Re: [PATCH v2] drm/loongson: stop to set get_vblank_timestamp for LS7A1000 CRTC
From: Icenowy Zheng
Date: Mon Mar 23 2026 - 07:42:46 EST
在 2026-03-23一的 11:31 +0100，Thomas Zimmermann写道：
> Hi
>
> Am 23.03.26 um 10:40 schrieb Icenowy Zheng:
> > 在 2026-03-23一的 11:40 +0800，Jianmin Lv写道：
> > >
> > > On 2026/3/21 下午5:29, Icenowy Zheng wrote:
> > > > 在 2026-03-21六的 17:28 +0800，Icenowy Zheng写道：
> > > > > As there's no known hardware capability about querying vblank
> > > > > on
> > > > > the
> > > > > LS7A1000 display controller, setting get_vblank_timestamp
> > > > > will
> > > > > mislead the
> > > > > kernel about the support of DC-backed high precision vblank
> > > > > query.
> > > > >
> > > > > Drop this function pointer in the CRTC function table for
> > > > > LS7A1000.
> > > > >
> > > > > This solves a kernel warning when booting Linux 7.0-rc3 on a
> > > > > Loongson-3A4000+LS7A1000 Haier Boyue G51 laptop (with
> > > > > injected
> > > > > EDID
> > > > > for
> > > > > replicating the display timing set by the firmware).
> > > > >
> > > The drm_crtc_vblank_helper_get_vblank_timestamp will call
> > > get_scanout_position hook to caculate timestamp, and the driver
> > > has
> > > get_scanout_position to support it (LS7A1000 has related
> > > register).
> > > So
> > > the get_vblank_timestamp hook is ok for LS7A1000, and we'd better
> > > look
> > > into the warning information.
> > The warning information is copied here:
> > ```
> > [ 0.334960] ------------[ cut here ]------------
> > [ 0.334960] WARNING: drivers/gpu/drm/drm_vblank.c:747 at
> > drm_crtc_vblank_helper_get_vblank_timestamp_internal+0x3c8/0x438,
> > CPU#0: kworker/0:1/11
> > [ 0.334960] loongson 0000:00:06.1: [drm]
> > drm_WARN_ON_ONCE(drm_drv_uses_atomic_modeset(dev))
>
> This is what you should investigate. mode->crtc_clock seems invalid.
> [1]
>
> [1]
> https://elixir.bootlin.com/linux/v6.19.9/source/drivers/gpu/drm/drm_vblank.c#L743
>
> Best regards
> Thomas
>
> > [ 0.334960] Modules linked in:
> > [ 0.334960] CPU: 0 UID: 0 PID: 11 Comm: kworker/0:1 Not tainted
> > 7.0.0-rc4-3a4000-laptop-test+ #31 PREEMPT
> > [ 0.334960] Hardware name: Haier Boyue G51/Boyue G51, BIOS
> > K3L3A4000.HR10.200909.T 09/09/2020
> > [ 0.334960] Workqueue: events drm_fb_helper_damage_work
> > [ 0.334960] Stack : ffffffff819dec03 0000000000000000
> > 0000000000000018 9800000100223938
> > [ 0.334960] 9800000100223938 9800000100223a68
> > 0000000000000000 0000000000000000
> > [ 0.334960] b8e8b3f2ec6b8200 98000000092bffa8
> > 9800000100223768 0000000000000000
> > [ 0.334960] ffffffff81632b08 0000000000000030
> > ffffffff813c9530 0000000000200b20
> > [ 0.334960] fffffffffffffffc 0000000000000000
> > 0000000000000000 ffffffff816c1d50
> > [ 0.334960] ffffffff81890000 00000000000002eb
> > 0000000000000000 ffffffff80c1b180
> > [ 0.334960] 9800000100ff8500 0000000000000000
> > 9800000005c965a8 0000000000000000
> > [ 0.334960] ffffffff81a70000 9800000100220000
> > 9800000100223930 0000000000000000
> > [ 0.334960] ffffffff802274dc 0000000000000000
> > 0000000000000000 0000000000000000
> > [ 0.334960] 0000000000000000 0000000000000000
> > ffffffff802274fc b8e8b3f2ec6b8200
> > [ 0.334960] ...
> > [ 0.334960] Call Trace:
> > [ 0.334960] [<ffffffff802274fc>] show_stack+0x64/0x158
> > [ 0.334960] [<ffffffff8021bac4>] dump_stack_lvl+0x70/0xb0
> > [ 0.334960] [<ffffffff8026a988>] __warn+0xa8/0x120
> > [ 0.334960] [<ffffffff8026ab60>] warn_slowpath_fmt+0x160/0x1f8
> > [ 0.334960] [<ffffffff80be5668>]
> > drm_crtc_vblank_helper_get_vblank_timestamp_internal+0x3c8/0x438
> > [ 0.334960] [<ffffffff80be48d8>]
> > drm_crtc_get_last_vbltimestamp+0x68/0xb0
> > [ 0.334960] [<ffffffff80be5e14>]
> > drm_update_vblank_count+0xac/0x418
> > [ 0.334960] [<ffffffff80be6270>] drm_vblank_enable+0xf0/0x288
> > [ 0.334960] [<ffffffff80be7010>] drm_vblank_get+0xf8/0x160
> > [ 0.334960] [<ffffffff80beb2e0>]
> > drm_client_modeset_wait_for_vblank+0x50/0xa0
I doubt the problem here is that the function above is called with
hardcoded 0 by drm_fb_helper_fb_dirty() (being inlined into
drm_fb_helper_damage_work() by the compiler). Then on my setup it
failed because my primary output (laptop LCD) is on CRTC 1 (CRTC 0 is
external HDMI instead).
This is new behavior introduced by d8c4bddcd8bc ("drm/fb-helper:
Synchronize dirty worker with vblank"), and the commit messages say
`This allows several screen updates to pile up and acts as a rate
limiter.` .
Should it wait for any vblanks instead of waiting for one on CRTC 0 ?
(Or check whether CRTC 0 is on before waiting on it?)
Thanks,
Icenowy
> > [ 0.334960] [<ffffffff80c0a144>]
> > drm_fb_helper_damage_work+0x8c/0x1d8
> > [ 0.334960] [<ffffffff8028fac8>] process_one_work+0x198/0x348
> > [ 0.334960] [<ffffffff802906c0>] worker_thread+0x238/0x390
> > [ 0.334960] [<ffffffff8029c0d8>] kthread+0x160/0x198
> > [ 0.334960] [<ffffffff8021fd8c>]
> > ret_from_kernel_thread+0x14/0x1c
> >
> > [ 0.334960] ---[ end trace 0000000000000000 ]---
> > ```
> >
> > > > > Signed-off-by: Icenowy Zheng <zhengxingda@xxxxxxxxxxx>
> > > > Oh forgot to attach when crafting v2:
> > > >
> > > > ```
> > > > Acked-by: Thomas Zimmermann <tzimmermann@xxxxxxx>
> > > > ```
> > > >
> > > > > ---
> > > > > Changes in v2:
> > > > > - Re-formatted Loongson product model numbers per request
> > > > > from
> > > > > Huacai.
> > > > >
> > > > > drivers/gpu/drm/loongson/lsdc_crtc.c | 1 -
> > > > > 1 file changed, 1 deletion(-)
> > > > >
> > > > > diff --git a/drivers/gpu/drm/loongson/lsdc_crtc.c
> > > > > b/drivers/gpu/drm/loongson/lsdc_crtc.c
> > > > > index 587fbe285e9ef..b3af8e0cdb15f 100644
> > > > > --- a/drivers/gpu/drm/loongson/lsdc_crtc.c
> > > > > +++ b/drivers/gpu/drm/loongson/lsdc_crtc.c
> > > > > @@ -721,7 +721,6 @@ static const struct drm_crtc_funcs
> > > > > ls7a1000_crtc_funcs = {
> > > > > .late_register = lsdc_crtc_late_register,
> > > > > .enable_vblank = lsdc_crtc_enable_vblank,
> > > > > .disable_vblank = lsdc_crtc_disable_vblank,
> > > > > - .get_vblank_timestamp =
> > > > > drm_crtc_vblank_helper_get_vblank_timestamp,
> > > > > .atomic_print_state = lsdc_crtc_atomic_print_state,
> > > > > };
> > > > >