Skip to contentSkip to conversation

drm/i915: Make i915 events part of uapi

PATCH Superseded Ben Widawsky 4 participants 11 messages Jul 16, 2013 – Jul 19, 2013

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/i915: Make i915 events part of uapi

Make the uevent strings part of the user API for people who wish to write their own listeners.

---
 drivers/gpu/drm/i915/i915_irq.c | 8 ++++----
 include/uapi/drm/i915_drm.h     | 3 +++
 2 files changed, 7 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_irq.c +4 -4
index 9910699..60fa513 100644
@@ -812,7 +812,7 @@ static void ivybridge_parity_work(struct work_struct *work)
 
 	mutex_unlock(&dev_priv->dev->struct_mutex);
 
-	parity_event[0] = "L3_PARITY_ERROR=1";
+	parity_event[0] = I915_L3_PARITY_EVENT"=1";
 	parity_event[1] = kasprintf(GFP_KERNEL, "ROW=%d", row);
 	parity_event[2] = kasprintf(GFP_KERNEL, "BANK=%d", bank);
 	parity_event[3] = kasprintf(GFP_KERNEL, "SUBBANK=%d", subbank);
@@ -1435,9 +1435,9 @@ static void i915_error_work_func(struct work_struct *work)
 						    gpu_error);
 	struct drm_device *dev = dev_priv->dev;
 	struct intel_ring_buffer *ring;
-	char *error_event[] = { "ERROR=1", NULL };
-	char *reset_event[] = { "RESET=1", NULL };
-	char *reset_done_event[] = { "ERROR=0", NULL };
+	char *error_event[] = { I915_ERROR_EVENT"=1", NULL };
+	char *reset_event[] = { I915_RESET_EVENT"=1", NULL };
+	char *reset_done_event[] = { I915_ERROR_EVENT"=0", NULL };
 	int i, ret;
 
 	kobject_uevent_env(&dev->primary->kdev.kobj, KOBJ_CHANGE, error_event);
include/uapi/drm/i915_drm.h +3 -0
index 923ed7f..25a3e74 100644
@@ -33,6 +33,9 @@
  * subject to backwards-compatibility constraints.
  */
 
+#define I915_L3_PARITY_EVENT "L3_PARITY_ERROR"
+#define I915_ERROR_EVENT "ERROR"
+#define I915_RESET_EVENT "RESET"
 
 /* Each region is a minimum of 16k, and there are at most 255 of them.
  */
Signature · 1 line
1.8.3.3
#1 lore raw
Quoted 5 lines · Ben Widawsky

Make the uevent strings part of the user API for people who wish to write their own listeners.

One thing I've toyed around with a bit is that we should add kerneldoc to our uapi headers and create a DocBook out of it (maybe as a subsection in the drm userspace api chapter). I guess the DocBook integration needs an overall approach, but we should start to add comments to each piece of userspace api to clearly spec them. See below for what I have in mind ...

Quoted 36 lines
---
 drivers/gpu/drm/i915/i915_irq.c | 8 ++++----
 include/uapi/drm/i915_drm.h     | 3 +++
 2 files changed, 7 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_irq.c +4 -4
index 9910699..60fa513 100644
@@ -812,7 +812,7 @@ static void ivybridge_parity_work(struct work_struct *work)
 
 	mutex_unlock(&dev_priv->dev->struct_mutex);
 
-	parity_event[0] = "L3_PARITY_ERROR=1";
+	parity_event[0] = I915_L3_PARITY_EVENT"=1";
 	parity_event[1] = kasprintf(GFP_KERNEL, "ROW=%d", row);
 	parity_event[2] = kasprintf(GFP_KERNEL, "BANK=%d", bank);
 	parity_event[3] = kasprintf(GFP_KERNEL, "SUBBANK=%d", subbank);
@@ -1435,9 +1435,9 @@ static void i915_error_work_func(struct work_struct *work)
 						    gpu_error);
 	struct drm_device *dev = dev_priv->dev;
 	struct intel_ring_buffer *ring;
-	char *error_event[] = { "ERROR=1", NULL };
-	char *reset_event[] = { "RESET=1", NULL };
-	char *reset_done_event[] = { "ERROR=0", NULL };
+	char *error_event[] = { I915_ERROR_EVENT"=1", NULL };
+	char *reset_event[] = { I915_RESET_EVENT"=1", NULL };
+	char *reset_done_event[] = { I915_ERROR_EVENT"=0", NULL };
 	int i, ret;
 
 	kobject_uevent_env(&dev->primary->kdev.kobj, KOBJ_CHANGE, error_event);
include/uapi/drm/i915_drm.h +0 -0
index 923ed7f..25a3e74 100644

include/uapi/drm/i915_drm.h line 35

Quoted 3 lines
include/uapi/drm/i915_drm.h +0 -0
@@ -33,6 +33,9 @@
  * subject to backwards-compatibility constraints.
  */
 

/**
* DOC uevents generated by i915 on it's device node
*
* These are events generated in addition to the standardized uevents
* genereated by the drm core (like signalling connector state changes). */

/**
* I915_L3_PARITY_EVENT - l3 parity event
*
* Generated when the driver receives a parity mismatch event from the gpu
* l3 cache. Additional information supplied is ROW, BANK, SUBBANK of the
* affected cacheline. Userspace should keep track of these events and if
* a specific cache-line seems to have a persistent error remap it with
* the l3 remapping tool supplied in intel-gpu-tools
*
* The value supplied with the event is always 1.
*/

include/uapi/drm/i915_drm.h line 38

Quoted 3 lines
include/uapi/drm/i915_drm.h +3 -0
+#define I915_L3_PARITY_EVENT "L3_PARITY_ERROR"
+#define I915_ERROR_EVENT "ERROR"
+#define I915_RESET_EVENT "RESET"

I'll let you cook up something nice for the other two ;-)

Cheers, Daniel

Quoted 10 lines
include/uapi/drm/i915_drm.h +0 -0
 /* Each region is a minimum of 16k, and there are at most 255 of them.
  */
Signature · 6 lines
1.8.3.3

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
#2 lore raw
Quoted 4 lines · Daniel Vetter
Quoted 3 lines · Ben Widawsky

Make the uevent strings part of the user API for people who wish to write their own listeners.

Quoted 8 lines
Quoted 2 lines

One thing I've toyed around with a bit is that we should add kerneldoc to our uapi headers and create a DocBook out of it (maybe as a subsection in the drm userspace api chapter). I guess the DocBook integration needs an overall approach, but we should start to add comments to each piece of userspace api to clearly spec them. See below for what I have in mind ...

Yes, docs please. I don't have the kernel-fu of a kernel-dev, so without docs I don't know what these events mean and what to expect from them.

Quoted 2 lines

drivers/gpu/drm/i915/i915_irq.c static void ivybridge_parity_work(struct work_struct *work) · line 815

Quoted 2 lines
drivers/gpu/drm/i915/i915_irq.c +1 -1
-	parity_event[0] = "L3_PARITY_ERROR=1";
+	parity_event[0] = I915_L3_PARITY_EVENT"=1";

Small nitpick. I usually find string concatenation more readable like this, with a space:

    parity_event[0] = I915_L3_PARITY_EVENT "=1";

But, that's just my preference, so feel free to ignore me.

Quoted 3 lines

include/uapi/drm/i915_drm.h line 38

Quoted 3 lines
include/uapi/drm/i915_drm.h +3 -0
+#define I915_L3_PARITY_EVENT "L3_PARITY_ERROR"
+#define I915_ERROR_EVENT "ERROR"
+#define I915_RESET_EVENT "RESET"

Maybe this is a dumb question... since these are uevents, do you think the names would be improved by given them a "UEVENT" suffix? Like I915_ERROR_UEVENT? Or is that dumb, because these tokens are intended to serve more purposes than uevents?

#3 lore raw

PATCH 1/3 drm/i915: Move error uevent to detection time

In reply to Ben Widawsky

We've recently deferred error handling with a workqueue for a number of reasons. However, when we're trying to pass the information to userspace, it's likely more interesting to know when the error was detected by the kernel, and not when we eventually get around to handling it in the workqueue.

