drm/i915: Add ppgtt init/release trace points
Subsystems (2)
PATCH 1/3 drm/i915: Add ppgtt init/release trace points
From: Daniele Ceraolo Spurio <[email protected]>
These tracepoints are useful for observing the creation and destruction of Full PPGTTs.
--- drivers/gpu/drm/i915/i915_gem_context.c | 5 +++++ drivers/gpu/drm/i915/i915_trace.h | 38 +++++++++++++++++++++++++++++++++ 2 files changed, 43 insertions(+)
drivers/gpu/drm/i915/i915_gem_context.c +5 -0
@@ -136,6 +136,8 @@ static void ppgtt_release(struct kref *kref)struct i915_hw_ppgtt *ppgtt = container_of(kref, struct i915_hw_ppgtt, ref); +trace_ppgtt_release (ppgtt); +do_ppgtt_cleanup (ppgtt);kfree (ppgtt); }
@@ -215,6 +217,9 @@ create_vm_for_ctx(struct drm_device *dev, struct intel_context *ctx) } ppgtt->ctx = ctx; + +trace_ppgtt_init (ppgtt); +return ppgtt; }
drivers/gpu/drm/i915/i915_trace.h +38 -0
@@ -587,6 +587,44 @@ TRACE_EVENT(intel_gpu_freq_change,TP_printk ("new_freq=%u" , __entry->freq) ); +TRACE_EVENT (ppgtt_init, + + TP_PROTO(struct i915_hw_ppgtt *ppgtt), + +TP_ARGS (ppgtt), + +TP_STRUCT__entry ( + __field(struct i915_hw_ppgtt*, trace_ppgtt) + __field(unsigned int, ppgtt_op_code) + ), + +TP_fast_assign ( + __entry->trace_ppgtt = ppgtt; + __entry->ppgtt_op_code = 0; + ), + +TP_printk ("ppgtt op: %u" , __entry->ppgtt_op_code) +); + +TRACE_EVENT (ppgtt_release, + + TP_PROTO(struct i915_hw_ppgtt *ppgtt), + +TP_ARGS (ppgtt), + +TP_STRUCT__entry ( + __field(struct i915_hw_ppgtt*, trace_ppgtt) + __field(unsigned int, ppgtt_op_code) + ), + +TP_fast_assign ( + __entry->trace_ppgtt = ppgtt; + __entry->ppgtt_op_code = 1; + ), + +TP_printk ("ppgtt op: %u" , __entry->ppgtt_op_code) +); +#endif /* _I915_TRACE_H_ */ /* This part must be outside protection */
Signature · 1 line
1.9.0
PATCH 2/3 drm/i915: Add ctx param to i915_gem_ring_dispatch trace point
From: Daniele Ceraolo Spurio <[email protected]>
The context used top execute a batchbuffer is becoming increasingly important.
--- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 2 +- drivers/gpu/drm/i915/i915_trace.h | 4 ++-- 2 files changed, 3 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/i915_gem_execbuffer.c +1 -1
@@ -1374,7 +1374,7 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data, goto err; } -trace_i915_gem_ring_dispatch (ring,intel_ring_get_seqno (ring), flags); +trace_i915_gem_ring_dispatch (ring,intel_ring_get_seqno (ring), flags, ctx);i915_gem_execbuffer_move_to_active (&eb->vmas, ring);i915_gem_execbuffer_retire_commands (dev, file, ring, batch_obj);
drivers/gpu/drm/i915/i915_trace.h +2 -2
@@ -352,8 +352,8 @@ TRACE_EVENT(i915_gem_ring_sync_to, ); TRACE_EVENT(i915_gem_ring_dispatch, - TP_PROTO(struct intel_engine_cs *ring, u32 seqno, u32 flags), - TP_ARGS(ring, seqno, flags), + TP_PROTO(struct intel_engine_cs *ring, u32 seqno, u32 flags, struct intel_context *ctx), + TP_ARGS(ring, seqno, flags, ctx), TP_STRUCT__entry( __field(u32, dev)
Signature · 1 line
1.9.0
PATCH 3/3 drm/i915: Trace point callbacks for validation
In reply to [email protected]From: Daniele Ceraolo Spurio <[email protected]>
These callbacks can be assigned to specific functions inside an external validation kernel module. This module can then extract run-time information to make sure everything is working as expected.
Specifically, these two callbacks extract information about ifull PPGGT and batch dispatch.
