| Message ID | 20191110184055.3979-1-hdegoede@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | drm: Add DRM_MODE_TYPE_USERDEF flag to probed modes matching a video= argument | expand |
On Sun, Nov 10, 2019 at 7:41 PM Hans de Goede <hdegoede@redhat.com> wrote: > > drm_helper_probe_add_cmdline_mode() prefers using a probed mode matching > a video= argument over calculating our own timings for the user specified > mode using CVT or GTF. > > But userspace code which is auto-configuring the mode may want to know that > the user has specified that mode on the kernel commandline so that it can > pick that mode over the mode which is marked as DRM_MODE_TYPE_PREFERRED. > > This commit sets the DRM_MODE_TYPE_USERDEF flag on the matching mode, just > as we would do on the user-specified mode when no matching probed mode is > found. > > Signed-off-by: Hans de Goede <hdegoede@redhat.com> Will existing userspace dtrt here with this? Some links to popular ones would be good (since essentially this is uapi fine tuning we need that anyway). With that will get my ack. -Danile > --- > drivers/gpu/drm/drm_probe_helper.c | 2 ++ > include/drm/drm_modes.h | 3 ++- > 2 files changed, 4 insertions(+), 1 deletion(-) > > diff --git a/drivers/gpu/drm/drm_probe_helper.c b/drivers/gpu/drm/drm_probe_helper.c > index ef2c468205a2..4fed64be11f9 100644 > --- a/drivers/gpu/drm/drm_probe_helper.c > +++ b/drivers/gpu/drm/drm_probe_helper.c > @@ -157,6 +157,8 @@ static int drm_helper_probe_add_cmdline_mode(struct drm_connector *connector) > continue; > } > > + /* Mark the matching mode as being preferred by the user */ > + mode->type |= DRM_MODE_TYPE_USERDEF; > return 0; > } > > diff --git a/include/drm/drm_modes.h b/include/drm/drm_modes.h > index e946e20c61d8..c7efb7487e9b 100644 > --- a/include/drm/drm_modes.h > +++ b/include/drm/drm_modes.h > @@ -256,7 +256,8 @@ struct drm_display_mode { > * - DRM_MODE_TYPE_DRIVER: Mode created by the driver, which is all of > * them really. Drivers must set this bit for all modes they create > * and expose to userspace. > - * - DRM_MODE_TYPE_USERDEF: Mode defined via kernel command line > + * - DRM_MODE_TYPE_USERDEF: Mode defined or selected via the kernel > + * command line. > * > * Plus a big list of flags which shouldn't be used at all, but are > * still around since these flags are also used in the userspace ABI. > -- > 2.23.0 > > _______________________________________________ > dri-devel mailing list > dri-devel@lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/dri-devel
Hi, On 11-11-2019 10:25, Daniel Vetter wrote: > On Sun, Nov 10, 2019 at 7:41 PM Hans de Goede <hdegoede@redhat.com> wrote: >> >> drm_helper_probe_add_cmdline_mode() prefers using a probed mode matching >> a video= argument over calculating our own timings for the user specified >> mode using CVT or GTF. >> >> But userspace code which is auto-configuring the mode may want to know that >> the user has specified that mode on the kernel commandline so that it can >> pick that mode over the mode which is marked as DRM_MODE_TYPE_PREFERRED. >> >> This commit sets the DRM_MODE_TYPE_USERDEF flag on the matching mode, just >> as we would do on the user-specified mode when no matching probed mode is >> found. >> >> Signed-off-by: Hans de Goede <hdegoede@redhat.com> > > Will existing userspace dtrt here with this? Some links to popular > ones would be good (since essentially this is uapi fine tuning we need > that anyway). With that will get my ack. A valid question, I've gone over what I consider the major userspace kms users: -Xorg xserver modesetting driver does not check for this: [hans@shalem xserver]$ ack DRM_MODE_TYPE_ hw/xfree86/drivers/modesetting/ hw/xfree86/drivers/modesetting/drmmode_display.c 1321: if (kmode->type & DRM_MODE_TYPE_DRIVER) 1323: if (kmode->type & DRM_MODE_TYPE_PREFERRED) -Other Xorg drivers: [hans@shalem driver]$ ls -d xf86-video-* xf86-video-amdgpu xf86-video-intel xf86-video-qxl xf86-video-armsoc xf86-video-modesetting xf86-video-sisusb xf86-video-ati xf86-video-nouveau xf86-video-vmware xf86-video-dummy xf86-video-omap xf86-video-voodoo xf86-video-geode xf86-video-opentegra These all only do the same checks as the Xorg modesetting driver -mutter: [hans@shalem mutter]$ ack DRM_MODE_TYPE_ src/backends/native/meta-output-kms.c 261: if (drm_mode->type & DRM_MODE_TYPE_PREFERRED) So it seems nothing (that I can quickly find) in userspace is using this atm. The reason I wrote this patch is because about a year ago plymouth used to fully rely on the kernel to setup the modes on monitors and would simply inherit the modes setup by the kernel. Basically plymouth was relying on fbcon to load first and setup modes. Deferred fbcon takeover (for flickerfree) means that this is no longer happening. So now plymouth picks a mode itself. When I submitted the plymouth change for plymouth to pick a mode itself the plymouth maintainer (Ray Strode) was afraid that would break plymouth honoring video= arguments. So currently plymouth still relies on the kernel to do the mode setup if a video= argument is present on the kernel commandline. My other recent series, which adds support for e.g. video=HDMI:panel_orientation=right_side_up, made me realize that relying on the kernel for this is no good since the fbcon code has various limitations wrt e.g. hotplug and this of course will not work when fbcon deferred takeover is used and fbcon never loads before plymouth. So I wrote a patch for plymouth to check the DRM_MODE_TYPE_USERDEF flag and prefer a mode with that flag over the PREFERRED mode: https://gitlab.freedesktop.org/plymouth/plymouth/merge_requests/84 And then I found out that for that code to work with modes which are already in the list of probed modes, I need something like the kernel patch we are discussing now. So if you want a link to an userspace consumer of this, I guess you want a v2 with a link to the plymouth MR added. + maybe a blurb in the commit message that to the best of my knowledge no userspace kms consumers are checking for the DRM_MODE_TYPE_USERDEF flag? Regards, Hans > -Danile > >> --- >> drivers/gpu/drm/drm_probe_helper.c | 2 ++ >> include/drm/drm_modes.h | 3 ++- >> 2 files changed, 4 