This was inspired as a result of the patch to move these events to the uapi.

---
 drivers/gpu/drm/i915/i915_irq.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/i915_irq.c +3 -3
index 9910699..9fe430a 100644
@@ -1435,13 +1435,10 @@ static void i915_error_work_func(struct work_struct *work)
 						    gpu_error);
 	struct drm_device *dev = dev_priv->dev;
 	struct intel_ring_buffer *ring;
-	char *error_event[] = { "ERROR=1", NULL };
 	char *reset_event[] = { "RESET=1", NULL };
 	char *reset_done_event[] = { "ERROR=0", NULL };
 	int i, ret;
 
-	kobject_uevent_env(&dev->primary->kdev.kobj, KOBJ_CHANGE, error_event);
-
 	/*
 	 * Note that there's only one work item which does gpu resets, so we
 	 * need not worry about concurrent gpu resets potentially incrementing
@@ -1594,11 +1591,14 @@ void i915_handle_error(struct drm_device *dev, bool wedged)
 {
 	struct drm_i915_private *dev_priv = dev->dev_private;
 	struct intel_ring_buffer *ring;
+	char *error_event[] = { "ERROR=1", NULL };
 	int i;
 
 	i915_capture_error_state(dev);
 	i915_report_and_clear_eir(dev);
 
+	kobject_uevent_env(&dev->primary->kdev.kobj, KOBJ_CHANGE, error_event);
+
 	if (wedged) {
 		atomic_set_mask(I915_RESET_IN_PROGRESS_FLAG,
 				&dev_priv->gpu_error.reset_counter);
Signature · 1 line
1.8.3.3
#4 lore raw

PATCH 2/3 drm/i915: Change uevent for reset completion

Prior to this patch, we used ERROR=0 as a way of signifying the reset was complete. The functionality dates back quite a ways to:

commit f316a42cc49eca73b33d85feb6177e32431747ff

Author: Ben Gamari <[email protected]>
Date:   Mon Sep 14 17:48:46 2009 -0400

drm/i915: Hookup chip reset in error handler

I'm not really sure what the motivation for this was originally, but to me it makes more sense to have a distinct event for error detection, and another event for reset start/finish (since reset is prone to failure).

---
 drivers/gpu/drm/i915/i915_irq.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/i915/i915_irq.c +1 -1
index 9fe430a..03071d7 100644
@@ -1436,7 +1436,7 @@ static void i915_error_work_func(struct work_struct *work)
 	struct drm_device *dev = dev_priv->dev;
 	struct intel_ring_buffer *ring;
 	char *reset_event[] = { "RESET=1", NULL };
-	char *reset_done_event[] = { "ERROR=0", NULL };
+	char *reset_done_event[] = { "RESET=0", NULL };
 	int i, ret;
 
 	/*
Signature · 1 line
1.8.3.3
#5 lore raw

Re: [PATCH 2/3] drm/i915: Change uevent for reset completion

Quoted 23 lines · Ben Widawsky

Prior to this patch, we used ERROR=0 as a way of signifying the reset was complete. The functionality dates back quite a ways to:

commit f316a42cc49eca73b33d85feb6177e32431747ff

Author: Ben Gamari <[email protected]>
Date:   Mon Sep 14 17:48:46 2009 -0400

drm/i915: Hookup chip reset in error handler

I'm not really sure what the motivation for this was originally, but to me it makes more sense to have a distinct event for error detection, and another event for reset start/finish (since reset is prone to failure).

---
 drivers/gpu/drm/i915/i915_irq.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/i915/i915_irq.c +0 -0
index 9fe430a..03071d7 100644

drivers/gpu/drm/i915/i915_irq.c static void i915_error_work_func(struct work_struct *work) · line 1439

Quoted 5 lines
drivers/gpu/drm/i915/i915_irq.c +1 -1
@@ -1436,7 +1436,7 @@ static void i915_error_work_func(struct work_struct *work)
 	struct drm_device *dev = dev_priv->dev;
 	struct intel_ring_buffer *ring;
 	char *reset_event[] = { "RESET=1", NULL };
-	char *reset_done_event[] = { "ERROR=0", NULL };
+	char *reset_done_event[] = { "RESET=0", NULL };

It's a kernel api change so we need to check all distros and everything else out there whether they don't use it. Imo not worth the effort since signalling that the ERROR condition has passed is somewhat sensible, too. Can you please respin with this part dropped?
-Daniel

Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
#6 lore raw

PATCH v2 3/3 drm/i915: Make i915 events part of uapi

In reply to Ben Widawsky

Make the uevent strings part of the user API for people who wish to write their own listeners.

v2: Make a space in the string concatenation. (Chad) Use the "UEVENT" suffix intead of "EVENT" (Chad) Make kernel-doc (Daniel)

Thanks to Daniel Vetter for a majority of the parity error comments.

---
 drivers/gpu/drm/i915/i915_irq.c |  8 ++++----
 include/uapi/drm/i915_drm.h     | 22 ++++++++++++++++++++++
 2 files changed, 26 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_irq.c +4 -4
index 03071d7..3e19ce0 100644
@@ -812,7 +812,7 @@ static void ivybridge_parity_work(struct work_struct *work)
 
 	mutex_unlock(&dev_priv->dev->struct_mutex);
 
-	parity_event[0] = "L3_PARITY_ERROR=1";
+	parity_event[0] = I915_L3_PARITY_UEVENT "=1";
 	parity_event[1] = kasprintf(GFP_KERNEL, "ROW=%d", row);
 	parity_event[2] = kasprintf(GFP_KERNEL, "BANK=%d", bank);
 	parity_event[3] = kasprintf(GFP_KERNEL, "SUBBANK=%d", subbank);
@@ -1435,8 +1435,8 @@ static void i915_error_work_func(struct work_struct *work)
 						    gpu_error);
 	struct drm_device *dev = dev_priv->dev;
 	struct intel_ring_buffer *ring;
-	char *reset_event[] = { "RESET=1", NULL };
-	char *reset_done_event[] = { "RESET=0", NULL };
+	char *reset_event[] = { I915_RESET_UEVENT "=1", NULL };
+	char *reset_done_event[] = { I915_RESET_UEVENT "=0", NULL };
 	int i, ret;
 
 	/*
@@ -1591,7 +1591,7 @@ void i915_handle_error(struct drm_device *dev, bool wedged)
 {
 	struct drm_i915_private *dev_priv = dev->dev_private;
 	struct intel_ring_buffer *ring;
-	char *error_event[] = { "ERROR=1", NULL };
+	char *error_event[] = { I915_ERROR_UEVENT "=1", NULL };
 	int i;
 
 	i915_capture_error_state(dev);
include/uapi/drm/i915_drm.h +22 -0
index 923ed7f..f5fe31b 100644
@@ -33,6 +33,28 @@
  * subject to backwards-compatibility constraints.
  */
 
+/**
+ * DOC: uevents generated by i915 on it's device node
+ *
+ * I915_L3_PARITY_UEVENT - Generated when the driver receives a parity mismatch
+ *	event from the gpu l3 cache. Additional information supplied is ROW,
+ *	BANK, SUBBANK of the affected cacheline. Userspace should keep track of
+ *	these events and if a specific cache-line seems to have a persistent
+ *	error remap it with the l3 remapping tool supplied in intel-gpu-tools.
+ *	The value supplied with the event is always 1.
+ *
+ * I915_ERROR_UEVENT - Generated upon error detection, currently only via
+ *	hangcheck. The error detection event is a good indicator of when things
+ *	began to go badly. The value supplied with the event is always 1.
+ *
+ * I915_RESET_UEVENT - Event is generated just before an attempt to reset the
+ *	the GPU. Reset has historically been a precarious attempt which the GPU
+ *	may or may not recover from. The value supplied with the event is 1
+ *	before the reset attempt, and a 0 after it has completed.
+ */
+#define I915_L3_PARITY_UEVENT "L3_PARITY_ERROR"
+#define I915_ERROR_UEVENT "ERROR"
+#define I915_RESET_UEVENT "RESET"
 
 /* Each region is a minimum of 16k, and there are at most 255 of them.
  */
Signature · 1 line
1.8.3.3
#7 lore raw

Re: [PATCH 1/3] drm/i915: Move error uevent to detection time

In reply to Ben Widawsky
Quoted 37 lines · Ben Widawsky

We've recently deferred error handling with a workqueue for a number of reasons. However, when we're trying to pass the information to userspace, it's likely more interesting to know when the error was detected by the kernel, and not when we eventually get around to handling it in the workqueue.

This was inspired as a result of the patch to move these events to the uapi.

---
 drivers/gpu/drm/i915/i915_irq.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/i915_irq.c +1 -3
index 9910699..9fe430a 100644
@@ -1435,13 +1435,10 @@ static void i915_error_work_func(struct work_struct *work)
 						    gpu_error);
 	struct drm_device *dev = dev_priv->dev;
 	struct intel_ring_buffer *ring;
-	char *error_event[] = { "ERROR=1", NULL };
 	char *reset_event[] = { "RESET=1", NULL };
 	char *reset_done_event[] = { "ERROR=0", NULL };
 	int i, ret;
 
-	kobject_uevent_env(&dev->primary->kdev.kobj, KOBJ_CHANGE, error_event);
-
 	/*
 	 * Note that there's only one work item which does gpu resets, so we
 	 * need not worry about concurrent gpu resets potentially incrementing
@@ -1594,11 +1591,14 @@ void i915_handle_error(struct drm_device *dev, bool wedged)
 {
 	struct drm_i915_private *dev_priv = dev->dev_private;
 	struct intel_ring_buffer *ring;
+	char *error_event[] = { "ERROR=1", NULL };

drivers/gpu/drm/i915/i915_irq.c void i915_handle_error(struct drm_device *dev, bool wedged) · line 1604

Quoted 10 lines
drivers/gpu/drm/i915/i915_irq.c +2 -0
 	int i;
 
 	i915_capture_error_state(dev);
 	i915_report_and_clear_eir(dev);
 
+	kobject_uevent_env(&dev->primary->kdev.kobj, KOBJ_CHANGE, error_event);
+
 	if (wedged) {
 		atomic_set_mask(I915_RESET_IN_PROGRESS_FLAG,
 				&dev_priv->gpu_error.reset_counter);

This leaves the reset counter unchanged prior to the uevent, if you move the uevent signalling down to just before the queue_work() there would be no userspace visible changes (other than timing, which is a moot point).

However, kobject_uevent_env() is not suitable for calling from irq-context.
-Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#8 lore raw

PATCH v3 1/1 drm/i915: Make i915 events part of uapi

In reply to Ben Widawsky

Make the uevent strings part of the user API for people who wish to write their own listeners.

v2: Make a space in the string concatenation. (Chad) Use the "UEVENT" suffix intead of "EVENT" (Chad) Make kernel-doc parseable Docbook comments (Daniel)

v3: Undid reset change introduced in last submission (Daniel) Fixed up comments to address removal changes.

Thanks to Daniel Vetter for a majority of the parity error comments.

---
 drivers/gpu/drm/i915/i915_irq.c |  8 ++++----
 include/uapi/drm/i915_drm.h     | 24 ++++++++++++++++++++++++
 2 files changed, 28 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_irq.c +4 -4
index 9910699..4e4fa5f 100644
@@ -812,7 +812,7 @@ static void ivybridge_parity_work(struct work_struct *work)
 
 	mutex_unlock(&dev_priv->dev->struct_mutex);
 
-	parity_event[0] = "L3_PARITY_ERROR=1";
+	parity_event[0] = I915_L3_PARITY_UEVENT "=1";
 	parity_event[1] = kasprintf(GFP_KERNEL, "ROW=%d", row);
 	parity_event[2] = kasprintf(GFP_KERNEL, "BANK=%d", bank);
 	parity_event[3] = kasprintf(GFP_KERNEL, "SUBBANK=%d", subbank);
@@ -1435,9 +1435,9 @@ static void i915_error_work_func(struct work_struct *work)
 						    gpu_error);
 	struct drm_device *dev = dev_priv->dev;
 	struct intel_ring_buffer *ring;
-	char *error_event[] = { "ERROR=1", NULL };
-	char *reset_event[] = { "RESET=1", NULL };
-	char *reset_done_event[] = { "ERROR=0", NULL };
+	char *error_event[] = { I915_ERROR_UEVENT "=1", NULL };
+	char *reset_event[] = { I915_RESET_UEVENT "=1", NULL };
+	char *reset_done_event[] = { I915_ERROR_UEVENT "=0", NULL };
 	int i, ret;
 
 	kobject_uevent_env(&dev->primary->kdev.kobj, KOBJ_CHANGE, error_event);
include/uapi/drm/i915_drm.h +24 -0
index 923ed7f..a1a7b6b 100644
@@ -33,6 +33,30 @@
  * subject to backwards-compatibility constraints.
  */
 
+/**
+ * DOC: uevents generated by i915 on it's device node
+ *
+ * I915_L3_PARITY_UEVENT - Generated when the driver receives a parity mismatch
+ *	event from the gpu l3 cache. Additional information supplied is ROW,
+ *	BANK, SUBBANK of the affected cacheline. Userspace should keep track of
+ *	these events and if a specific cache-line seems to have a persistent
+ *	error remap it with the l3 remapping tool supplied in intel-gpu-tools.
+ *	The value supplied with the event is always 1.
+ *
+ * I915_ERROR_UEVENT - Generated upon error detection, currently only via
+ *	hangcheck. The error detection event is a good indicator of when things
+ *	began to go badly. The value supplied with the event is a 1 upon error
+ *	detection, and a 0 upon reset completion, signifying no more error
+ *	exists. NOTE: Disabling hangcheck or reset via module parameter will
+ *	cause the related events to not be seen.
+ *
+ * I915_RESET_UEVENT - Event is generated just before an attempt to reset the
+ *	the GPU. The value supplied with the event is always 1. NOTE: Disable
+ *	reset via module parameter will cause this event to not be seen.
+ */
+#define I915_L3_PARITY_UEVENT		"L3_PARITY_ERROR"
+#define I915_ERROR_UEVENT		"ERROR"
+#define I915_RESET_UEVENT		"RESET"
 
 /* Each region is a minimum of 16k, and there are at most 255 of them.
  */
Signature · 1 line
1.8.3.3
#9 lore raw
Quoted 7 lines · Ben Widawsky

Make the uevent strings part of the user API for people who wish to write their own listeners.

v2: Make a space in the string concatenation. (Chad) Use the "UEVENT" suffix intead of "EVENT" (Chad) Make kernel-doc parseable Docbook comments (Daniel)

Quoted 7 lines

v3: Undid reset change introduced in last submission (Daniel) Fixed up comments to address removal changes.

Thanks to Daniel Vetter for a majority of the parity error comments.

Queued for -next, thanks for the patch.
-Daniel

Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
Reviewed-by on v3 1/1
#10 lore raw
In reply to Ben Widawsky
Quoted 10 lines · Ben Widawsky

Make the uevent strings part of the user API for people who wish to write their own listeners.

v2: Make a space in the string concatenation. (Chad) Use the "UEVENT" suffix intead of "EVENT" (Chad) Make kernel-doc parseable Docbook comments (Daniel)

v3: Undid reset change introduced in last submission (Daniel) Fixed up comments to address removal changes.

Quoted 8 lines

Thanks to Daniel Vetter for a majority of the parity error comments.

---
  drivers/gpu/drm/i915/i915_irq.c |  8 ++++----
  include/uapi/drm/i915_drm.h     | 24 ++++++++++++++++++++++++
  2 files changed, 28 insertions(+), 4 deletions(-)

For what it's worth,