Use no_vblank property for drivers without VBLANK
Subsystems (12)
Thomas Zimmermann
maintainerv3 cover Use no_vblank property for drivers without VBLANK
Instead of faking VBLANK events by themselves, drivers without VBLANK support can enable drm_crtc_vblank.no_vblank and let DRM do the rest. The patchset makes this official and converts over drivers.
The current implementation looks at the number of initialized CRTCs wrt vblanking. If vblanking has been initialized for a CRTC, the driver is responsible for sending out VBLANK events. Otherwise, DRM will send out the event. The behaviour selected by initializing no_vblank as part of drm_atomic_helper_check_modeset().
I went through all drivers, looking for those that call send out VBLANK events but do not call drm_vblank_init(). These are converted to the new semantics. This affects tiny drivers; drivers for virtual hardware; and a few others, which do not support interrupts. Xen comes with its own VBLANK logic and disables no_vblank explictly.
For now, I left out Hans' R-b on v2 of the series, as the patches changed quite a bit.
v3:
* reorder and squash patches
* set no_vblank in drm_atomic_helper_check_modeset() for *all* drivers (Daniel)
* convert all drivers to new semnatics as necessary
v2:
* document functionality (Daniel)
* cleanup ast (Daniel)
* let simple-kms handle no_vblank where possible
Thomas Zimmermann (4):
drm: Add drm_crtc_has_vblank()
drm: Initialize struct drm_crtc_state.no_vblank from device settings drm/ast: Don't set struct drm_crtc_state.no_vblank explictly drm/udl: Don't set struct drm_crtc_state.no_vblank explictly
drivers/gpu/drm/arc/arcpgu_crtc.c | 16 ------------- drivers/gpu/drm/ast/ast_mode.c | 2 -- drivers/gpu/drm/bochs/bochs_kms.c | 9 ------- drivers/gpu/drm/cirrus/cirrus.c | 8 ------- drivers/gpu/drm/drm_atomic_helper.c | 10 +++++++- drivers/gpu/drm/drm_mipi_dbi.c | 9 ------- drivers/gpu/drm/drm_vblank.c | 30 ++++++++++++++++++++++++ drivers/gpu/drm/qxl/qxl_display.c | 14 ----------- drivers/gpu/drm/tiny/gm12u320.c | 9 ------- drivers/gpu/drm/tiny/ili9225.c | 9 ------- drivers/gpu/drm/tiny/repaper.c | 9 ------- drivers/gpu/drm/tiny/st7586.c | 9 ------- drivers/gpu/drm/udl/udl_modeset.c | 11 --------- drivers/gpu/drm/vboxvideo/vbox_mode.c | 12 ---------- drivers/gpu/drm/virtio/virtgpu_display.c | 8 ------- drivers/gpu/drm/xen/xen_drm_front_kms.c | 13 ++++++++++ include/drm/drm_crtc.h | 27 +++++++++++++++------ include/drm/drm_simple_kms_helper.h | 7 ++++-- include/drm/drm_vblank.h | 1 + 19 files changed, 78 insertions(+), 135 deletions(-)
--
2.24.1
Thomas Zimmermann
maintainerPATCH v3 1/4 drm: Add drm_crtc_has_vblank()
The new interface drm_crtc_has_vblank() return true if vblanking has been initialized for a certain CRTC, or false otherwise. This function will be useful for initializing CRTC state.
--- drivers/gpu/drm/drm_vblank.c | 21 +++++++++++++++++++++ include/drm/drm_vblank.h | 1 + 2 files changed, 22 insertions(+)
drivers/gpu/drm/drm_vblank.c +21 -0
@@ -501,6 +501,27 @@ int drm_vblank_init(struct drm_device *dev, unsigned int num_crtcs) }EXPORT_SYMBOL (drm_vblank_init); +/** +* drm_crtc_has_vblank - test if vblanking has been initialized for +* a CRTC +* @crtc: the CRTC +* +* Drivers may call this function to test if vblank support is +* initialized for a CRTC. For most hardware this means that vblanking +* can also be enabled on the CRTC. +* +* Returns: +* True if vblanking has been initialized for the given CRTC, false +* otherwise. +*/ +bool drm_crtc_has_vblank (const struct drm_crtc *crtc) +{ +struct drm_device *dev = crtc->dev; + +return crtc->index < dev->num_crtcs; +} +EXPORT_SYMBOL (drm_crtc_has_vblank); +/** * drm_crtc_vblank_waitqueue - get vblank waitqueue for the CRTC * @crtc: which CRTC's vblank waitqueue to retrieve
include/drm/drm_vblank.h +1 -0
@@ -206,6 +206,7 @@ struct drm_vblank_crtc { };int drm_vblank_init (struct drm_device *dev,unsigned int num_crtcs); +bool drm_crtc_has_vblank (const struct drm_crtc *crtc);u64 drm_crtc_vblank_count (struct drm_crtc *crtc);u64 drm_crtc_vblank_count_and_time (struct drm_crtc *crtc,ktime_t *vblanktime);
Signature · 1 line
2.24.1
Re: [PATCH v3 1/4] drm: Add drm_crtc_has_vblank()
Quoted 22 lines · Thomas Zimmermann
The new interface
drm_crtc_has_vblank()return true if vblanking has been initialized for a certain CRTC, or false otherwise. This function will be useful for initializing CRTC state.Signed-off-by Thomas Zimmermann <[email protected]>--- drivers/gpu/drm/drm_vblank.c | 21 +++++++++++++++++++++ include/drm/drm_vblank.h | 1 + 2 files changed, 22 insertions(+)drivers/gpu/drm/drm_vblank.c +4
-0@@ -501,6 +501,27 @@ int drm_vblank_init(struct drm_device *dev, unsigned int num_crtcs) }EXPORT_SYMBOL (drm_vblank_init); +/** +* drm_crtc_has_vblank - test if vblanking has been initialized for +* a CRTC +* @crtc: the CRTC
drivers/gpu/drm/drm_vblank.c int drm_vblank_init(struct drm_device *dev, unsigned int num_crtcs) · line 517
Quoted 10 lines
drivers/gpu/drm/drm_vblank.c +10
-0+* +* Drivers may call this function to test if vblank support is +* initialized for a CRTC. For most hardware this means that vblanking +* can also be enabled on the CRTC. +* +* Returns: +* True if vblanking has been initialized for the given CRTC, false +* otherwise. +*/ +bool drm_crtc_has_vblank (const struct drm_crtc *crtc)
So making this specific to a CRTC sounds like a good idea. But it's not the reality, drm_vblank.c assumes that either everything or nothing supports vblanks.
The reason for dev->num_crtcs is historical baggage, it predates kms by a few years. For kms drivers the only two valid values are either 0 or dev->mode_config.num_crtcs. Yes that's an entire different can of worms that's been irking me since forever (ideally drm_vblank_init would somehow loose the num_crtcs argument for kms drivers, but some drivers call this before they've done all the drm_crtc_init calls so it's complicated).
Hence drm_dev_has_vblank as I suggested. That would also allow you to replace a bunch of if (dev->num_crtcs) checks in drm_vblank.c, which should help quite a bit in code readability.
Cheers, Daniel
Quoted 25 lines
drivers/gpu/drm/drm_vblank.c +7
-0+{ +struct drm_device *dev = crtc->dev; + +return crtc->index < dev->num_crtcs; +} +EXPORT_SYMBOL (drm_crtc_has_vblank); +/** * drm_crtc_vblank_waitqueue - get vblank waitqueue for the CRTC * @crtc: which CRTC's vblank waitqueue to retrieve include/drm/drm_vblank.h +1
-0@@ -206,6 +206,7 @@ struct drm_vblank_crtc { };int drm_vblank_init (struct drm_device *dev,unsigned int num_crtcs); +bool drm_crtc_has_vblank (const struct drm_crtc *crtc);u64 drm_crtc_vblank_count (struct drm_crtc *crtc);u64 drm_crtc_vblank_count_and_time (struct drm_crtc *crtc,ktime_t *vblanktime);Signature · 1 line
2.24.1
Signature · 8 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Thomas Zimmermann
maintainerRe: [PATCH v3 1/4] drm: Add drm_crtc_has_vblank()
Hi
Am 22.01.20 um 09:31 schrieb Daniel Vetter:
Quoted 38 lines · Thomas Zimmermann
Quoted 22 lines
The new interface
drm_crtc_has_vblank()return true if vblanking has been initialized for a certain CRTC, or false otherwise. This function will be useful for initializing CRTC state.Signed-off-by Thomas Zimmermann <[email protected]>--- drivers/gpu/drm/drm_vblank.c | 21 +++++++++++++++++++++ include/drm/drm_vblank.h | 1 + 2 files changed, 22 insertions(+)drivers/gpu/drm/drm_vblank.c +4
-0@@ -501,6 +501,27 @@ int drm_vblank_init(struct drm_device *dev, unsigned int num_crtcs) }EXPORT_SYMBOL (drm_vblank_init); +/** +* drm_crtc_has_vblank - test if vblanking has been initialized for +* a CRTC +* @crtc: the CRTC drivers/gpu/drm/drm_vblank.c int drm_vblank_init(struct drm_device *dev, unsigned int num_crtcs) · line 517
Quoted 10 lines
drivers/gpu/drm/drm_vblank.c +10
-0+* +* Drivers may call this function to test if vblank support is +* initialized for a CRTC. For most hardware this means that vblanking +* can also be enabled on the CRTC. +* +* Returns: +* True if vblanking has been initialized for the given CRTC, false +* otherwise. +*/ +bool drm_crtc_has_vblank (const struct drm_crtc *crtc)So making this specific to a CRTC sounds like a good idea. But it's not the reality,
drm_vblank.cassumes that either everything or nothing supports vblanks.
