| Message ID | 20230427232848.5200-1-quic_abhinavk@quicinc.com |
|---|---|
| State | Accepted |
| Commit | b78c77273a5648eddc02f275f582b681a5127711 |
| Headers | show |
| Series | drm/msm/dp: add module parameter for PSR | expand |
On 28/04/2023 02:28, Abhinav Kumar wrote: > On sc7280 where eDP is the primary display, PSR is causing > IGT breakage even for basic test cases like kms_atomic and > kms_atomic_transition. Most often the issue starts with below > stack so providing that as reference > > Call trace: > dpu_encoder_assign_crtc+0x64/0x6c > dpu_crtc_enable+0x188/0x204 > drm_atomic_helper_commit_modeset_enables+0xc0/0x274 > msm_atomic_commit_tail+0x1a8/0x68c > commit_tail+0xb0/0x160 > drm_atomic_helper_commit+0x11c/0x124 > drm_atomic_commit+0xb0/0xdc > drm_atomic_connector_commit_dpms+0xf4/0x110 > drm_mode_obj_set_property_ioctl+0x16c/0x3b0 > drm_connector_property_set_ioctl+0x4c/0x74 > drm_ioctl_kernel+0xec/0x15c > drm_ioctl+0x264/0x408 > __arm64_sys_ioctl+0x9c/0xd4 > invoke_syscall+0x4c/0x110 > el0_svc_common+0x94/0xfc > do_el0_svc+0x3c/0xb0 > el0_svc+0x2c/0x7c > el0t_64_sync_handler+0x48/0x114 > el0t_64_sync+0x190/0x194 > ---[ end trace 0000000000000000 ]--- > [drm-dp] dp_ctrl_push_idle: PUSH_IDLE pattern timedout > > Other basic use-cases still seem to work fine hence add a > a module parameter to allow toggling psr enable/disable till > PSR related issues are hashed out with IGT. For the reference: Bjorn reported that he has issues with VT on a PSR-enabled laptops. This patch fixes the issue for him Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org> > > Signed-off-by: Abhinav Kumar <quic_abhinavk@quicinc.com> > --- > drivers/gpu/drm/msm/dp/dp_display.c | 6 +++++- > 1 file changed, 5 insertions(+), 1 deletion(-) > > diff --git a/drivers/gpu/drm/msm/dp/dp_display.c b/drivers/gpu/drm/msm/dp/dp_display.c > index 628b0e248db6..dba43167de66 100644 > --- a/drivers/gpu/drm/msm/dp/dp_display.c > +++ b/drivers/gpu/drm/msm/dp/dp_display.c > @@ -28,6 +28,10 @@ > #include "dp_audio.h" > #include "dp_debug.h" > > +static bool psr_enabled = false; > +module_param(psr_enabled, bool, 0); > +MODULE_PARM_DESC(psr_enabled, "enable PSR for eDP and DP displays"); > + > #define HPD_STRING_SIZE 30 > > enum { > @@ -407,7 +411,7 @@ static int dp_display_process_hpd_high(struct dp_display_private *dp) > > edid = dp->panel->edid; > > - dp->dp_display.psr_supported = dp->panel->psr_cap.version; > + dp->dp_display.psr_supported = dp->panel->psr_cap.version && psr_enabled; > > dp->audio_supported = drm_detect_monitor_audio(edid); > dp_panel_handle_sink_request(dp->panel);
On Fri, May 12, 2023 at 09:13:04PM +0300, Dmitry Baryshkov wrote: > On 28/04/2023 02:28, Abhinav Kumar wrote: > > On sc7280 where eDP is the primary display, PSR is causing > > IGT breakage even for basic test cases like kms_atomic and > > kms_atomic_transition. Most often the issue starts with below > > stack so providing that as reference > > > > Call trace: > > dpu_encoder_assign_crtc+0x64/0x6c > > dpu_crtc_enable+0x188/0x204 > > drm_atomic_helper_commit_modeset_enables+0xc0/0x274 > > msm_atomic_commit_tail+0x1a8/0x68c > > commit_tail+0xb0/0x160 > > drm_atomic_helper_commit+0x11c/0x124 > > drm_atomic_commit+0xb0/0xdc > > drm_atomic_connector_commit_dpms+0xf4/0x110 > > drm_mode_obj_set_property_ioctl+0x16c/0x3b0 > > drm_connector_property_set_ioctl+0x4c/0x74 > > drm_ioctl_kernel+0xec/0x15c > > drm_ioctl+0x264/0x408 > > __arm64_sys_ioctl+0x9c/0xd4 > > invoke_syscall+0x4c/0x110 > > el0_svc_common+0x94/0xfc > > do_el0_svc+0x3c/0xb0 > > el0_svc+0x2c/0x7c > > el0t_64_sync_handler+0x48/0x114 > > el0t_64_sync+0x190/0x194 > > ---[ end trace 0000000000000000 ]--- > > [drm-dp] dp_ctrl_push_idle: PUSH_IDLE pattern timedout > > > > Other basic use-cases still seem to work fine hence add a > > a module parameter to allow toggling psr enable/disable till > > PSR related issues are hashed out with IGT. > > For the reference: Bjorn reported that he has issues with VT on a > PSR-enabled laptops. This patch fixes the issue for him Module parameters are almost never warranted, and it is definitely not the right way to handle a broken implementation. I've just sent a revert that unconditionally disables PSR support until the implementation has been fixed: https://lore.kernel.org/lkml/20230523151646.28366-1-johan+linaro@kernel.org/ Johan
On 5/23/2023 8:24 AM, Johan Hovold wrote: > On Fri, May 12, 2023 at 09:13:04PM +0300, Dmitry Baryshkov wrote: >> On 28/04/2023 02:28, Abhinav Kumar wrote: >>> On sc7280 where eDP is the primary display, PSR is causing >>> IGT breakage even for basic test cases like kms_atomic and >>> kms_atomic_transition. Most often the issue starts with below >>> stack so providing that as reference >>> >>> Call trace: >>> dpu_encoder_assign_crtc+0x64/0x6c >>> dpu_crtc_enable+0x188/0x204 >>> drm_atomic_helper_commit_modeset_enables+0xc0/0x274 >>> msm_atomic_commit_tail+0x1a8/0x68c >>> commit_tail+0xb0/0x160 >>> drm_atomic_helper_commit+0x11c/0x124 >>> drm_atomic_commit+0xb0/0xdc >>> drm_atomic_connector_commit_dpms+0xf4/0x110 >>> drm_mode_obj_set_property_ioctl+0x16c/0x3b0 >>> drm_connector_property_set_ioctl+0x4c/0x74 >>> drm_ioctl_kernel+0xec/0x15c >>> drm_ioctl+0x264/0x408 >>> __arm64_sys_ioctl+0x9c/0xd4 >>> invoke_syscall+0x4c/0x110 >>> el0_svc_common+0x94/0xfc >>> do_el0_svc+0x3c/0xb0 >>> el0_svc+0x2c/0x7c >>> el0t_64_sync_handler+0x48/0x114 >>> el0t_64_sync+0x190/0x194 >>> ---[ end trace 0000000000000000 ]--- >>> [drm-dp] dp_ctrl_push_idle: PUSH_IDLE pattern timedout >>> >>> Other basic use-cases still seem to work fine hence add a >>> a module parameter to allow toggling psr enable/disable till >>> PSR related issues are hashed out with IGT. >> >> For the reference: Bjorn reported that he has issues with VT on a >> PSR-enabled laptops. This patch fixes the issue for him > > Module parameters are almost never warranted, and it is definitely not > the right way to handle a broken implementation. > > I've just sent a revert that unconditionally disables PSR support until > the implementation has been fixed: > > https://lore.kernel.org/lkml/20230523151646.28366-1-johan+linaro@kernel.org/ > > Johan I dont completely agree with this. Even the virtual terminal case was reported to be fixed by one user but not the other. So it was probably something missed out either in validation or reproduction steps of the user who reported it to be fixed OR the user who reported it not fixed. That needs to be investigated now. We should have ideally gone with the modparam with the feature patches itself knowing that it gets enabled for all sinks if PSR is supported. I had discussed with Rob that till we have some more confidence with the reported issues we would go with the modparam so as to not do the full revert. In this particular case, the one line revert is not really a deal breaker. In some other implementations, it might not really be so trivial to revert the feature with a one line change. So I would like to understand what is the concern with the mod param if the maintainers are onboard with it.
