| Message ID | 20230425070330.8520-1-jammy_huang@aspeedtech.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | drm/ast: Fix modeset failed on DisplayPort | expand |
Hi Am 25.04.23 um 09:03 schrieb Jammy Huang: > If we switch display and update cursor together, it could lead to > modeset failed because of concurrent access to IO registers. > > Add lock protection in DP's edid access to avoid this problem. Thanks for the patch. I thought I fixed this issue already, but that apparently only happened for SIL164 and VGA. What about ast_dp501_connector_helper_get_modes()? Does it require the locking as well? > > Signed-off-by: Jammy Huang <jammy_huang@aspeedtech.com> > --- > drivers/gpu/drm/ast/ast_mode.c | 11 +++++++++++ > 1 file changed, 11 insertions(+) > > diff --git a/drivers/gpu/drm/ast/ast_mode.c b/drivers/gpu/drm/ast/ast_mode.c > index 984ec590a7e7..fe5f1fd61361 100644 > --- a/drivers/gpu/drm/ast/ast_mode.c > +++ b/drivers/gpu/drm/ast/ast_mode.c > @@ -1635,6 +1635,8 @@ static int ast_dp501_output_init(struct ast_private *ast) > static int ast_astdp_connector_helper_get_modes(struct drm_connector *connector) > { > void *edid; > + struct drm_device *dev = connector->dev; > + struct ast_private *ast = to_ast_private(dev); We've meanwhile renamed ast_private to ast_device. Could you please provide an updated patch for the drm-misc-next tree? Best regards Thomas > > int succ; > int count; > @@ -1643,10 +1645,18 @@ static int ast_astdp_connector_helper_get_modes(struct drm_connector *connector) > if (!edid) > goto err_drm_connector_update_edid_property; > > + /* > + * Protect access to I/O registers from concurrent modesetting > + * by acquiring the I/O-register lock. > + */ > + mutex_lock(&ast->ioregs_lock); > + > succ = ast_astdp_read_edid(connector->dev, edid); > if (succ < 0) > goto err_kfree; > > + mutex_unlock(&ast->ioregs_lock); > + > drm_connector_update_edid_property(connector, edid); > count = drm_add_edid_modes(connector, edid); > kfree(edid); > @@ -1654,6 +1664,7 @@ static int ast_astdp_connector_helper_get_modes(struct drm_connector *connector) > return count; > > err_kfree: > + mutex_unlock(&ast->ioregs_lock); > kfree(edid); > err_drm_connector_update_edid_property: > drm_connector_update_edid_property(connector, NULL); > > base-commit: 61d325dcbc05d8fef88110d35ef7776f3ac3f68b
Hi Thomas, I think DP501 is OK. It doesn't use ioregs in ast_dp501_read_edid(). On 2023/4/25 下午 03:27, Thomas Zimmermann wrote: > Hi > > Am 25.04.23 um 09:03 schrieb Jammy Huang: >> If we switch display and update cursor together, it could lead to >> modeset failed because of concurrent access to IO registers. >> >> Add lock protection in DP's edid access to avoid this problem. > > Thanks for the patch. I thought I fixed this issue already, but that > apparently only happened for SIL164 and VGA. > > What about ast_dp501_connector_helper_get_modes()? Does it require the > locking as well? > >> >> Signed-off-by: Jammy Huang <jammy_huang@aspeedtech.com> >> --- >> drivers/gpu/drm/ast/ast_mode.c | 11 +++++++++++ >> 1 file changed, 11 insertions(+) >> >> diff --git a/drivers/gpu/drm/ast/ast_mode.c >> b/drivers/gpu/drm/ast/ast_mode.c >> index 984ec590a7e7..fe5f1fd61361 100644 >> --- a/drivers/gpu/drm/ast/ast_mode.c >> +++ b/drivers/gpu/drm/ast/ast_mode.c >> @@ -1635,6 +1635,8 @@ static int ast_dp501_output_init(struct >> ast_private *ast) >> static int ast_astdp_connector_helper_get_modes(struct >> drm_connector *connector) >> { >> void *edid; >> + struct drm_device *dev = connector->dev; >> + struct ast_private *ast = to_ast_private(dev); > > We've meanwhile renamed ast_private to ast_device. Could you please > provide an updated patch for the drm-misc-next tree? > > Best regards > Thomas > >> int succ; >> int count; >> @@ -1643,10 +1645,18 @@ static int >> ast_astdp_connector_helper_get_modes(struct drm_connector *connector) >> if (!edid) >> goto err_drm_connector_update_edid_property; >> + /* >> + * Protect access to I/O registers from concurrent modesetting >> + * by acquiring the I/O-register lock. >> + */ >> + mutex_lock(&ast->ioregs_lock); >> + >> succ = ast_astdp_read_edid(connector->dev, edid); >> if (succ < 0) >> goto err_kfree; >> + mutex_unlock(&ast->ioregs_lock); >> + >> drm_connector_update_edid_property(connector, edid); >> count = drm_add_edid_modes(connector, edid); >> kfree(edid); >> @@ -1654,6 +1664,7 @@ static int >> ast_astdp_connector_helper_get_modes(struct drm_connector *connector) >> return count; >> err_kfree: >> + mutex_unlock(&ast->ioregs_lock); >> kfree(edid); >> err_drm_connector_update_edid_property: >> drm_connector_update_edid_property(connector, NULL); >> >> base-commit: 61d325dcbc05d8fef88110d35ef7776f3ac3f68b >
Hi Thomas, Do you have other suggestion for this patch?? Please kindly advise. On 2023/4/25 下午 03:39, Jammy Huang wrote: > Hi Thomas, > > I think DP501 is OK. It doesn't use ioregs in ast_dp501_read_edid(). > > On 2023/4/25 下午 03:27, Thomas Zimmermann wrote: >> Hi >> >> Am 25.04.23 um 09:03 schrieb Jammy Huang: >>> If we switch display and update cursor together, it could lead to >>> modeset failed because of concurrent access to IO registers. >>> >>> Add lock protection in DP's edid access to avoid this problem. >> >> Thanks for the patch. I thought I fixed this issue already, but that >> apparently only happened for SIL164 and VGA. >> >> What about ast_dp501_connector_helper_get_modes()? Does it require >> the locking as well? >> >>> >>> Signed-off-by: Jammy Huang <jammy_huang@aspeedtech.com> >>> --- >>> drivers/gpu/drm/ast/ast_mode.c | 11 +++++++++++ >>> 1 file changed, 11 insertions(+) >>> >>> diff --git a/drivers/gpu/drm/ast/ast_mode.c >>> b/drivers/gpu/drm/ast/ast_mode.c >>> index 984ec590a7e7..fe5f1fd61361 100644 >>> --- a/drivers/gpu/drm/ast/ast_mode.c >>> +++ b/drivers/gpu/drm/ast/ast_mode.c >>> @@ -1635,6 +1635,8 @@ static int ast_dp501_output_init(struct >>> ast_private *ast) >>> static int ast_astdp_connector_helper_get_modes(struct >>> drm_connector *connector) >>> { >>> void *edid; >>> + struct drm_device *dev = connector->dev; >>> + struct ast_private *ast = to_ast_private(dev); >> >> We've meanwhile renamed ast_private to ast_device. Could you please >> provide an updated patch for the drm-misc-next tree? >> >> Best regards >> Thomas >> >>> int succ; >>> int count; >>> @@ -1643,10 +1645,18 @@ static int >>> ast_astdp_connector_helper_get_modes(struct drm_connector *connector) >>> if (!edid) >>> goto err_drm_connector_update_edid_property; >>> + /* >>> + * Protect access to I/O registers from concurrent modesetting >>> + * by acquiring the I/O-register lock. >>> + */ >>> + mutex_lock(&ast->ioregs_lock); >>> + >>> succ = ast_astdp_read_edid(connector->dev, edid); >>> if (succ < 0) >>> goto err_kfree; >>> + mutex_unlock(&ast->ioregs_lock); >>> + >>> drm_connector_update_edid_property(connector, edid); >>> count = drm_add_edid_modes(connector, edid); >>> kfree(edid); >>> @@ -1654,6 +1664,7 @@ static int >>> ast_astdp_connector_helper_get_modes(struct drm_connector *connector) >>> return count; >>> err_kfree: >>> + mutex_unlock(&ast->ioregs_lock); >>> kfree(edid); >>> err_drm_connector_update_edid_property: >>> drm_connector_update_edid_property(connector, NULL); >>> >>> base-commit: 61d325dcbc05d8fef88110d35ef7776f3ac3f68b >>
Hi, this patch also fell through the cracks. Apologies. Am 24.05.23 um 04:29 schrieb Jammy Huang: > Hi Thomas, > > Do you have other suggestion for this patch?? The main issue was that struct ast_private is now called struct ast_device. So the current patch cannot be applied. > > Please kindly advise. > > On 2023/4/25 下午 03:39, Jammy Huang wrote: >> Hi Thomas, >> >> I think DP501 is OK. It doesn't use ioregs in ast_dp501_read_edid(). >> >> On 2023/4/25 下午 03:27, Thomas Zimmermann wrote: >>> Hi >>> >>> Am 25.04.23 um 09:03 schrieb Jammy Huang: >>>> If we switch display and update cursor together, it could lead to >>>> modeset failed because of concurrent access to IO registers. >>>> >>>> Add lock protection in DP's edid access to avoid this problem. >>> >>> Thanks for the patch. I thought I fixed this issue already, but that >>> apparently only happened for SIL164 and VGA. >>> >>> What about ast_dp501_connector_helper_get_modes()? Does it require >>> the locking as well? >>> >>>> >>>> Signed-off-by: Jammy Huang <jammy_huang@aspeedtech.com> >>>> --- >>>> drivers/gpu/drm/ast/ast_mode.c | 11 +++++++++++ >>>> 1 file changed, 11 insertions(+) >>>> >>>> diff --git a/drivers/gpu/drm/ast/ast_mode.c >>>> b/drivers/gpu/drm/ast/ast_mode.c >>>> index 984ec590a7e7..fe5f1fd61361 100644 >>>> --- a/drivers/gpu/drm/ast/ast_mode.c >>>> +++ b/drivers/gpu/drm/ast/ast_mode.c >>>> @@ -1635,6 +1635,8 @@ static int ast_dp501_output_init(struct >>>> ast_private *ast) >>>> static int ast_astdp_connector_helper_get_modes(struct >>>> drm_connector *connector) >>>> { >>>> void *edid; >>>> + struct drm_device *dev = connector->dev; >>>> + struct ast_private *ast = to_ast_private(dev); >>> >>> We've meanwhile renamed ast_private to ast_device. Could you please >>> provide an updated patch for the drm-misc-next tree? >>> >>> Best regards >>> Thomas >>> >>>> int succ; >>>> int count; >>>> @@ -1643,10 +1645,18 @@ static int >>>> ast_astdp_connector_helper_get_modes(struct drm_connector *connector) >>>> if (!edid) >>>> goto err_drm_connector_update_edid_property; >>>> + /* >>>> + * Protect access to I/O registers from concurrent modesetting >>>> + * by acquiring the I/O-register lock. >>>> + */ >>>> + mutex_lock(&ast->ioregs_lock); >>>> + >>>> succ = ast_astdp_read_edid(connector->dev, edid); >>>> if (succ < 0) >>>> goto err_kfree; >>>> + mutex_unlock(&ast->ioregs_lock); >>>> + >>>> drm_connector_update_edid_property(connector, edid); >>>> count = drm_add_edid_modes(connector, edid); >>>> kfree(edid); >>>> @@ -1654,6 +1664,7 @@ static int >>>> ast_astdp_connector_helper_get_modes(struct drm_connector *connector) >>>> return count; >>>> err_kfree: Here's a minor issue that the goto label should now be called err_mutex_unlock. Best regards Thomas >>>> + mutex_unlock(&ast->ioregs_lock); >>>> kfree(edid); >>>> err_drm_connector_update_edid_property: >>>> drm_connector_update_edid_property(connector, NULL); >>>> >>>> base-commit: 61d325dcbc05d8fef88110d35ef7776f3ac3f68b >>>
Hi Thomas, Thanks, I will modify the patch accordingly. On 2023/5/24 下午 06:47, Thomas Zimmermann wrote: > Hi, > > this patch also fell through the cracks. Apologies. > > Am 24.05.23 um 04:29 schrieb Jammy Huang: >> Hi Thomas, >> >> Do you have other suggestion for this patch?? > > The main issue was that struct ast_private is now called struct > ast_device. So the current patch cannot be applied. > >> >> Please kindly advise. >> >> On 2023/4/25 下午 03:39, Jammy Huang wrote: >>> Hi Thomas, >>> >>> I think DP501 is OK. It doesn't use ioregs in ast_dp501_read_edid(). >>> >>> On 2023/4/25 下午 03:27, Thomas Zimmermann wrote: >>>> Hi >>>> >>>> Am 25.04.23 um 09:03 schrieb Jammy Huang: >>>>> If we switch display and update cursor together, it could lead to >>>>> modeset failed because of concurrent access to IO registers. >>>>> >>>>> Add lock protection in DP's edid access to avoid this problem. >>>> >>>> Thanks for the patch. I thought I fixed this issue already, but >>>> that apparently only happened for SIL164 and VGA. >>>> >>>> What about ast_dp501_connector_helper_get_modes()? Does it require >>>> the locking as well? >>>> >>>>> >>>>> Signed-off-by: Jammy Huang <jammy_huang@aspeedtech.com> >>>>> --- >>>>> drivers/gpu/drm/ast/ast_mode.c | 11 +++++++++++ >>>>> 1 file changed, 11 insertions(+) >>>>> >>>>> diff --git a/drivers/gpu/drm/ast/ast_mode.c >>>>> b/drivers/gpu/drm/ast/ast_mode.c >>>>> index 984ec590a7e7..fe5f1fd61361 100644 >>>>> --- a/drivers/gpu/drm/ast/ast_mode.c >>>>> +++ b/drivers/gpu/drm/ast/ast_mode.c >>>>> @@ -1635,6 +1635,8 @@ static int ast_dp501_output_init(struct >>>>> ast_private *ast) >>>>> static int ast_astdp_connector_helper_get_modes(struct >>>>> drm_connector *connector) >>>>> { >>>>> void *edid; >>>>> + struct drm_device *dev = connector->dev; >>>>> + struct ast_private *ast = to_ast_private(dev); >>>> >>>> We've meanwhile renamed ast_private to ast_device. Could you please >>>> provide an updated patch for the drm-misc-next tree? >>>> >>>> Best regards >>>> Thomas >>>> >>>>> int succ; >>>>> int count; >>>>> @@ -1643,10 +1645,18 @@ static int >>>>> ast_astdp_connector_helper_get_modes(struct drm_connector *connector) >>>>> if (!edid) >>>>> goto err_drm_connector_update_edid_property; >>>>> + /* >>>>> + * Protect access to I/O registers from concurrent modesetting >>>>> + * by acquiring the I/O-register lock. >>>>> + */ >>>>> + mutex_lock(&ast->ioregs_lock); >>>>> + >>>>> succ = ast_astdp_read_edid(connector->dev, edid); >>>>> if (succ < 0) >>>>> goto err_kfree; >>>>> + mutex_unlock(&ast->ioregs_lock); >>>>> + >>>>> drm_connector_update_edid_property(connector, edid); >>>>> count = drm_add_edid_modes(connector, edid); >>>>> kfree(edid); >>>>> @@ -1654,6 +1664,7 @@ static int >>>>> ast_astdp_connector_helper_get_modes(struct drm_connector *connector) >>>>> return count; >>>>> err_kfree: > > Here's a minor issue that the goto label should now be called > err_mutex_unlock. > > Best regards > Thomas > >>>>> + mutex_unlock(&ast->ioregs_lock); >>>>> kfree(edid); >>>>> err_drm_connector_update_edid_property: >>>>> drm_connector_update_edid_property(connector, NULL); >>>>> >>>>> base-commit: 61d325dcbc05d8fef88110d35ef7776f3ac3f68b >>>> >
diff --git a/drivers/gpu/drm/ast/ast_mode.c b/drivers/gpu/drm/ast/ast_mode.c index 984ec590a7e7..fe5f1fd61361 100644 --- a/drivers/gpu/drm/ast/ast_mode.c +++ b/drivers/gpu/drm/ast/ast_mode.c @@ -1635,6 +1635,8 @@ static int ast_dp501_output_init(struct ast_private *ast) static int ast_astdp_connector_helper_get_modes(struct drm_connector *connector) { void *edid; + struct drm_device *dev = connector->dev; + struct ast_private *ast = to_ast_private(dev); int succ; int count; @@ -1643,10 +1645,18 @@ static int ast_astdp_connector_helper_get_modes(struct drm_connector *connector) if (!edid) goto err_drm_connector_update_edid_property; + /* + * Protect access to I/O registers from concurrent modesetting + * by acquiring the I/O-register lock. + */ + mutex_lock(&ast->ioregs_lock); + succ = ast_astdp_read_edid(connector->dev, edid); if (succ < 0) goto err_kfree; + mutex_unlock(&ast->ioregs_lock); + drm_connector_update_edid_property(connector, edid); count = drm_add_edid_modes(connector, edid); kfree(edid); @@ -1654,6 +1664,7 @@ static int ast_astdp_connector_helper_get_modes(struct drm_connector *connector) return count; err_kfree: + mutex_unlock(&ast->ioregs_lock); kfree(edid); err_drm_connector_update_edid_property: drm_connector_update_edid_property(connector, NULL);
If we switch display and update cursor together, it could lead to modeset failed because of concurrent access to IO registers. Add lock protection in DP's edid access to avoid this problem. Signed-off-by: Jammy Huang <jammy_huang@aspeedtech.com> --- drivers/gpu/drm/ast/ast_mode.c | 11 +++++++++++ 1 file changed, 11 insertions(+) base-commit: 61d325dcbc05d8fef88110d35ef7776f3ac3f68b