Skip to contentSkip to conversation

Expanding drm_mode_modeinfo flags

RFC New Jeykumar Sankaran 8 participants 12 messages Jul 11, 2019 – Jul 28, 2019 3 messages not in the archive

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

Hello All,
drm_mode_modeinfo::flags is a 32 bit field currently used to
describe the properties of a connector mode. I see the least order 22 bits are already in use. Posting this RFC to discuss on any potential plans to expand the bit range support of this field for growing mode properties and ways to handle one such property needed by the msm dpu driver.

msm drivers support panels which can dynamically switch between video(active) and command(smart) modes. Within video mode, they also support switching between resolutions seamlessly i.e. glitch free resolution switch. But they cannot do a seamless switch from a resolutions from video to command or vice versa. Clients need to be aware for these capablities before they switch between the resolutions. Since these capabilities are identified per drm_mode, we are considering the below two approaches to handle this use case.

Option 1:
Attached patch adds flag values to associate a drm_mode to video/command mode and to indicate its capability to do a seamless switch.

Option 2:
drm_mode_modeinfo can expose a new "private_flags" field to handle vendor specific mode flags. Besides the above mentioned use case, we are also expoloring methods to handle some of our display port resolution switch use cases where the DP ports can be operated in a tiled/detiled modes. This approach will provide a standard channel for drm driver vendors for their growing need for drm_mode specific capabilities.

Please provide your inputs on the options or any upstream friendly recommendation to handle such custom use cases.

Thanks and Regards,
Jeykumar S.

Jeykumar Sankaran (1):
drm: add mode flags in uapi for seamless mode switch

 include/uapi/drm/drm_mode.h | 5 +++++
 1 file changed, 5 insertions(+)
Signature · 7 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#1 lore raw

RFC PATCH 1/1 drm: add mode flags in uapi for seamless mode switch

Add drm mode flag values to expose mode capabilities to perform dynamic seamless mode switch. This change also exposes the backing panel type associated with a mode for panels which can dynamically switch between video and command display modes.

---
 include/uapi/drm/drm_mode.h | 5 +++++
 1 file changed, 5 insertions(+)
include/uapi/drm/drm_mode.h +5 -0
index 5ab331e..b76f1bf 100644
@@ -124,6 +124,11 @@
 #define  DRM_MODE_FLAG_PIC_AR_256_135 \
 			(DRM_MODE_PICTURE_ASPECT_256_135<<19)
 
+#define DRM_MODE_FLAG_SEAMLESS_SWITCH (1<<23)
+
+#define DRM_MODE_FLAG_VIDEO_MODE   (1<<24)
+#define DRM_MODE_FLAG_COMMAND_MODE (1<<25)
+
 #define  DRM_MODE_FLAG_ALL	(DRM_MODE_FLAG_PHSYNC |		\
 				 DRM_MODE_FLAG_NHSYNC |		\
 				 DRM_MODE_FLAG_PVSYNC |		\
Signature · 7 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

A message that is not in the archive (sent off-list or not yet received): 1562870805-32314-1-git-send-email-jsanka-sgV2jX0FEOL9JmXXK+q4OQ@public.gmane.org

#3 lore raw
Quoted 28 lines · Jeykumar Sankaran

Hello All,
drm_mode_modeinfo::flags is a 32 bit field currently used to
describe the properties of a connector mode. I see the least order 22 bits are already in use. Posting this RFC to discuss on any potential plans to expand the bit range support of this field for growing mode properties and ways to handle one such property needed by the msm dpu driver.

msm drivers support panels which can dynamically switch between video(active) and command(smart) modes. Within video mode, they also support switching between resolutions seamlessly i.e. glitch free resolution switch. But they cannot do a seamless switch from a resolutions from video to command or vice versa. Clients need to be aware for these capablities before they switch between the resolutions. Since these capabilities are identified per drm_mode, we are considering the below two approaches to handle this use case.

Option 1:
Attached patch adds flag values to associate a drm_mode to video/command mode and to indicate its capability to do a seamless switch.

Option 2:
drm_mode_modeinfo can expose a new "private_flags" field to handle vendor specific mode flags. Besides the above mentioned use case, we are also expoloring methods to handle some of our display port resolution switch use cases where the DP ports can be operated in a tiled/detiled modes. This approach will provide a standard channel for drm driver vendors for their growing need for drm_mode specific capabilities.

Quoted 8 lines

Please provide your inputs on the options or any upstream friendly recommendation to handle such custom use cases.

Thanks and Regards,
Jeykumar S.

Jeykumar Sankaran (1):
drm: add mode flags in uapi for seamless mode switch

I think the uapi is the trivial part here, the real deal is how userspace uses this. Can you pls post the patches for your compositor?

Also note that we already allow userspace to tell the kernel whether flickering is ok or not for a modeset. msm driver could use that to at least tell userspace whether a modeset change is possible. So you can already implement glitch-free modeset changes for at least video mode. -Daniel

Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Freedreno mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/freedreno

A message that is not in the archive (sent off-list or not yet received): 20190716090712.GY15868-dv86pmgwkMBes7Z6vYuT8azUEOm+Xw19@public.gmane.org

#5 lore raw
Quoted 57 lines · Daniel Vetter
Quoted 3 lines · Jeykumar Sankaran

Hello All,
drm_mode_modeinfo::flags is a 32 bit field currently used to
describe the properties of a connector mode. I see the least order

22 bits

Quoted 1 line

are already in use. Posting this RFC to discuss on any potential

plans to

Quoted 1 line

expand the bit range support of this field for growing mode

properties and

Quoted 5 lines

ways to handle one such property needed by the msm dpu driver.

msm drivers support panels which can dynamically switch between video(active) and command(smart) modes. Within video mode, they
also

support

Quoted 2 lines

switching between resolutions seamlessly i.e. glitch free
resolution

switch.

Quoted 1 line

But they cannot do a seamless switch from a resolutions from video

to

Quoted 1 line

command or vice versa. Clients need to be aware for these

capablities before

Quoted 1 line

they switch between the resolutions. Since these capabilities are

identified

Quoted 2 lines

per drm_mode, we are considering the below two approaches to
handle

this

Quoted 4 lines

use case.

Option 1:
Attached patch adds flag values to associate a drm_mode to

video/command

Quoted 4 lines

mode and to indicate its capability to do a seamless switch.

Option 2:
drm_mode_modeinfo can expose a new "private_flags" field to handle

vendor

Quoted 1 line

specific mode flags. Besides the above mentioned use case, we are

also

Quoted 1 line

expoloring methods to handle some of our display port resolution

switch use

Quoted 1 line

cases where the DP ports can be operated in a tiled/detiled modes.

This

Quoted 2 lines

approach will provide a standard channel for drm driver vendors
for

their

Quoted 10 lines

growing need for drm_mode specific capabilities.

Please provide your inputs on the options or any upstream friendly recommendation to handle such custom use cases.

Thanks and Regards,
Jeykumar S.

Jeykumar Sankaran (1):
drm: add mode flags in uapi for seamless mode switch

Quoted 8 lines

I think the uapi is the trivial part here, the real deal is how userspace uses this. Can you pls post the patches for your compositor?

Also note that we already allow userspace to tell the kernel whether flickering is ok or not for a modeset. msm driver could use that to at least tell userspace whether a modeset change is possible. So you can already implement glitch-free modeset changes for at least video mode. -Daniel

I believe you are referring to the below tv property of the connector.

/**
* @tv_flicker_reduction_property: Optional TV property to control the
* flicker reduction mode.
*/
struct drm_property *tv_flicker_reduction_property;

Sure. We can use this to indicate whether the connector representing the panel can support the dynamic glitch-free switch. But when the panel supports both video and command mode resolutions and such glitch-free switch is possible only between video mode resolutions, we need per resolution(drm_mode_modeinfo) information to identify the resolutions enumerated for video mode.

Below is an example of the compositor utility function which can use the tv_flicker_property and the proposed modeinfo flags to identify glitch-free switches.

bool is_seamless_switch_supported(struct drm_mode_modeinfo src_mode,
struct
drm_mode_modeinfo *dst_mode, bool
flicker_reduction_supported)
{
if (!flicker_reduction_supported) {
printf("flicker reduction prop not set for the

connector\n");
                 return false;
         }

         /*
          * Seamless switch(if supported) is possible only between video 
mode
          * resolutions
          */
         if (src_mode->flags & DRM_MODE_FLAG_VIDEO_MODE &&
                         dst_mode->flags & DRM_MODE_FLAG_VIDEO_MODE)
                 return true;
         else
                 return false;

}

Signature · 5 lines
Jeykumar S
_______________________________________________
Freedreno mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/freedreno
#6 lore raw
Quoted 68 lines · Jeykumar Sankaran
Quoted 61 lines · Daniel Vetter
Quoted 3 lines · Jeykumar Sankaran

Hello All,
drm_mode_modeinfo::flags is a 32 bit field currently used to
describe the properties of a connector mode. I see the least order

22 bits

Quoted 1 line

are already in use. Posting this RFC to discuss on any potential

plans to

Quoted 1 line

expand the bit range support of this field for growing mode

properties and

Quoted 5 lines

ways to handle one such property needed by the msm dpu driver.

msm drivers support panels which can dynamically switch between video(active) and command(smart) modes. Within video mode, they
also

support

Quoted 2 lines

switching between resolutions seamlessly i.e. glitch free
resolution

switch.

Quoted 1 line

But they cannot do a seamless switch from a resolutions from video

to

Quoted 1 line

command or vice versa. Clients need to be aware for these

capablities before

Quoted 1 line

they switch between the resolutions. Since these capabilities are

identified

Quoted 2 lines

per drm_mode, we are considering the below two approaches to
handle

this

Quoted 4 lines

use case.

Option 1:
Attached patch adds flag values to associate a drm_mode to

video/command

Quoted 4 lines

mode and to indicate its capability to do a seamless switch.

Option 2:
drm_mode_modeinfo can expose a new "private_flags" field to handle

vendor

Quoted 1 line

specific mode flags. Besides the above mentioned use case, we are

also

Quoted 1 line

expoloring methods to handle some of our display port resolution

switch use

Quoted 1 line

cases where the DP ports can be operated in a tiled/detiled modes.

This

Quoted 2 lines

approach will provide a standard channel for drm driver vendors
for

their

Quoted 10 lines

growing need for drm_mode specific capabilities.

Please provide your inputs on the options or any upstream friendly recommendation to handle such custom use cases.

Thanks and Regards,
Jeykumar S.

Jeykumar Sankaran (1):
drm: add mode flags in uapi for seamless mode switch

I think the uapi is the trivial part here, the real deal is how userspace
uses this. Can you pls post the patches for your compositor?

Quoted 5 lines

Also note that we already allow userspace to tell the kernel whether flickering is ok or not for a modeset. msm driver could use that to at least tell userspace whether a modeset change is possible. So you can already implement glitch-free modeset changes for at least video mode. -Daniel

Quoted 7 lines

I believe you are referring to the below tv property of the connector.

/**
* @tv_flicker_reduction_property: Optional TV property to control the
* flicker reduction mode.
*/
struct drm_property *tv_flicker_reduction_property;

Not even close :-)

I mean the DRM_MODE_ATOMIC_ALLOW_MODESET flag for the atomic ioctl. This is not a property of a mode, this is a property of a _transition_ between configurations. Some transitions can be done flicker free, others can't.

