On Thu, Jun 23, 2016 at 01:45:06PM +0200, Maarten Lankhorst wrote:
> Atomic updates may acquire more state than initially locked through
> drm_modeset_lock_crtc, running with heavy stress can cause a
> WARN_ON(crtc->acquire_ctx) in drm_modeset_lock_crtc:
>
> [ 601.491296] [ cut here ]
> [ 601.491366] WARNING: CPU: 0 PID: 2411 at
> drivers/gpu/drm/drm_modeset_lock.c:191 drm_modeset_lock_crtc+0xeb/0xf0 [drm]
> [ 601.491369] Modules linked in: drm i915 drm_kms_helper
> [ 601.491414] CPU: 0 PID: 2411 Comm: kms_cursor_lega Tainted: G U
> 4.7.0-rc4-patser+ #4798
> [ 601.491417] Hardware name: Intel Corporation Skylake Client
> [ 601.491420] 88044d153c98 812ead28
>
> [ 601.491425] 88044d153cd8 810868e6
> 00bf58058030
> [ 601.491431] 880088b415e8 880458058030 88008a271548
> 88008a271568
> [ 601.491436] Call Trace:
> [ 601.491443] [] dump_stack+0x4d/0x65
> [ 601.491447] [] __warn+0xc6/0xe0
> [ 601.491452] [] warn_slowpath_null+0x18/0x20
> [ 601.491472] [] drm_modeset_lock_crtc+0xeb/0xf0 [drm]
> [ 601.491491] [] drm_mode_cursor_common+0x66/0x180 [drm]
> [ 601.491509] [] drm_mode_cursor_ioctl+0x3c/0x40 [drm]
> [ 601.491524] [] drm_ioctl+0x14d/0x530 [drm]
> [ 601.491540] [] ? drm_mode_setcrtc+0x520/0x520 [drm]
> [ 601.491545] [] ? handle_mm_fault+0x106b/0x1430
> [ 601.491550] [] ? stop_one_cpu+0x61/0x70
> [ 601.491556] [] do_vfs_ioctl+0x8d/0x570
> [ 601.491560] [] ? security_file_ioctl+0x3e/0x60
> [ 601.491565] [] SyS_ioctl+0x74/0x80
> [ 601.491571] [] ? posix_get_monotonic_raw+0xc/0x10
> [ 601.491576] [] entry_SYSCALL_64_fastpath+0x13/0x8f
> [ 601.491581] ---[ end trace 56f3d3d85f000d00 ]---
>
> For good measure, test mode_config.acquire_ctx too, although this should
> never happen.
>
> Testcase: kms_cursor_legacy
> Signed-off-by: Maarten Lankhorst
It's kinda a bug in the skl wm code which unecessarily looks at other
plane state.
Reviewed-by: Daniel Vetter
Given that this is fairly hard to hit, but will blow in really strange
ways in practice I think also justified for backporting.
Cc: stable at vger.kernel.org
Dave, can you pls pick up for -fixes directly?
Thanks, Daniel
> ---
> drivers/gpu/drm/drm_atomic.c | 27 ++-
> 1 file changed, 26 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/drm_atomic.c b/drivers/gpu/drm/drm_atomic.c
> index d99ab2f6663f..3cee084e9d28 100644
> --- a/drivers/gpu/drm/drm_atomic.c
> +++ b/drivers/gpu/drm/drm_atomic.c
> @@ -1299,14 +1299,39 @@ EXPORT_SYMBOL(drm_atomic_add_affected_planes);
> */
> void drm_atomic_legacy_backoff(struct drm_atomic_state *state)
> {
> + struct drm_device *dev = state->dev;
> + unsigned crtc_mask = 0;
> + struct drm_crtc *crtc;
> int ret;
> + bool global = false;
> +
> + drm_for_each_crtc(crtc, dev) {
> + if (crtc->acquire_ctx != state->acquire_ctx)
> + continue;
> +
> + crtc_mask |= drm_crtc_mask(crtc);
> + crtc->acquire_ctx = NULL;
> + }
> +
> + if (WARN_ON(dev->mode_config.acquire_ctx == state->acquire_ctx)) {
> + global = true;
> +
> + dev->mode_config.acquire_ctx = NULL;
> + }
>
> retry:
> drm_modeset_backoff(state->acquire_ctx);
>
> - ret = drm_modeset_lock_all_ctx(state->dev, state->acquire_ctx);
> + ret = drm_modeset_lock_all_ctx(dev, state->acquire_ctx);
> if (ret)
> goto retry;
> +
> + drm_for_each_crtc(crtc, dev)
> + if (drm_crtc_mask(crtc) & crtc_mask)
> + crtc->acquire_ctx = state->acquire_ctx;
> +
> + if (global)
> + dev->mode_config.acquire_ctx = state->acquire_ctx;
> }
> EXPORT_SYMBOL(drm_atomic_legacy_backoff);
>
> --
> 2.5.5
>
> ___
> Intel-gfx mailing list
> Intel-gfx at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
