diff mbox series

[RFC,5/6] drm/i915: Fail too long user submissions by default

Message ID 20210312154622.1767865-6-tvrtko.ursulin@linux.intel.com (mailing list archive)
State New, archived
Headers show
Series Default request/fence expiry + watchdog | expand

Commit Message

Tvrtko Ursulin March 12, 2021, 3:46 p.m. UTC 
From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

A new Kconfig option CONFIG_DRM_I915_REQUEST_TIMEOUT is added, defaulting
to 10s, and this timeout is applied to _all_ contexts using the previously
added watchdog facility.

Result of this is that any user submission will simply fail after this
time, either causing a reset (for non-preemptable) or incomplete results.

This can have an effect that workloads which used to work fine will
suddenly start failing.

When the default expiry is active userspace will not be allowed to
decrease the timeout using the context param setting.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
---
 drivers/gpu/drm/i915/Kconfig.profile        |  8 ++++
 drivers/gpu/drm/i915/gem/i915_gem_context.c | 47 ++++++++++++++++++---
 2 files changed, 48 insertions(+), 7 deletions(-)

Comments

Daniel Vetter March 16, 2021, 10:10 a.m. UTC | #1 
On Fri, Mar 12, 2021 at 03:46:21PM +0000, Tvrtko Ursulin wrote:
> From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
> 
> A new Kconfig option CONFIG_DRM_I915_REQUEST_TIMEOUT is added, defaulting
> to 10s, and this timeout is applied to _all_ contexts using the previously
> added watchdog facility.
> 
> Result of this is that any user submission will simply fail after this
> time, either causing a reset (for non-preemptable) or incomplete results.
> 
> This can have an effect that workloads which used to work fine will
> suddenly start failing.
> 
> When the default expiry is active userspace will not be allowed to
> decrease the timeout using the context param setting.
> 
> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
> Cc: Daniel Vetter <daniel.vetter@ffwll.ch>

I think this should explain that it will break long running compute
workloads, and that maybe the modparam in the next patch can paper over
that until we've implemented proper long running compute workload support
in upstream. Which is unfortunately still some ways off.

Otherwise makes all sense to me. Maybe if you want also copy some of the
discussion from your cover letter into this commit message, and think
there's some good stuff there.
-Daniel

> ---
>  drivers/gpu/drm/i915/Kconfig.profile        |  8 ++++
>  drivers/gpu/drm/i915/gem/i915_gem_context.c | 47 ++++++++++++++++++---
>  2 files changed, 48 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile
> index 35bbe2b80596..55e157ffff73 100644
> --- a/drivers/gpu/drm/i915/Kconfig.profile
> +++ b/drivers/gpu/drm/i915/Kconfig.profile
> @@ -1,3 +1,11 @@
> +config DRM_I915_REQUEST_TIMEOUT
> +	int "Default timeout for requests (ms)"
> +	default 10000 # milliseconds
> +	help
> +	  ...
> +
> +	  May be 0 to disable the timeout.
> +
>  config DRM_I915_FENCE_TIMEOUT
>  	int "Timeout for unsignaled foreign fences (ms, jiffy granularity)"
>  	default 10000 # milliseconds
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context.c b/drivers/gpu/drm/i915/gem/i915_gem_context.c
> index 32b05af4fc8f..21c0176e27a0 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_context.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_context.c
> @@ -854,6 +854,25 @@ static void __assign_timeline(struct i915_gem_context *ctx,
>  	context_apply_all(ctx, __apply_timeline, timeline);
>  }
>  
> +static int
> +__set_watchdog(struct i915_gem_context *ctx, unsigned long timeout_us);
> +
> +static void __set_default_fence_expiry(struct i915_gem_context *ctx)
> +{
> +	struct drm_i915_private *i915 = ctx->i915;
> +	int ret;
> +
> +	if (!IS_ACTIVE(CONFIG_DRM_I915_REQUEST_TIMEOUT))
> +		return;
> +
> +	/* Default expiry for user fences. */
> +	ret = __set_watchdog(ctx, CONFIG_DRM_I915_REQUEST_TIMEOUT * 1000);
> +	if (ret)
> +		drm_notice(&i915->drm,
> +			   "Failed to configure default fence expiry! (%d)",
> +			   ret);
> +}
> +
>  static struct i915_gem_context *
>  i915_gem_create_context(struct drm_i915_private *i915, unsigned int flags)
>  {
> @@ -898,6 +917,8 @@ i915_gem_create_context(struct drm_i915_private *i915, unsigned int flags)
>  		intel_timeline_put(timeline);
>  	}
>  
> +	__set_default_fence_expiry(ctx);
> +
>  	trace_i915_context_create(ctx);
>  
>  	return ctx;
> @@ -1404,23 +1425,35 @@ static int __apply_watchdog(struct intel_context *ce, void *timeout_us)
>  	return intel_context_set_watchdog_us(ce, (uintptr_t)timeout_us);
>  }
>  
> -static int set_watchdog(struct i915_gem_context *ctx,
> -			struct drm_i915_gem_context_param *args)
> +static int
> +__set_watchdog(struct i915_gem_context *ctx, unsigned long timeout_us)
>  {
>  	int ret;
>  
> -	if (args->size)
> -		return -EINVAL;
> -
>  	ret = context_apply_all(ctx, __apply_watchdog,
> -				(void *)(uintptr_t)args->value);
> +				(void *)(uintptr_t)timeout_us);
>  
>  	if (!ret)
> -		ctx->watchdog.timeout_us = args->value;
> +		ctx->watchdog.timeout_us = timeout_us;
>  
>  	return ret;
>  }
>  
> +static int set_watchdog(struct i915_gem_context *ctx,
> +			struct drm_i915_gem_context_param *args)
> +{
> +	if (args->size)
> +		return -EINVAL;
> +
> +	/* Disallow disabling or configuring longer watchdog than default. */
> +	if (IS_ACTIVE(CONFIG_DRM_I915_REQUEST_TIMEOUT) &&
> +	    (!args->value ||
> +	     args->value > CONFIG_DRM_I915_REQUEST_TIMEOUT * 1000))
> +		return -EPERM;
> +
> +	return __set_watchdog(ctx, args->value);
> +}
> +
>  static int __get_ringsize(struct intel_context *ce, void *arg)
>  {
>  	long sz;
> -- 
> 2.27.0
>
diff mbox series

