digport_work_func [i915]
> [283.406020] RIP: 0010:cpu_latency_qos_update_request+0x2d/0x100
> ...
> [283.406040] Call Trace:
> [283.406041]  <TASK>
> [283.406044]  intel_dp_aux_xfer+0x60e/0x8e0 [i915]
> [283.406131]  ? finish_swait+0x80/0x80
> [283.406139]  intel_dp_aux_transfer+0xc5/0x2b0 [i915]
> [283.406218]  drm_dp_dpcd_access+0x79/0x130 [drm_display_helper]
> [283.406227]  drm_dp_dpcd_read+0xe2/0xf0 [drm_display_helper]
> [283.406233]  intel_dp_hpd_pulse+0x134/0x570 [i915]
> [283.406308]  ? __down_killable+0x70/0x140
> [283.406313]  i915_digport_work_func+0xba/0x150 [i915]
> 
> Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/4586
> Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/5558
> Signed-off-by: Andrzej Hajda <andrzej.hajda at intel.com>
> Reviewed-by: Arun R Murthy <arun.r.murthy at intel.com>
> ---
>  drivers/gpu/drm/i915/display/intel_display.c | 3 +++
>  drivers/gpu/drm/i915/i915_irq.c              | 1 -
>  2 files changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c
> index a0f84cbe974fc3..f1c765ac7ab8aa 100644
> --- a/drivers/gpu/drm/i915/display/intel_display.c
> +++ b/drivers/gpu/drm/i915/display/intel_display.c
> @@ -9021,6 +9021,9 @@ void intel_modeset_driver_remove_noirq(struct drm_i915_private *i915)
>  	 */
>  	intel_dp_mst_suspend(i915);
>  
> +	/* MST is the last user of HPD work */
> +	intel_hpd_cancel_work(i915);
> +

MST still requires AUX and short HPD interrupts and during shutdown and
suspend the order is suspend-MST -> disable-IRQs. So I think it makes
more sense to move intel_dp_mst_suspend() to i915_driver_remove() before
intel_irq_uninstall().

>  	/* poll work can call into fbdev, hence clean that up afterwards */
>  	intel_fbdev_fini(i915);
>  
> diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
> index 73cebc6aa65072..db14787aef95dd 100644
> --- a/drivers/gpu/drm/i915/i915_irq.c
> +++ b/drivers/gpu/drm/i915/i915_irq.c
> @@ -4597,7 +4597,6 @@ void intel_irq_uninstall(struct drm_i915_private *dev_priv)
>  
>  	free_irq(irq, dev_priv);
>  
> -	intel_hpd_cancel_work(dev_priv);
>  	dev_priv->runtime_pm.irqs_enabled = false;
>  }
>  
> -- 
> 2.25.1
>