| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <b21aa2e6-1b47-cdc1-307c-21fb331d4afb@amd.com>
Date: Fri, 23 Sep 2022 16:33:49 +0200
From: Christian König <christian.koenig@....com>
To: Daniel Gomez <daniel@...c.com>
Cc: dagmcr@...il.com, Harry Wentland <harry.wentland@....com>,
Leo Li <sunpeng.li@....com>,
Rodrigo Siqueira <Rodrigo.Siqueira@....com>,
Alex Deucher <alexander.deucher@....com>,
"Pan, Xinhui" <Xinhui.Pan@....com>,
David Airlie <airlied@...ux.ie>,
Daniel Vetter <daniel@...ll.ch>,
Alan Liu <HaoPing.Liu@....com>, Melissa Wen <mwen@...lia.com>,
Aurabindo Pillai <aurabindo.pillai@....com>,
Bernard Zhao <bernard@...o.com>,
Isabella Basso <isabbasso@...eup.net>,
Becle Lee <becle.lee@....com>,
Bas Nieuwenhuizen <bas@...nieuwenhuizen.nl>,
amd-gfx@...ts.freedesktop.org, dri-devel@...ts.freedesktop.org,
linux-kernel@...r.kernel.org
Subject: Re: [RFC][PATCH] drm/amd/display: Restore DC_FP_* wrapper in
dml/calcs
Am 23.09.22 um 16:17 schrieb Daniel Gomez:
> Commit [1] removes DC_FP_* wrappers from dml. However, this generates
> the BUG [2] on the amdgpu driver. Restore DC_FP_* wrappers in dml/calcs
> but only for the functions dcn_bw_update_from_pplib and
> dcn_bw_notify_pplib_of_wm_ranges.
That's a really bad idea. The wrappers must be *outside* of the code
file which uses floating point code.
The code should probably just be re-aranged to not try to lock the mutex
inside the critical section.
Regards,
Christian.
>
> [1] 9696679bf7ac40a8fb6a488a75bd66d4414cd3c3 drm/amd/display: remove
> DC_FP_* wrapper from dml folder
>
> [2] BUG: sleeping function called from invalid context at
> kernel/locking/mutex.c:283
>
> Signed-off-by: Daniel Gomez <daniel@...c.com>
> ---
>
> Hi,
>
> The patch [1] introduces BUG [2] since linux 5.18. The reason seems to be
> wrapping entirely the functions dcn_bw_update_from_pplib and
> dcn_bw_notify_pplib_of_wm_ranges in the dcn10.
>
> On dcn_bw_update_from_pplib function the problem seems to be the
> dm_pp_get_clock_levels_by_type_with_voltage call.
>
> Any suggestions on what should we do here?
>
> BUG: sleeping function called from invalid context at kernel/locking/mutex.c:283
> in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 227, name: systemd-udevd
> preempt_count: 1, expected: 0
> CPU: 4 PID: 227 Comm: systemd-udevd Not tainted 6.0.0-rc6-qtec-standard #2
> Hardware name: Qtechnology A/S QT5222/QT5221, BIOS v1.0.1 06/07/2021
> Call Trace:
> <TASK>
> dump_stack_lvl+0x33/0x42
> __might_resched.cold.172+0xa5/0xb3
> mutex_lock+0x1a/0x40
> amdgpu_dpm_get_clock_by_type_with_voltage+0x38/0x70 [amdgpu]
> dm_pp_get_clock_levels_by_type_with_voltage+0x64/0xa0 [amdgpu]
> dcn_bw_update_from_pplib+0x70/0x340 [amdgpu]
> dcn10_create_resource_pool+0x8c8/0xd20 [amdgpu]
> ? __kmalloc+0x1c7/0x4a0
> dc_create_resource_pool+0xe7/0x190 [amdgpu]
> dc_create+0x212/0x5d0 [amdgpu]
> amdgpu_dm_init+0x246/0x370 [amdgpu]
> ? schedule_hrtimeout_range_clock+0x93/0x120
> ? phm_wait_for_register_unequal.part.1+0x4a/0x80 [amdgpu]
> dm_hw_init+0xe/0x20 [amdgpu]
> amdgpu_device_init.cold.56+0x1324/0x1653 [amdgpu]
> ? pci_bus_read_config_word+0x43/0x80
> amdgpu_driver_load_kms+0x15/0x120 [amdgpu]
> amdgpu_pci_probe+0x116/0x320 [amdgpu]
> pci_device_probe+0x97/0x110
> really_probe+0xdd/0x340
> __driver_probe_device+0x80/0x170
> driver_probe_device+0x1f/0x90
> __driver_attach+0xdc/0x180
> ? __device_attach_driver+0x100/0x100
> ? __device_attach_driver+0x100/0x100
> bus_for_each_dev+0x74/0xc0
> bus_add_driver+0x19e/0x210
> ? kset_find_obj+0x30/0xa0