Quoted 6 lines
The reason for
dev->num_crtcsis historical baggage, it predates kms by a few years. For kms drivers the only two valid values are either 0 ordev->mode_config.num_crtcs. Yes that's an entire different can of worms that's been irking me since forever (ideallydrm_vblank_initwould somehow loose thenum_crtcsargument for kms drivers, but some drivers call this before they've done all thedrm_crtc_initcalls so it's complicated).
Maybe as a first step, drm_vblank_init() could use dev->mode_config.num_crtcs if the supplied number of CRTCs is zero.
Quoted 4 lines
Hence
drm_dev_has_vblankas I suggested. That would also allow you to replace a bunch of if (dev->num_crtcs) checks indrm_vblank.c, which should help quite a bit in code readability.
OK, but I still don't understand why this interface is better overall. We don't loose anything by passing in the crtc instead of the device structure. And if there's ever a per-crtc vblank initialization, we'd have the interface in place already. The tests with "if (dev->num_crtcs)" could probably be removed in most places in any case.
We should also consider forking the vblank code for non-KMS drivers. While working in this, I found the support for legacy drivers is getting in the way at times. With such a fork, legacy drivers could continue using struct drm_vblank_crtc, while modern drivers could maybe store vblank state directly in struct drm_crtc.
Anyway, all this is for another patch. Unless you change your mind, I'll replace drm_crtc_has_vblank() with drm_dev_has_vblank() for the patchset's next iteration.
Best regards
Thomas
Quoted 29 lines
Cheers, Daniel
Quoted 25 lines
drivers/gpu/drm/drm_vblank.c +7
-0+{ +struct drm_device *dev = crtc->dev; + +return crtc->index < dev->num_crtcs; +} +EXPORT_SYMBOL (drm_crtc_has_vblank); +/** * drm_crtc_vblank_waitqueue - get vblank waitqueue for the CRTC * @crtc: which CRTC's vblank waitqueue to retrieve include/drm/drm_vblank.h +1
-0@@ -206,6 +206,7 @@ struct drm_vblank_crtc { };int drm_vblank_init (struct drm_device *dev,unsigned int num_crtcs); +bool drm_crtc_has_vblank (const struct drm_crtc *crtc);u64 drm_crtc_vblank_count (struct drm_crtc *crtc);u64 drm_crtc_vblank_count_and_time (struct drm_crtc *crtc,ktime_t *vblanktime);Signature · 1 line
2.24.1
Signature · 6 lines
Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Re: [PATCH v3 1/4] drm: Add drm_crtc_has_vblank()
Quoted 56 lines · Thomas Zimmermann
Hi
Am 22.01.20 um 09:31 schrieb Daniel Vetter:
Quoted 38 lines · Thomas Zimmermann
Quoted 22 lines
The new interface
drm_crtc_has_vblank()return true if vblanking has been initialized for a certain CRTC, or false otherwise. This function will be useful for initializing CRTC state.Signed-off-by Thomas Zimmermann <[email protected]>--- drivers/gpu/drm/drm_vblank.c | 21 +++++++++++++++++++++ include/drm/drm_vblank.h | 1 + 2 files changed, 22 insertions(+)drivers/gpu/drm/drm_vblank.c +4
-0@@ -501,6 +501,27 @@ int drm_vblank_init(struct drm_device *dev, unsigned int num_crtcs) }EXPORT_SYMBOL (drm_vblank_init); +/** +* drm_crtc_has_vblank - test if vblanking has been initialized for +* a CRTC +* @crtc: the CRTC drivers/gpu/drm/drm_vblank.c int drm_vblank_init(struct drm_device *dev, unsigned int num_crtcs) · line 517
Quoted 10 lines
drivers/gpu/drm/drm_vblank.c +10
-0+* +* Drivers may call this function to test if vblank support is +* initialized for a CRTC. For most hardware this means that vblanking +* can also be enabled on the CRTC. +* +* Returns: +* True if vblanking has been initialized for the given CRTC, false +* otherwise. +*/ +bool drm_crtc_has_vblank (const struct drm_crtc *crtc)So making this specific to a CRTC sounds like a good idea. But it's not the reality,
drm_vblank.cassumes that either everything or nothing supports vblanks.Quoted 6 lines
The reason for
dev->num_crtcsis historical baggage, it predates kms by a few years. For kms drivers the only two valid values are either 0 ordev->mode_config.num_crtcs. Yes that's an entire different can of worms that's been irking me since forever (ideallydrm_vblank_initwould somehow loose thenum_crtcsargument for kms drivers, but some drivers call this before they've done all thedrm_crtc_initcalls so it's complicated).Maybe as a first step,
drm_vblank_init()could usedev->mode_config.num_crtcsif the supplied number of CRTCs is zero.Quoted 4 lines
Hence
drm_dev_has_vblankas I suggested. That would also allow you to replace a bunch of if (dev->num_crtcs) checks indrm_vblank.c, which should help quite a bit in code readability.
Quoted 5 lines
OK, but I still don't understand why this interface is better overall. We don't loose anything by passing in the crtc instead of the device structure. And if there's ever a per-crtc vblank initialization, we'd have the interface in place already. The tests with "if (
dev->num_crtcs)" could probably be removed in most places in any case.
You can't use it in drm_vblank.c code, because we only have the drm_device, not the drm_crtc (in most places at least). Your other patch series to deprecate the drm_device callbacks for vblanks is a huge step into the direction to fix that, but still more work needed: We'd essentially need to copypaste drm_vblank.c into drm_crtc_vblank.c for kms drivers, and in that copy switch from (dev, pipe) to crtc everywhere. Plus then move the drm_vblank structure into struct drm_crtc.
Wrt removing the check: In a pile of cases it changes the return value, which matters both for vblank usage in helper code and the ioctl itself. From a quick look most of the checks that don't matter are already wrapped in a WARN.
Quoted 5 lines
We should also consider forking the vblank code for non-KMS drivers. While working in this, I found the support for legacy drivers is getting in the way at times. With such a fork, legacy drivers could continue using struct
drm_vblank_crtc, while modern drivers could maybe store vblank state directly in structdrm_crtc.
Hm if you want to do all that then the drm_crtc_has_vblank makes sense. But only after we've done the full split. So maybe make the public function drm_crtc_has_vblank, which calls the internal-only drm_has_vblank, and use that internally in drm_vblank.c?
btw I still think a sub-struct for vblank stuff in drm_crtc makes sense, and drm_vblank_crtc seems to mostly fit the bill.
That way we're at least not adding the the conversion pain of switching the vblank code over to drm_crtc fully.
Thoughts?