insertions(+), 1 deletion(-) >> >> diff --git a/drivers/gpu/drm/drm_probe_helper.c b/drivers/gpu/drm/drm_probe_helper.c >> index ef2c468205a2..4fed64be11f9 100644 >> --- a/drivers/gpu/drm/drm_probe_helper.c >> +++ b/drivers/gpu/drm/drm_probe_helper.c >> @@ -157,6 +157,8 @@ static int drm_helper_probe_add_cmdline_mode(struct drm_connector *connector) >> continue; >> } >> >> + /* Mark the matching mode as being preferred by the user */ >> + mode->type |= DRM_MODE_TYPE_USERDEF; >> return 0; >> } >> >> diff --git a/include/drm/drm_modes.h b/include/drm/drm_modes.h >> index e946e20c61d8..c7efb7487e9b 100644 >> --- a/include/drm/drm_modes.h >> +++ b/include/drm/drm_modes.h >> @@ -256,7 +256,8 @@ struct drm_display_mode { >> * - DRM_MODE_TYPE_DRIVER: Mode created by the driver, which is all of >> * them really. Drivers must set this bit for all modes they create >> * and expose to userspace. >> - * - DRM_MODE_TYPE_USERDEF: Mode defined via kernel command line >> + * - DRM_MODE_TYPE_USERDEF: Mode defined or selected via the kernel >> + * command line. >> * >> * Plus a big list of flags which shouldn't be used at all, but are >> * still around since these flags are also used in the userspace ABI. >> -- >> 2.23.0 >> >> _______________________________________________ >> dri-devel mailing list >> dri-devel@lists.freedesktop.org >> https://lists.freedesktop.org/mailman/listinfo/dri-devel > > >
On Mon, Nov 11, 2019 at 10:50 AM Hans de Goede <hdegoede@redhat.com> wrote: > > Hi, > > On 11-11-2019 10:25, Daniel Vetter wrote: > > On Sun, Nov 10, 2019 at 7:41 PM Hans de Goede <hdegoede@redhat.com> wrote: > >> > >> drm_helper_probe_add_cmdline_mode() prefers using a probed mode matching > >> a video= argument over calculating our own timings for the user specified > >> mode using CVT or GTF. > >> > >> But userspace code which is auto-configuring the mode may want to know that > >> the user has specified that mode on the kernel commandline so that it can > >> pick that mode over the mode which is marked as DRM_MODE_TYPE_PREFERRED. > >> > >> This commit sets the DRM_MODE_TYPE_USERDEF flag on the matching mode, just > >> as we would do on the user-specified mode when no matching probed mode is > >> found. > >> > >> Signed-off-by: Hans de Goede <hdegoede@redhat.com> > > > > Will existing userspace dtrt here with this? Some links to popular > > ones would be good (since essentially this is uapi fine tuning we need > > that anyway). With that will get my ack. > > A valid question, I've gone over what I consider the major userspace kms users: > -Xorg xserver modesetting driver does not check for this: > [hans@shalem xserver]$ ack DRM_MODE_TYPE_ hw/xfree86/drivers/modesetting/ > hw/xfree86/drivers/modesetting/drmmode_display.c > 1321: if (kmode->type & DRM_MODE_TYPE_DRIVER) > 1323: if (kmode->type & DRM_MODE_TYPE_PREFERRED) > -Other Xorg drivers: > [hans@shalem driver]$ ls -d xf86-video-* > xf86-video-amdgpu xf86-video-intel xf86-video-qxl > xf86-video-armsoc xf86-video-modesetting xf86-video-sisusb > xf86-video-ati xf86-video-nouveau xf86-video-vmware > xf86-video-dummy xf86-video-omap xf86-video-voodoo > xf86-video-geode xf86-video-opentegra > These all only do the same checks as the Xorg modesetting driver > -mutter: > [hans@shalem mutter]$ ack DRM_MODE_TYPE_ > src/backends/native/meta-output-kms.c > 261: if (drm_mode->type & DRM_MODE_TYPE_PREFERRED) > > So it seems nothing (that I can quickly find) in userspace is using this atm. > > The reason I wrote this patch is because about a year ago plymouth used to > fully rely on the kernel to setup the modes on monitors and would simply > inherit the modes setup by the kernel. Basically plymouth was relying on > fbcon to load first and setup modes. > > Deferred fbcon takeover (for flickerfree) means that this is no longer > happening. So now plymouth picks a mode itself. When I submitted the > plymouth change for plymouth to pick a mode itself the plymouth maintainer > (Ray Strode) was afraid that would break plymouth honoring video= arguments. > So currently plymouth still relies on the kernel to do the mode setup if > a video= argument is present on the kernel commandline. > > My other recent series, which adds support for e.g. > video=HDMI:panel_orientation=right_side_up, made me realize that relying > on the kernel for this is no good since the fbcon code has various > limitations wrt e.g. hotplug and this of course will not work when > fbcon deferred takeover is used and fbcon never loads before plymouth. > > So I wrote a patch for plymouth to check the DRM_MODE_TYPE_USERDEF > flag and prefer a mode with that flag over the PREFERRED mode: > https://gitlab.freedesktop.org/plymouth/plymouth/merge_requests/84 > > And then I found out that for that code to work with modes which > are already in the list of probed modes, I need something like > the kernel patch we are discussing now. > > So if you want a link to an userspace consumer of this, I guess you > want a v2 with a link to the plymouth MR added. + maybe a blurb in > the commit message that to the best of my knowledge no userspace > kms consumers are checking for the DRM_MODE_TYPE_USERDEF flag? And how does just relying on plymouth fix this? Given what you figured out in your quick search I think we should set both USERDEF and PREFERRED, otherwise not much will happen here. Plus I guess we need some fixup code to make sure that only the cmdline mode is preferred, and no other mode? Except if all compositors pick the first preferred mode consistently, then I guess we just need to make sure it's first. Changing the meaning of USERDEF to also mean "preferred mode, pls use that" could be fraught with peril, e.g. you run a game on an old compositor (or with direct display) at lower refresh, maybe with a custom mode, then it crashes. And from then on all compositors insist on using your "preferred" mode. -Daniel > > Regards, > > Hans > > > > > > > > -Danile > > > >> --- > >> drivers/gpu/drm/drm_probe_helper.c | 2 ++ > >> include/drm/drm_modes.h | 3 ++- > >> 2 files changed, 4 insertions(+), 1 deletion(-) > >> > >> diff --git a/drivers/gpu/drm/drm_probe_helper.c b/drivers/gpu/drm/drm_probe_helper.c > >> index