There's then still the question of how to pick video vs command mode, but imo better to start with implementing the transition behaviour correctly first.
-Daniel

Quoted 40 lines

Sure. We can use this to indicate whether the connector representing the panel
can support the dynamic glitch-free switch. But when the panel supports both video and command mode resolutions and such glitch-free switch is possible only between
video mode resolutions, we need per resolution(drm_mode_modeinfo) information
to identify the resolutions enumerated for video mode.

Below is an example of the compositor utility function which can use the tv_flicker_property
and the proposed modeinfo flags to identify glitch-free switches.

bool is_seamless_switch_supported(struct drm_mode_modeinfo src_mode, struct
drm_mode_modeinfo *dst_mode, bool
flicker_reduction_supported)
{
if (!flicker_reduction_supported) {
printf("flicker reduction prop not set for the

connector\n");
                 return false;
         }

         /*
          * Seamless switch(if supported) is possible only between video
mode
          * resolutions
          */
         if (src_mode->flags & DRM_MODE_FLAG_VIDEO_MODE &&
                         dst_mode->flags & DRM_MODE_FLAG_VIDEO_MODE)
                 return true;
         else
                 return false;

}

Signature · 1 line
Jeykumar S
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#7 lore raw
Quoted 77 lines · Daniel Vetter
Quoted 68 lines · Jeykumar Sankaran
Quoted 61 lines · Daniel Vetter
Quoted 3 lines · Jeykumar Sankaran

Hello All,
drm_mode_modeinfo::flags is a 32 bit field currently used to
describe the properties of a connector mode. I see the least order

22 bits

Quoted 1 line

are already in use. Posting this RFC to discuss on any potential

plans to

Quoted 1 line

expand the bit range support of this field for growing mode

properties and

Quoted 5 lines

ways to handle one such property needed by the msm dpu driver.

msm drivers support panels which can dynamically switch between video(active) and command(smart) modes. Within video mode, they
also

support

Quoted 2 lines

switching between resolutions seamlessly i.e. glitch free
resolution

switch.

Quoted 1 line

But they cannot do a seamless switch from a resolutions from video

to

Quoted 1 line

command or vice versa. Clients need to be aware for these

capablities before

Quoted 1 line

they switch between the resolutions. Since these capabilities are

identified

Quoted 2 lines

per drm_mode, we are considering the below two approaches to
handle

this

Quoted 4 lines

use case.

Option 1:
Attached patch adds flag values to associate a drm_mode to

video/command

Quoted 4 lines

mode and to indicate its capability to do a seamless switch.

Option 2:
drm_mode_modeinfo can expose a new "private_flags" field to handle

vendor

Quoted 1 line

specific mode flags. Besides the above mentioned use case, we are

also

Quoted 1 line

expoloring methods to handle some of our display port resolution

switch use

Quoted 1 line

cases where the DP ports can be operated in a tiled/detiled modes.

This

Quoted 2 lines

approach will provide a standard channel for drm driver vendors
for

their

Quoted 10 lines

growing need for drm_mode specific capabilities.

Please provide your inputs on the options or any upstream friendly recommendation to handle such custom use cases.

Thanks and Regards,
Jeykumar S.

Jeykumar Sankaran (1):
drm: add mode flags in uapi for seamless mode switch

I think the uapi is the trivial part here, the real deal is how userspace
uses this. Can you pls post the patches for your compositor?

Quoted 5 lines

Also note that we already allow userspace to tell the kernel whether flickering is ok or not for a modeset. msm driver could use that to at least tell userspace whether a modeset change is possible. So you can already implement glitch-free modeset changes for at least video mode. -Daniel

Quoted 7 lines

I believe you are referring to the below tv property of the connector.

/**
* @tv_flicker_reduction_property: Optional TV property to control the
* flicker reduction mode.
*/
struct drm_property *tv_flicker_reduction_property;

Quoted 5 lines

Not even close :-)

I mean the DRM_MODE_ATOMIC_ALLOW_MODESET flag for the atomic ioctl. This is not a property of a mode, this is a property of a _transition_ between configurations. Some transitions can be done flicker free, others can't.

Agree that an atomic flag on a commit is the way to accomplish this. It's pretty similar to the psr transitions, where we want to reuse most of the atomic circuitry, but in a specialized way. We'd also have to be careful to only involve the drm objects which are seamless modeset aware (you could imagine a bridge chain where the bridges downstream of the first bridge don't care).

Quoted 4 lines

There's then still the question of how to pick video vs command mode, but imo better to start with implementing the transition behaviour correctly first.

Connector property? Possibly a terrible idea, but I wonder if we could [re]use the vrr properties for command mode. The docs state that the driver has the option of putting upper and lower bounds on the refresh rate.

Sean

Quoted 51 lines

-Daniel

Quoted 40 lines

Sure. We can use this to indicate whether the connector representing the panel
can support the dynamic glitch-free switch. But when the panel supports both video and command mode resolutions and such glitch-free switch is possible only between
video mode resolutions, we need per resolution(drm_mode_modeinfo) information
to identify the resolutions enumerated for video mode.

Below is an example of the compositor utility function which can use the tv_flicker_property
and the proposed modeinfo flags to identify glitch-free switches.

bool is_seamless_switch_supported(struct drm_mode_modeinfo src_mode, struct
drm_mode_modeinfo *dst_mode, bool
flicker_reduction_supported)
{
if (!flicker_reduction_supported) {
printf("flicker reduction prop not set for the

connector\n");
                 return false;
         }

         /*
          * Seamless switch(if supported) is possible only between video
mode
          * resolutions
          */
         if (src_mode->flags & DRM_MODE_FLAG_VIDEO_MODE &&
                         dst_mode->flags & DRM_MODE_FLAG_VIDEO_MODE)
                 return true;
         else
                 return false;

}

Signature · 1 line
Jeykumar S
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Signature · 5 lines
Sean Paul, Software Engineer, Google / Chromium OS
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#8 lore raw
Quoted 91 lines · Sean Paul
Quoted 77 lines · Daniel Vetter
Quoted 68 lines · Jeykumar Sankaran
Quoted 61 lines · Daniel Vetter
Quoted 3 lines · Jeykumar Sankaran

Hello All,
drm_mode_modeinfo::flags is a 32 bit field currently used to
describe the properties of a connector mode. I see the least order

22 bits

Quoted 1 line

are already in use. Posting this RFC to discuss on any potential

plans to

Quoted 1 line

expand the bit range support of this field for growing mode

properties and

Quoted 5 lines

ways to handle one such property needed by the msm dpu driver.

msm drivers support panels which can dynamically switch between video(active) and command(smart) modes. Within video mode, they
also

support

Quoted 2 lines

switching between resolutions seamlessly i.e. glitch free
resolution

switch.

Quoted 1 line

But they cannot do a seamless switch from a resolutions from video

to

Quoted 1 line

command or vice versa. Clients need to be aware for these

capablities before

Quoted 1 line

they switch between the resolutions. Since these capabilities are

identified

Quoted 2 lines

per drm_mode, we are considering the below two approaches to
handle

this

Quoted 4 lines

use case.

Option 1:
Attached patch adds flag values to associate a drm_mode to

video/command

Quoted 4 lines

mode and to indicate its capability to do a seamless switch.

Option 2:
drm_mode_modeinfo can expose a new "private_flags" field to handle

vendor

Quoted 1 line

specific mode flags. Besides the above mentioned use case, we are

also

Quoted 1 line

expoloring methods to handle some of our display port resolution

switch use

Quoted 1 line

cases where the DP ports can be operated in a tiled/detiled modes.

This

Quoted 2 lines

approach will provide a standard channel for drm driver vendors
for

their

Quoted 10 lines

growing need for drm_mode specific capabilities.

Please provide your inputs on the options or any upstream friendly recommendation to handle such custom use cases.

Thanks and Regards,
Jeykumar S.

Jeykumar Sankaran (1):
drm: add mode flags in uapi for seamless mode switch

I think the uapi is the trivial part here, the real deal is how userspace
uses this. Can you pls post the patches for your compositor?

Quoted 5 lines

Also note that we already allow userspace to tell the kernel whether flickering is ok or not for a modeset. msm driver could use that to at least tell userspace whether a modeset change is possible. So you can already implement glitch-free modeset changes for at least video mode. -Daniel

Quoted 7 lines

I believe you are referring to the below tv property of the connector.

/**
* @tv_flicker_reduction_property: Optional TV property to control the
* flicker reduction mode.
*/
struct drm_property *tv_flicker_reduction_property;

Quoted 5 lines

Not even close :-)

I mean the DRM_MODE_ATOMIC_ALLOW_MODESET flag for the atomic ioctl. This is not a property of a mode, this is a property of a _transition_ between configurations. Some transitions can be done flicker free, others can't.

Agree that an atomic flag on a commit is the way to accomplish this. It's pretty similar to the psr transitions, where we want to reuse most of the atomic circuitry, but in a specialized way. We'd also have to be careful to only involve the drm objects which are seamless modeset aware (you could imagine a bridge chain where the bridges downstream of the first bridge don't care).

Quoted 1 line
Quoted 7 lines
Quoted 3 lines

There's then still the question of how to pick video vs command mode, but imo better to start with implementing the transition behaviour correctly first.

Connector property? Possibly a terrible idea, but I wonder if we could [re]use the vrr properties for command mode. The docs state that the driver has the option of putting upper and lower bounds on the refresh rate.

Not really sure why this needs new props and whatnot. This is kinda what the i915 "fastset" stuff already does:
1. userspace asks for something to be changed via atomic
2. driver calculates whether a modeset is actually required
3. atomic validates need_modeset() vs. DRM_MODE_ATOMIC_ALLOW_MODESET
4. if (need_modeset) heavyweight_commit() else lightweight_commit()

Ie. why should userspace really care about anything except the "flickers are OK" vs. "flickers not wanted" thing?

Also what's the benefit of using video mode if your panel supportes command mode? Can you turn off the memory in the panel and actually save power that way? And if there is a benefit can't the driver just automagically switch between the two based on how often things are getting updated? That would match how eDP PSR already works.

Signature · 6 lines
Ville Syrjälä
Intel
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

A message that is not in the archive (sent off-list or not yet received): [email protected]

#10 lore raw
Quoted 5 lines · Ville Syrjälä
Quoted 4 lines · Sean Paul
Quoted 3 lines · Daniel Vetter
Quoted 2 lines · Jeykumar Sankaran
Quoted 1 line · Daniel Vetter

On Thu, Jul 11, 2019 at 11:46:44AM -0700, Jeykumar Sankaran wrote:

/snip

Quoted 48 lines
Quoted 33 lines
Quoted 20 lines
Quoted 12 lines
Quoted 11 lines
Quoted 1 line

drm: add mode flags in uapi for seamless mode switch

I think the uapi is the trivial part here, the real deal is how userspace
uses this. Can you pls post the patches for your compositor?

Also note that we already allow userspace to tell the kernel whether flickering is ok or not for a modeset. msm driver could use that to at least tell userspace whether a modeset change is possible. So you can already implement glitch-free modeset changes for at least video mode. -Daniel

Quoted 7 lines

I believe you are referring to the below tv property of the connector.

/**
* @tv_flicker_reduction_property: Optional TV property to control the
* flicker reduction mode.
*/
struct drm_property *tv_flicker_reduction_property;

Quoted 5 lines

Not even close :-)

I mean the DRM_MODE_ATOMIC_ALLOW_MODESET flag for the atomic ioctl. This is not a property of a mode, this is a property of a _transition_ between configurations. Some transitions can be done flicker free, others can't.

Agree that an atomic flag on a commit is the way to accomplish this. It's pretty similar to the psr transitions, where we want to reuse most of the atomic circuitry, but in a specialized way. We'd also have to be careful to only involve the drm objects which are seamless modeset aware (you could imagine a bridge chain where the bridges downstream of the first bridge don't care).

Quoted 1 line
Quoted 7 lines
Quoted 3 lines

There's then still the question of how to pick video vs command mode, but imo better to start with implementing the transition behaviour correctly first.

Connector property? Possibly a terrible idea, but I wonder if we could [re]use the vrr properties for command mode. The docs state that the driver has the option of putting upper and lower bounds on the refresh rate.

Not really sure why this needs new props and whatnot. This is kinda what the i915 "fastset" stuff already does:
1. userspace asks for something to be changed via atomic
2. driver calculates whether a modeset is actually required
3. atomic validates need_modeset() vs. DRM_MODE_ATOMIC_ALLOW_MODESET
4. if (need_modeset) heavyweight_commit() else lightweight_commit()

Quoted 2 lines

Ie. why should userspace really care about anything except the "flickers are OK" vs. "flickers not wanted" thing?

Agree, I don't think the seamless modeset (ie: changing resolution without flicker) needs a property. Just need to test the commit without ALLOW_MODESET and commit it if the test passes.

Quoted 6 lines

Also what's the benefit of using video mode if your panel supportes command mode? Can you turn off the memory in the panel and actually save power that way? And if there is a benefit can't the driver just automagically switch between the two based on how often things are getting updated? That would match how eDP PSR already works.

I'm guessing video mode might have some latency benefits over command mode?

Sean

Quoted 4 lines
Signature · 3 lines
-- 
Ville Syrjälä
Intel
Signature · 5 lines
Sean Paul, Software Engineer, Google / Chromium OS
_______________________________________________
Freedreno mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/freedreno
#11 lore raw
Quoted 57 lines · Sean Paul
Quoted 5 lines · Ville Syrjälä
Quoted 4 lines · Sean Paul
Quoted 3 lines · Daniel Vetter
Quoted 2 lines · Jeykumar Sankaran
Quoted 1 line · Daniel Vetter

On Thu, Jul 11, 2019 at 11:46:44AM -0700, Jeykumar Sankaran wrote:

/snip

Quoted 48 lines
Quoted 33 lines
Quoted 20 lines
Quoted 12 lines
Quoted 11 lines
Quoted 1 line

drm: add mode flags in uapi for seamless mode switch

I think the uapi is the trivial part here, the real deal is how userspace
uses this. Can you pls post the patches for your compositor?

Also note that we already allow userspace to tell the kernel whether flickering is ok or not for a modeset. msm driver could use that to at least tell userspace whether a modeset change is possible. So you can already implement glitch-free modeset changes for at least video mode. -Daniel

Quoted 7 lines

I believe you are referring to the below tv property of the connector.

/**
* @tv_flicker_reduction_property: Optional TV property to control the
* flicker reduction mode.
*/
struct drm_property *tv_flicker_reduction_property;

Quoted 5 lines

Not even close :-)

I mean the DRM_MODE_ATOMIC_ALLOW_MODESET flag for the atomic ioctl. This is not a property of a mode, this is a property of a _transition_ between configurations. Some transitions can be done flicker free, others can't.

Agree that an atomic flag on a commit is the way to accomplish this. It's pretty similar to the psr transitions, where we want to reuse most of the atomic circuitry, but in a specialized way. We'd also have to be careful to only involve the drm objects which are seamless modeset aware (you could imagine a bridge chain where the bridges downstream of the first bridge don't care).

Quoted 1 line
Quoted 7 lines
Quoted 3 lines

There's then still the question of how to pick video vs command mode, but imo better to start with implementing the transition behaviour correctly first.

Connector property? Possibly a terrible idea, but I wonder if we could [re]use the vrr properties for command mode. The docs state that the driver has the option of putting upper and lower bounds on the refresh rate.

Not really sure why this needs new props and whatnot. This is kinda what the i915 "fastset" stuff already does:
1. userspace asks for something to be changed via atomic
2. driver calculates whether a modeset is actually required
3. atomic validates need_modeset() vs. DRM_MODE_ATOMIC_ALLOW_MODESET
4. if (need_modeset) heavyweight_commit() else lightweight_commit()

Quoted 7 lines
Quoted 2 lines

Ie. why should userspace really care about anything except the "flickers are OK" vs. "flickers not wanted" thing?

Agree, I don't think the seamless modeset (ie: changing resolution without flicker) needs a property. Just need to test the commit without ALLOW_MODESET and commit it if the test passes.

Agreed that a TEST_ONLY commit without ALLOW_MODESET flag can be used to check whether the modeset can be done seamless. But since there are no error code returns, the client cannot distinguish the test_only commit failures from other invalid config failures.

Also, note that when the client sees two 1080p modes (vid/cmd) and it is interested only to do *only* seamless switches, without any additional flag it cannot distinguish between these two 1080p modes. The client has to invoke two test_only commits with these modes to identify the seamless one. Is that a preferred approach?

Intel's "fastset" calculates the need for modeset internally based on the configuration and chooses the best commit path. But the requirement here is to expose the information up-front since the use case cannot afford to fall back to the normal modeset when it has requested for a seamless one.

Quoted 10 lines
Quoted 6 lines

Also what's the benefit of using video mode if your panel supportes command mode? Can you turn off the memory in the panel and actually save power that way? And if there is a benefit can't the driver just automagically switch between the two based on how often things are getting updated? That would match how eDP PSR already works.

I'm guessing video mode might have some latency benefits over command mode?

Sean

Yes. Pretty much those are reasons we need to switch to video mode. But instead of making the decision internal to the driver based on the frequency of frame updates, we have proprietary use cases where the client has to trigger the switch explicitly. So we are trying to find ways to represent the same resolution in both video/cmd modes.

Thanks and Regards,
Jeykumar S.

Quoted 5 lines
Quoted 5 lines

--
Ville Syrjälä
Intel

Signature · 5 lines
Jeykumar S
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#12 lore raw
Quoted 104 lines · Jeykumar Sankaran
Quoted 58 lines · Sean Paul
Quoted 5 lines · Ville Syrjälä
Quoted 4 lines · Sean Paul
Quoted 3 lines · Daniel Vetter
Quoted 2 lines · Jeykumar Sankaran
Quoted 1 line · Daniel Vetter

On Thu, Jul 11, 2019 at 11:46:44AM -0700, Jeykumar Sankaran wrote:

/snip

Quoted 49 lines
Quoted 33 lines
Quoted 20 lines
Quoted 12 lines
Quoted 11 lines
Quoted 1 line

drm: add mode flags in uapi for seamless mode switch

I think the uapi is the trivial part here, the real deal is how userspace
uses this. Can you pls post the patches for your compositor?

Also note that we already allow userspace to tell the kernel whether flickering is ok or not for a modeset. msm driver could use that to at least tell userspace whether a modeset change is possible. So you can already implement glitch-free modeset changes for at least video mode. -Daniel

Quoted 7 lines

I believe you are referring to the below tv property of the connector.

/**
* @tv_flicker_reduction_property: Optional TV property to control the
* flicker reduction mode.
*/
struct drm_property *tv_flicker_reduction_property;

Quoted 5 lines

Not even close :-)

I mean the DRM_MODE_ATOMIC_ALLOW_MODESET flag for the atomic ioctl. This is not a property of a mode, this is a property of a _transition_ between configurations. Some transitions can be done flicker free, others can't.

Agree that an atomic flag on a commit is the way to accomplish this. It's pretty similar to the psr transitions, where we want to reuse most of the atomic circuitry, but in a specialized way. We'd also have to be careful to only involve the drm objects which are seamless modeset aware (you could imagine a bridge chain where the bridges downstream of the first bridge don't care).

Quoted 1 line
Quoted 7 lines
Quoted 3 lines

There's then still the question of how to pick video vs command mode, but imo better to start with implementing the transition behaviour correctly first.

Connector property? Possibly a terrible idea, but I wonder if we could [re]use the vrr properties for command mode. The docs state that the driver has the option of putting upper and lower bounds on the refresh rate.

Not really sure why this needs new props and whatnot. This is kinda what
the i915 "fastset" stuff already does:
1. userspace asks for something to be changed via atomic
2. driver calculates whether a modeset is actually required
3. atomic validates need_modeset() vs. DRM_MODE_ATOMIC_ALLOW_MODESET
4. if (need_modeset) heavyweight_commit() else lightweight_commit()

Quoted 9 lines
Quoted 2 lines

Ie. why should userspace really care about anything except the "flickers are OK" vs. "flickers not wanted" thing?

Agree, I don't think the seamless modeset (ie: changing resolution without
flicker) needs a property. Just need to test the commit without ALLOW_MODESET
and commit it if the test passes.

Agreed that a TEST_ONLY commit without ALLOW_MODESET flag can be used to check
whether the modeset can be done seamless. But since there are no error code returns,
the client cannot distinguish the test_only commit failures from other invalid config failures.

Also, note that when the client sees two 1080p modes (vid/cmd) and it is interested only
to do *only* seamless switches, without any additional flag it cannot distinguish between
these two 1080p modes. The client has to invoke two test_only commits with these
modes to identify the seamless one. Is that a preferred approach?

Intel's "fastset" calculates the need for modeset internally based on the
configuration and chooses the best commit path. But the requirement here is to expose the information up-front since the use case cannot afford to fall back to the normal modeset when it has requested for a seamless one.

Quoted 11 lines
Quoted 6 lines

Also what's the benefit of using video mode if your panel supportes command mode? Can you turn off the memory in the panel and actually save power that way? And if there is a benefit can't the driver just automagically switch between the two based on how often things are getting updated? That would match how eDP PSR already works.

I'm guessing video mode might have some latency benefits over command mode?

Sean

Quoted 8 lines

Yes. Pretty much those are reasons we need to switch to video mode. But instead
of making the decision internal to the driver based on the frequency of frame updates,
we have proprietary use cases where the client has to trigger the switch explicitly.
So we are trying to find ways to represent the same resolution in both video/cmd modes.

If "proprietary" here means closed source or not upstream, then that's the part you need to fix first. See

https://dri.freedesktop.org/docs/drm/gpu/drm-uapi.html#open-source-userspace-requirements

Cheers, Daniel

Quoted 12 lines

Thanks and Regards,
Jeykumar S.

Quoted 5 lines
Quoted 5 lines

--
Ville Syrjälä
Intel

--
Jeykumar S

Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#13 lore raw
In reply to Jeykumar Sankaran
Quoted 76 lines · Jeykumar Sankaran
Quoted 58 lines · Sean Paul
Quoted 5 lines · Ville Syrjälä
Quoted 4 lines · Sean Paul
Quoted 3 lines · Daniel Vetter
Quoted 2 lines · Jeykumar Sankaran
Quoted 1 line · Daniel Vetter

On Thu, Jul 11, 2019 at 11:46:44AM -0700, Jeykumar Sankaran wrote:

/snip

Quoted 49 lines
Quoted 33 lines
Quoted 20 lines
Quoted 12 lines
Quoted 11 lines
Quoted 1 line

drm: add mode flags in uapi for seamless mode switch

I think the uapi is the trivial part here, the real deal is how userspace
uses this. Can you pls post the patches for your compositor?

Also note that we already allow userspace to tell the kernel whether flickering is ok or not for a modeset. msm driver could use that to at least tell userspace whether a modeset change is possible. So you can already implement glitch-free modeset changes for at least video mode. -Daniel

Quoted 7 lines

I believe you are referring to the below tv property of the connector.

/**
* @tv_flicker_reduction_property: Optional TV property to control the
* flicker reduction mode.
*/
struct drm_property *tv_flicker_reduction_property;

Quoted 5 lines

Not even close :-)

I mean the DRM_MODE_ATOMIC_ALLOW_MODESET flag for the atomic ioctl. This is not a property of a mode, this is a property of a _transition_ between configurations. Some transitions can be done flicker free, others can't.

Agree that an atomic flag on a commit is the way to accomplish this. It's pretty similar to the psr transitions, where we want to reuse most of the atomic circuitry, but in a specialized way. We'd also have to be careful to only involve the drm objects which are seamless modeset aware (you could imagine a bridge chain where the bridges downstream of the first bridge don't care).

Quoted 1 line
Quoted 7 lines
Quoted 3 lines

There's then still the question of how to pick video vs command mode, but imo better to start with implementing the transition behaviour correctly first.

Connector property? Possibly a terrible idea, but I wonder if we could [re]use the vrr properties for command mode. The docs state that the driver has the option of putting upper and lower bounds on the refresh rate.

Not really sure why this needs new props and whatnot. This is kinda what
the i915 "fastset" stuff already does:
1. userspace asks for something to be changed via atomic
2. driver calculates whether a modeset is actually required
3. atomic validates need_modeset() vs. DRM_MODE_ATOMIC_ALLOW_MODESET
4. if (need_modeset) heavyweight_commit() else lightweight_commit()

Quoted 9 lines
Quoted 2 lines

Ie. why should userspace really care about anything except the "flickers are OK" vs. "flickers not wanted" thing?

Agree, I don't think the seamless modeset (ie: changing resolution without
flicker) needs a property. Just need to test the commit without ALLOW_MODESET
and commit it if the test passes.

Agreed that a TEST_ONLY commit without ALLOW_MODESET flag can be used to check
whether the modeset can be done seamless. But since there are no error code returns,
the client cannot distinguish the test_only commit failures from other invalid config failures.

Quoted 7 lines

Also, note that when the client sees two 1080p modes (vid/cmd) and it is interested only
to do *only* seamless switches, without any additional flag it cannot distinguish between
these two 1080p modes. The client has to invoke two test_only commits with these
modes to identify the seamless one. Is that a preferred approach?

Hi Jey!
Yeah, pretty much. Stepping back a bit though, why is the kernel exposing 2 1080p modes in the first place? If you just expose one mode and then use a property to enter "low-latency operation" (or overloading vrr for cmd mode), you shouldn't need to do the mode switch, just flip the property and let the kernel figure out how to transition to video/cmd mode.

Quoted 16 lines

Intel's "fastset" calculates the need for modeset internally based on the configuration and chooses the best commit path. But the requirement here is to expose the information up-front since the use case cannot afford to fall back to the normal modeset when it has requested for a seamless one.

Quoted 10 lines
Quoted 6 lines

Also what's the benefit of using video mode if your panel supportes command mode? Can you turn off the memory in the panel and actually save power that way? And if there is a benefit can't the driver just automagically switch between the two based on how often things are getting updated? That would match how eDP PSR already works.

I'm guessing video mode might have some latency benefits over command mode?

Quoted 8 lines
Quoted 1 line

Sean

Yes. Pretty much those are reasons we need to switch to video mode. But instead
of making the decision internal to the driver based on the frequency of frame updates,
we have proprietary use cases where the client has to trigger the switch explicitly.

Unsolicited advice: if you find yourself typing "proprietary" while justifying an upstream proposal, reword your argument immediately :-)

Now that I've filled my awful joke quota for the mail, I can move on. Generally speaking, it's better to you expand on why userspace needs this. There's a very good chance that if it makes sense for Qualcomm, it makes sense for others. If we have an open discussion about it, others might chime in with "me too" and that will help your case (and you might even hook someone into doing the work for you!). Alternatively, we might be able to solve the problem in a different way that makes everyone happy (or most everyone happy).

There's 100% a case for allowing userspace to trigger a low-latency mode at the expense of power (we should have it disable the psr entry timer as well). Let's steer this conversation in the direction of "can we use something that already exists, or should we add something new?"

Sean

Quoted 14 lines

So we are trying to find ways to represent the same resolution in both video/cmd modes.

Thanks and Regards,
Jeykumar S.

Quoted 5 lines
Quoted 5 lines

--
Ville Syrjälä
Intel

Signature · 1 line
Jeykumar S
Signature · 5 lines
Sean Paul, Software Engineer, Google / Chromium OS
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#14 lore raw
Quoted 171 lines · Sean Paul
Quoted 5 lines · Jeykumar Sankaran
Quoted 4 lines · Sean Paul
Quoted 3 lines · Ville Syrjälä
Quoted 2 lines · Sean Paul
Quoted 1 line · Daniel Vetter

On Thu, Jul 18, 2019 at 11:18:42AM -0700, Jeykumar Sankaran

Quoted 2 lines · wrote:
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines · Daniel Vetter
Quoted 1 line

On Thu, Jul 11, 2019 at 11:46:44AM -0700, Jeykumar Sankaran

Quoted 6 lines · wrote:
Quoted 6 lines

/snip

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 1 line

drm: add mode flags in uapi for seamless mode switch

I think the uapi is the trivial part here, the real deal is

how

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

userspace
uses this. Can you pls post the patches for your compositor?