On Tue, May 23, 2023 at 12:23 PM Abhinav Kumar <quic_abhinavk@quicinc.com> wrote: > > > > On 5/23/2023 8:24 AM, Johan Hovold wrote: > > On Fri, May 12, 2023 at 09:13:04PM +0300, Dmitry Baryshkov wrote: > >> On 28/04/2023 02:28, Abhinav Kumar wrote: > >>> On sc7280 where eDP is the primary display, PSR is causing > >>> IGT breakage even for basic test cases like kms_atomic and > >>> kms_atomic_transition. Most often the issue starts with below > >>> stack so providing that as reference > >>> > >>> Call trace: > >>> dpu_encoder_assign_crtc+0x64/0x6c > >>> dpu_crtc_enable+0x188/0x204 > >>> drm_atomic_helper_commit_modeset_enables+0xc0/0x274 > >>> msm_atomic_commit_tail+0x1a8/0x68c > >>> commit_tail+0xb0/0x160 > >>> drm_atomic_helper_commit+0x11c/0x124 > >>> drm_atomic_commit+0xb0/0xdc > >>> drm_atomic_connector_commit_dpms+0xf4/0x110 > >>> drm_mode_obj_set_property_ioctl+0x16c/0x3b0 > >>> drm_connector_property_set_ioctl+0x4c/0x74 > >>> drm_ioctl_kernel+0xec/0x15c > >>> drm_ioctl+0x264/0x408 > >>> __arm64_sys_ioctl+0x9c/0xd4 > >>> invoke_syscall+0x4c/0x110 > >>> el0_svc_common+0x94/0xfc > >>> do_el0_svc+0x3c/0xb0 > >>> el0_svc+0x2c/0x7c > >>> el0t_64_sync_handler+0x48/0x114 > >>> el0t_64_sync+0x190/0x194 > >>> ---[ end trace 0000000000000000 ]--- > >>> [drm-dp] dp_ctrl_push_idle: PUSH_IDLE pattern timedout > >>> > >>> Other basic use-cases still seem to work fine hence add a > >>> a module parameter to allow toggling psr enable/disable till > >>> PSR related issues are hashed out with IGT. > >> > >> For the reference: Bjorn reported that he has issues with VT on a > >> PSR-enabled laptops. This patch fixes the issue for him > > > > Module parameters are almost never warranted, and it is definitely not > > the right way to handle a broken implementation. > > > > I've just sent a revert that unconditionally disables PSR support until > > the implementation has been fixed: > > > > https://lore.kernel.org/lkml/20230523151646.28366-1-johan+linaro@kernel.org/ > > > > Johan > > I dont completely agree with this. Even the virtual terminal case was > reported to be fixed by one user but not the other. So it was probably > something missed out either in validation or reproduction steps of the > user who reported it to be fixed OR the user who reported it not fixed. > That needs to be investigated now. > > We should have ideally gone with the modparam with the feature patches > itself knowing that it gets enabled for all sinks if PSR is supported. > > I had discussed with Rob that till we have some more confidence with the > reported issues we would go with the modparam so as to not do the full > revert. > > In this particular case, the one line revert is not really a deal > breaker. In some other implementations, it might not really be so > trivial to revert the feature with a one line change. > > So I would like to understand what is the concern with the mod param if > the maintainers are onboard with it. Tbf, I'd go further in the modparam direction, ie. add a default disabled modparam, but then _also_ enable the modparam in CI and add the failing tests to xfails. I'd rather have xfails in CI than skips. Acked-by: Rob Clark <robdclark@gmail.com> BR, -R
On Tue, May 23, 2023 at 12:23:04PM -0700, Abhinav Kumar wrote: > On 5/23/2023 8:24 AM, Johan Hovold wrote: > > On Fri, May 12, 2023 at 09:13:04PM +0300, Dmitry Baryshkov wrote: > >> On 28/04/2023 02:28, Abhinav Kumar wrote: > >>> On sc7280 where eDP is the primary display, PSR is causing > >>> IGT breakage even for basic test cases like kms_atomic and > >>> kms_atomic_transition. Most often the issue starts with below > >>> stack so providing that as reference > >>> > >>> Call trace: > >>> ---[ end trace 0000000000000000 ]--- > >>> [drm-dp] dp_ctrl_push_idle: PUSH_IDLE pattern timedout > >>> > >>> Other basic use-cases still seem to work fine hence add a > >>> a module parameter to allow toggling psr enable/disable till > >>> PSR related issues are hashed out with IGT. > >> > >> For the reference: Bjorn reported that he has issues with VT on a > >> PSR-enabled laptops. This patch fixes the issue for him > > > > Module parameters are almost never warranted, and it is definitely not > > the right way to handle a broken implementation. > > > > I've just sent a revert that unconditionally disables PSR support until > > the implementation has been fixed: > > > > https://lore.kernel.org/lkml/20230523151646.28366-1-johan+linaro@kernel.org/ > > I dont completely agree with this. Even the virtual terminal case was > reported to be fixed by one user but not the other. So it was probably > something missed out either in validation or reproduction steps of the > user who reported it to be fixed OR the user who reported it not fixed. > That needs to be investigated now. Yes, there may still be some time left to fix it, but it's pretty damn annoying to find that an issue reported two months ago still is not fixed at 6.4-rc3. (I even waited to make the switch to 6.4 so that I would not have to spend time on this.) I didn't see any mail from Bjorn saying that the series that claimed to fix the VT issue actually did fix the VT issue. There's only the comment above from Dmitry suggesting that disabling this feature is the only way to get a working terminal back. Regressions happen and sometimes there are corner cases that are harder to find, but this is a breakage of a fundamental feature that was reported before the code was even merged into mainline. > We should have ideally gone with the modparam with the feature patches > itself knowing that it gets enabled for all sinks if PSR is supported. Modparams are things of the past should not be used to enable broken features so that some vendor can tick of their internal lists of features that have been "mainlined". You can carry that single patch out-of-tree to enable this if you need it for some particular use case where you don't care about VTs. But hopefully you can just get this sorted quickly. If not, the revert I posted is the way to go rather than adding random module parameters. Johan
On 24/05/2023 09:59, Johan Hovold wrote: > On Tue, May 23, 2023 at 12:23:04PM -0700, Abhinav Kumar wrote: >> On 5/23/2023 8:24 AM, Johan Hovold wrote: >>> On Fri, May 12, 2023 at 09:13:04PM +0300, Dmitry Baryshkov wrote: >>>> On 28/04/2023 02:28, Abhinav Kumar wrote: >>>>> On sc7280 where eDP is the primary display, PSR is causing >>>>> IGT breakage even for basic test cases like kms_atomic and >>>>> kms_atomic_transition. Most often the issue starts with below >>>>> stack so providing that as reference >>>>> >>>>> Call trace: > >>>>> ---[ end trace 0000000000000000 ]--- >>>>> [drm-dp] dp_ctrl_push_idle: PUSH_IDLE pattern timedout >>>>> >>>>> Other basic use-cases still seem to work fine hence add a >>>>> a module parameter to allow toggling psr enable/disable till >>>>> PSR related issues are hashed out with IGT. >>>> >>>> For the reference: Bjorn reported that he has issues with VT on a >>>> PSR-enabled laptops. This patch fixes the issue for him >>> >>> Module parameters are almost never warranted, and it is definitely not >>> the right way to handle a broken implementation. >>> >>> I've just sent a revert that unconditionally disables PSR support until >>> the implementation has been fixed: >>> >>> https://lore.kernel.org/lkml/20230523151646.28366-1-johan+linaro@kernel.org/ >> >> I dont completely agree with this. Even the virtual terminal case was >> reported to be fixed by one user but not the other. So it was probably >> something missed out either in validation or reproduction steps of the >> user who reported it to be fixed OR the user who reported it not fixed. >> That needs to be investigated now. > > Yes, there may still be some time left to fix it, but it's pretty damn > annoying to find that an issue reported two months ago still is not > fixed at 6.4-rc3. (I even waited to make the switch to 6.4 so that I > would not have to spend time on this.) > > I didn't see any mail from Bjorn saying that the series that claimed to > fix the VT issue actually did fix the VT issue. There's only the comment > above from Dmitry suggesting that disabling this feature is the only way > to get a working terminal back. Originally this issue was reported by Doug, and at [1] he reported that an issue is fixed for him. So, for me it looks like we have hardware where VT works and hardware where it doesn't. Doug, can you please confirm whether you can reproduce the PSR+VT issue on 6.4-rc (without extra patches) or if the issue is fixed for you? [1] https://lore.kernel.org/dri-devel/CAD=FV=VSHmQPtsQfWjviEZeErms-VEOTmfozejASUC9zsMjAbA@mail.gmail.com/ > > Regressions happen and sometimes there are corner cases that are harder > to find, but this is a breakage of a fundamental feature that was > reported before the code was even merged into mainline. > >> We should have ideally gone with the modparam with the feature patches >> itself knowing that it gets enabled for all sinks if PSR is supported. > > Modparams are things of the past should not be used to enable broken > features so that some vendor can tick of their internal lists of > features that have been "mainlined". We have had a history of using modparam with i915 and IIRC amdgpu / radeon drivers to allow users to easily check whether new feature works for their hardware. My current understanding is that PSR+VT works for on some laptops and doesn't on some other laptops, which makes it a valid case. > > You can carry that single patch out-of-tree to enable this if you need > it for some particular use case where you don't care about VTs. > > But hopefully you can just get this sorted quickly. If not, the revert I > posted is the way to go rather than adding random module parameters. > > Johan
On Wed, May 24, 2023 at 11:06:03AM +0300, Dmitry Baryshkov wrote: > On 24/05/2023 09:59, Johan Hovold wrote: > > Regressions happen and sometimes there are corner cases that are harder > > to find, but this is a breakage of a fundamental feature that was > > reported before the code was even merged into mainline. > > > >> We should have ideally gone with the modparam with the feature patches > >> itself knowing that it gets enabled for all sinks if PSR is supported. > > > > Modparams are things of the past should not be used to enable broken > > features so that some vendor can tick of their internal lists of > > features that have been "mainlined". > > We have had a history of using modparam with i915 and IIRC amdgpu / > radeon drivers to allow users to easily check whether new feature works > for their hardware. My current understanding is that PSR+VT works for on > some laptops and doesn't on some other laptops, which makes it a valid case. But here it does not seem to be the hardware that's the issue, but rather that the implementation is incorrect or incomplete. Johan
diff --git a/drivers/gpu/drm/msm/dp/dp_display.c b/drivers/gpu/drm/msm/dp/dp_display.c index 628b0e248db6..dba43167de66 100644 --- a/drivers/gpu/drm/msm/dp/dp_display.c +++ b/drivers/gpu/drm/msm/dp/dp_display.c @@ -28,6 +28,10 @@ #include "dp_audio.h" #include "dp_debug.h" +static bool psr_enabled = false; +module_param(psr_enabled, bool, 0); +MODULE_PARM_DESC(psr_enabled, "enable PSR for eDP and DP displays"); + #define HPD_STRING_SIZE 30 enum { @@ -407,7 +411,7 @@ static int dp_display_process_hpd_high(struct dp_display_private *dp) edid = dp->panel->edid; - dp->dp_display.psr_supported = dp->panel->psr_cap.version; + dp->dp_display.psr_supported = dp->panel->psr_cap.version && psr_enabled; dp->audio_supported = drm_detect_monitor_audio(edid); dp_panel_handle_sink_request(dp->panel);
On sc7280 where eDP is the primary display, PSR is causing IGT breakage even for basic test cases like kms_atomic and kms_atomic_transition. Most often the issue starts with below stack so providing that as reference Call trace: dpu_encoder_assign_crtc+0x64/0x6c dpu_crtc_enable+0x188/0x204 drm_atomic_helper_commit_modeset_enables+0xc0/0x274 msm_atomic_commit_tail+0x1a8/0x68c commit_tail+0xb0/0x160 drm_atomic_helper_commit+0x11c/0x124 drm_atomic_commit+0xb0/0xdc drm_atomic_connector_commit_dpms+0xf4/0x110 drm_mode_obj_set_property_ioctl+0x16c/0x3b0 drm_connector_property_set_ioctl+0x4c/0x74 drm_ioctl_kernel+0xec/0x15c drm_ioctl+0x264/0x408 __arm64_sys_ioctl+0x9c/0xd4 invoke_syscall+0x4c/0x110 el0_svc_common+0x94/0xfc do_el0_svc+0x3c/0xb0 el0_svc+0x2c/0x7c el0t_64_sync_handler+0x48/0x114 el0t_64_sync+0x190/0x194 ---[ end trace 0000000000000000 ]--- [drm-dp] dp_ctrl_push_idle: PUSH_IDLE pattern timedout Other basic use-cases still seem to work fine hence add a a module parameter to allow toggling psr enable/disable till PSR related issues are hashed out with IGT. Signed-off-by: Abhinav Kumar <quic_abhinavk@quicinc.com> --- drivers/gpu/drm/msm/dp/dp_display.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)