ef2c468205a2..4fed64be11f9 100644 > >> --- a/drivers/gpu/drm/drm_probe_helper.c > >> +++ b/drivers/gpu/drm/drm_probe_helper.c > >> @@ -157,6 +157,8 @@ static int drm_helper_probe_add_cmdline_mode(struct drm_connector *connector) > >> continue; > >> } > >> > >> + /* Mark the matching mode as being preferred by the user */ > >> + mode->type |= DRM_MODE_TYPE_USERDEF; > >> return 0; > >> } > >> > >> diff --git a/include/drm/drm_modes.h b/include/drm/drm_modes.h > >> index e946e20c61d8..c7efb7487e9b 100644 > >> --- a/include/drm/drm_modes.h > >> +++ b/include/drm/drm_modes.h > >> @@ -256,7 +256,8 @@ struct drm_display_mode { > >> * - DRM_MODE_TYPE_DRIVER: Mode created by the driver, which is all of > >> * them really. Drivers must set this bit for all modes they create > >> * and expose to userspace. > >> - * - DRM_MODE_TYPE_USERDEF: Mode defined via kernel command line > >> + * - DRM_MODE_TYPE_USERDEF: Mode defined or selected via the kernel > >> + * command line. > >> * > >> * Plus a big list of flags which shouldn't be used at all, but are > >> * still around since these flags are also used in the userspace ABI. > >> -- > >> 2.23.0 > >> > >> _______________________________________________ > >> dri-devel mailing list > >> dri-devel@lists.freedesktop.org > >> https://lists.freedesktop.org/mailman/listinfo/dri-devel > > > > > > >
Hi Daniel, On 11-11-2019 11:26, Daniel Vetter wrote: > On Mon, Nov 11, 2019 at 10:50 AM Hans de Goede <hdegoede@redhat.com> wrote: >> >> Hi, >> >> On 11-11-2019 10:25, Daniel Vetter wrote: >>> On Sun, Nov 10, 2019 at 7:41 PM Hans de Goede <hdegoede@redhat.com> wrote: >>>> >>>> drm_helper_probe_add_cmdline_mode() prefers using a probed mode matching >>>> a video= argument over calculating our own timings for the user specified >>>> mode using CVT or GTF. >>>> >>>> But userspace code which is auto-configuring the mode may want to know that >>>> the user has specified that mode on the kernel commandline so that it can >>>> pick that mode over the mode which is marked as DRM_MODE_TYPE_PREFERRED. >>>> >>>> This commit sets the DRM_MODE_TYPE_USERDEF flag on the matching mode, just >>>> as we would do on the user-specified mode when no matching probed mode is >>>> found. >>>> >>>> Signed-off-by: Hans de Goede <hdegoede@redhat.com> >>> >>> Will existing userspace dtrt here with this? Some links to popular >>> ones would be good (since essentially this is uapi fine tuning we need >>> that anyway). With that will get my ack. >> >> A valid question, I've gone over what I consider the major userspace kms users: >> -Xorg xserver modesetting driver does not check for this: >> [hans@shalem xserver]$ ack DRM_MODE_TYPE_ hw/xfree86/drivers/modesetting/ >> hw/xfree86/drivers/modesetting/drmmode_display.c >> 1321: if (kmode->type & DRM_MODE_TYPE_DRIVER) >> 1323: if (kmode->type & DRM_MODE_TYPE_PREFERRED) >> -Other Xorg drivers: >> [hans@shalem driver]$ ls -d xf86-video-* >> xf86-video-amdgpu xf86-video-intel xf86-video-qxl >> xf86-video-armsoc xf86-video-modesetting xf86-video-sisusb >> xf86-video-ati xf86-video-nouveau xf86-video-vmware >> xf86-video-dummy xf86-video-omap xf86-video-voodoo >> xf86-video-geode xf86-video-opentegra >> These all only do the same checks as the Xorg modesetting driver >> -mutter: >> [hans@shalem mutter]$ ack DRM_MODE_TYPE_ >> src/backends/native/meta-output-kms.c >> 261: if (drm_mode->type & DRM_MODE_TYPE_PREFERRED) >> >> So it seems nothing (that I can quickly find) in userspace is using this atm. >> >> The reason I wrote this patch is because about a year ago plymouth used to >> fully rely on the kernel to setup the modes on monitors and would simply >> inherit the modes setup by the kernel. Basically plymouth was relying on >> fbcon to load first and setup modes. >> >> Deferred fbcon takeover (for flickerfree) means that this is no longer >> happening. So now plymouth picks a mode itself. When I submitted the >> plymouth change for plymouth to pick a mode itself the plymouth maintainer >> (Ray Strode) was afraid that would break plymouth honoring video= arguments. >> So currently plymouth still relies on the kernel to do the mode setup if >> a video= argument is present on the kernel commandline. >> >> My other recent series, which adds support for e.g. >> video=HDMI:panel_orientation=right_side_up, made me realize that relying >> on the kernel for this is no good since the fbcon code has various >> limitations wrt e.g. hotplug and this of course will not work when >> fbcon deferred takeover is used and fbcon never loads before plymouth. >> >> So I wrote a patch for plymouth to check the DRM_MODE_TYPE_USERDEF >> flag and prefer a mode with that flag over the PREFERRED mode: >> https://gitlab.freedesktop.org/plymouth/plymouth/merge_requests/84 >> >> And then I found out that for that code to work with modes which >> are already in the list of probed modes, I need something like >> the kernel patch we are discussing now. >> >> So if you want a link to an userspace consumer of this, I guess you >> want a v2 with a link to the plymouth MR added. + maybe a blurb in >> the commit message that to the best of my knowledge no userspace >> kms consumers are checking for the DRM_MODE_TYPE_USERDEF flag? > > And how does just relying on plymouth fix this? The idea is to not automatically fix this, all I want to do is provide enough info to userspace to make an informed decision, plymouth honored the video= resolution, we want to preserve that behavior. Likewise historically and currently compositors / Xorg pretty much ignore video=, I do not want to introduce changes on the kernel side which unilateral force a change here. If compositors want to start honoring a video= provided/selected mode then that is up to them. > Given what you figured out in your quick search I think we should set > both USERDEF and PREFERRED, otherwise not much will happen here. Plus > I guess we need some fixup code to make sure that only the cmdline > mode is preferred, and no other mode? Except if all compositors pick > the first preferred mode consistently, then I guess we just need to > make sure it's first. Changing the meaning of USERDEF to also mean > "preferred mode, pls use that" could be fraught