Also note that we already allow userspace to tell the kernel

whether

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

flickering is ok or not for a modeset. msm driver could use

that to at

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

least tell userspace whether a modeset change is possible.

So you can

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

already implement glitch-free modeset changes for at least

video mode.

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 1 line

-Daniel

I believe you are referring to the below tv property of the

connector.

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

/**
* @tv_flicker_reduction_property: Optional TV property to

control the

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 3 lines

* flicker reduction mode.
*/
struct drm_property *tv_flicker_reduction_property;

Not even close :-)

I mean the DRM_MODE_ATOMIC_ALLOW_MODESET flag for the atomic

ioctl. This

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

is not a property of a mode, this is a property of a

_transition_ between

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

configurations. Some transitions can be done flicker free,

others can't.

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

Agree that an atomic flag on a commit is the way to accomplish

this. It's pretty

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

similar to the psr transitions, where we want to reuse most of the

atomic

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

circuitry, but in a specialized way. We'd also have to be careful

to only

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

involve the drm objects which are seamless modeset aware (you

could imagine

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

a bridge chain where the bridges downstream of the first bridge

don't care).

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

There's then still the question of how to pick video vs command

mode, but

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

imo better to start with implementing the transition behaviour

correctly

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 1 line

first.

Connector property? Possibly a terrible idea, but I wonder if we

could [re]use

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

the vrr properties for command mode. The docs state that the

driver has the

Quoted 20 lines
Quoted 10 lines
Quoted 10 lines
Quoted 1 line

option of putting upper and lower bounds on the refresh rate.

Not really sure why this needs new props and whatnot. This is kinda what
the i915 "fastset" stuff already does:
1. userspace asks for something to be changed via atomic
2. driver calculates whether a modeset is actually required
3. atomic validates need_modeset() vs. DRM_MODE_ATOMIC_ALLOW_MODESET
4. if (need_modeset) heavyweight_commit() else lightweight_commit()

Quoted 9 lines
Quoted 2 lines

Ie. why should userspace really care about anything except the "flickers are OK" vs. "flickers not wanted" thing?

Agree, I don't think the seamless modeset (ie: changing resolution without
flicker) needs a property. Just need to test the commit without ALLOW_MODESET
and commit it if the test passes.

Quoted 3 lines

Agreed that a TEST_ONLY commit without ALLOW_MODESET flag can be used to check
whether the modeset can be done seamless. But since there are no error

code

Quoted 9 lines

returns,
the client cannot distinguish the test_only commit failures from other invalid config failures.

Also, note that when the client sees two 1080p modes (vid/cmd) and it is interested only
to do *only* seamless switches, without any additional flag it cannot distinguish between
these two 1080p modes. The client has to invoke two test_only commits

with

Quoted 2 lines

these
modes to identify the seamless one. Is that a preferred approach?

Hi Jey!
Yeah, pretty much. Stepping back a bit though, why is the kernel exposing 2
1080p modes in the first place? If you just expose one mode and then use a property to enter "low-latency operation" (or overloading vrr for cmd mode), you
shouldn't need to do the mode switch, just flip the property and let the kernel
figure out how to transition to video/cmd mode.

Quoted 2 lines

Intel's "fastset" calculates the need for modeset internally based on

the

Quoted 3 lines

configuration and chooses the best commit path. But the requirement here is to expose the information up-front since the use case cannot afford to fall back to the normal modeset when it has requested for a seamless

one.

Quoted 11 lines
Quoted 11 lines
Quoted 7 lines

Also what's the benefit of using video mode if your panel supportes command mode? Can you turn off the memory in the panel and actually save power that way? And if there is a benefit can't the driver just automagically switch between the two based on how often things are getting updated? That would match how eDP PSR already works.

I'm guessing video mode might have some latency benefits over command mode?

Quoted 8 lines
Quoted 1 line

Sean

Yes. Pretty much those are reasons we need to switch to video mode. But instead
of making the decision internal to the driver based on the frequency of frame updates,
we have proprietary use cases where the client has to trigger the switch explicitly.

Unsolicited advice: if you find yourself typing "proprietary" while justifying
an upstream proposal, reword your argument immediately :-)

Now that I've filled my awful joke quota for the mail, I can move on. Generally
speaking, it's better to you expand on why userspace needs this. There's a very
good chance that if it makes sense for Qualcomm, it makes sense for others. If
we have an open discussion about it, others might chime in with "me too" and
that will help your case (and you might even hook someone into doing the work
for you!). Alternatively, we might be able to solve the problem in a different
way that makes everyone happy (or most everyone happy).

There's 100% a case for allowing userspace to trigger a low-latency mode at the
expense of power (we should have it disable the psr entry timer as well). Let's
steer this conversation in the direction of "can we use something that already
exists, or should we add something new?"

Quoted 2 lines

Sean

Thank you Sean for the suggestion! I can use the proposed property approach to switch between vid/cmd mode if every resolution is capable of switching between these two modes. But we have uses cases where the switch is restricted only between a set of resolutions.

Also, appreciate your guidance for formating closed usecase questions. Let me check how much more details I can provide on why H/W has to enforce such restrictions and will revive the thread if needed.

Thanks and Regards,
Jeykumar S.

Quoted 18 lines
Quoted 9 lines

So we are trying to find ways to represent the same resolution in both video/cmd modes.

Thanks and Regards,
Jeykumar S.

Quoted 2 lines
Quoted 2 lines
Quoted 6 lines
Quoted 3 lines
Quoted 3 lines

--
Ville Syrjälä
Intel

--
Jeykumar S

--
Sean Paul, Software Engineer, Google / Chromium OS

Signature · 5 lines
Jeykumar S
_______________________________________________
Freedreno mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/freedreno