It was recently introduced a change that enables driver to disable streams if pixel clock changes. Consequently, the code path executed in the disable vbios function expanded to an encoder verification part. The encoder loop is nested inside the pipe count loop, and both loops share the 'i' variable in control of their flow. This situation may lead to an infinite loop because the encoder loop constantly updates the `i` variable, making the first loop always positive. As a result, we can see a soft hang during the module load (modprobe amdgpu) and a series of dmesg log that looks like this:
kernel:[ 124.538727] watchdog: BUG: soft lockup - CPU#2 stuck for 22s! [modprobe:1000] RSP: 0018:ffffabbf419bf0e8 EFLAGS: 00000282 RAX: ffffffffc0809de0 RBX: ffff93b35ccc0000 RCX: ffff93b366c21800 RDX: 0000000000000000 RSI: 0000000000000141 RDI: ffff93b35ccc0000 RBP: ffffabbf419bf108 R08: ffffabbf419bf164 R09: 0000000000000001 R10: 0000000000000003 R11: 0000000000000003 R12: 0000000008677d40 R13: 0000000000000141 R14: ffff93b35cfc0000 R15: ffff93b35abc0000 FS: 00007f1400717540(0000) GS:ffff93b37f680000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00005649b66b0968 CR3: 00000003e0fec000 CR4: 0000000000350ee0
Call Trace: amdgpu_device_rreg+0x17/0x20 [amdgpu] amdgpu_cgs_read_register+0x14/0x20 [amdgpu] dm_read_reg_func+0x3a/0xb0 [amdgpu] get_pixel_clk_frequency_100hz+0x30/0x50 [amdgpu] dc_commit_state+0x8f1/0xae0 [amdgpu] ? drm_calc_timestamping_constants+0x101/0x160 [drm] amdgpu_dm_atomic_commit_tail+0x39d/0x21a0 [amdgpu] ? dcn21_validate_bandwidth+0xe5/0x290 [amdgpu] ? kfree+0xc3/0x390 ? dcn21_validate_bandwidth+0xe5/0x290 [amdgpu] ... RSP: 002b:00007fff26009bd8 EFLAGS: 00000246 ORIG_RAX: 0000000000000139 RAX: ffffffffffffffda RBX: 000055a8025bea50 RCX: 00007f140085c89d RDX: 0000000000000000 RSI: 000055a8025b8290 RDI: 000000000000000c RBP: 0000000000040000 R08: 0000000000000000 R09: 0000000000000000 R10: 000000000000000c R11: 0000000000000246 R12: 000055a8025b8290 R13: 0000000000000000 R14: 000055a8025bead0 R15: 000055a8025bea50
This issue was fixed by introducing a second variable for the internal loop.
Fixes: 32933ac6ec3a ("drm/amd/display: disable stream if pixel clock changed with link active") Signed-off-by: Rodrigo Siqueira Rodrigo.Siqueira@amd.com --- drivers/gpu/drm/amd/display/dc/core/dc.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)
diff --git a/drivers/gpu/drm/amd/display/dc/core/dc.c b/drivers/gpu/drm/amd/display/dc/core/dc.c index 8d96ef157987..3f7d04b8956a 100644 --- a/drivers/gpu/drm/amd/display/dc/core/dc.c +++ b/drivers/gpu/drm/amd/display/dc/core/dc.c @@ -848,7 +848,7 @@ static void disable_vbios_mode_if_required( struct dc *dc, struct dc_state *context) { - unsigned int i; + unsigned int i, j;
/* check if timing_changed, disable stream*/ for (i = 0; i < dc->res_pool->pipe_count; i++) { @@ -872,10 +872,10 @@ static void disable_vbios_mode_if_required(
enc_inst = link->link_enc->funcs->get_dig_frontend(link->link_enc); if (enc_inst != ENGINE_ID_UNKNOWN) { - for (i = 0; i < dc->res_pool->stream_enc_count; i++) { - if (dc->res_pool->stream_enc[i]->id == enc_inst) { - tg_inst = dc->res_pool->stream_enc[i]->funcs->dig_source_otg( - dc->res_pool->stream_enc[i]); + for (j = 0; j < dc->res_pool->stream_enc_count; j++) { + if (dc->res_pool->stream_enc[j]->id == enc_inst) { + tg_inst = dc->res_pool->stream_enc[j]->funcs->dig_source_otg( + dc->res_pool->stream_enc[j]); break; } }
[AMD Public Use]
Reviewed-by: Roman Li Roman.Li@amd.com
-----Original Message----- From: Siqueira, Rodrigo Rodrigo.Siqueira@amd.com Sent: Friday, October 9, 2020 5:18 PM To: dri-devel@lists.freedesktop.org Cc: Wentland, Harry Harry.Wentland@amd.com; Li, Sun peng (Leo) Sunpeng.Li@amd.com; Deucher, Alexander Alexander.Deucher@amd.com; Li, Roman Roman.Li@amd.com; Kazlauskas, Nicholas Nicholas.Kazlauskas@amd.com Subject: [PATCH] drm/amd/display: Fix module load hangs when connected to an eDP
It was recently introduced a change that enables driver to disable streams if pixel clock changes. Consequently, the code path executed in the disable vbios function expanded to an encoder verification part. The encoder loop is nested inside the pipe count loop, and both loops share the 'i' variable in control of their flow. This situation may lead to an infinite loop because the encoder loop constantly updates the `i` variable, making the first loop always positive. As a result, we can see a soft hang during the module load (modprobe amdgpu) and a series of dmesg log that looks like this:
kernel:[ 124.538727] watchdog: BUG: soft lockup - CPU#2 stuck for 22s! [modprobe:1000] RSP: 0018:ffffabbf419bf0e8 EFLAGS: 00000282 RAX: ffffffffc0809de0 RBX: ffff93b35ccc0000 RCX: ffff93b366c21800 RDX: 0000000000000000 RSI: 0000000000000141 RDI: ffff93b35ccc0000 RBP: ffffabbf419bf108 R08: ffffabbf419bf164 R09: 0000000000000001 R10: 0000000000000003 R11: 0000000000000003 R12: 0000000008677d40 R13: 0000000000000141 R14: ffff93b35cfc0000 R15: ffff93b35abc0000 FS: 00007f1400717540(0000) GS:ffff93b37f680000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00005649b66b0968 CR3: 00000003e0fec000 CR4: 0000000000350ee0
Call Trace: amdgpu_device_rreg+0x17/0x20 [amdgpu] amdgpu_cgs_read_register+0x14/0x20 [amdgpu] dm_read_reg_func+0x3a/0xb0 [amdgpu] get_pixel_clk_frequency_100hz+0x30/0x50 [amdgpu] dc_commit_state+0x8f1/0xae0 [amdgpu] ? drm_calc_timestamping_constants+0x101/0x160 [drm] amdgpu_dm_atomic_commit_tail+0x39d/0x21a0 [amdgpu] ? dcn21_validate_bandwidth+0xe5/0x290 [amdgpu] ? kfree+0xc3/0x390 ? dcn21_validate_bandwidth+0xe5/0x290 [amdgpu] ... RSP: 002b:00007fff26009bd8 EFLAGS: 00000246 ORIG_RAX: 0000000000000139 RAX: ffffffffffffffda RBX: 000055a8025bea50 RCX: 00007f140085c89d RDX: 0000000000000000 RSI: 000055a8025b8290 RDI: 000000000000000c RBP: 0000000000040000 R08: 0000000000000000 R09: 0000000000000000 R10: 000000000000000c R11: 0000000000000246 R12: 000055a8025b8290 R13: 0000000000000000 R14: 000055a8025bead0 R15: 000055a8025bea50
This issue was fixed by introducing a second variable for the internal loop.
Fixes: 32933ac6ec3a ("drm/amd/display: disable stream if pixel clock changed with link active") Signed-off-by: Rodrigo Siqueira Rodrigo.Siqueira@amd.com --- drivers/gpu/drm/amd/display/dc/core/dc.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)
diff --git a/drivers/gpu/drm/amd/display/dc/core/dc.c b/drivers/gpu/drm/amd/display/dc/core/dc.c index 8d96ef157987..3f7d04b8956a 100644 --- a/drivers/gpu/drm/amd/display/dc/core/dc.c +++ b/drivers/gpu/drm/amd/display/dc/core/dc.c @@ -848,7 +848,7 @@ static void disable_vbios_mode_if_required( struct dc *dc, struct dc_state *context) { - unsigned int i; + unsigned int i, j;
/* check if timing_changed, disable stream*/ for (i = 0; i < dc->res_pool->pipe_count; i++) { @@ -872,10 +872,10 @@ static void disable_vbios_mode_if_required(
enc_inst = link->link_enc->funcs->get_dig_frontend(link->link_enc); if (enc_inst != ENGINE_ID_UNKNOWN) { - for (i = 0; i < dc->res_pool->stream_enc_count; i++) { - if (dc->res_pool->stream_enc[i]->id == enc_inst) { - tg_inst = dc->res_pool->stream_enc[i]->funcs->dig_source_otg( - dc->res_pool->stream_enc[i]); + for (j = 0; j < dc->res_pool->stream_enc_count; j++) { + if (dc->res_pool->stream_enc[j]->id == enc_inst) { + tg_inst = dc->res_pool->stream_enc[j]->funcs->dig_source_otg( + dc->res_pool->stream_enc[j]); break; } } -- 2.28.0
dri-devel@lists.freedesktop.org
-
Li, Roman -
Rodrigo Siqueira