with peril, e.g. you > run a game on an old compositor (or with direct display) at lower > refresh, maybe with a custom mode, then it crashes. And from then on > all compositors insist on using your "preferred" mode. Right, I considered using PREFERRED instead (and clearing it on other modes) and I deliberately did not do this. You say: "Changing the meaning of USERDEF to also mean "preferred mode, pls use that" could be fraught with peril" But since no userspace currently checks for USERDEF (*) it means that adding the flag will not cause anything to change. Where as adding PREFERRED will all of a sudden cause compositors/Xorg to pick a different default resolution. Arguably that is desirable, but it is an ABI breaks of sort and I'm sure some users will be using video=XXXxYYY to set a lower-resolution for fbcon (e.g. for readability on HiDPI) and will not appreciate it if their compositor now also picks this resolution. So as said the purpose of this patch is not too change anything, but only to pass info to userspace that this mode was set for the connector through video= on the kernel commandline. Alternatively we could add a new USERSELECTED flag and set that when we are not adding a new CVT/GTF mode since one of the probed modes already provides the requested resolution. Basically my goal is to get the info about the mode passed for the connector through video=name:XXXxYYY to userspace without userspace having to have its own copy of the video= parsing code. I hope this clarifies things. Regards, Hans *) For the pending plymouth patch the behavior change is desirable, as it keeps plymouth's historical behavior of honoring video= > -Daniel > >> >> Regards, >> >> Hans >> >> >> >> >> >> >>> -Danile >>> >>>> --- >>>> drivers/gpu/drm/drm_probe_helper.c | 2 ++ >>>> include/drm/drm_modes.h | 3 ++- >>>> 2 files changed, 4 insertions(+), 1 deletion(-) >>>> >>>> diff --git a/drivers/gpu/drm/drm_probe_helper.c b/drivers/gpu/drm/drm_probe_helper.c >>>> index ef2c468205a2..4fed64be11f9 100644 >>>> --- a/drivers/gpu/drm/drm_probe_helper.c >>>> +++ b/drivers/gpu/drm/drm_probe_helper.c >>>> @@ -157,6 +157,8 @@ static int drm_helper_probe_add_cmdline_mode(struct drm_connector *connector) >>>> continue; >>>> } >>>> >>>> + /* Mark the matching mode as being preferred by the user */ >>>> + mode->type |= DRM_MODE_TYPE_USERDEF; >>>> return 0; >>>> } >>>> >>>> diff --git a/include/drm/drm_modes.h b/include/drm/drm_modes.h >>>> index e946e20c61d8..c7efb7487e9b 100644 >>>> --- a/include/drm/drm_modes.h >>>> +++ b/include/drm/drm_modes.h >>>> @@ -256,7 +256,8 @@ struct drm_display_mode { >>>> * - DRM_MODE_TYPE_DRIVER: Mode created by the driver, which is all of >>>> * them really. Drivers must set this bit for all modes they create >>>> * and expose to userspace. >>>> - * - DRM_MODE_TYPE_USERDEF: Mode defined via kernel command line >>>> + * - DRM_MODE_TYPE_USERDEF: Mode defined or selected via the kernel >>>> + * command line. >>>> * >>>> * Plus a big list of flags which shouldn't be used at all, but are >>>> * still around since these flags are also used in the userspace ABI. >>>> -- >>>> 2.23.0 >>>> >>>> _______________________________________________ >>>> dri-devel mailing list >>>> dri-devel@lists.freedesktop.org >>>> https://lists.freedesktop.org/mailman/listinfo/dri-devel >>> >>> >>> >> > >
On Mon, Nov 11, 2019 at 10:50:42AM +0100, Hans de Goede wrote: > Hi, > > On 11-11-2019 10:25, Daniel Vetter wrote: > > On Sun, Nov 10, 2019 at 7:41 PM Hans de Goede <hdegoede@redhat.com> wrote: > >> > >> drm_helper_probe_add_cmdline_mode() prefers using a probed mode matching > >> a video= argument over calculating our own timings for the user specified > >> mode using CVT or GTF. > >> > >> But userspace code which is auto-configuring the mode may want to know that > >> the user has specified that mode on the kernel commandline so that it can > >> pick that mode over the mode which is marked as DRM_MODE_TYPE_PREFERRED. > >> > >> This commit sets the DRM_MODE_TYPE_USERDEF flag on the matching mode, just > >> as we would do on the user-specified mode when no matching probed mode is > >> found. > >> > >> Signed-off-by: Hans de Goede <hdegoede@redhat.com> > > > > Will existing userspace dtrt here with this? Some links to popular > > ones would be good (since essentially this is uapi fine tuning we need > > that anyway). With that will get my ack. > > A valid question, I've gone over what I consider the major userspace kms users: > -Xorg xserver modesetting driver does not check for this: > [hans@shalem xserver]$ ack DRM_MODE_TYPE_ hw/xfree86/drivers/modesetting/ > hw/xfree86/drivers/modesetting/drmmode_display.c > 1321: if (kmode->type & DRM_MODE_TYPE_DRIVER) > 1323: if (kmode->type & DRM_MODE_TYPE_PREFERRED) > -Other Xorg drivers: > [hans@shalem driver]$ ls -d xf86-video-* > xf86-video-amdgpu xf86-video-intel xf86-video-qxl > xf86-video-armsoc xf86-video-modesetting xf86-video-sisusb > xf86-video-ati xf86-video-nouveau xf86-video-vmware > xf86-video-dummy xf86-video-omap xf86-video-voodoo > xf86-video-geode xf86-video-opentegra > These all only do the same checks as the Xorg modesetting driver > -mutter: > [hans@shalem mutter]$ ack DRM_MODE_TYPE_ > src/backends/native/meta-output-kms.c > 261: if (drm_mode->type & DRM_MODE_TYPE_PREFERRED) > > So it seems nothing (that I can quickly find) in userspace is using this atm. > > The reason I wrote this patch is because about a year ago plymouth used to > fully rely on the kernel to setup the modes on monitors and would simply > inherit the modes setup by the kernel. Basically plymouth was relying on > fbcon to load first and setup modes. > > Deferred fbcon takeover (for flickerfree) means that this is no longer > happening. So now plymouth picks a mode itself. When I submitted the > plymouth change for plymouth to pick a mode itself the plymouth maintainer > (Ray Strode) was afraid that would break plymouth honoring video= arguments. > So currently plymouth still relies on the kernel to do the mode setup if > a video= argument is present on the kernel commandline. Why can't plymouth just keep using the current mode if the crtc is already active and otherwise pick a mode itself?
