| Message ID | 20190726210658.GA6299@ravnborg.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v3,1/1] drm/vblank: drop use of DRM_WAIT_ON() | expand |
On 2019-07-26 11:06 p.m., Sam Ravnborg wrote: > DRM_WAIT_ON() is from the deprecated drm_os_linux header and > the modern replacement is the wait_event_*. > > The return values differ, so a conversion is needed to > keep the original interface towards userspace. > Introduced a switch/case to make code obvious. > > Analysis from Michel Dänzer: > > The waiting condition rely on all relevant places where vblank_count > is modified calls wake_up(&vblank->queue). > > drm_handle_vblank(): > - Calls wake_up(&vblank->queue) > > drm_vblank_enable(): > - There is no need here because there can be no sleeping waiters > in the queue, because vblank->enabled == false immediately > terminates any waits. > > drm_crtc_accurate_vblank_count(): > - This is called from interrupt handlers, at least from > amdgpu_dm.c:dm_pflip_high_irq(). Not sure it needs to wake up > the queue though, the driver should call > drm_(crtc_)_handle_vblank anyway. > > drm_vblank_disable_and_save(): > - It can be called from an interrupt, via drm_handle_vblank -> > vblank_disable_fn. However, the only place where > drm_vblank_disable_and_save can be called with sleeping waiters > in the queue is in drm_crtc_vblank_off, which wakes up the queue > afterwards (which terminates all waits, because > vblank->enabled == false at this point). > > v3: > - Added analysis to changelog from Michel Dänzer > - Moved return result handling inside if (req_seq != seq) (Daniel V) > - Reused more of the former logic - resulting in simpler code > - Dropped Reviewed-by from Sean Paul as this is a new implmentation > > v2: > - Fix so the case where req_seq equals seq was handled properly > - quick hack to check if IGT became happy > - Only sent to igt, not to dri-devel > > Signed-off-by: Sam Ravnborg <sam@ravnborg.org> > Cc: "Michel Dänzer" <michel@daenzer.net> > Cc: Sean Paul <sean@poorly.run> > Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> > Cc: Maxime Ripard <maxime.ripard@bootlin.com> > Cc: David Airlie <airlied@linux.ie> > Cc: Daniel Vetter <daniel@ffwll.ch> Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Hi all. On Fri, Jul 26, 2019 at 11:06:58PM +0200, Sam Ravnborg wrote: > DRM_WAIT_ON() is from the deprecated drm_os_linux header and > the modern replacement is the wait_event_*. > > The return values differ, so a conversion is needed to > keep the original interface towards userspace. > Introduced a switch/case to make code obvious. Patch passes igt testing, and Michel was happy with it. It is now applied to drm-misc-next and pushed out. With this we have one less user of drm_os_linux.h (only very few remains) And one less cryptic macro use. Sam > > Analysis from Michel Dänzer: > > The waiting condition rely on all relevant places where vblank_count > is modified calls wake_up(&vblank->queue). > > drm_handle_vblank(): > - Calls wake_up(&vblank->queue) > > drm_vblank_enable(): > - There is no need here because there can be no sleeping waiters > in the queue, because vblank->enabled == false immediately > terminates any waits. > > drm_crtc_accurate_vblank_count(): > - This is called from interrupt handlers, at least from > amdgpu_dm.c:dm_pflip_high_irq(). Not sure it needs to wake up > the queue though, the driver should call > drm_(crtc_)_handle_vblank anyway. > > drm_vblank_disable_and_save(): > - It can be called from an interrupt, via drm_handle_vblank -> > vblank_disable_fn. However, the only place where > drm_vblank_disable_and_save can be called with sleeping waiters > in the queue is in drm_crtc_vblank_off, which wakes up the queue > afterwards (which terminates all waits, because > vblank->enabled == false at this point). > > v3: > - Added analysis to changelog from Michel Dänzer > - Moved return result handling inside if (req_seq != seq) (Daniel V) > - Reused more of the former logic - resulting in simpler code > - Dropped Reviewed-by from Sean Paul as this is a new implmentation > > v2: > - Fix so the case where req_seq equals seq was handled properly > - quick hack to check if IGT became happy > - Only sent to igt, not to dri-devel > > Signed-off-by: Sam Ravnborg <sam@ravnborg.org> > Cc: "Michel Dänzer" <michel@daenzer.net> > Cc: Sean Paul <sean@poorly.run> > Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> > Cc: Maxime Ripard <maxime.ripard@bootlin.com> > Cc: David Airlie <airlied@linux.ie> > Cc: Daniel Vetter <daniel@ffwll.ch> > --- > > This patch survives my testing here. > vbltest spits out the same value as before, and I can see something on > the screen. > Added intel-gfx@lists.freedesktop.org, as IGT have identified troubles > with this