-Daniel
Quoted 45 lines
Anyway, all this is for another patch. Unless you change your mind, I'll replace
drm_crtc_has_vblank()withdrm_dev_has_vblank()for the patchset's next iteration.Best regards
ThomasQuoted 29 lines
Cheers, Daniel
Quoted 25 lines
drivers/gpu/drm/drm_vblank.c +7
-0+{ +struct drm_device *dev = crtc->dev; + +return crtc->index < dev->num_crtcs; +} +EXPORT_SYMBOL (drm_crtc_has_vblank); +/** * drm_crtc_vblank_waitqueue - get vblank waitqueue for the CRTC * @crtc: which CRTC's vblank waitqueue to retrieve include/drm/drm_vblank.h +1
-0@@ -206,6 +206,7 @@ struct drm_vblank_crtc { };int drm_vblank_init (struct drm_device *dev,unsigned int num_crtcs); +bool drm_crtc_has_vblank (const struct drm_crtc *crtc);u64 drm_crtc_vblank_count (struct drm_crtc *crtc);u64 drm_crtc_vblank_count_and_time (struct drm_crtc *crtc,ktime_t *vblanktime);Signature · 1 line
2.24.1Signature · 6 lines
Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Thomas Zimmermann
maintainerRe: [PATCH v3 1/4] drm: Add drm_crtc_has_vblank()
Hi
Am 22.01.20 um 10:04 schrieb Daniel Vetter:
Quoted 87 lines · Thomas Zimmermann
Quoted 56 lines
Hi
Am 22.01.20 um 09:31 schrieb Daniel Vetter:
Quoted 38 lines · Thomas Zimmermann
Quoted 22 lines
The new interface
drm_crtc_has_vblank()return true if vblanking has been initialized for a certain CRTC, or false otherwise. This function will be useful for initializing CRTC state.Signed-off-by Thomas Zimmermann <[email protected]>--- drivers/gpu/drm/drm_vblank.c | 21 +++++++++++++++++++++ include/drm/drm_vblank.h | 1 + 2 files changed, 22 insertions(+)drivers/gpu/drm/drm_vblank.c +4
-0@@ -501,6 +501,27 @@ int drm_vblank_init(struct drm_device *dev, unsigned int num_crtcs) }EXPORT_SYMBOL (drm_vblank_init); +/** +* drm_crtc_has_vblank - test if vblanking has been initialized for +* a CRTC +* @crtc: the CRTC drivers/gpu/drm/drm_vblank.c int drm_vblank_init(struct drm_device *dev, unsigned int num_crtcs) · line 517
Quoted 10 lines
drivers/gpu/drm/drm_vblank.c +10
-0+* +* Drivers may call this function to test if vblank support is +* initialized for a CRTC. For most hardware this means that vblanking +* can also be enabled on the CRTC. +* +* Returns: +* True if vblanking has been initialized for the given CRTC, false +* otherwise. +*/ +bool drm_crtc_has_vblank (const struct drm_crtc *crtc)So making this specific to a CRTC sounds like a good idea. But it's not the reality,
drm_vblank.cassumes that either everything or nothing supports vblanks.Quoted 6 lines
The reason for
dev->num_crtcsis historical baggage, it predates kms by a few years. For kms drivers the only two valid values are either 0 ordev->mode_config.num_crtcs. Yes that's an entire different can of worms that's been irking me since forever (ideallydrm_vblank_initwould somehow loose thenum_crtcsargument for kms drivers, but some drivers call this before they've done all thedrm_crtc_initcalls so it's complicated).Maybe as a first step,
drm_vblank_init()could usedev->mode_config.num_crtcsif the supplied number of CRTCs is zero.Quoted 4 lines
Hence
drm_dev_has_vblankas I suggested. That would also allow you to replace a bunch of if (dev->num_crtcs) checks indrm_vblank.c, which should help quite a bit in code readability.Quoted 5 lines
OK, but I still don't understand why this interface is better overall. We don't loose anything by passing in the crtc instead of the device structure. And if there's ever a per-crtc vblank initialization, we'd have the interface in place already. The tests with "if (
dev->num_crtcs)" could probably be removed in most places in any case.You can't use it in
drm_vblank.ccode, because we only have thedrm_device, not thedrm_crtc(in most places at least). Your other patch series to deprecate thedrm_devicecallbacks for vblanks is a huge step into the direction to fix that, but still more work needed: We'd essentially need to copypastedrm_vblank.cintodrm_crtc_vblank.cfor kms drivers, and in that copy switch from (dev, pipe) to crtc everywhere. Plus then move thedrm_vblankstructure into structdrm_crtc.Wrt removing the check: In a pile of cases it changes the return value, which matters both for vblank usage in helper code and the ioctl itself. From a quick look most of the checks that don't matter are already wrapped in a WARN.
Quoted 5 lines
We should also consider forking the vblank code for non-KMS drivers. While working in this, I found the support for legacy drivers is getting in the way at times. With such a fork, legacy drivers could continue using struct
drm_vblank_crtc, while modern drivers could maybe store vblank state directly in structdrm_crtc.Hm if you want to do all that then the
drm_crtc_has_vblankmakes sense. But only after we've done the full split. So maybe make the public functiondrm_crtc_has_vblank, which calls the internal-onlydrm_has_vblank, and use that internally indrm_vblank.c?
Quoted 7 lines
btw I still think a sub-struct for vblank stuff in
drm_crtcmakes sense, anddrm_vblank_crtcseems to mostly fit the bill.That way we're at least not adding the the conversion pain of switching the vblank code over to
drm_crtcfully.Thoughts?
That all sounds good. Using struct drm_vblank_crtc with legacy and modern vblank functions, might allow us to continue to share some of the implementation.
Wrt the current interface, drm_dev_has_vblank() is only called in a single place, so switching to drm_crtc_has_vblank() later would not be hard.
Best regards
Thomas
Quoted 51 lines
-Daniel
Quoted 45 lines
Anyway, all this is for another patch. Unless you change your mind, I'll replace
drm_crtc_has_vblank()withdrm_dev_has_vblank()for the patchset's next iteration.Best regards
ThomasQuoted 29 lines
Cheers, Daniel
Quoted 25 lines
drivers/gpu/drm/drm_vblank.c +7
-0+{ +struct drm_device *dev = crtc->dev; + +return crtc->index < dev->num_crtcs; +} +EXPORT_SYMBOL (drm_crtc_has_vblank); +/** * drm_crtc_vblank_waitqueue - get vblank waitqueue for the CRTC * @crtc: which CRTC's vblank waitqueue to retrieve include/drm/drm_vblank.h +1
-0@@ -206,6 +206,7 @@ struct drm_vblank_crtc { };int drm_vblank_init (struct drm_device *dev,unsigned int num_crtcs); +bool drm_crtc_has_vblank (const struct drm_crtc *crtc);u64 drm_crtc_vblank_count (struct drm_crtc *crtc);u64 drm_crtc_vblank_count_and_time (struct drm_crtc *crtc,ktime_t *vblanktime);Signature · 1 line
2.24.1Signature · 6 lines
Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer
Signature · 6 lines
Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Thomas Zimmermann
maintainerPATCH v3 2/4 drm: Initialize struct drm_crtc_state.no_vblank from device settings
In reply to Thomas ZimmermannAt the end of a commit, atomic helpers can generate a VBLANK event automatically. Originally implemented for writeback connectors, the functionality can be used by any driver and/or hardware without proper VBLANK interrupt.
First of all, the patch updates the documentation to make this behaviour
official: settings struct drm_crtc_state.no_vblank to true enables automatic VBLANK generation.
Atomic modesetting helper set the initial value of no_vblank in drm_atomic_helper_check_modeset(). If vblanking has been initialized for a CRTC, no_blank is disabled. Otherwise it's enabled. Hence, atomic helpers will automatically send out VBLANK events with any driver that did not initialize vblanking.
As drivers previously send out VBLANK events by themselves, all affected drivers have to be updated as well. Usually, deleting the driver's vblanking code is sufficient. Xen implements its own logic for generating events and therefore needs to override no_vblank with a value of false.