Hi, On 11-11-2019 15:11, Ville Syrjälä wrote: > On Mon, Nov 11, 2019 at 10:50:42AM +0100, Hans de Goede wrote: >> Hi, >> >> On 11-11-2019 10:25, Daniel Vetter wrote: >>> On Sun, Nov 10, 2019 at 7:41 PM Hans de Goede <hdegoede@redhat.com> wrote: >>>> >>>> drm_helper_probe_add_cmdline_mode() prefers using a probed mode matching >>>> a video= argument over calculating our own timings for the user specified >>>> mode using CVT or GTF. >>>> >>>> But userspace code which is auto-configuring the mode may want to know that >>>> the user has specified that mode on the kernel commandline so that it can >>>> pick that mode over the mode which is marked as DRM_MODE_TYPE_PREFERRED. >>>> >>>> This commit sets the DRM_MODE_TYPE_USERDEF flag on the matching mode, just >>>> as we would do on the user-specified mode when no matching probed mode is >>>> found. >>>> >>>> Signed-off-by: Hans de Goede <hdegoede@redhat.com> >>> >>> Will existing userspace dtrt here with this? Some links to popular >>> ones would be good (since essentially this is uapi fine tuning we need >>> that anyway). With that will get my ack. >> >> A valid question, I've gone over what I consider the major userspace kms users: >> -Xorg xserver modesetting driver does not check for this: >> [hans@shalem xserver]$ ack DRM_MODE_TYPE_ hw/xfree86/drivers/modesetting/ >> hw/xfree86/drivers/modesetting/drmmode_display.c >> 1321: if (kmode->type & DRM_MODE_TYPE_DRIVER) >> 1323: if (kmode->type & DRM_MODE_TYPE_PREFERRED) >> -Other Xorg drivers: >> [hans@shalem driver]$ ls -d xf86-video-* >> xf86-video-amdgpu xf86-video-intel xf86-video-qxl >> xf86-video-armsoc xf86-video-modesetting xf86-video-sisusb >> xf86-video-ati xf86-video-nouveau xf86-video-vmware >> xf86-video-dummy xf86-video-omap xf86-video-voodoo >> xf86-video-geode xf86-video-opentegra >> These all only do the same checks as the Xorg modesetting driver >> -mutter: >> [hans@shalem mutter]$ ack DRM_MODE_TYPE_ >> src/backends/native/meta-output-kms.c >> 261: if (drm_mode->type & DRM_MODE_TYPE_PREFERRED) >> >> So it seems nothing (that I can quickly find) in userspace is using this atm. >> >> The reason I wrote this patch is because about a year ago plymouth used to >> fully rely on the kernel to setup the modes on monitors and would simply >> inherit the modes setup by the kernel. Basically plymouth was relying on >> fbcon to load first and setup modes. >> >> Deferred fbcon takeover (for flickerfree) means that this is no longer >> happening. So now plymouth picks a mode itself. When I submitted the >> plymouth change for plymouth to pick a mode itself the plymouth maintainer >> (Ray Strode) was afraid that would break plymouth honoring video= arguments. >> So currently plymouth still relies on the kernel to do the mode setup if >> a video= argument is present on the kernel commandline. > > Why can't plymouth just keep using the current mode if the crtc is > already active and otherwise pick a mode itself? Well for one the firmware may have setup an output with a non native mode because it always uses e.g 1024x768, we really want to switch to the preferred mode in that case. We could only pick the current mode if the crtc is already active when video= is present on the kernel cmdline in an attempt to honor video= arguments, but that will only work if fbcon has already setup the modes according to the video= arguments and with flickerfree boot we defer loading fbcon until there are kernel are initrd errors to show. So in order to honor video= in a flickerfree setup plymouth needs to pick the requested mode itself. Actually knowing which mode is requested makes this a lot easier. Regards, Hans
On Mon, Nov 11, 2019 at 12:06 PM Hans de Goede <hdegoede@redhat.com> wrote: > > Hi Daniel, > > On 11-11-2019 11:26, Daniel Vetter wrote: > > On Mon, Nov 11, 2019 at 10:50 AM Hans de Goede <hdegoede@redhat.com> wrote: > >> > >> Hi, > >> > >> On 11-11-2019 10:25, Daniel Vetter wrote: > >>> On Sun, Nov 10, 2019 at 7:41 PM Hans de Goede <hdegoede@redhat.com> wrote: > >>>> > >>>> drm_helper_probe_add_cmdline_mode() prefers using a probed mode matching > >>>> a video= argument over calculating our own timings for the user specified > >>>> mode using CVT or GTF. > >>>> > >>>> But userspace code which is auto-configuring the mode may want to know that > >>>> the user has specified that mode on the kernel commandline so that it can > >>>> pick that mode over the mode which is marked as DRM_MODE_TYPE_PREFERRED. > >>>> > >>>> This commit sets the DRM_MODE_TYPE_USERDEF flag on the matching mode, just > >>>> as we would do on the user-specified mode when no matching probed mode is > >>>> found. > >>>> > >>>> Signed-off-by: Hans de Goede <hdegoede@redhat.com> > >>> > >>> Will existing userspace dtrt here with this? Some links to popular > >>> ones would be good (since essentially this is uapi fine tuning we need > >>> that anyway). With that will get my ack. > >> > >> A valid question, I've gone over what I consider the major userspace kms users: > >> -Xorg xserver modesetting driver does not check for this: > >> [hans@shalem xserver]$ ack DRM_MODE_TYPE_ hw/xfree86/drivers/modesetting/ > >> hw/xfree86/drivers/modesetting/drmmode_display.c > >> 1321: if (kmode->type & DRM_MODE_TYPE_DRIVER) > >> 1323: if (kmode->type & DRM_MODE_TYPE_PREFERRED) > >> -Other Xorg drivers: > >> [hans@shalem driver]$ ls -d xf86-video-* > >> xf86-video-amdgpu xf86-video-intel xf86-video-qxl > >> xf86-video-armsoc xf86-video-modesetting xf86-video-sisusb > >> xf86-video-ati xf86-video-nouveau xf86-video-vmware > >> xf86-video-dummy xf86-video-omap xf86-video-voodoo > >> xf86-video-geode xf86-video-opentegra > >> These all only do the same checks as the Xorg modesetting driver > >> -mutter: > >> [hans@shalem mutter]$ ack DRM_MODE_TYPE_ > >> src/backends/native/meta-output-kms.c > >> 261: if (drm_mode->type & DRM_MODE_TYPE_PREFERRED) > >> > >> So it seems nothing (that I can quickly find) in userspace is using this atm. > >> > >> The reason I wrote this patch is because about a year ago plymouth used to > >> fully rely on the kernel to setup the modes on monitors and would simply > >> inherit the modes setup by the kernel. Basically plymouth was relying on > >> fbcon to load first and setup modes. > >> > >> Deferred fbcon takeover (for flickerfree) means that this is no longer > >> happening. So now plymouth picks a mode itself. When I submitted the > >> plymouth change for plymouth to pick a mode itself the plymouth maintainer > >> (Ray Strode) was afraid that would break plymouth honoring video= arguments. > >> So currently plymouth still relies on the kernel to do the mode setup if > >> a video= argument is present on the kernel commandline. > >> > >> My other recent series, which adds support for e.g. > >> video=HDMI:panel_orientation=right_side_up, made me realize that relying > >> on the kernel for this is no good since the fbcon code has various > >> limitations wrt e.g. hotplug and this of course will not work when > >> fbcon deferred takeover is used and fbcon never loads before plymouth. > >> > >> So I wrote a patch for plymouth to check the DRM_MODE_TYPE_USERDEF > >> flag and prefer a mode with that flag over the PREFERRED mode: > >> https://gitlab.freedesktop.org/plymouth/plymouth/merge_requests/84 > >> > >> And then I found out that for that code to work with modes which > >> are already in the list of probed modes, I need something like > >> the kernel patch we are discussing now. > >> > >> So if you want a link to an userspace consumer of this, I guess you > >> want a v2 with a link to the plymouth MR added. + maybe a blurb in > >> the commit message that to the best of my knowledge no userspace > >> kms consumers are checking for the DRM_MODE_TYPE_USERDEF flag? > > > > And how does just relying on plymouth fix this? > > The idea is to not automatically fix this, all I want to do is > provide enough info to userspace to make an informed decision, > plymouth honored the video= resolution, we want to > preserve that behavior. > > Likewise historically and currently compositors / Xorg pretty > much ignore video=, I do not want to introduce changes on the > kernel side which unilateral force a change here. If compositors > want to start honoring a video= provided/selected mode then that > is up to them. > > > Given what you figured out in your quick search I think we should set > > both USERDEF and PREFERRED, otherwise not much will happen here. Plus > > I guess we need some fixup code to make sure that only the cmdline > > mode is preferred, and no other mode? Except if all compositors pick > > the first preferred mode consistently, then I guess we just need to > > make sure it's first. Changing the meaning of USERDEF to also mean > > "preferred mode, pls use that" could be fraught with peril, e.g. you > > run a game on an old compositor (or with direct display) at lower > > refresh, maybe with a custom mode, then it crashes. And from then on > > all compositors insist on using your "preferred" mode. > > Right, I considered using PREFERRED instead (and clearing it on other > modes) and I deliberately did not do this. You say: > > "Changing the meaning of USERDEF to also mean > "preferred mode, pls use that" could be fraught with peril" > > But since no userspace currently checks for USERDEF (*) it means that > adding the flag will not cause anything to change. Where as adding > PREFERRED will all of a sudden cause compositors/Xorg to pick a > different default resolution. Arguably that is desirable, but it > is an ABI breaks of sort and I'm sure some users will be using > video=XXXxYYY to set a lower-resolution for fbcon (e.g. for readability > on HiDPI) and will not appreciate it if their compositor now also > picks this resolution. But we already have other modes that get the USERDEF flag. That's my worry, your new userspace code wont just fire on this, but on any user created mode? Or am I missing something here? Ok I rechecked, I did miss something, this stuff is unused. So yeah I think this is ok, but we really also need good documentation for this somewhere ... at least kerneldoc for the relevant ioctl structures needs to be fully covering all these flags and their meanings. -Daniel > So as said the purpose of this patch is not too change anything, > but only to pass info to userspace that this mode was set for the > connector through video= on the kernel commandline. > > Alternatively we could add a new USERSELECTED flag and set that > when we are not adding a new CVT/GTF mode since one of the probed > modes already provides the requested resolution. > > Basically my goal is to get the info about the mode passed > for the connector through video=name:XXXxYYY to userspace without > userspace having to have its own copy of the video= parsing code. > > I hope this clarifies things. > > Regards, > > Hans > > > > *) For the pending plymouth patch the behavior change is desirable, > as it keeps plymouth's historical behavior of honoring video= > > > > > > -Daniel > > > >> > >> Regards, > >> > >> Hans > >> > >> > >> > >> > >> > >> > >>> -Danile > >>> > >>>> --- > >>>> drivers/gpu/drm/drm_probe_helper.c | 2 ++ > >>>> include/drm/drm_modes.h | 3 ++- > >>>> 2 files changed, 4 insertions(+), 1 deletion(-) > >>>> > >>>> diff --git a/drivers/gpu/drm/drm_probe_helper.c b/drivers/gpu/drm/drm_probe_helper.c > >>>> index ef2c468205a2..4fed64be11f9 100644 > >>>> --- a/drivers/gpu/drm/drm_probe_helper.c > >>>> +++ b/drivers/gpu/drm/drm_probe_helper.c > >>>> @@ -157,6 +157,8 @@ static int drm_helper_probe_add_cmdline_mode(struct drm_connector *connector) > >>>> continue; > >>>> } > >>>> > >>>> + /* Mark the matching