in v1. > > Sam > > drivers/gpu/drm/drm_vblank.c | 25 ++++++++++++++++++++----- > 1 file changed, 20 insertions(+), 5 deletions(-) > > diff --git a/drivers/gpu/drm/drm_vblank.c b/drivers/gpu/drm/drm_vblank.c > index 603ab105125d..fd1fbc77871f 100644 > --- a/drivers/gpu/drm/drm_vblank.c > +++ b/drivers/gpu/drm/drm_vblank.c > @@ -31,7 +31,6 @@ > #include <drm/drm_drv.h> > #include <drm/drm_framebuffer.h> > #include <drm/drm_print.h> > -#include <drm/drm_os_linux.h> > #include <drm/drm_vblank.h> > > #include "drm_internal.h" > @@ -1670,12 +1669,28 @@ int drm_wait_vblank_ioctl(struct drm_device *dev, void *data, > } > > if (req_seq != seq) { > + int wait; > + > DRM_DEBUG("waiting on vblank count %llu, crtc %u\n", > req_seq, pipe); > - DRM_WAIT_ON(ret, vblank->queue, 3 * HZ, > - vblank_passed(drm_vblank_count(dev, pipe), > - req_seq) || > - !READ_ONCE(vblank->enabled)); > + wait = wait_event_interruptible_timeout(vblank->queue, > + vblank_passed(drm_vblank_count(dev, pipe), req_seq) || > + !READ_ONCE(vblank->enabled), > + msecs_to_jiffies(3000)); > + > + switch (wait) { > + case 0: > + /* timeout */ > + ret = -EBUSY; > + break; > + case -ERESTARTSYS: > + /* interrupted by signal */ > + ret = -EINTR; > + break; > + default: > + ret = 0; > + break; > + } > } > > if (ret != -EINTR) { > -- > 2.20.1 > > _______________________________________________ > dri-devel mailing list > dri-devel@lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/dri-devel
diff --git a/drivers/gpu/drm/drm_vblank.c b/drivers/gpu/drm/drm_vblank.c index 603ab105125d..fd1fbc77871f 100644 --- a/drivers/gpu/drm/drm_vblank.c +++ b/drivers/gpu/drm/drm_vblank.c @@ -31,7 +31,6 @@ #include <drm/drm_drv.h> #include <drm/drm_framebuffer.h> #include <drm/drm_print.h> -#include <drm/drm_os_linux.h> #include <drm/drm_vblank.h> #include "drm_internal.h" @@ -1670,12 +1669,28 @@ int drm_wait_vblank_ioctl(struct drm_device *dev, void *data, } if (req_seq != seq) { + int wait; + DRM_DEBUG("waiting on vblank count %llu, crtc %u\n", req_seq, pipe); - DRM_WAIT_ON(ret, vblank->queue, 3 * HZ, - vblank_passed(drm_vblank_count(dev, pipe), - req_seq) || - !READ_ONCE(vblank->enabled)); + wait = wait_event_interruptible_timeout(vblank->queue, + vblank_passed(drm_vblank_count(dev, pipe), req_seq) || + !READ_ONCE(vblank->enabled), + msecs_to_jiffies(3000)); + + switch (wait) { + case 0: + /* timeout */ + ret = -EBUSY; + break; + case -ERESTARTSYS: + /* interrupted by signal */ + ret = -EINTR; + break; + default: + ret = 0; + break; + } } if (ret != -EINTR) {
DRM_WAIT_ON() is from the deprecated drm_os_linux header and the modern replacement is the wait_event_*. The return values differ, so a conversion is needed to keep the original interface towards userspace. Introduced a switch/case to make code obvious. Analysis from Michel Dänzer: The waiting condition rely on all relevant places where vblank_count is modified calls wake_up(&vblank->queue). drm_handle_vblank(): - Calls wake_up(&vblank->queue) drm_vblank_enable(): - There is no need here because there can be no sleeping waiters in the queue, because vblank->enabled == false immediately terminates any waits. drm_crtc_accurate_vblank_count(): - This is called from interrupt handlers, at least from amdgpu_dm.c:dm_pflip_high_irq(). Not sure it needs to wake up the queue though, the driver should call drm_(crtc_)_handle_vblank anyway. drm_vblank_disable_and_save(): - It can be called from an interrupt, via drm_handle_vblank -> vblank_disable_fn. However, the only place where drm_vblank_disable_and_save can be called with sleeping waiters in the queue is in drm_crtc_vblank_off, which wakes up the queue afterwards (which terminates all waits, because vblank->enabled == false at this point). v3: - Added analysis to changelog from Michel Dänzer - Moved return result handling inside if (req_seq != seq) (Daniel V) - Reused more of the former logic - resulting in simpler code - Dropped Reviewed-by from Sean Paul as this is a new implmentation v2: - Fix so the case where req_seq equals seq was handled properly - quick hack to check if IGT became happy - Only sent to igt, not to dri-devel Signed-off-by: Sam Ravnborg <sam@ravnborg.org> Cc: "Michel Dänzer" <michel@daenzer.net> Cc: Sean Paul <sean@poorly.run> Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Cc: Maxime Ripard <maxime.ripard@bootlin.com> Cc: David Airlie <airlied@linux.ie> Cc: Daniel Vetter <daniel@ffwll.ch> --- This patch survives my testing here. vbltest spits out the same value as before, and I can see something on the screen. Added intel-gfx@lists.freedesktop.org, as IGT have identified troubles with this in v1. Sam drivers/gpu/drm/drm_vblank.c | 25 ++++++++++++++++++++----- 1 file changed, 20 insertions(+), 5 deletions(-)