vmwgfx: use monotonic event timestamps
diff mbox

Message ID 20171127111629.1682286-1-arnd@arndb.de
State New
Headers show

Commit Message

Arnd Bergmann Nov. 27, 2017, 11:16 a.m. UTC
DRM_VMW_EVENT_FENCE_SIGNALED (struct drm_vmw_event_fence) and
DRM_EVENT_VBLANK (struct drm_event_vblank) pass timestamps in 32-bit
seconds/microseconds format.

As of commit c61eef726a78 ("drm: add support for monotonic vblank
timestamps"), other DRM drivers use monotonic times for drm_event_vblank,
but vmwgfx still uses CLOCK_REALTIME for both events, which suffers from
the y2038/y2106 overflow as well as time jumps.

For consistency, this changes vmwgfx to use ktime_get_ts64 as well,
which solves those problems and avoids the deprecated do_gettimeofday()
function.

This should be transparent to to user space, as long as it doesn't
compare the time against the result of gettimeofday().

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_fence.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

Comments

Sinclair Yeh Dec. 13, 2017, 8:56 p.m. UTC | #1
This looks okay to me.  Although we should change eaction->tv_* type
to 64-bit as well.

I'll roll this in to our next pull request.

thanks,

Sinclair


On Mon, Nov 27, 2017 at 12:16:19PM +0100, Arnd Bergmann wrote:
> DRM_VMW_EVENT_FENCE_SIGNALED (struct drm_vmw_event_fence) and
> DRM_EVENT_VBLANK (struct drm_event_vblank) pass timestamps in 32-bit
> seconds/microseconds format.
> 
> As of commit c61eef726a78 ("drm: add support for monotonic vblank
> timestamps"), other DRM drivers use monotonic times for drm_event_vblank,
> but vmwgfx still uses CLOCK_REALTIME for both events, which suffers from
> the y2038/y2106 overflow as well as time jumps.
> 
> For consistency, this changes vmwgfx to use ktime_get_ts64 as well,
> which solves those problems and avoids the deprecated do_gettimeofday()
> function.
> 
> This should be transparent to to user space, as long as it doesn't
> compare the time against the result of gettimeofday().
> 
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> ---
>  drivers/gpu/drm/vmwgfx/vmwgfx_fence.c | 9 +++++----
>  1 file changed, 5 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
> index d6b1c509ae01..55214d0da66e 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
> @@ -897,11 +897,12 @@ static void vmw_event_fence_action_seq_passed(struct vmw_fence_action *action)
>  	spin_lock_irq(&dev->event_lock);
>  
>  	if (likely(eaction->tv_sec != NULL)) {
> -		struct timeval tv;
> +		struct timespec64 ts;
>  
> -		do_gettimeofday(&tv);
> -		*eaction->tv_sec = tv.tv_sec;
> -		*eaction->tv_usec = tv.tv_usec;
> +		ktime_get_ts64(&ts);
> +		/* monotonic time, so no y2038 overflow */
> +		*eaction->tv_sec = ts.tv_sec;
> +		*eaction->tv_usec = ts.tv_nsec / NSEC_PER_USEC;
>  	}
>  
>  	drm_send_event_locked(dev, eaction->event);
> -- 
> 2.9.0
>
Arnd Bergmann Dec. 14, 2017, 9:05 a.m. UTC | #2
On Wed, Dec 13, 2017 at 9:56 PM, Sinclair Yeh <syeh@vmware.com> wrote:
> This looks okay to me.  Although we should change eaction->tv_* type
> to 64-bit as well.

I thought about it but it would add significant complication without real gain,
as the  eaction->tv_* pointers point into uapi structures (drm_vmw_event_fence
and drm_event_vblank) that are defined as 32-bit times and cannot be
changed. Changing the temporary pointer to 64 bit would require adding
a temporary variable to point to and then doing the truncation in the
callers. With monotonic times, we know that 32-bit times are sufficient,
they are good for 136 years after boot (68 years when interpreted as
signed).

       Arnd

Patch
diff mbox

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
index d6b1c509ae01..55214d0da66e 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
@@ -897,11 +897,12 @@  static void vmw_event_fence_action_seq_passed(struct vmw_fence_action *action)
 	spin_lock_irq(&dev->event_lock);
 
 	if (likely(eaction->tv_sec != NULL)) {
-		struct timeval tv;
+		struct timespec64 ts;
 
-		do_gettimeofday(&tv);
-		*eaction->tv_sec = tv.tv_sec;
-		*eaction->tv_usec = tv.tv_usec;
+		ktime_get_ts64(&ts);
+		/* monotonic time, so no y2038 overflow */
+		*eaction->tv_sec = ts.tv_sec;
+		*eaction->tv_usec = ts.tv_nsec / NSEC_PER_USEC;
 	}
 
 	drm_send_event_locked(dev, eaction->event);