mode as being preferred by the user */ > >>>> + mode->type |= DRM_MODE_TYPE_USERDEF; > >>>> return 0; > >>>> } > >>>> > >>>> diff --git a/include/drm/drm_modes.h b/include/drm/drm_modes.h > >>>> index e946e20c61d8..c7efb7487e9b 100644 > >>>> --- a/include/drm/drm_modes.h > >>>> +++ b/include/drm/drm_modes.h > >>>> @@ -256,7 +256,8 @@ struct drm_display_mode { > >>>> * - DRM_MODE_TYPE_DRIVER: Mode created by the driver, which is all of > >>>> * them really. Drivers must set this bit for all modes they create > >>>> * and expose to userspace. > >>>> - * - DRM_MODE_TYPE_USERDEF: Mode defined via kernel command line > >>>> + * - DRM_MODE_TYPE_USERDEF: Mode defined or selected via the kernel > >>>> + * command line. > >>>> * > >>>> * Plus a big list of flags which shouldn't be used at all, but are > >>>> * still around since these flags are also used in the userspace ABI. > >>>> -- > >>>> 2.23.0 > >>>> > >>>> _______________________________________________ > >>>> dri-devel mailing list > >>>> dri-devel@lists.freedesktop.org > >>>> https://lists.freedesktop.org/mailman/listinfo/dri-devel > >>> > >>> > >>> > >> > > > > >
Hi, On 11-11-2019 16:40, Daniel Vetter wrote: > On Mon, Nov 11, 2019 at 12:06 PM Hans de Goede <hdegoede@redhat.com> wrote: >> >> Hi Daniel, >> >> On 11-11-2019 11:26, Daniel Vetter wrote: >>> On Mon, Nov 11, 2019 at 10:50 AM Hans de Goede <hdegoede@redhat.com> wrote: >>>> >>>> Hi, >>>> >>>> On 11-11-2019 10:25, Daniel Vetter wrote: >>>>> On Sun, Nov 10, 2019 at 7:41 PM Hans de Goede <hdegoede@redhat.com> wrote: >>>>>> >>>>>> drm_helper_probe_add_cmdline_mode() prefers using a probed mode matching >>>>>> a video= argument over calculating our own timings for the user specified >>>>>> mode using CVT or GTF. >>>>>> >>>>>> But userspace code which is auto-configuring the mode may want to know that >>>>>> the user has specified that mode on the kernel commandline so that it can >>>>>> pick that mode over the mode which is marked as DRM_MODE_TYPE_PREFERRED. >>>>>> >>>>>> This commit sets the DRM_MODE_TYPE_USERDEF flag on the matching mode, just >>>>>> as we would do on the user-specified mode when no matching probed mode is >>>>>> found. >>>>>> >>>>>> Signed-off-by: Hans de Goede <hdegoede@redhat.com> >>>>> >>>>> Will existing userspace dtrt here with this? Some links to popular >>>>> ones would be good (since essentially this is uapi fine tuning we need >>>>> that anyway). With that will get my ack. >>>> >>>> A valid question, I've gone over what I consider the major userspace kms users: >>>> -Xorg xserver modesetting driver does not check for this: >>>> [hans@shalem xserver]$ ack DRM_MODE_TYPE_ hw/xfree86/drivers/modesetting/ >>>> hw/xfree86/drivers/modesetting/drmmode_display.c >>>> 1321: if (kmode->type & DRM_MODE_TYPE_DRIVER) >>>> 1323: if (kmode->type & DRM_MODE_TYPE_PREFERRED) >>>> -Other Xorg drivers: >>>> [hans@shalem driver]$ ls -d xf86-video-* >>>> xf86-video-amdgpu xf86-video-intel xf86-video-qxl >>>> xf86-video-armsoc xf86-video-modesetting xf86-video-sisusb >>>> xf86-video-ati xf86-video-nouveau xf86-video-vmware >>>> xf86-video-dummy xf86-video-omap xf86-video-voodoo >>>> xf86-video-geode xf86-video-opentegra >>>> These all only do the same checks as the Xorg modesetting driver >>>> -mutter: >>>> [hans@shalem mutter]$ ack DRM_MODE_TYPE_ >>>> src/backends/native/meta-output-kms.c >>>> 261: if (drm_mode->type & DRM_MODE_TYPE_PREFERRED) >>>> >>>> So it seems nothing (that I can quickly find) in userspace is using this atm. >>>> >>>> The reason I wrote this patch is because about a year ago plymouth used to >>>> fully rely on the kernel to setup the modes on monitors and would simply >>>> inherit the modes setup by the kernel. Basically plymouth was relying on >>>> fbcon to load first and setup modes. >>>> >>>> Deferred fbcon takeover (for flickerfree) means that this is no longer >>>> happening. So now plymouth picks a mode itself. When I submitted the >>>> plymouth change for plymouth to pick a mode itself the plymouth maintainer >>>> (Ray Strode) was afraid that would break plymouth honoring video= arguments. >>>> So currently plymouth still relies on the kernel to do the mode setup if >>>> a video= argument is present on the kernel commandline. >>>> >>>> My other recent series, which adds support for e.g. >>>> video=HDMI:panel_orientation=right_side_up, made me realize that relying >>>> on the kernel for this is no good since the fbcon code has various >>>> limitations wrt e.g. hotplug and this of course will not work when >>>> fbcon deferred takeover is used and fbcon never loads before plymouth. >>>> >>>> So I wrote a patch for plymouth to check the DRM_MODE_TYPE_USERDEF >>>> flag and prefer a mode with that flag over the PREFERRED mode: >>>> https://gitlab.freedesktop.org/plymouth/plymouth/merge_requests/84 >>>> >>>> And then I found out that for that code to work with modes which >>>> are already in the list of probed modes, I need something like >>>> the kernel patch we are discussing now. >>>> >>>> So if you want a link to an userspace consumer of this, I guess you >>>> want a v2 with a link to the plymouth MR added. + maybe a blurb in >>>> the commit message that to the best of my knowledge no userspace >>>> kms consumers are checking for the DRM_MODE_TYPE_USERDEF flag? >>> >>> And how does just relying on plymouth fix this? >> >> The idea is to not automatically fix this, all I want to do is >> provide enough info to userspace to make an informed decision, >> plymouth honored the video= resolution, we want to >> preserve that behavior. >> >> Likewise historically and currently compositors / Xorg pretty >> much ignore video=, I do not want to introduce changes on the >> kernel side which unilateral force a change here. If compositors >> want to start honoring a video= provided/selected mode then that >> is up to them. >> >>> Given what you figured out in your quick search I think we should set >>> both USERDEF and PREFERRED, otherwise not much will happen here. Plus >>> I guess we need some fixup code to make sure that only the cmdline >>> mode is preferred, and no other mode? Except if all compositors pick >>> the first preferred mode consistently, then I guess we just need to >>> make sure it's first. Changing the meaning of USERDEF to also mean >>> "preferred mode, pls use that" could be fraught with peril, e.g. you >>> run a game on an old compositor (or with direct display) at lower >>> refresh, maybe with a custom mode, then it crashes. And from then on >>> all compositors insist on using your "preferred" mode. >> >> Right, I considered using PREFERRED instead (and clearing it on other >> modes) and I deliberately did not do this. You say: >> >> "Changing the meaning of USERDEF to also mean >> "preferred mode, pls use that" could be fraught with peril" >> >> But since no userspace currently checks for USERDEF (*) it means that >> adding the flag will not cause anything to change. Where as adding >> PREFERRED will all of a sudden cause compositors/Xorg to pick a >> different default resolution. Arguably that is desirable, but it >> is an ABI breaks of sort and I'm sure some users will be using >> video=XXXxYYY to set a lower-resolution for fbcon (e.g. for readability >> on HiDPI) and will not appreciate it if their compositor now also >> picks this resolution. > > But we already have other modes that get the USERDEF flag. That's my > worry, your new userspace code wont just fire on this, but on any user > created mode? Or am I missing something here? > > Ok I rechecked, I did miss something, this stuff is unused. So yeah I > think this is ok, but we really also need good documentation for this > somewhere ... at least kerneldoc for the relevant ioctl structures > needs to be fully covering all these flags and their meanings. The flags are already documented here: https://www.kernel.org/doc/html/latest/gpu/drm-kms.html#c.drm_display_mode And before this commit the documentation for the USERDEF flag reads: "DRM_MODE_TYPE_USERDEF: Mode defined via kernel command line" So it not being used for anything else matches the docs, note that this patch updates the docs to match the new use: "DRM_MODE_TYPE_USERDEF: Mode defined or selected via the kernel command line." So I'm not entirely sure what more documentation you want for this ? Regards, Hans >>>>>> diff --git a/drivers/gpu/drm/drm_probe_helper.c b/drivers/gpu/drm/drm_probe_helper.c >>>>>> index ef2c468205a2..4fed64be11f9 100644 >>>>>> --- a/drivers/gpu/drm/drm_probe_helper.c >>>>>> +++ b/drivers/gpu/drm/drm_probe_helper.c >>>>>> @@ -157,6 +157,8 @@ static int drm_helper_probe_add_cmdline_mode(struct drm_connector *connector) >>>>>> continue; >>>>>> } >>>>>> >>>>>> + /* Mark the matching mode as being preferred by the user */ >>>>>> + mode->type |= DRM_MODE_TYPE_USERDEF; >>>>>> return 0; >>>>>> } >>>>>> >>>>>> diff --git a/include/drm/drm_modes.h b/include/drm/drm_modes.h >>>>>> index e946e20c61d8..c7efb7487e9b 100644 >>>>>> --- a/include/drm/drm_modes.h >>>>>> +++ b/include/drm/drm_modes.h >>>>>> @@ -256,7 +256,8 @@ struct drm_display_mode { >>>>>> * - DRM_MODE_TYPE_DRIVER: Mode created by the driver, which is all of >>>>>> * them really. Drivers must set this bit for all modes they create >>>>>> * and expose to userspace. >>>>>> - * - DRM_MODE_TYPE_USERDEF: Mode defined via kernel command line >>>>>> + * - DRM_MODE_TYPE_USERDEF: Mode defined or selected via the kernel >>>>>> + * command line. >>>>>> * >>>>>> * Plus a big list of flags which shouldn't be used at all, but are >>>>>> * still around since these flags are also used in the userspace ABI. >>>>>> -- >>>>>> 2.23.0 >>>>>> >>>>>> _______________________________________________ >>>>>> dri-devel mailing list >>>>>> dri-devel@lists.freedesktop.org >>>>>> https://lists.freedesktop.org/mailman/listinfo/dri-devel >>>>> >>>>> >>>>> >>>> >>> >>> >> > >
diff --git a/drivers/gpu/drm/drm_probe_helper.c b/drivers/gpu/drm/drm_probe_helper.c index ef2c468205a2..4fed64be11f9 100644 --- a/drivers/gpu/drm/drm_probe_helper.c +++ b/drivers/gpu/drm/drm_probe_helper.c @@ -157,6 +157,8 @@ static int drm_helper_probe_add_cmdline_mode(struct drm_connector *connector) continue; } + /* Mark the matching mode as being preferred by the user */ + mode->type |= DRM_MODE_TYPE_USERDEF; return 0; } diff --git a/include/drm/drm_modes.h b/include/drm/drm_modes.h index e946e20c61d8..c7efb7487e9b 100644 --- a/include/drm/drm_modes.h +++ b/include/drm/drm_modes.h @@ -256,7 +256,8 @@ struct drm_display_mode { * - DRM_MODE_TYPE_DRIVER: Mode created by the driver, which is all of * them really. Drivers must set this bit for all modes they create * and expose to userspace. - * - DRM_MODE_TYPE_USERDEF: Mode defined via kernel command line + * - DRM_MODE_TYPE_USERDEF: Mode defined or selected via the kernel + * command line. * * Plus a big list of flags which shouldn't be used at all, but are * still around since these flags are also used in the userspace ABI.
drm_helper_probe_add_cmdline_mode() prefers using a probed mode matching a video= argument over calculating our own timings for the user specified mode using CVT or GTF. But userspace code which is auto-configuring the mode may want to know that the user has specified that mode on the kernel commandline so that it can pick that mode over the mode which is marked as DRM_MODE_TYPE_PREFERRED. This commit sets the DRM_MODE_TYPE_USERDEF flag on the matching mode, just as we would do on the user-specified mode when no matching probed mode is found. Signed-off-by: Hans de Goede <hdegoede@redhat.com> --- drivers/gpu/drm/drm_probe_helper.c | 2 ++ include/drm/drm_modes.h | 3 ++- 2 files changed, 4 insertions(+), 1 deletion(-)