v3:
* squash all related changes patches into this patch
--- drivers/gpu/drm/arc/arcpgu_crtc.c | 16 -------------- drivers/gpu/drm/bochs/bochs_kms.c | 9 -------- drivers/gpu/drm/cirrus/cirrus.c | 8 ------- drivers/gpu/drm/drm_atomic_helper.c | 10 ++++++++- drivers/gpu/drm/drm_mipi_dbi.c | 9 -------- drivers/gpu/drm/drm_vblank.c | 9 ++++++++ drivers/gpu/drm/qxl/qxl_display.c | 14 ------------ drivers/gpu/drm/tiny/gm12u320.c | 9 -------- drivers/gpu/drm/tiny/ili9225.c | 9 -------- drivers/gpu/drm/tiny/repaper.c | 9 -------- drivers/gpu/drm/tiny/st7586.c | 9 -------- drivers/gpu/drm/vboxvideo/vbox_mode.c | 12 ----------- drivers/gpu/drm/virtio/virtgpu_display.c | 8 ------- drivers/gpu/drm/xen/xen_drm_front_kms.c | 13 ++++++++++++ include/drm/drm_crtc.h | 27 ++++++++++++++++++------ include/drm/drm_simple_kms_helper.h | 7 ++++-- 16 files changed, 56 insertions(+), 122 deletions(-)
drivers/gpu/drm/arc/arcpgu_crtc.c +0 -16
@@ -9,7 +9,6 @@#include <drm/drm_device.h> #include <drm/drm_fb_cma_helper.h> #include <drm/drm_gem_cma_helper.h> -#include <drm/drm_vblank.h> #include <drm/drm_plane_helper.h> #include <drm/drm_probe_helper.h> #include <linux/clk.h>
@@ -138,24 +137,9 @@ static void arc_pgu_crtc_atomic_disable(struct drm_crtc *crtc, ~ARCPGU_CTRL_ENABLE_MASK); } -static void arc_pgu_crtc_atomic_begin (struct drm_crtc *crtc, -struct drm_crtc_state *state) -{ -struct drm_pending_vblank_event *event = crtc->state->event; - -if (event) { - crtc->state->event =NULL ; - -spin_lock_irq (&crtc->dev->event_lock); -drm_crtc_send_vblank_event (crtc, event); -spin_unlock_irq (&crtc->dev->event_lock); - } -} -static const struct drm_crtc_helper_funcs arc_pgu_crtc_helper_funcs = { .mode_valid = arc_pgu_crtc_mode_valid, .mode_set_nofb = arc_pgu_crtc_mode_set_nofb, - .atomic_begin = arc_pgu_crtc_atomic_begin, .atomic_enable = arc_pgu_crtc_atomic_enable, .atomic_disable = arc_pgu_crtc_atomic_disable, };
drivers/gpu/drm/bochs/bochs_kms.c +0 -9
@@ -7,7 +7,6 @@#include <drm/drm_atomic_helper.h> #include <drm/drm_gem_framebuffer_helper.h> #include <drm/drm_probe_helper.h> -#include <drm/drm_vblank.h> #include "bochs.h"
@@ -57,16 +56,8 @@ static void bochs_pipe_update(struct drm_simple_display_pipe *pipe, struct drm_plane_state *old_state) {struct bochs_device *bochs = pipe->crtc.dev->dev_private; -struct drm_crtc *crtc = &pipe->crtc;bochs_plane_update (bochs, pipe->plane.state); - -if (crtc->state->event) { -spin_lock_irq (&crtc->dev->event_lock); -drm_crtc_send_vblank_event (crtc, crtc->state->event); - crtc->state->event =NULL ; -spin_unlock_irq (&crtc->dev->event_lock); - } }static const struct drm_simple_display_pipe_funcs bochs_pipe_funcs = {
drivers/gpu/drm/cirrus/cirrus.c +0 -8
@@ -38,7 +38,6 @@#include <drm/drm_modeset_helper_vtables.h> #include <drm/drm_probe_helper.h> #include <drm/drm_simple_kms_helper.h> -#include <drm/drm_vblank.h> #define DRIVER_NAME "cirrus"#define DRIVER_DESC "qemu cirrus vga"
@@ -434,13 +433,6 @@ static void cirrus_pipe_update(struct drm_simple_display_pipe *pipe,if (drm_atomic_helper_damage_merged (old_state, state, &rect))cirrus_fb_blit_rect (pipe->plane.state->fb, &rect); - -if (crtc->state->event) { -spin_lock_irq (&crtc->dev->event_lock); -drm_crtc_send_vblank_event (crtc, crtc->state->event); - crtc->state->event =NULL ; -spin_unlock_irq (&crtc->dev->event_lock); - } }static const struct drm_simple_display_pipe_funcs cirrus_pipe_funcs = {
drivers/gpu/drm/drm_atomic_helper.c +9 -1
@@ -583,6 +583,7 @@ mode_valid(struct drm_atomic_state *state)* &drm_crtc_state.connectors_changed is set when a connector is added or * removed from the CRTC. &drm_crtc_state.active_changed is set when * &drm_crtc_state.active changes, which is used for DPMS. +* &drm_crtc_state.no_vblank is set from the result of drm_crtc_has_vblank(). * See also: drm_atomic_crtc_needs_modeset() * * IMPORTANT:
@@ -649,6 +650,11 @@ drm_atomic_helper_check_modeset(struct drm_device *dev,return -EINVAL ; } + +if (drm_crtc_has_vblank (crtc)) + new_crtc_state->no_vblank = false; +else + new_crtc_state->no_vblank = true; } ret =handle_conflicting_encoders (state, false);
@@ -2215,7 +2221,9 @@ EXPORT_SYMBOL(drm_atomic_helper_wait_for_dependencies);* when a job is queued, and any change to the pipeline that does not touch the * connector is leading to timeouts when calling * drm_atomic_helper_wait_for_vblanks() or -* drm_atomic_helper_wait_for_flip_done(). +* drm_atomic_helper_wait_for_flip_done(). In addition to writeback +* connectors, this function can also fake VBLANK events for CRTCs without +* VBLANK interrupt. * * This is part of the atomic helper support for nonblocking commits, see * drm_atomic_helper_setup_commit() for an overview.
drivers/gpu/drm/drm_mipi_dbi.c +0 -9
@@ -24,7 +24,6 @@#include <drm/drm_modes.h> #include <drm/drm_probe_helper.h> #include <drm/drm_rect.h> -#include <drm/drm_vblank.h> #include <video/mipi_display.h> #define MIPI_DBI_MAX_SPI_READ_SPEED 2000000/* 2MHz */
@@ -299,18 +298,10 @@ void mipi_dbi_pipe_update(struct drm_simple_display_pipe *pipe, struct drm_plane_state *old_state) {struct drm_plane_state *state = pipe->plane.state; -struct drm_crtc *crtc = &pipe->crtc;struct drm_rect rect;if (drm_atomic_helper_damage_merged (old_state, state, &rect))mipi_dbi_fb_dirty (state->fb, &rect); - -if (crtc->state->event) { -spin_lock_irq (&crtc->dev->event_lock); -drm_crtc_send_vblank_event (crtc, crtc->state->event); -spin_unlock_irq (&crtc->dev->event_lock); - crtc->state->event =NULL ; - } }EXPORT_SYMBOL (mipi_dbi_pipe_update);
drivers/gpu/drm/drm_vblank.c +9 -0
@@ -69,6 +69,12 @@* &drm_driver.max_vblank_count. In that case the vblank core only disables the * vblanks after a timer has expired, which can be configured through the * ``vblankoffdelay`` module parameter. +* +* Drivers for hardware without support for vertical-blanking interrupts +* must not call drm_vblank_init(). For such drivers, atomic helpers will +* automatically generate vblank events as part of the display update. This +* functionality also can be controlled by the driver by enabling and disabling +* struct drm_crtc_state.no_vblank. */ /* Retry timestamp calculation up to 3 times to satisfy
@@ -510,6 +516,9 @@ EXPORT_SYMBOL(drm_vblank_init);* initialized for a CRTC. For most hardware this means that vblanking * can also be enabled on the CRTC. * +* Atomic helpers use this function to initialize +* &drm_crtc_state.no_vblank. See also drm_atomic_helper_check_modeset(). +* * Returns: * True if vblanking has been initialized for the given CRTC, false * otherwise.
drivers/gpu/drm/qxl/qxl_display.c +0 -14
@@ -31,7 +31,6 @@#include <drm/drm_gem_framebuffer_helper.h> #include <drm/drm_plane_helper.h> #include <drm/drm_probe_helper.h> -#include <drm/drm_vblank.h> #include "qxl_drv.h" #include "qxl_object.h"
@@ -372,19 +371,6 @@ static void qxl_crtc_update_monitors_config(struct drm_crtc *crtc,static void qxl_crtc_atomic_flush (struct drm_crtc *crtc,struct drm_crtc_state *old_crtc_state) { -struct drm_device *dev = crtc->dev; -struct drm_pending_vblank_event *event; -unsigned long flags; - -if (crtc->state && crtc->state->event) { - event = crtc->state->event; - crtc->state->event =NULL ; - -spin_lock_irqsave (&dev->event_lock, flags); -drm_crtc_send_vblank_event (crtc, event); -spin_unlock_irqrestore (&dev->event_lock, flags); - } -qxl_crtc_update_monitors_config (crtc,"flush" ); }
drivers/gpu/drm/tiny/gm12u320.c +0 -9
@@ -22,7 +22,6 @@#include <drm/drm_modeset_helper_vtables.h> #include <drm/drm_probe_helper.h> #include <drm/drm_simple_kms_helper.h> -#include <drm/drm_vblank.h> static bool eco_mode;module_param (eco_mode, bool,0644 );
@@ -610,18 +609,10 @@ static void gm12u320_pipe_update(struct drm_simple_display_pipe *pipe, struct drm_plane_state *old_state) {struct drm_plane_state *state = pipe->plane.state; -struct drm_crtc *crtc = &pipe->crtc;struct drm_rect rect;if (drm_atomic_helper_damage_merged (old_state, state, &rect))gm12u320_fb_mark_dirty (pipe->plane.state->fb, &rect); - -if (crtc->state->event) { -spin_lock_irq (&crtc->dev->event_lock); -drm_crtc_send_vblank_event (crtc, crtc->state->event); - crtc->state->event =NULL ; -spin_unlock_irq (&crtc->dev->event_lock); - } }static const struct drm_simple_display_pipe_funcs gm12u320_pipe_funcs = {
drivers/gpu/drm/tiny/ili9225.c +0 -9
@@ -26,7 +26,6 @@#include <drm/drm_gem_framebuffer_helper.h> #include <drm/drm_mipi_dbi.h> #include <drm/drm_rect.h> -#include <drm/drm_vblank.h> #define ILI9225_DRIVER_READ_CODE 0x00#define ILI9225_DRIVER_OUTPUT_CONTROL 0x01
@@ -165,18 +164,10 @@ static void ili9225_pipe_update(struct drm_simple_display_pipe *pipe, struct drm_plane_state *old_state) {struct drm_plane_state *state = pipe->plane.state; -struct drm_crtc *crtc = &pipe->crtc;struct drm_rect rect;if (drm_atomic_helper_damage_merged (old_state, state, &rect))ili9225_fb_dirty (state->fb, &rect); - -if (crtc->state->event) { -spin_lock_irq (&crtc->dev->event_lock); -drm_crtc_send_vblank_event (crtc, crtc->state->event); -spin_unlock_irq (&crtc->dev->event_lock); - crtc->state->event =NULL ; - } } static void ili9225_pipe_enable(struct drm_simple_display_pipe *pipe,
drivers/gpu/drm/tiny/repaper.c +0 -9
@@ -33,7 +33,6 @@#include <drm/drm_gem_framebuffer_helper.h> #include <drm/drm_modes.h> #include <drm/drm_rect.h> -#include <drm/drm_vblank.h> #include <drm/drm_probe_helper.h> #include <drm/drm_simple_kms_helper.h>
@@ -856,18 +855,10 @@ static void repaper_pipe_update(struct drm_simple_display_pipe *pipe, struct drm_plane_state *old_state) {struct drm_plane_state *state = pipe->plane.state; -struct drm_crtc *crtc = &pipe->crtc;struct drm_rect rect;if (drm_atomic_helper_damage_merged (old_state, state, &rect))repaper_fb_dirty (state->fb); - -if (crtc->state->event) { -spin_lock_irq (&crtc->dev->event_lock); -drm_crtc_send_vblank_event (crtc, crtc->state->event); -spin_unlock_irq (&crtc->dev->event_lock); - crtc->state->event =NULL ; - } }static const struct drm_simple_display_pipe_funcs repaper_pipe_funcs = {
drivers/gpu/drm/tiny/st7586.c +0 -9
@@ -23,7 +23,6 @@#include <drm/drm_gem_framebuffer_helper.h> #include <drm/drm_mipi_dbi.h> #include <drm/drm_rect.h> -#include <drm/drm_vblank.h> /* controller-specific commands */ #define ST7586_DISP_MODE_GRAY 0x38
@@ -159,18 +158,10 @@ static void st7586_pipe_update(struct drm_simple_display_pipe *pipe, struct drm_plane_state *old_state) {struct drm_plane_state *state = pipe->plane.state; -struct drm_crtc *crtc = &pipe->crtc;struct drm_rect rect;if (drm_atomic_helper_damage_merged (old_state, state, &rect))st7586_fb_dirty (state->fb, &rect); - -if (crtc->state->event) { -spin_lock_irq (&crtc->dev->event_lock); -drm_crtc_send_vblank_event (crtc, crtc->state->event); -spin_unlock_irq (&crtc->dev->event_lock); - crtc->state->event =NULL ; - } } static void st7586_pipe_enable(struct drm_simple_display_pipe *pipe,
drivers/gpu/drm/vboxvideo/vbox_mode.c +0 -12
@@ -18,7 +18,6 @@#include <drm/drm_gem_framebuffer_helper.h> #include <drm/drm_plane_helper.h> #include <drm/drm_probe_helper.h> -#include <drm/drm_vblank.h> #include "hgsmi_channels.h" #include "vbox_drv.h"
@@ -226,17 +225,6 @@ static void vbox_crtc_atomic_disable(struct drm_crtc *crtc,static void vbox_crtc_atomic_flush (struct drm_crtc *crtc,struct drm_crtc_state *old_crtc_state) { -struct drm_pending_vblank_event *event; -unsigned long flags; - -if (crtc->state && crtc->state->event) { - event = crtc->state->event; - crtc->state->event =NULL ; - -spin_lock_irqsave (&crtc->dev->event_lock, flags); -drm_crtc_send_vblank_event (crtc, event); -spin_unlock_irqrestore (&crtc->dev->event_lock, flags); - } }static const struct drm_crtc_helper_funcs vbox_crtc_helper_funcs = {
drivers/gpu/drm/virtio/virtgpu_display.c +0 -8
@@ -30,7 +30,6 @@#include <drm/drm_fourcc.h> #include <drm/drm_gem_framebuffer_helper.h> #include <drm/drm_probe_helper.h> -#include <drm/drm_vblank.h> #include "virtgpu_drv.h"
@@ -121,13 +120,6 @@ static int virtio_gpu_crtc_atomic_check(struct drm_crtc *crtc,static void virtio_gpu_crtc_atomic_flush (struct drm_crtc *crtc,struct drm_crtc_state *old_state) { -unsigned long flags; - -spin_lock_irqsave (&crtc->dev->event_lock, flags); -if (crtc->state->event) -drm_crtc_send_vblank_event (crtc, crtc->state->event); - crtc->state->event =NULL ; -spin_unlock_irqrestore (&crtc->dev->event_lock, flags); }static const struct drm_crtc_helper_funcs virtio_gpu_crtc_helper_funcs = {
drivers/gpu/drm/xen/xen_drm_front_kms.c +13 -0
@@ -220,6 +220,18 @@ static bool display_send_page_flip(struct drm_simple_display_pipe *pipe,return false; } +static int display_check (struct drm_simple_display_pipe *pipe, +struct drm_plane_state *plane_state, +struct drm_crtc_state *crtc_state) +{ +/* Make sure that DRM helpers don't send VBLANK events +* automatically. Xen has it's own logic to do so. +*/ + crtc_state->no_vblank = false; + +return 0 ; +} +static void display_update (struct drm_simple_display_pipe *pipe,struct drm_plane_state *old_plane_state) {
@@ -284,6 +296,7 @@ static const struct drm_simple_display_pipe_funcs display_funcs = { .enable = display_enable, .disable = display_disable, .prepare_fb = drm_gem_fb_simple_display_pipe_prepare_fb, + .check = display_check, .update = display_update, };
include/drm/drm_crtc.h +20 -7
@@ -174,12 +174,22 @@ struct drm_crtc_state {* @no_vblank: * * Reflects the ability of a CRTC to send VBLANK events. This state -* usually depends on the pipeline configuration, and the main usuage -* is CRTCs feeding a writeback connector operating in oneshot mode. -* In this case the VBLANK event is only generated when a job is queued -* to the writeback connector, and we want the core to fake VBLANK -* events when this part of the pipeline hasn't changed but others had -* or when the CRTC and connectors are being disabled. +* usually depends on the pipeline configuration. If set to true, DRM +* atomic helpers will sendout a fake VBLANK event during display +* updates. +* +* One usage is for drivers and/or hardware without support for VBLANK +* interrupts. Such drivers typically do not initialize vblanking +* (i.e., call drm_vblank_init() wit the number of CRTCs). For CRTCs +* without initialized vblanking, the field is initialized to true and +* a VBLANK event will be send out on each update of the display +* pipeline. +* +* Another usage is CRTCs feeding a writeback connector operating in +* oneshot mode. In this case the VBLANK event is only generated when +* a job is queued to the writeback connector, and we want the core +* to fake VBLANK events when this part of the pipeline hasn't changed +* but others had or when the CRTC and connectors are being disabled. * * __drm_atomic_helper_crtc_duplicate_state() will not reset the value * from the current state, the CRTC driver is then responsible for
@@ -335,7 +345,10 @@ struct drm_crtc_state {* - Events for disabled CRTCs are not allowed, and drivers can ignore * that case. * -* This can be handled by the drm_crtc_send_vblank_event() function, +* For very simple hardware without VBLANK interrupt, enabling +* &struct drm_crtc_state.no_vblank makes DRM's atomic commit helpers +* send the event at an appropriate time. For more complex hardware this +* can be handled by the drm_crtc_send_vblank_event() function, * which the driver should call on the provided event upon completion of * the atomic commit. Note that if the driver supports vblank signalling * and timestamping the vblank counters and timestamps must agree with
include/drm/drm_simple_kms_helper.h +5 -2
@@ -100,8 +100,11 @@ struct drm_simple_display_pipe_funcs {* This is the function drivers should submit the * &drm_pending_vblank_event from. Using either * drm_crtc_arm_vblank_event(), when the driver supports vblank -* interrupt handling, or drm_crtc_send_vblank_event() directly in case -* the hardware lacks vblank support entirely. +* interrupt handling, or drm_crtc_send_vblank_event() for more +* complex case. In case the hardware lacks vblank support entirely, +* drivers can set &struct drm_crtc_state.no_vblank in +* &struct drm_simple_display_pipe_funcs.check and let DRM's +* atomic helper fake a vblank event. */ void (*update)(struct drm_simple_display_pipe *pipe,struct drm_plane_state *old_plane_state);
Signature · 1 line
2.24.1
Re: [PATCH v3 2/4] drm: Initialize struct drm_crtc_state.no_vblank from device settings
Quoted 15 lines · Thomas Zimmermann
At the end of a commit, atomic helpers can generate a VBLANK event automatically. Originally implemented for writeback connectors, the functionality can be used by any driver and/or hardware without proper VBLANK interrupt.
