[14/15] drm/i915: Only query timestamp when measuring elapsed time

Submitted by Chris Wilson on Nov. 29, 2015, 8:48 a.m.

Details

Message ID 1448786893-2522-15-git-send-email-chris@chris-wilson.co.uk
State New
Headers show
Series "Series without cover letter" ( rev: 8 7 6 5 4 3 2 1 ) in Intel GFX

Not browsing as part of any series.

Commit Message

Chris Wilson Nov. 29, 2015, 8:48 a.m.
Avoid the two calls to ktime_get_raw_ns() (at best it reads the TSC) as
we only need to compute the elapsed time for a timed wait.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem.c | 13 +++++--------
 1 file changed, 5 insertions(+), 8 deletions(-)

Patch hide | download patch | download mbox

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 2c3e36e19cb0..871201713c73 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -1227,7 +1227,6 @@  int __i915_wait_request(struct drm_i915_gem_request *req,
 	int state = interruptible ? TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE;
 	DEFINE_WAIT(wait);
 	unsigned long timeout_remain;
-	s64 before, now;
 	int ret;
 
 	if (list_empty(&req->list))
@@ -1244,13 +1243,12 @@  int __i915_wait_request(struct drm_i915_gem_request *req,
 		if (*timeout == 0)
 			return -ETIME;
 
+		/* Record current time in case interrupted, or wedged */
 		timeout_remain = nsecs_to_jiffies_timeout(*timeout);
+		*timeout += ktime_get_raw_ns();
 	}
 
-	/* Record current time in case interrupted by signal, or wedged */
 	trace_i915_gem_request_wait_begin(req);
-	before = ktime_get_raw_ns();
-
 	if (INTEL_INFO(req->i915)->gen >= 6)
 		gen6_rps_boost(req->i915, rps, req->emitted_jiffies);
 
@@ -1286,14 +1284,13 @@  int __i915_wait_request(struct drm_i915_gem_request *req,
 	}
 	finish_wait(&req->wait, &wait);
 out:
-	now = ktime_get_raw_ns();
 	intel_breadcrumbs_remove_waiter(req);
 	trace_i915_gem_request_wait_end(req);
 
 	if (timeout) {
-		s64 tres = *timeout - (now - before);
-
-		*timeout = tres < 0 ? 0 : tres;
+		*timeout -= ktime_get_raw_ns();
+		if (*timeout < 0)
+			*timeout = 0;
 
 		/*
 		 * Apparently ktime isn't accurate enough and occasionally has a

Comments

On 29/11/15 08:48, Chris Wilson wrote:
> Avoid the two calls to ktime_get_raw_ns() (at best it reads the TSC) as
> we only need to compute the elapsed time for a timed wait.
>
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>   drivers/gpu/drm/i915/i915_gem.c | 13 +++++--------
>   1 file changed, 5 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index 2c3e36e19cb0..871201713c73 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -1227,7 +1227,6 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
>   	int state = interruptible ? TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE;
>   	DEFINE_WAIT(wait);
>   	unsigned long timeout_remain;
> -	s64 before, now;
>   	int ret;
>
>   	if (list_empty(&req->list))
> @@ -1244,13 +1243,12 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
>   		if (*timeout == 0)
>   			return -ETIME;
>
> +		/* Record current time in case interrupted, or wedged */
>   		timeout_remain = nsecs_to_jiffies_timeout(*timeout);
> +		*timeout += ktime_get_raw_ns();

Don't really like this one, how you use the passed in pointer to store 
the intermediate local state.

It works etc but just feels too hacky.

Regards,

Tvrtko


>   	}
>
> -	/* Record current time in case interrupted by signal, or wedged */
>   	trace_i915_gem_request_wait_begin(req);
> -	before = ktime_get_raw_ns();
> -
>   	if (INTEL_INFO(req->i915)->gen >= 6)
>   		gen6_rps_boost(req->i915, rps, req->emitted_jiffies);
>
> @@ -1286,14 +1284,13 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
>   	}
>   	finish_wait(&req->wait, &wait);
>   out:
> -	now = ktime_get_raw_ns();
>   	intel_breadcrumbs_remove_waiter(req);
>   	trace_i915_gem_request_wait_end(req);
>
>   	if (timeout) {
> -		s64 tres = *timeout - (now - before);
> -
> -		*timeout = tres < 0 ? 0 : tres;
> +		*timeout -= ktime_get_raw_ns();
> +		if (*timeout < 0)
> +			*timeout = 0;
>
>   		/*
>   		 * Apparently ktime isn't accurate enough and occasionally has a
>
On Mon, Nov 30, 2015 at 10:19:40AM +0000, Tvrtko Ursulin wrote:
> 
> On 29/11/15 08:48, Chris Wilson wrote:
> >Avoid the two calls to ktime_get_raw_ns() (at best it reads the TSC) as
> >we only need to compute the elapsed time for a timed wait.
> >
> >Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> >---
> >  drivers/gpu/drm/i915/i915_gem.c | 13 +++++--------
> >  1 file changed, 5 insertions(+), 8 deletions(-)
> >
> >diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> >index 2c3e36e19cb0..871201713c73 100644
> >--- a/drivers/gpu/drm/i915/i915_gem.c
> >+++ b/drivers/gpu/drm/i915/i915_gem.c
> >@@ -1227,7 +1227,6 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
> >  	int state = interruptible ? TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE;
> >  	DEFINE_WAIT(wait);
> >  	unsigned long timeout_remain;
> >-	s64 before, now;
> >  	int ret;
> >
> >  	if (list_empty(&req->list))
> >@@ -1244,13 +1243,12 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
> >  		if (*timeout == 0)
> >  			return -ETIME;
> >
> >+		/* Record current time in case interrupted, or wedged */
> >  		timeout_remain = nsecs_to_jiffies_timeout(*timeout);
> >+		*timeout += ktime_get_raw_ns();
> 
> Don't really like this one, how you use the passed in pointer to
> store the intermediate local state.
> 
> It works etc but just feels too hacky.

Bah, it's already in the CPU cache I might as well take advantage of the
space. :)
-Chris