Patch

diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile
index 35bbe2b80596..55e157ffff73 100644
--- a/drivers/gpu/drm/i915/Kconfig.profile
+++ b/drivers/gpu/drm/i915/Kconfig.profile
@@ -1,3 +1,11 @@ 
+config DRM_I915_REQUEST_TIMEOUT
+	int "Default timeout for requests (ms)"
+	default 10000 # milliseconds
+	help
+	  ...
+
+	  May be 0 to disable the timeout.
+
 config DRM_I915_FENCE_TIMEOUT
 	int "Timeout for unsignaled foreign fences (ms, jiffy granularity)"
 	default 10000 # milliseconds
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context.c b/drivers/gpu/drm/i915/gem/i915_gem_context.c
index 32b05af4fc8f..21c0176e27a0 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_context.c
@@ -854,6 +854,25 @@  static void __assign_timeline(struct i915_gem_context *ctx,
 	context_apply_all(ctx, __apply_timeline, timeline);
 }
 
+static int
+__set_watchdog(struct i915_gem_context *ctx, unsigned long timeout_us);
+
+static void __set_default_fence_expiry(struct i915_gem_context *ctx)
+{
+	struct drm_i915_private *i915 = ctx->i915;
+	int ret;
+
+	if (!IS_ACTIVE(CONFIG_DRM_I915_REQUEST_TIMEOUT))
+		return;
+
+	/* Default expiry for user fences. */
+	ret = __set_watchdog(ctx, CONFIG_DRM_I915_REQUEST_TIMEOUT * 1000);
+	if (ret)
+		drm_notice(&i915->drm,
+			   "Failed to configure default fence expiry! (%d)",
+			   ret);
+}
+
 static struct i915_gem_context *
 i915_gem_create_context(struct drm_i915_private *i915, unsigned int flags)
 {
@@ -898,6 +917,8 @@  i915_gem_create_context(struct drm_i915_private *i915, unsigned int flags)
 		intel_timeline_put(timeline);
 	}
 
+	__set_default_fence_expiry(ctx);
+
 	trace_i915_context_create(ctx);
 
 	return ctx;
@@ -1404,23 +1425,35 @@  static int __apply_watchdog(struct intel_context *ce, void *timeout_us)
 	return intel_context_set_watchdog_us(ce, (uintptr_t)timeout_us);
 }
 
-static int set_watchdog(struct i915_gem_context *ctx,
-			struct drm_i915_gem_context_param *args)
+static int
+__set_watchdog(struct i915_gem_context *ctx, unsigned long timeout_us)
 {
 	int ret;
 
-	if (args->size)
-		return -EINVAL;
-
 	ret = context_apply_all(ctx, __apply_watchdog,
-				(void *)(uintptr_t)args->value);
+				(void *)(uintptr_t)timeout_us);
 
 	if (!ret)
-		ctx->watchdog.timeout_us = args->value;
+		ctx->watchdog.timeout_us = timeout_us;
 
 	return ret;
 }
 
+static int set_watchdog(struct i915_gem_context *ctx,
+			struct drm_i915_gem_context_param *args)
+{
+	if (args->size)
+		return -EINVAL;
+
+	/* Disallow disabling or configuring longer watchdog than default. */
+	if (IS_ACTIVE(CONFIG_DRM_I915_REQUEST_TIMEOUT) &&
+	    (!args->value ||
+	     args->value > CONFIG_DRM_I915_REQUEST_TIMEOUT * 1000))
+		return -EPERM;
+
+	return __set_watchdog(ctx, args->value);
+}
+
 static int __get_ringsize(struct intel_context *ce, void *arg)
 {
 	long sz;