First of all, the patch updates the documentation to make this behaviour
official: settings struct drm_crtc_state.no_vblank to true enables automatic VBLANK generation.Atomic modesetting helper set the initial value of
no_vblankindrm_atomic_helper_check_modeset(). If vblanking has been initialized for a CRTC,no_blankis disabled. Otherwise it's enabled. Hence, atomic helpers will automatically send out VBLANK events with any driver that did not initialize vblanking.
Quoted 8 lines
As drivers previously send out VBLANK events by themselves, all affected drivers have to be updated as well. Usually, deleting the driver's vblanking code is sufficient. Xen implements its own logic for generating events and therefore needs to override
no_vblankwith a value of false.v3:
* squash all related changes patches into this patch
Hm, since the fall-back only happens when the driver hasn't sent out the even I think it'd be safe to split the driver cleanups into a separate patch. Makes the core/helper changes stand out more properly.
Even the xen hunk I think isn't strictly needed, since that pick up the event correctly and clears state->event to NULL.
Quoted 126 lines
Signed-off-by Thomas Zimmermann <[email protected]>--- drivers/gpu/drm/arc/arcpgu_crtc.c | 16 -------------- drivers/gpu/drm/bochs/bochs_kms.c | 9 -------- drivers/gpu/drm/cirrus/cirrus.c | 8 ------- drivers/gpu/drm/drm_atomic_helper.c | 10 ++++++++- drivers/gpu/drm/drm_mipi_dbi.c | 9 -------- drivers/gpu/drm/drm_vblank.c | 9 ++++++++ drivers/gpu/drm/qxl/qxl_display.c | 14 ------------ drivers/gpu/drm/tiny/gm12u320.c | 9 -------- drivers/gpu/drm/tiny/ili9225.c | 9 -------- drivers/gpu/drm/tiny/repaper.c | 9 -------- drivers/gpu/drm/tiny/st7586.c | 9 -------- drivers/gpu/drm/vboxvideo/vbox_mode.c | 12 ----------- drivers/gpu/drm/virtio/virtgpu_display.c | 8 ------- drivers/gpu/drm/xen/xen_drm_front_kms.c | 13 ++++++++++++ include/drm/drm_crtc.h | 27 ++++++++++++++++++------ include/drm/drm_simple_kms_helper.h | 7 ++++-- 16 files changed, 56 insertions(+), 122 deletions(-)drivers/gpu/drm/arc/arcpgu_crtc.c +0
-16@@ -9,7 +9,6 @@#include <drm/drm_device.h> #include <drm/drm_fb_cma_helper.h> #include <drm/drm_gem_cma_helper.h> -#include <drm/drm_vblank.h> #include <drm/drm_plane_helper.h> #include <drm/drm_probe_helper.h> #include <linux/clk.h> @@ -138,24 +137,9 @@ static void arc_pgu_crtc_atomic_disable(struct drm_crtc *crtc, ~ARCPGU_CTRL_ENABLE_MASK); } -static void arc_pgu_crtc_atomic_begin (struct drm_crtc *crtc, -struct drm_crtc_state *state) -{ -struct drm_pending_vblank_event *event = crtc->state->event; - -if (event) { - crtc->state->event =NULL ; - -spin_lock_irq (&crtc->dev->event_lock); -drm_crtc_send_vblank_event (crtc, event); -spin_unlock_irq (&crtc->dev->event_lock); - } -} -static const struct drm_crtc_helper_funcs arc_pgu_crtc_helper_funcs = { .mode_valid = arc_pgu_crtc_mode_valid, .mode_set_nofb = arc_pgu_crtc_mode_set_nofb, - .atomic_begin = arc_pgu_crtc_atomic_begin, .atomic_enable = arc_pgu_crtc_atomic_enable, .atomic_disable = arc_pgu_crtc_atomic_disable, };drivers/gpu/drm/bochs/bochs_kms.c +0
-9@@ -7,7 +7,6 @@#include <drm/drm_atomic_helper.h> #include <drm/drm_gem_framebuffer_helper.h> #include <drm/drm_probe_helper.h> -#include <drm/drm_vblank.h> #include "bochs.h" @@ -57,16 +56,8 @@ static void bochs_pipe_update(struct drm_simple_display_pipe *pipe, struct drm_plane_state *old_state) {struct bochs_device *bochs = pipe->crtc.dev->dev_private; -struct drm_crtc *crtc = &pipe->crtc;bochs_plane_update (bochs, pipe->plane.state); - -if (crtc->state->event) { -spin_lock_irq (&crtc->dev->event_lock); -drm_crtc_send_vblank_event (crtc, crtc->state->event); - crtc->state->event =NULL ; -spin_unlock_irq (&crtc->dev->event_lock); - } }static const struct drm_simple_display_pipe_funcs bochs_pipe_funcs = {drivers/gpu/drm/cirrus/cirrus.c +0
-8@@ -38,7 +38,6 @@#include <drm/drm_modeset_helper_vtables.h> #include <drm/drm_probe_helper.h> #include <drm/drm_simple_kms_helper.h> -#include <drm/drm_vblank.h> #define DRIVER_NAME "cirrus"#define DRIVER_DESC "qemu cirrus vga"@@ -434,13 +433,6 @@ static void cirrus_pipe_update(struct drm_simple_display_pipe *pipe,if (drm_atomic_helper_damage_merged (old_state, state, &rect))cirrus_fb_blit_rect (pipe->plane.state->fb, &rect); - -if (crtc->state->event) { -spin_lock_irq (&crtc->dev->event_lock); -drm_crtc_send_vblank_event (crtc, crtc->state->event); - crtc->state->event =NULL ; -spin_unlock_irq (&crtc->dev->event_lock); - } }static const struct drm_simple_display_pipe_funcs cirrus_pipe_funcs = {drivers/gpu/drm/drm_atomic_helper.c +1
-0@@ -583,6 +583,7 @@ mode_valid(struct drm_atomic_state *state)* &drm_crtc_state.connectors_changed is set when a connector is added or * removed from the CRTC. &drm_crtc_state.active_changed is set when * &drm_crtc_state.active changes, which is used for DPMS. +* &drm_crtc_state.no_vblank is set from the result of drm_crtc_has_vblank(). * See also: drm_atomic_crtc_needs_modeset() * * IMPORTANT:
drivers/gpu/drm/drm_atomic_helper.c drm_atomic_helper_check_modeset(struct drm_device *dev, · line 657
Quoted 8 lines
drivers/gpu/drm/drm_atomic_helper.c +5
-0@@ -649,6 +650,11 @@ drm_atomic_helper_check_modeset(struct drm_device *dev,return -EINVAL ; } + +if (drm_crtc_has_vblank (crtc)) + new_crtc_state->no_vblank = false; +else + new_crtc_state->no_vblank = true;
Yeah this looks much better than my hack :-)
drivers/gpu/drm/drm_atomic_helper.c EXPORT_SYMBOL(drm_atomic_helper_wait_for_dependencies); · line 2226
Quoted 11 lines
drivers/gpu/drm/drm_atomic_helper.c +3
-1} ret =handle_conflicting_encoders (state, false);@@ -2215,7 +2221,9 @@ EXPORT_SYMBOL(drm_atomic_helper_wait_for_dependencies);* when a job is queued, and any change to the pipeline that does not touch the * connector is leading to timeouts when calling * drm_atomic_helper_wait_for_vblanks() or -* drm_atomic_helper_wait_for_flip_done(). +* drm_atomic_helper_wait_for_flip_done(). In addition to writeback +* connectors, this function can also fake VBLANK events for CRTCs without +* VBLANK interrupt.
I still think we should reword this entire paragraph to make the "hw has no vblank" the main use-case, with writeback connectors as the "Also used for ..." special case.