--- drivers/gpu/drm/i915/i915_gem_context.c | 3 +++ drivers/gpu/drm/i915/i915_gem_execbuffer.c | 3 +++ drivers/gpu/drm/i915/i915_trace.h | 21 +++++++++++++++++++++ 3 files changed, 27 insertions(+)
drivers/gpu/drm/i915/i915_gem_context.c +3 -0
@@ -89,6 +89,9 @@#include <drm/i915_drm.h> #include "i915_drv.h" +i915_ppgtt_validation_callback_t ppgtt_validation_callback =NULL ; +EXPORT_SYMBOL_GPL (ppgtt_validation_callback); +/* This is a HW constraint. The value below is the largest known requirement * I've seen in a spec to date, and that was a workaround for a non-shipping * part. It should be safe to decrease this, but it's more future proof as is.
drivers/gpu/drm/i915/i915_gem_execbuffer.c +3 -0
@@ -33,6 +33,9 @@#include "intel_drv.h" #include <linux/dma_remapping.h> +i915_gem_ring_dispatch_callback_t i915_gem_ring_dispatch_validation_callback =NULL ; +EXPORT_SYMBOL_GPL (i915_gem_ring_dispatch_validation_callback); +#define __EXEC_OBJECT_HAS_PIN (1<<31)#define __EXEC_OBJECT_HAS_FENCE (1<<30)#define __EXEC_OBJECT_NEEDS_BIAS (1<<28)
drivers/gpu/drm/i915/i915_trace.h +21 -0
@@ -15,6 +15,17 @@#define TRACE_SYSTEM_STRING __stringify(TRACE_SYSTEM)#define TRACE_INCLUDE_FILE i915_trace +/* validation callbacks */ + +typedef void (*i915_ppgtt_validation_callback_t )(unsigned int action_code, +struct i915_hw_ppgtt *ppgtt); +extern i915_ppgtt_validation_callback_t ppgtt_validation_callback; + +typedef void (*i915_gem_ring_dispatch_callback_t )(struct intel_engine_cs *ring, +u32 seqno,u32 flags, +struct intel_context *ctx); +extern i915_gem_ring_dispatch_callback_t i915_gem_ring_dispatch_validation_callback; +/* pipe updates */ TRACE_EVENT(i915_pipe_update_start,
@@ -368,6 +379,10 @@ TRACE_EVENT(i915_gem_ring_dispatch, __entry->seqno = seqno; __entry->flags = flags;i915_trace_irq_get (ring, seqno); + +if (i915_gem_ring_dispatch_validation_callback) +i915_gem_ring_dispatch_validation_callback (ring, + seqno, flags, ctx); ), TP_printk("dev=%u, ring=%u, seqno=%u, flags=%x",
@@ -601,6 +616,9 @@ TRACE_EVENT(ppgtt_init,TP_fast_assign ( __entry->trace_ppgtt = ppgtt; __entry->ppgtt_op_code =0 ; + +if (ppgtt_validation_callback) +ppgtt_validation_callback (0 , ppgtt); ),TP_printk ("ppgtt op: %u" , __entry->ppgtt_op_code)
@@ -620,6 +638,9 @@ TRACE_EVENT(ppgtt_release,TP_fast_assign ( __entry->trace_ppgtt = ppgtt; __entry->ppgtt_op_code =1 ; + +if (ppgtt_validation_callback) +ppgtt_validation_callback (1 , ppgtt); ),TP_printk ("ppgtt op: %u" , __entry->ppgtt_op_code)
Signature · 1 line
1.9.0
Quoted 49 lines · On Wed, Jun 18, 2014 at 05:16:39PM +0100, [email protected] wrote:
From: Daniele Ceraolo Spurio <[email protected]>
These tracepoints are useful for observing the creation and destruction of Full PPGTTs.
Signed-off-by Oscar Mateo <[email protected]>--- drivers/gpu/drm/i915/i915_gem_context.c | 5 +++++ drivers/gpu/drm/i915/i915_trace.h | 38 +++++++++++++++++++++++++++++++++ 2 files changed, 43 insertions(+)drivers/gpu/drm/i915/i915_gem_context.c +5
-0@@ -136,6 +136,8 @@ static void ppgtt_release(struct kref *kref)struct i915_hw_ppgtt *ppgtt = container_of(kref, struct i915_hw_ppgtt, ref); +trace_ppgtt_release (ppgtt); +do_ppgtt_cleanup (ppgtt);kfree (ppgtt); }@@ -215,6 +217,9 @@ create_vm_for_ctx(struct drm_device *dev, struct intel_context *ctx) } ppgtt->ctx = ctx; + +trace_ppgtt_init (ppgtt); +return ppgtt; }drivers/gpu/drm/i915/i915_trace.h +7
-0@@ -587,6 +587,44 @@ TRACE_EVENT(intel_gpu_freq_change,TP_printk ("new_freq=%u" , __entry->freq) ); +TRACE_EVENT(ppgtt_init, + + TP_PROTO(struct i915_hw_ppgtt *ppgtt), + + TP_ARGS(ppgtt), + + TP_STRUCT__entry(
drivers/gpu/drm/i915/i915_trace.h TRACE_EVENT(intel_gpu_freq_change, · line 606
Quoted 10 lines
drivers/gpu/drm/i915/i915_trace.h +10
-0+ __field(struct i915_hw_ppgtt*, trace_ppgtt) +__field (unsigned int , ppgtt_op_code) + ), + +TP_fast_assign ( + __entry->trace_ppgtt = ppgtt; + __entry->ppgtt_op_code =0 ; + ), + +TP_printk ("ppgtt op: %u" , __entry->ppgtt_op_code)
op is redundant and unuseful since it is already encoded into the tracepoint itself. I'm not happy with the tracepoint name either, but it is close. Note, that we do like to pretend that our driver can coexist with itself. That means we have to pass along the dev->minor here so that listeners can distinguish events between imaginary devices. You need to say which client created the ppgtt and include some method for identifying the ppgtt.
-Chris
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
Quoted 72 lines
-----Original Message-----
From: Chris Wilson [mailto:[email protected]]
Sent: Thursday, June 19, 2014 8:57 AM
To: Mateo Lozano, Oscar
Cc: [email protected]
Subject: Re: [Intel-gfx] [PATCH 1/3] drm/i915: Add ppgtt init/release trace pointsQuoted 9 lines · On Wed, Jun 18, 2014 at 05:16:39PM +0100, [email protected] wrote:
From: Daniele Ceraolo Spurio <[email protected]>
These tracepoints are useful for observing the creation and destruction of Full PPGTTs.
Signed-off-by Oscar Mateo <[email protected]>--- drivers/gpu/drm/i915/i915_gem_context.c | 5 +++++ drivers/gpu/drm/i915/i915_trace.h | 38+++++++++++++++++++++++++++++++++
drivers/gpu/drm/i915/i915_gem_context.c create_vm_for_ctx(struct drm_device *dev, struct
Quoted 17 lines
2 files changed, 43 insertions(+)drivers/gpu/drm/i915/i915_gem_context.c +2
-0@@ -136,6 +136,8 @@ static void ppgtt_release(struct kref *kref)struct i915_hw_ppgtt *ppgtt = container_of(kref, struct i915_hw_ppgtt, ref); +trace_ppgtt_release (ppgtt); +do_ppgtt_cleanup (ppgtt);kfree (ppgtt); }@@ -215,6 +217,9 @@ create_vm_for_ctx(struct drm_device *dev, struct
intel_context*ctx)Quoted 24 lines
drivers/gpu/drm/i915/i915_gem_context.c +3
-0} ppgtt->ctx = ctx; + +trace_ppgtt_init (ppgtt); +return ppgtt; }drivers/gpu/drm/i915/i915_trace.h +7
-0@@ -587,6 +587,44 @@ TRACE_EVENT(intel_gpu_freq_change, TP_printk("new_freq=%u", __entry->freq) ); +TRACE_EVENT(ppgtt_init, + + TP_PROTO(struct i915_hw_ppgtt *ppgtt), + + TP_ARGS(ppgtt), + + TP_STRUCT__entry(drivers/gpu/drm/i915/i915_trace.h TRACE_EVENT(intel_gpu_freq_change, · line 605
Quoted 10 lines
drivers/gpu/drm/i915/i915_trace.h +10
-0+ __field(struct i915_hw_ppgtt*, trace_ppgtt) +__field (unsigned int , ppgtt_op_code) + ), + +TP_fast_assign ( + __entry->trace_ppgtt = ppgtt; + __entry->ppgtt_op_code =0 ; + ), + +TP_printk ("ppgtt op: %u" , __entry->ppgtt_op_code)
Quoted 7 lines
op is redundant and unuseful since it is already encoded into the tracepoint itself. I'm not happy with the tracepoint name either, but it is close. Note, that we do like to pretend that our driver can coexist with itself. That means we have to pass along the
dev->minorhere so that listeners can distinguish events between imaginary devices. You need to say which client created the ppgtt and include some method for identifying the ppgtt.
-Chris
Hmmm... I was submitting this on behalf of our validation team but in retrospect that was a bad idea: it´s much better if they participate in the discussion. Daniele, care to join?
-- Oscar
Quoted 67 lines · Oscar
Quoted 67 lines
-----Original Message-----
From: Chris Wilson [mailto:[email protected]]
Sent: Thursday, June 19, 2014 8:57 AM
To: Mateo Lozano, Oscar
Cc: [email protected]
Subject: Re: [Intel-gfx] [PATCH 1/3] drm/i915: Add ppgtt init/release trace pointsQuoted 9 lines · On Wed, Jun 18, 2014 at 05:16:39PM +0100, [email protected] wrote:
From: Daniele Ceraolo Spurio <[email protected]>
These tracepoints are useful for observing the creation and destruction of Full PPGTTs.
Signed-off-by Oscar Mateo <[email protected]>--- drivers/gpu/drm/i915/i915_gem_context.c | 5 +++++ drivers/gpu/drm/i915/i915_trace.h | 38+++++++++++++++++++++++++++++++++
drivers/gpu/drm/i915/i915_gem_context.c create_vm_for_ctx(struct drm_device *dev, struct
Quoted 17 lines
2 files changed, 43 insertions(+)drivers/gpu/drm/i915/i915_gem_context.c +2
-0@@ -136,6 +136,8 @@ static void ppgtt_release(struct kref *kref)struct i915_hw_ppgtt *ppgtt = container_of(kref, struct i915_hw_ppgtt, ref); +trace_ppgtt_release (ppgtt); +do_ppgtt_cleanup (ppgtt);kfree (ppgtt); }@@ -215,6 +217,9 @@ create_vm_for_ctx(struct drm_device *dev, struct
intel_context*ctx)Quoted 29 lines
drivers/gpu/drm/i915/i915_gem_context.c +3
-0} ppgtt->ctx = ctx; + +trace_ppgtt_init (ppgtt); +return ppgtt; }drivers/gpu/drm/i915/i915_trace.h +12
-0@@ -587,6 +587,44 @@ TRACE_EVENT(intel_gpu_freq_change, TP_printk("new_freq=%u", __entry->freq) ); +TRACE_EVENT(ppgtt_init, + + TP_PROTO(struct i915_hw_ppgtt *ppgtt), + + TP_ARGS(ppgtt), + + TP_STRUCT__entry( + __field(struct i915_hw_ppgtt*, trace_ppgtt) + __field(unsigned int, ppgtt_op_code) + ), + + TP_fast_assign(
Quoted 8 lines
Quoted 8 lines
drivers/gpu/drm/i915/i915_trace.h TRACE_EVENT(intel_gpu_freq_change, · line 606
Quoted 5 lines
drivers/gpu/drm/i915/i915_trace.h +5
-0+ __entry->trace_ppgtt = ppgtt; + __entry->ppgtt_op_code =0 ; + ), + +TP_printk ("ppgtt op: %u" , __entry->ppgtt_op_code)op is redundant and unuseful since it is already encoded into the tracepoint itself. I'm not happy with the tracepoint name either, but it is close. Note,
Would something like "i915_create_ppgtt" be ok?
Quoted 4 lines
Quoted 4 lines
that we do like to pretend that our driver can coexist with itself. That means we have to pass along the
dev->minorhere so that listeners can distinguish events between imaginary devices. You need to say which client created the ppgtt and include some method for identifying the ppgtt.
There are several things that we could use to identify the ppgtt. My choice would be the VM pointer, which is unique for each ppgtt and is also printed by the i915_vma_bind and i915_vma_unbind traces, but we could also use the context id. The problem with using the context id is that we won't be able to print it in the ppgtt_release trace. Regarding the client, my idea is to print the task pid and/or the process name; would that look good for you?
Quoted 5 lines
Quoted 1 line
-Chris
Hmmm... I was submitting this on behalf of our validation team but in retrospect that was a bad idea: it´s much better if they participate in the discussion. Daniele, care to join?
-- Oscar
Thanks,
-- Daniele
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoted 2 lines · Daniele Ceraolo Spurio
Regarding the client, my idea is to print the task pid and/or the process name; would that look good for you?
Going off an tangent. We have a problem with file->pid. In the dri3, or rendernodes, future all pids belong to the master. :(
Suggestions? Complete solutions?
-Chris
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre