From: Rui Wang rui.y.wang@intel.com
There are still some places in the fb helper that need to avoid sleeping in panic context. Here's an example:
[ 65.615496] bad: scheduling from the idle thread! [ 65.620747] CPU: 92 PID: 0 Comm: swapper/92 Tainted: G M E 3.18.0-rc4-7-default+ #20
[ 65.630364] Hardware name: Intel Corporation BRICKLAND/BRICKLAND, BIOS BRHSXSD1.86B.0056.R01.1409242327 09/24/2014 [ 65.641923] ffff88087f693d80 ffff88087f689878 ffffffff81566db9 0000000000000000 [ 65.650226] ffff88087f693d80 ffff88087f689898 ffffffff810871ff ffff88046eb3e0d0 [ 65.658527] ffff88087f693d80 ffff88087f6898c8 ffffffff8107c1fa 000000017f6898b8 [ 65.666830] Call Trace: [ 65.669557] <#MC> [<ffffffff81566db9>] dump_stack+0x46/0x58 [ 65.675994] [<ffffffff810871ff>] dequeue_task_idle+0x2f/0x40 [ 65.682412] [<ffffffff8107c1fa>] dequeue_task+0x5a/0x80 [ 65.688345] [<ffffffff810804f3>] deactivate_task+0x23/0x30 [ 65.694569] [<ffffffff81569050>] __schedule+0x580/0x7f0 [ 65.700502] [<ffffffff81569739>] schedule_preempt_disabled+0x29/0x70 [ 65.707696] [<ffffffff8156abb6>] __ww_mutex_lock_slowpath+0xb8/0x162 [ 65.714891] [<ffffffff8156acb3>] __ww_mutex_lock+0x53/0x85 [ 65.721125] [<ffffffffa00b3a5d>] drm_modeset_lock+0x3d/0x110 [drm] [ 65.728132] [<ffffffffa00b3c2a>] __drm_modeset_lock_all+0x8a/0x120 [drm] [ 65.735721] [<ffffffffa00b3cd0>] drm_modeset_lock_all+0x10/0x30 [drm] [ 65.743015] [<ffffffffa01af8bf>] drm_fb_helper_pan_display+0x2f/0xf0 [drm_kms_helper] [ 65.751857] [<ffffffff8132bd21>] fb_pan_display+0xd1/0x1a0 [ 65.758081] [<ffffffff81326010>] bit_update_start+0x20/0x50 [ 65.764400] [<ffffffff813259f2>] fbcon_switch+0x3a2/0x550 [ 65.770528] [<ffffffff813a01c9>] redraw_screen+0x189/0x240 [ 65.776750] [<ffffffff81322f8a>] fbcon_blank+0x20a/0x2d0 [ 65.782778] [<ffffffff8137d359>] ? erst_writer+0x209/0x330 [ 65.789002] [<ffffffff810ba2f3>] ? internal_add_timer+0x63/0x80 [ 65.795710] [<ffffffff810bc137>] ? mod_timer+0x127/0x1e0 [ 65.801740] [<ffffffff813a0cd8>] do_unblank_screen+0xa8/0x1d0 [ 65.808255] [<ffffffff813a0e10>] unblank_screen+0x10/0x20 [ 65.814381] [<ffffffff812ca0d9>] bust_spinlocks+0x19/0x40 [ 65.820508] [<ffffffff81561ca7>] panic+0x106/0x1f5 [ 65.825955] [<ffffffff8102336c>] mce_panic+0x2ac/0x2e0 [ 65.831789] [<ffffffff812c796a>] ? delay_tsc+0x4a/0x80 [ 65.837625] [<ffffffff81024e1f>] do_machine_check+0xbaf/0xbf0 [ 65.844138] [<ffffffff813365d7>] ? intel_idle+0xc7/0x150 [ 65.850166] [<ffffffff8156f03f>] machine_check+0x1f/0x30 [ 65.856195] [<ffffffff813365d7>] ? intel_idle+0xc7/0x150 [ 65.862222] <<EOE>> [<ffffffff814283d5>] cpuidle_enter_state+0x55/0x170 [ 65.869823] [<ffffffff814285a7>] cpuidle_enter+0x17/0x20 [ 65.875852] [<ffffffff81097b08>] cpu_startup_entry+0x2d8/0x370 [ 65.882467] [<ffffffff8102fe29>] start_secondary+0x159/0x180
There's __drm_modeset_lock_all() which Daniel Vetter introduced for this purpose. We can leverage that without reinventing anything. This patch works with the latest kernel.
Reviewed-by: Rob Clark robdclark@gmail.com Tested-by: Tony Luck tony.luck@intel.com Signed-off-by: Rui Wang rui.y.wang@intel.com ---
Dave: I don't see this in any of your trees yet. Bug is really annoying when testing machine check code because the above stack dumps just repeat endlessly until I manually hit the reset button
drivers/gpu/drm/drm_fb_helper.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/drm_fb_helper.c b/drivers/gpu/drm/drm_fb_helper.c index 0c0c39bac23d..70dd2f4eb299 100644 --- a/drivers/gpu/drm/drm_fb_helper.c +++ b/drivers/gpu/drm/drm_fb_helper.c @@ -732,7 +732,9 @@ int drm_fb_helper_setcmap(struct fb_cmap *cmap, struct fb_info *info) int i, j, rc = 0; int start;
- drm_modeset_lock_all(dev); + if (__drm_modeset_lock_all(dev, !!oops_in_progress)) { + return -EBUSY; + } if (!drm_fb_helper_is_bound(fb_helper)) { drm_modeset_unlock_all(dev); return -EBUSY; @@ -910,7 +912,9 @@ int drm_fb_helper_pan_display(struct fb_var_screeninfo *var, int ret = 0; int i;
- drm_modeset_lock_all(dev); + if (__drm_modeset_lock_all(dev, !!oops_in_progress)) { + return -EBUSY; + } if (!drm_fb_helper_is_bound(fb_helper)) { drm_modeset_unlock_all(dev); return -EBUSY;