Quoted 316 lines
drivers/gpu/drm/drm_atomic_helper.c +0
-0* * This is part of the atomic helper support for nonblocking commits, see * drm_atomic_helper_setup_commit() for an overview. drivers/gpu/drm/drm_mipi_dbi.c +0
-9@@ -24,7 +24,6 @@#include <drm/drm_modes.h> #include <drm/drm_probe_helper.h> #include <drm/drm_rect.h> -#include <drm/drm_vblank.h> #include <video/mipi_display.h> #define MIPI_DBI_MAX_SPI_READ_SPEED 2000000/* 2MHz */ @@ -299,18 +298,10 @@ void mipi_dbi_pipe_update(struct drm_simple_display_pipe *pipe, struct drm_plane_state *old_state) {struct drm_plane_state *state = pipe->plane.state; -struct drm_crtc *crtc = &pipe->crtc;struct drm_rect rect;if (drm_atomic_helper_damage_merged (old_state, state, &rect))mipi_dbi_fb_dirty (state->fb, &rect); - -if (crtc->state->event) { -spin_lock_irq (&crtc->dev->event_lock); -drm_crtc_send_vblank_event (crtc, crtc->state->event); -spin_unlock_irq (&crtc->dev->event_lock); - crtc->state->event =NULL ; - } }EXPORT_SYMBOL (mipi_dbi_pipe_update);drivers/gpu/drm/drm_vblank.c +9
-0@@ -69,6 +69,12 @@* &drm_driver.max_vblank_count. In that case the vblank core only disables the * vblanks after a timer has expired, which can be configured through the * ``vblankoffdelay`` module parameter. +* +* Drivers for hardware without support for vertical-blanking interrupts +* must not call drm_vblank_init(). For such drivers, atomic helpers will +* automatically generate vblank events as part of the display update. This +* functionality also can be controlled by the driver by enabling and disabling +* struct drm_crtc_state.no_vblank. */ /* Retry timestamp calculation up to 3 times to satisfy @@ -510,6 +516,9 @@ EXPORT_SYMBOL(drm_vblank_init);* initialized for a CRTC. For most hardware this means that vblanking * can also be enabled on the CRTC. * +* Atomic helpers use this function to initialize +* &drm_crtc_state.no_vblank. See also drm_atomic_helper_check_modeset(). +* * Returns: * True if vblanking has been initialized for the given CRTC, false * otherwise. drivers/gpu/drm/qxl/qxl_display.c +0
-14@@ -31,7 +31,6 @@#include <drm/drm_gem_framebuffer_helper.h> #include <drm/drm_plane_helper.h> #include <drm/drm_probe_helper.h> -#include <drm/drm_vblank.h> #include "qxl_drv.h" #include "qxl_object.h" @@ -372,19 +371,6 @@ static void qxl_crtc_update_monitors_config(struct drm_crtc *crtc,static void qxl_crtc_atomic_flush (struct drm_crtc *crtc,struct drm_crtc_state *old_crtc_state) { -struct drm_device *dev = crtc->dev; -struct drm_pending_vblank_event *event; -unsigned long flags; - -if (crtc->state && crtc->state->event) { - event = crtc->state->event; - crtc->state->event =NULL ; - -spin_lock_irqsave (&dev->event_lock, flags); -drm_crtc_send_vblank_event (crtc, event); -spin_unlock_irqrestore (&dev->event_lock, flags); - } -qxl_crtc_update_monitors_config (crtc,"flush" ); }drivers/gpu/drm/tiny/gm12u320.c +0
-9@@ -22,7 +22,6 @@#include <drm/drm_modeset_helper_vtables.h> #include <drm/drm_probe_helper.h> #include <drm/drm_simple_kms_helper.h> -#include <drm/drm_vblank.h> static bool eco_mode;module_param (eco_mode, bool,0644 );@@ -610,18 +609,10 @@ static void gm12u320_pipe_update(struct drm_simple_display_pipe *pipe, struct drm_plane_state *old_state) {struct drm_plane_state *state = pipe->plane.state; -struct drm_crtc *crtc = &pipe->crtc;struct drm_rect rect;if (drm_atomic_helper_damage_merged (old_state, state, &rect))gm12u320_fb_mark_dirty (pipe->plane.state->fb, &rect); - -if (crtc->state->event) { -spin_lock_irq (&crtc->dev->event_lock); -drm_crtc_send_vblank_event (crtc, crtc->state->event); - crtc->state->event =NULL ; -spin_unlock_irq (&crtc->dev->event_lock); - } }static const struct drm_simple_display_pipe_funcs gm12u320_pipe_funcs = {drivers/gpu/drm/tiny/ili9225.c +0
-9@@ -26,7 +26,6 @@#include <drm/drm_gem_framebuffer_helper.h> #include <drm/drm_mipi_dbi.h> #include <drm/drm_rect.h> -#include <drm/drm_vblank.h> #define ILI9225_DRIVER_READ_CODE 0x00#define ILI9225_DRIVER_OUTPUT_CONTROL 0x01@@ -165,18 +164,10 @@ static void ili9225_pipe_update(struct drm_simple_display_pipe *pipe, struct drm_plane_state *old_state) {struct drm_plane_state *state = pipe->plane.state; -struct drm_crtc *crtc = &pipe->crtc;struct drm_rect rect;if (drm_atomic_helper_damage_merged (old_state, state, &rect))ili9225_fb_dirty (state->fb, &rect); - -if (crtc->state->event) { -spin_lock_irq (&crtc->dev->event_lock); -drm_crtc_send_vblank_event (crtc, crtc->state->event); -spin_unlock_irq (&crtc->dev->event_lock); - crtc->state->event =NULL ; - } } static void ili9225_pipe_enable(struct drm_simple_display_pipe *pipe,drivers/gpu/drm/tiny/repaper.c +0
-9@@ -33,7 +33,6 @@#include <drm/drm_gem_framebuffer_helper.h> #include <drm/drm_modes.h> #include <drm/drm_rect.h> -#include <drm/drm_vblank.h> #include <drm/drm_probe_helper.h> #include <drm/drm_simple_kms_helper.h> @@ -856,18 +855,10 @@ static void repaper_pipe_update(struct drm_simple_display_pipe *pipe, struct drm_plane_state *old_state) {struct drm_plane_state *state = pipe->plane.state; -struct drm_crtc *crtc = &pipe->crtc;struct drm_rect rect;if (drm_atomic_helper_damage_merged (old_state, state, &rect))repaper_fb_dirty (state->fb); - -if (crtc->state->event) { -spin_lock_irq (&crtc->dev->event_lock); -drm_crtc_send_vblank_event (crtc, crtc->state->event); -spin_unlock_irq (&crtc->dev->event_lock); - crtc->state->event =NULL ; - } }static const struct drm_simple_display_pipe_funcs repaper_pipe_funcs = {drivers/gpu/drm/tiny/st7586.c +0
-9@@ -23,7 +23,6 @@#include <drm/drm_gem_framebuffer_helper.h> #include <drm/drm_mipi_dbi.h> #include <drm/drm_rect.h> -#include <drm/drm_vblank.h> /* controller-specific commands */ #define ST7586_DISP_MODE_GRAY 0x38@@ -159,18 +158,10 @@ static void st7586_pipe_update(struct drm_simple_display_pipe *pipe, struct drm_plane_state *old_state) {struct drm_plane_state *state = pipe->plane.state; -struct drm_crtc *crtc = &pipe->crtc;struct drm_rect rect;if (drm_atomic_helper_damage_merged (old_state, state, &rect))st7586_fb_dirty (state->fb, &rect); - -if (crtc->state->event) { -spin_lock_irq (&crtc->dev->event_lock); -drm_crtc_send_vblank_event (crtc, crtc->state->event); -spin_unlock_irq (&crtc->dev->event_lock); - crtc->state->event =NULL ; - } } static void st7586_pipe_enable(struct drm_simple_display_pipe *pipe,drivers/gpu/drm/vboxvideo/vbox_mode.c +0
-12@@ -18,7 +18,6 @@#include <drm/drm_gem_framebuffer_helper.h> #include <drm/drm_plane_helper.h> #include <drm/drm_probe_helper.h> -#include <drm/drm_vblank.h> #include "hgsmi_channels.h" #include "vbox_drv.h" @@ -226,17 +225,6 @@ static void vbox_crtc_atomic_disable(struct drm_crtc *crtc,static void vbox_crtc_atomic_flush (struct drm_crtc *crtc,struct drm_crtc_state *old_crtc_state) { -struct drm_pending_vblank_event *event; -unsigned long flags; - -if (crtc->state && crtc->state->event) { - event = crtc->state->event; - crtc->state->event =NULL ; - -spin_lock_irqsave (&crtc->dev->event_lock, flags); -drm_crtc_send_vblank_event (crtc, event); -spin_unlock_irqrestore (&crtc->dev->event_lock, flags); - } }static const struct drm_crtc_helper_funcs vbox_crtc_helper_funcs = {drivers/gpu/drm/virtio/virtgpu_display.c +0
-8@@ -30,7 +30,6 @@#include <drm/drm_fourcc.h> #include <drm/drm_gem_framebuffer_helper.h> #include <drm/drm_probe_helper.h> -#include <drm/drm_vblank.h> #include "virtgpu_drv.h" @@ -121,13 +120,6 @@ static int virtio_gpu_crtc_atomic_check(struct drm_crtc *crtc,static void virtio_gpu_crtc_atomic_flush (struct drm_crtc *crtc,struct drm_crtc_state *old_state) { -unsigned long flags; - -spin_lock_irqsave (&crtc->dev->event_lock, flags); -if (crtc->state->event) -drm_crtc_send_vblank_event (crtc, crtc->state->event); - crtc->state->event =NULL ; -spin_unlock_irqrestore (&crtc->dev->event_lock, flags); }static const struct drm_crtc_helper_funcs virtio_gpu_crtc_helper_funcs = {drivers/gpu/drm/xen/xen_drm_front_kms.c +13
-0@@ -220,6 +220,18 @@ static bool display_send_page_flip(struct drm_simple_display_pipe *pipe,return false; } +static int display_check (struct drm_simple_display_pipe *pipe, +struct drm_plane_state *plane_state, +struct drm_crtc_state *crtc_state) +{ +/* Make sure that DRM helpers don't send VBLANK events +* automatically. Xen has it's own logic to do so. +*/ + crtc_state->no_vblank = false; + +return 0 ; +} +static void display_update (struct drm_simple_display_pipe *pipe,struct drm_plane_state *old_plane_state) {@@ -284,6 +296,7 @@ static const struct drm_simple_display_pipe_funcs display_funcs = { .enable = display_enable, .disable = display_disable, .prepare_fb = drm_gem_fb_simple_display_pipe_prepare_fb, + .check = display_check, .update = display_update, };include/drm/drm_crtc.h +0
-4@@ -174,12 +174,22 @@ struct drm_crtc_state {* @no_vblank: * * Reflects the ability of a CRTC to send VBLANK events. This state -* usually depends on the pipeline configuration, and the main usuage -* is CRTCs feeding a writeback connector operating in oneshot mode. -* In this case the VBLANK event is only generated when a job is queued -* to the writeback connector, and we want the core to fake VBLANK
include/drm/drm_crtc.h struct drm_crtc_state { · line 184
Quoted 10 lines
include/drm/drm_crtc.h +8
-2-* events when this part of the pipeline hasn't changed but others had -* or when the CRTC and connectors are being disabled. +* usually depends on the pipeline configuration. If set to true, DRM +* atomic helpers will sendout a fake VBLANK event during display +* updates. +* +* One usage is for drivers and/or hardware without support for VBLANK +* interrupts. Such drivers typically do not initialize vblanking +* (i.e., call drm_vblank_init() wit the number of CRTCs). For CRTCs +* without initialized vblanking, the field is initialized to true and
... is initialized to true in the atomic helpers by calling drm_atomic_helper_check_modeset() and ...
