mirror of
https://github.com/nxp-imx/linux-imx.git
synced 2025-09-03 02:16:09 +02:00
c6e5ded297
The stress test is blocked in dma_fence_default_wait and missing dpu_be_fence signaling leads to weston hang. The root cause is unreasonable calling sequence of irq_handler and atomic_inc. We use refcnt in dpu_be_fence to sync dual dpu on 8QM. To avoid this issue, we set refcnt = dpu_num when get_fence. Then each dpu blitter will decrease the refcnt in irq_handler when doing sync and finally signal the fence. The same issue with LF-11223. [ 1212.771073] sysrq: Show Blocked State [ 1212.774848] task:kworker/u12:36 state stack:0 pid:975 tgid:975 ppid:2 flags:0x00000208 [ 1212.784142] Workqueue: events_unbound commit_work [ 1212.788846] Call trace: [ 1212.791282] __switch_to+0xe4/0x128 [ 1212.794765] __schedule+0x2d0/0x83c [ 1212.798249] schedule+0x34/0xc8 [ 1212.801386] schedule_timeout+0xfc/0x108 [ 1212.805304] dma_fence_default_wait+0x1b8/0x1ec [ 1212.809828] dma_fence_wait_timeout+0x84/0xa4 [ 1212.814178] drm_atomic_helper_wait_for_fences+0x174/0x234 [ 1212.819656] commit_tail+0x38/0x18c [ 1212.823131] commit_work+0x14/0x20 [ 1212.826519] process_one_work+0x138/0x248 [ 1212.830524] worker_thread+0x320/0x438 [ 1212.834268] kthread+0x110/0x114 [ 1212.837492] ret_from_fork+0x10/0x20 Signed-off-by: Guangliu Ding <guangliu.ding@nxp.com> Reviewed-by: Jiyu Yang <jiyu.yang@nxp.com> Tested-by: Jared Hu <jared.hu@nxp.com> |
||
|---|---|---|
| .. | ||
| dpu-blit-registers.h | ||
| dpu-blit.c | ||
| dpu-blit.h | ||
| Kconfig | ||
| Makefile | ||