> ? 0xffffffffa0a5b000
> driver_register+0x6b/0xc0
> ? 0xffffffffa0a5b000
> do_one_initcall+0x4a/0x1f0
> ? __vunmap+0x28e/0x2f0
> ? __cond_resched+0x15/0x30
> ? kmem_cache_alloc_trace+0x3d/0x440
> do_init_module+0x4a/0x1e0
> load_module+0x1cba/0x1e10
> ? __do_sys_finit_module+0xb7/0x120
> __do_sys_finit_module+0xb7/0x120
> do_syscall_64+0x3c/0x80
> entry_SYSCALL_64_after_hwframe+0x63/0xcd
> RIP: 0033:0x7ff2b5f5422d
> Code: 5d c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d c3 ab 0e 00 f7 d8 64 89 01 48
> RSP: 002b:00007ffc44ab28e8 EFLAGS: 00000246 ORIG_RAX: 0000000000000139
> RAX: ffffffffffffffda RBX: 0000555c566a9240 RCX: 00007ff2b5f5422d
> RDX: 0000000000000000 RSI: 00007ff2b60bb353 RDI: 0000000000000019
> RBP: 00007ff2b60bb353 R08: 0000000000000000 R09: 0000555c566a9240
> R10: 0000000000000019 R11: 0000000000000246 R12: 0000000000000000
> R13: 0000000000020000 R14: 0000000000000000 R15: 0000000000000000
> </TASK>
>
> drivers/gpu/drm/amd/display/dc/dcn10/dcn10_resource.c | 2 --
> drivers/gpu/drm/amd/display/dc/dml/calcs/dcn_calcs.c | 10 ++++++++++
> 2 files changed, 10 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_resource.c b/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_resource.c
> index 174eebbe8b4f..a6ef20b43f3a 100644
> --- a/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_resource.c
> +++ b/drivers/gpu/drm/amd/display/dc/dcn10/dcn10_resource.c
> @@ -1505,7 +1505,6 @@ static bool dcn10_resource_construct(
> && pool->base.pp_smu->rv_funcs.set_pme_wa_enable != NULL)
> dc->debug.az_endpoint_mute_only = false;
>
> - DC_FP_START();
> if (!dc->debug.disable_pplib_clock_request)
> dcn_bw_update_from_pplib(dc);
> dcn_bw_sync_calcs_and_dml(dc);
> @@ -1513,7 +1512,6 @@ static bool dcn10_resource_construct(
> dc->res_pool = &pool->base;
> dcn_bw_notify_pplib_of_wm_ranges(dc);
> }
> - DC_FP_END();
>
> {
> struct irq_service_init_data init_data;
> diff --git a/drivers/gpu/drm/amd/display/dc/dml/calcs/dcn_calcs.c b/drivers/gpu/drm/amd/display/dc/dml/calcs/dcn_calcs.c
> index db3b16b77034..a3c71d875adb 100644
> --- a/drivers/gpu/drm/amd/display/dc/dml/calcs/dcn_calcs.c
> +++ b/drivers/gpu/drm/amd/display/dc/dml/calcs/dcn_calcs.c
> @@ -1490,6 +1490,8 @@ void dcn_bw_update_from_pplib(struct dc *dc)
> res = dm_pp_get_clock_levels_by_type_with_voltage(
> ctx, DM_PP_CLOCK_TYPE_FCLK, &fclks);
>
> + DC_FP_START();
> +
> if (res)
> res = verify_clock_values(&fclks);
>
> @@ -1519,9 +1521,13 @@ void dcn_bw_update_from_pplib(struct dc *dc)
> } else
> BREAK_TO_DEBUGGER();
>
> + DC_FP_END();
> +
> res = dm_pp_get_clock_levels_by_type_with_voltage(
> ctx, DM_PP_CLOCK_TYPE_DCFCLK, &dcfclks);
>
> + DC_FP_START();
> +
> if (res)
> res = verify_clock_values(&dcfclks);
>
> @@ -1532,6 +1538,8 @@ void dcn_bw_update_from_pplib(struct dc *dc)
> dc->dcn_soc->dcfclkv_max0p9 = dcfclks.data[dcfclks.num_levels - 1].clocks_in_khz / 1000.0;
> } else
> BREAK_TO_DEBUGGER();
> +
> + DC_FP_END();
> }
>
> void dcn_bw_notify_pplib_of_wm_ranges(struct dc *dc)
> @@ -1546,9 +1554,11 @@ void dcn_bw_notify_pplib_of_wm_ranges(struct dc *dc)
> if (!pp || !pp->set_wm_ranges)
> return;
>
> + DC_FP_START();
> min_fclk_khz = dc->dcn_soc->fabric_and_dram_bandwidth_vmin0p65 * 1000000 / 32;
> min_dcfclk_khz = dc->dcn_soc->dcfclkv_min0p65 * 1000;
> socclk_khz = dc->dcn_soc->socclk * 1000;
> + DC_FP_END();
>
> /* Now notify PPLib/SMU about which Watermarks sets they should select
> * depending on DPM state they are in. And update BW MGR GFX Engine and
> --
> 2.35.1
>
Powered by blists - more mailing lists