Just to clarify that this isn't done for everyone by default, but a helper thing.
Quoted 31 lines
include/drm/drm_crtc.h +12
-1+* a VBLANK event will be send out on each update of the display +* pipeline. +* +* Another usage is CRTCs feeding a writeback connector operating in +* oneshot mode. In this case the VBLANK event is only generated when +* a job is queued to the writeback connector, and we want the core +* to fake VBLANK events when this part of the pipeline hasn't changed +* but others had or when the CRTC and connectors are being disabled. * * __drm_atomic_helper_crtc_duplicate_state() will not reset the value * from the current state, the CRTC driver is then responsible for @@ -335,7 +345,10 @@ struct drm_crtc_state {* - Events for disabled CRTCs are not allowed, and drivers can ignore * that case. * -* This can be handled by the drm_crtc_send_vblank_event() function, +* For very simple hardware without VBLANK interrupt, enabling +* &struct drm_crtc_state.no_vblank makes DRM's atomic commit helpers +* send the event at an appropriate time. For more complex hardware this +* can be handled by the drm_crtc_send_vblank_event() function, * which the driver should call on the provided event upon completion of * the atomic commit. Note that if the driver supports vblank signalling * and timestamping the vblank counters and timestamps must agree with include/drm/drm_simple_kms_helper.h +0
-0@@ -100,8 +100,11 @@ struct drm_simple_display_pipe_funcs {* This is the function drivers should submit the * &drm_pending_vblank_event from. Using either * drm_crtc_arm_vblank_event(), when the driver supports vblank
include/drm/drm_simple_kms_helper.h struct drm_simple_display_pipe_funcs { · line 110
Quoted 10 lines
include/drm/drm_simple_kms_helper.h +5
-2-* interrupt handling, or drm_crtc_send_vblank_event() directly in case -* the hardware lacks vblank support entirely. +* interrupt handling, or drm_crtc_send_vblank_event() for more +* complex case. In case the hardware lacks vblank support entirely, +* drivers can set &struct drm_crtc_state.no_vblank in +* &struct drm_simple_display_pipe_funcs.check and let DRM's +* atomic helper fake a vblank event. */ void (*update)(struct drm_simple_display_pipe *pipe,struct drm_plane_state *old_plane_state);
With the kerneldoc fixed, for the core parts of this patch (driver changes all split out):
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Thomas Zimmermann
maintainerPATCH v3 3/4 drm/ast: Don't set struct drm_crtc_state.no_vblank explictly
In reply to Thomas ZimmermannAs ast does not initialize vblanking, atomic helpers initialize the value of struct drm_crtc_state.no_vblank to be true. No need to set it from within the driver.
--- drivers/gpu/drm/ast/ast_mode.c | 2 -- 1 file changed, 2 deletions(-)
drivers/gpu/drm/ast/ast_mode.c +0 -2
@@ -833,8 +833,6 @@ static void ast_crtc_helper_atomic_flush(struct drm_crtc *crtc,struct ast_vbios_mode_info *vbios_mode_info;struct drm_display_mode *adjusted_mode; - crtc->state->no_vblank = true; - ast_state =to_ast_crtc_state (crtc->state); format = ast_state->format;
Signature · 1 line
2.24.1
Thomas Zimmermann
maintainerPATCH v3 4/4 drm/udl: Don't set struct drm_crtc_state.no_vblank explictly
In reply to Thomas ZimmermannAs udl does not initialize vblanking, atomic helpers initialize the value of struct drm_crtc_state.no_vblank to be true. No need to set it from within the driver.
--- drivers/gpu/drm/udl/udl_modeset.c | 11 ----------- 1 file changed, 11 deletions(-)
drivers/gpu/drm/udl/udl_modeset.c +0 -11
@@ -375,8 +375,6 @@ udl_simple_display_pipe_enable(struct drm_simple_display_pipe *pipe,char *wrptr;int color_depth =UDL_COLOR_DEPTH_16BPP ; - crtc_state->no_vblank = true; - buf = (char *)udl->mode_buf;/* This first section has to do with setting the base address on the
@@ -428,14 +426,6 @@ udl_simple_display_pipe_disable(struct drm_simple_display_pipe *pipe)udl_submit_urb (dev, urb, buf - (char *)urb->transfer_buffer); } -static int -udl_simple_display_pipe_check (struct drm_simple_display_pipe *pipe, -struct drm_plane_state *plane_state, -struct drm_crtc_state *crtc_state) -{ -return 0 ; -} -static void udl_simple_display_pipe_update (struct drm_simple_display_pipe *pipe,struct drm_plane_state *old_plane_state)
@@ -457,7 +447,6 @@ struct drm_simple_display_pipe_funcs udl_simple_display_pipe_funcs = { .mode_valid = udl_simple_display_pipe_mode_valid, .enable = udl_simple_display_pipe_enable, .disable = udl_simple_display_pipe_disable, - .check = udl_simple_display_pipe_check, .update = udl_simple_display_pipe_update, .prepare_fb = drm_gem_fb_simple_display_pipe_prepare_fb, };
Signature · 1 line
2.24.1
Quoted 10 lines · Thomas Zimmermann
Instead of faking VBLANK events by themselves, drivers without VBLANK support can enable drm_crtc_vblank.no_vblank and let DRM do the rest. The patchset makes this official and converts over drivers.
The current implementation looks at the number of initialized CRTCs wrt vblanking. If vblanking has been initialized for a CRTC, the driver is responsible for sending out VBLANK events. Otherwise, DRM will send out the event. The behaviour selected by initializing
no_vblankas part ofdrm_atomic_helper_check_modeset().
Quoted 5 lines
I went through all drivers, looking for those that call send out VBLANK events but do not call
drm_vblank_init(). These are converted to the new semantics. This affects tiny drivers; drivers for virtual hardware; and a few others, which do not support interrupts. Xen comes with its own VBLANK logic and disablesno_vblankexplictly.
Thomas Zimmermann
maintainerHi
Am 21.01.20 um 10:36 schrieb Gerd Hoffmann:
Quoted 11 lines · Thomas Zimmermann
Quoted 10 lines
Instead of faking VBLANK events by themselves, drivers without VBLANK support can enable drm_crtc_vblank.no_vblank and let DRM do the rest. The patchset makes this official and converts over drivers.
The current implementation looks at the number of initialized CRTCs wrt vblanking. If vblanking has been initialized for a CRTC, the driver is responsible for sending out VBLANK events. Otherwise, DRM will send out the event. The behaviour selected by initializing
no_vblankas part ofdrm_atomic_helper_check_modeset().
Quoted 7 lines
Quoted 5 lines
I went through all drivers, looking for those that call send out VBLANK events but do not call
drm_vblank_init(). These are converted to the new semantics. This affects tiny drivers; drivers for virtual hardware; and a few others, which do not support interrupts. Xen comes with its own VBLANK logic and disablesno_vblankexplictly.Acked-by Gerd Hoffmann <[email protected]>
Thanks a lot. v4 will mostly reorganize the patches so I'll keep your A-b.
Best regards
Thomas
Quoted 6 lines
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Signature · 6 lines
Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization