Skip to contentSkip to conversation

Enable Gen 7 Observation Architecture

PATCH In review Robert Bragg 3 participants 13 messages Feb 3, 2016 – Feb 4, 2016

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

cover Enable Gen 7 Observation Architecture

Compared to the last revision, this is just rebased on drm-intel-nightly

Robert Bragg (8):
drm/i915: Add i915 perf infrastructure drm/i915: rename OACONTROL GEN7_OACONTROL drm/i915: Add 'render basic' Haswell OA unit config drm/i915: Add i915 perf event for Haswell OA unit drm/i915: advertise available metrics via sysfs drm/i915: Add dev.i915.perf_event_paranoid sysctl option drm/i915: add oa_event_min_timer_exponent sysctl drm/i915: Add more Haswell OA metric sets

 drivers/gpu/drm/i915/Makefile           |    4 +
 drivers/gpu/drm/i915/i915_cmd_parser.c  |    4 +-
 drivers/gpu/drm/i915/i915_dma.c         |    7 +
 drivers/gpu/drm/i915/i915_drv.h         |  154 ++++
 drivers/gpu/drm/i915/i915_gem_context.c |   23 +-
 drivers/gpu/drm/i915/i915_oa_hsw.c      |  658 ++++++++++++++++
 drivers/gpu/drm/i915/i915_oa_hsw.h      |   38 +
 drivers/gpu/drm/i915/i915_perf.c        | 1283 +++++++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_reg.h         |  340 +++++++-
 include/uapi/drm/i915_drm.h             |  121 +++
 10 files changed, 2625 insertions(+), 7 deletions(-)
 create mode 100644 drivers/gpu/drm/i915/i915_oa_hsw.c
 create mode 100644 drivers/gpu/drm/i915/i915_oa_hsw.h
 create mode 100644 drivers/gpu/drm/i915/i915_perf.c
Signature · 6 lines
2.7.0

_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel
#1 lore raw

PATCH 1/8 drm/i915: Add i915 perf infrastructure

Adds base i915 perf infrastructure for Gen performance metrics.

This adds a DRM_IOCTL_I915_PERF_OPEN ioctl that takes an array of uint64 properties to configure a stream of metrics and returns a new fd usable with standard VFS system calls including read() to read typed and sized records; ioctl() to enable or disable capture and poll() to wait for data.

A stream is opened something like:

  uint64_t properties[] = {
      /* Single context sampling */
      DRM_I915_PERF_CTX_HANDLE_PROP,        ctx_handle,
      /* Include OA reports in samples */
      DRM_I915_PERF_SAMPLE_OA_PROP,         true,

      /* OA unit configuration */
      DRM_I915_PERF_OA_METRICS_SET_PROP,    metrics_set_id,
      DRM_I915_PERF_OA_FORMAT_PROP,         report_format,
      DRM_I915_PERF_OA_EXPONENT_PROP,       period_exponent,
   };
   struct drm_i915_perf_open_param parm = {
      .flags = I915_PERF_FLAG_FD_CLOEXEC |
               I915_PERF_FLAG_FD_NONBLOCK |
               I915_PERF_FLAG_DISABLED,
      .properties = properties,
      .n_properties = sizeof(properties) / 16,

};
int fd = drmIoctl(drm_fd, DRM_IOCTL_I915_PERF_OPEN, &param);

Records read all start with a common { type, size } header with DRM_I915_PERF_RECORD_SAMPLE being of most interest. Sample records contain an extensible number of fields and it's the DRM_I915_PERF_SAMPLE_xyz properties given when opening that determine what's included in every sample.

No specific streams are supported yet so any attempt to open a stream will return an error.

---
 drivers/gpu/drm/i915/Makefile    |   3 +
 drivers/gpu/drm/i915/i915_dma.c  |   7 +
 drivers/gpu/drm/i915/i915_drv.h  |  88 ++++++++
 drivers/gpu/drm/i915/i915_perf.c | 442 +++++++++++++++++++++++++++++++++++++++
 include/uapi/drm/i915_drm.h      |  69 ++++++
 5 files changed, 609 insertions(+)
 create mode 100644 drivers/gpu/drm/i915/i915_perf.c
drivers/gpu/drm/i915/Makefile +3 -0
index 0851de07..af8ecbb 100644
@@ -94,6 +94,9 @@ i915-y += dvo_ch7017.o \
 # virtual gpu code
 i915-y += i915_vgpu.o
 
+# perf code
+i915-y += i915_perf.o
+
 # legacy horrors
 i915-y += i915_dma.o
 
drivers/gpu/drm/i915/i915_dma.c +7 -0
index a42eb58..4f4c9ba 100644
@@ -1012,6 +1012,11 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags)
 	if (ret < 0)
 		goto out_free_priv;
 
+	/* Must at least be initialized before trying to pin any context
+	 * which i915_perf hooks into.
+	 */
+	i915_perf_init(dev);
+
 	intel_pm_setup(dev);
 
 	intel_runtime_pm_get(dev_priv);
@@ -1203,6 +1208,7 @@ int i915_driver_unload(struct drm_device *dev)
 		return ret;
 	}
 
+	i915_perf_fini(dev);
 	intel_power_domains_fini(dev_priv);
 
 	intel_gpu_ips_teardown();
@@ -1379,6 +1385,7 @@ const struct drm_ioctl_desc i915_ioctls[] = {
 	DRM_IOCTL_DEF_DRV(I915_GEM_USERPTR, i915_gem_userptr_ioctl, DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(I915_GEM_CONTEXT_GETPARAM, i915_gem_context_getparam_ioctl, DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(I915_GEM_CONTEXT_SETPARAM, i915_gem_context_setparam_ioctl, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(I915_PERF_OPEN, i915_perf_open_ioctl, DRM_RENDER_ALLOW),
 };
 
 int i915_max_ioctl = ARRAY_SIZE(i915_ioctls);
drivers/gpu/drm/i915/i915_drv.h +88 -0

112 lines not on this page · raw message

drivers/gpu/drm/i915/i915_perf.c new file +442 -0

442 lines not on this page · raw message

include/uapi/drm/i915_drm.h +69 -0

85 lines not on this page · raw message

Signature · 6 lines
2.7.0

_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel
#2 lore raw

Re: [PATCH 1/8] drm/i915: Add i915 perf infrastructure

Quoted 10 lines · Robert Bragg

index a5524cc..68ca26e 100644

include/uapi/drm/i915_drm.h +2 -0
@@ -1170,4 +1172,71 @@ struct drm_i915_gem_context_param {
        __u64 value;
 };
 
+#define I915_PERF_FLAG_FD_CLOEXEC      (1<<0)
+#define I915_PERF_FLAG_FD_NONBLOCK     (1<<1)

include/uapi/drm/i915_drm.h struct drm_i915_gem_context_param { · line 1186

Quoted 10 lines
include/uapi/drm/i915_drm.h +10 -0
+#define I915_PERF_FLAG_DISABLED                (1<<2)
+
+enum drm_i915_perf_property_id {
+       /**
+        * Open the stream for a specific context handle (as used with
+        * execbuffer2). A stream opened for a specific context this way
+        * won't typically require root privileges.
+        */
+       DRM_I915_PERF_CTX_HANDLE_PROP = 1,
+

Wouldn't DRM_I915_PERF_PROP_CTX_HANDLE be a better name ? It's more obvious at least :-P

include/uapi/drm/i915_drm.h struct drm_i915_gem_context_param { · line 1187

Quoted 1 line
include/uapi/drm/i915_drm.h +1 -0
+       DRM_I915_PERF_PROP_MAX /* non-ABI */

Isn't the use of enums (in drm UAPI) discouraged ? Afaics only the old DRI1 i915 code has them.
Same question applies throughout the patch/series.

include/uapi/drm/i915_drm.h struct drm_i915_gem_context_param { · line 1191

Quoted 4 lines
include/uapi/drm/i915_drm.h +4 -0
+};
+
+struct drm_i915_perf_open_param {
+       /** CLOEXEC, NONBLOCK... */

Some other places in i915 define the macros just after the __u32 flags. This will allow you to drop the comment ;-)

include/uapi/drm/i915_drm.h struct drm_i915_gem_context_param { · line 1193

Quoted 2 lines
include/uapi/drm/i915_drm.h +2 -0
+       __u32 flags;
+

And ... we broke 32 bit userspare on 64 bit kernels. Please check for holes and other issues as described in Daniel Vetter's article/documentation [1]

[1] https://www.kernel.org/doc/Documentation/ioctl/botching-up-ioctls.txt

include/uapi/drm/i915_drm.h struct drm_i915_gem_context_param { · line 1201

Quoted 8 lines
include/uapi/drm/i915_drm.h +8 -0
+       /**
+        * Pointer to array of u64 (id, value) pairs configuring the stream
+        * to open.
+        */
+       __u64 __user properties;
+
+       /** The number of u64 (id, value) pairs */
+       __u32 n_properties;

The rest of the file uses num_foo or foo_count. Can we opt for one of those ?

include/uapi/drm/i915_drm.h struct drm_i915_gem_context_param { · line 1212

Quoted 11 lines
include/uapi/drm/i915_drm.h +11 -0
+};
+
+#define I915_PERF_IOCTL_ENABLE _IO('i', 0x0)
+#define I915_PERF_IOCTL_DISABLE        _IO('i', 0x1)
+
+/**
+ * Common to all i915 perf records
+ */
+struct drm_i915_perf_record_header {
+       __u32 type;
+       __u16 pad;

Move pad at the end ? This way we can (maybe?) reuse it in the future.

include/uapi/drm/i915_drm.h struct drm_i915_gem_context_param { · line 1215

Quoted 3 lines
include/uapi/drm/i915_drm.h +3 -0
+       __u16 size;
+};
+

Daniel, is there a consensus about zeroing the pad from userspace and checking it for garbage in the ioctl ? The documentation says "do it", yet I have a hunch that we're missing a few. Also what error should the ioctl return in such a case - EINVAL or EFAULT ? Worth documenting, just in case ?

-Emil
_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel

#3 lore raw

Re: [PATCH 1/8] drm/i915: Add i915 perf infrastructure

Quoted 10 lines · Emil Velikov
Quoted 8 lines · Robert Bragg

index a5524cc..68ca26e 100644

include/uapi/drm/i915_drm.h +0 -0
@@ -1170,4 +1172,71 @@ struct drm_i915_gem_context_param {
        __u64 value;
 };
 
Quoted 15 lines

include/uapi/drm/i915_drm.h struct drm_i915_gem_context_param { · line 1186

Quoted 12 lines
include/uapi/drm/i915_drm.h +12 -0
+#define I915_PERF_FLAG_FD_CLOEXEC      (1<<0)
+#define I915_PERF_FLAG_FD_NONBLOCK     (1<<1)
+#define I915_PERF_FLAG_DISABLED                (1<<2)
+
+enum drm_i915_perf_property_id {
+       /**
+        * Open the stream for a specific context handle (as used with
+        * execbuffer2). A stream opened for a specific context this way
+        * won't typically require root privileges.
+        */
+       DRM_I915_PERF_CTX_HANDLE_PROP = 1,
+

Wouldn't DRM_I915_PERF_PROP_CTX_HANDLE be a better name ? It's more obvious at least :-P

Yeah would be more consistent to keep the common namespacing at the front.

Quoted 6 lines

include/uapi/drm/i915_drm.h struct drm_i915_gem_context_param { · line 1187

Quoted 2 lines
include/uapi/drm/i915_drm.h +1 -0
+       DRM_I915_PERF_PROP_MAX /* non-ABI */

Isn't the use of enums (in drm UAPI) discouraged ? Afaics only the old DRI1 i915 code has them.
Same question applies throughout the patch/series.

An enum here seems like a pretty good technical fit. I think the potential for different enums to have different sizes is a likely reason for being cautious about using them as part of a kernel ABI (so probably unwise to have enum based members in an ioctl structure) but in this case these IDs are always passed to the kernel as a u64. We benefit from the compiler reminding us to handle all properties in the driver if we switch on this enum which I like, as well as having the _MAX value to refer to in the driver which is perhaps a little more reliable at the end of an enum vs a #define which needs to be explicitly updated for each addition.

For reference, the first iteration of this driver was based on the core pref infrastructure and in this case the enum + _MAX /* non-ABI */ approach is borrowed from there (ref: include/uapi/linux/perf_event.h)

Quoted 7 lines

include/uapi/drm/i915_drm.h struct drm_i915_gem_context_param { · line 1191

Quoted 4 lines
include/uapi/drm/i915_drm.h +4 -0
+};
+
+struct drm_i915_perf_open_param {
+       /** CLOEXEC, NONBLOCK... */

Some other places in i915 define the macros just after the __u32 flags. This will allow you to drop the comment ;-)

Quoted 7 lines

include/uapi/drm/i915_drm.h struct drm_i915_gem_context_param { · line 1193

Quoted 2 lines
include/uapi/drm/i915_drm.h +2 -0
+       __u32 flags;
+

And ... we broke 32 bit userspare on 64 bit kernels. Please check for holes and other issues as described in Daniel Vetter's article/documentation [1]

[1] https://www.kernel.org/doc/Documentation/ioctl/botching-up-ioctls.txt

Ah yeah, don't think this would break 32bit userspace, but still would be good to remove that hole, this has been through a few iterations and there used to be a __u32 type here, well spotted thanks.

I think I'll bump the flags to be 64bit.

Quoted 5 lines

include/uapi/drm/i915_drm.h struct drm_i915_gem_context_param { · line 1195

Quoted 4 lines
include/uapi/drm/i915_drm.h +2 -0
+       /**
+        * Pointer to array of u64 (id, value) pairs configuring the

stream

Quoted 9 lines

include/uapi/drm/i915_drm.h struct drm_i915_gem_context_param { · line 1201

Quoted 6 lines
include/uapi/drm/i915_drm.h +6 -0
+        * to open.
+        */
+       __u64 __user properties;
+
+       /** The number of u64 (id, value) pairs */
+       __u32 n_properties;

The rest of the file uses num_foo or foo_count. Can we opt for one of those ?

Ah yep, num_properties looks good to me.

Quoted 14 lines

include/uapi/drm/i915_drm.h struct drm_i915_gem_context_param { · line 1212

Quoted 12 lines
include/uapi/drm/i915_drm.h +11 -0
+};
+
+#define I915_PERF_IOCTL_ENABLE _IO('i', 0x0)
+#define I915_PERF_IOCTL_DISABLE        _IO('i', 0x1)
+
+/**
+ * Common to all i915 perf records
+ */
+struct drm_i915_perf_record_header {
+       __u32 type;
+       __u16 pad;

Move pad at the end ? This way we can (maybe?) reuse it in the future.

This header was originally based on struct perf_event_header which is layed out with the padding in the middle too. I can't currently think of a strong reason to maintain a record header that's ABI compatible with perf, but /maybe/ it could be convenient in some case to have a common layout for profiling tools that deal with both. I think we can consider it reserved for future use either way.

Quoted 11 lines

include/uapi/drm/i915_drm.h struct drm_i915_gem_context_param { · line 1215

Quoted 4 lines
include/uapi/drm/i915_drm.h +3 -0
+       __u16 size;
+};
+

Daniel, is there a consensus about zeroing the pad from userspace and checking it for garbage in the ioctl ? The documentation says "do it", yet I have a hunch that we're missing a few. Also what error should the ioctl return in such a case - EINVAL or EFAULT ? Worth documenting, just in case ?

In case you're wondering about the padding in the header above, note that this header never gets passed in from userspace, it's a header for data read by userspace and the driver zeros the padding.

For the hole in the ioctl I'll probably fill that by extending the flags member and the driver currently returns -EINVAL if any unknown flag bits are set by userspace.

Thanks for looking through this,
- Robert

Quoted 3 lines

-Emil

_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel

#4 lore raw

Re: [PATCH 1/8] drm/i915: Add i915 perf infrastructure

Quoted 21 lines · Robert Bragg
Quoted 17 lines · Emil Velikov
Quoted 13 lines · Robert Bragg

include/uapi/drm/i915_drm.h struct drm_i915_gem_context_param { · line 1193

Quoted 7 lines
include/uapi/drm/i915_drm.h +6 -0
+};
+
+struct drm_i915_perf_open_param {
+       /** CLOEXEC, NONBLOCK... */
+       __u32 flags;
+

And ... we broke 32 bit userspare on 64 bit kernels. Please check for holes and other issues as described in Daniel Vetter's article/documentation [1]

[1]
https://www.kernel.org/doc/Documentation/ioctl/botching-up-ioctls.txt

Quoted 8 lines
Quoted 7 lines

Ah yeah, don't think this would break 32bit userspace, but still would be good to remove that hole, this has been through a few iterations and there used to be a __u32 type here, well spotted thanks.

I think I'll bump the flags to be 64bit.

Actually, just noticed that since the structure has a u32 hole at the end too I can move the trailing u32 num_properties up into here instead.

Am also renaming properties to properties_ptr which seems the norm in i915_drm.h.

- Robert

_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel

#5 lore raw

PATCH 2/8 drm/i915: rename OACONTROL GEN7_OACONTROL

In reply to Robert Bragg

OACONTROL changes quite a bit for gen8, with some bits split out into a per-context OACTXCONTROL register. Rename now before add more gen7 OA registers

---
 drivers/gpu/drm/i915/i915_cmd_parser.c | 4 ++--
 drivers/gpu/drm/i915/i915_reg.h        | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/i915_cmd_parser.c +2 -2
index 814d894..920b75f 100644
@@ -444,7 +444,7 @@ static const struct drm_i915_reg_descriptor gen7_render_regs[] = {
 	REG64(CL_PRIMITIVES_COUNT),
 	REG64(PS_INVOCATION_COUNT),
 	REG64(PS_DEPTH_COUNT),
-	REG32(OACONTROL), /* Only allowed for LRI and SRM. See below. */
+	REG32(GEN7_OACONTROL), /* Only allowed for LRI and SRM. See below. */
 	REG64(MI_PREDICATE_SRC0),
 	REG64(MI_PREDICATE_SRC1),
 	REG32(GEN7_3DPRIM_END_OFFSET),
@@ -1028,7 +1028,7 @@ static bool check_cmd(const struct intel_engine_cs *ring,
 			 * to the register. Hence, limit OACONTROL writes to
 			 * only MI_LOAD_REGISTER_IMM commands.
 			 */
-			if (reg_addr == i915_mmio_reg_offset(OACONTROL)) {
+			if (reg_addr == i915_mmio_reg_offset(GEN7_OACONTROL)) {
 				if (desc->cmd.value == MI_LOAD_REGISTER_MEM) {
 					DRM_DEBUG_DRIVER("CMD: Rejected LRM to OACONTROL\n");
 					return false;
drivers/gpu/drm/i915/i915_reg.h +1 -1
index c0bd691..119401c 100644
@@ -586,7 +586,7 @@ static inline bool i915_mmio_reg_valid(i915_reg_t reg)
 #define GEN7_GPGPU_DISPATCHDIMY         _MMIO(0x2504)
 #define GEN7_GPGPU_DISPATCHDIMZ         _MMIO(0x2508)
 
-#define OACONTROL _MMIO(0x2360)
+#define GEN7_OACONTROL _MMIO(0x2360)
 
 #define _GEN7_PIPEA_DE_LOAD_SL	0x70068
 #define _GEN7_PIPEB_DE_LOAD_SL	0x71068
Signature · 6 lines
2.7.0

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#6 lore raw

PATCH 3/8 drm/i915: Add 'render basic' Haswell OA unit config

In reply to Robert Bragg

Adds a static OA unit, MUX + B Counter configuration for basic render metrics on Haswell. This is autogenerated from an internal XML description of metric sets.

---
 drivers/gpu/drm/i915/Makefile      |   3 +-
 drivers/gpu/drm/i915/i915_drv.h    |  14 ++++
 drivers/gpu/drm/i915/i915_oa_hsw.c | 132 +++++++++++++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_oa_hsw.h |  34 ++++++++++
 4 files changed, 182 insertions(+), 1 deletion(-)
 create mode 100644 drivers/gpu/drm/i915/i915_oa_hsw.c
 create mode 100644 drivers/gpu/drm/i915/i915_oa_hsw.h
drivers/gpu/drm/i915/Makefile +2 -1
index af8ecbb..a0131da 100644
@@ -95,7 +95,8 @@ i915-y += dvo_ch7017.o \
 i915-y += i915_vgpu.o
 
 # perf code
-i915-y += i915_perf.o
+i915-y += i915_perf.o \
+	  i915_oa_hsw.o
 
 # legacy horrors
 i915-y += i915_dma.o
drivers/gpu/drm/i915/i915_drv.h +14 -0
index 7748383..86da0ae 100644
@@ -1732,6 +1732,11 @@ struct intel_wm_config {
 	bool sprites_scaled;
 };
 
+struct i915_oa_reg {
+	u32 addr;
+	u32 value;
+};
+
 struct i915_perf_read_state {
 	int count;
 	ssize_t read;
@@ -2062,6 +2067,15 @@ struct drm_i915_private {
 		bool initialized;
 		struct mutex lock;
 		struct list_head streams;
+
+		struct {
+			u32 metrics_set;
+
+			const struct i915_oa_reg *mux_regs;
+			int mux_regs_len;
+			const struct i915_oa_reg *b_counter_regs;
+			int b_counter_regs_len;
+		} oa;
 	} perf;
 
 	/* Abstract the submission mechanism (legacy ringbuffer or execlists) away */
drivers/gpu/drm/i915/i915_oa_hsw.c new file +132 -0

132 lines not on this page · raw message

drivers/gpu/drm/i915/i915_oa_hsw.h new file +34 -0
new file mode 100644
index 0000000..b618a1f
@@ -0,0 +1,34 @@
+/*
+ * Autogenerated file, DO NOT EDIT manually!
+ *
+ * Copyright (c) 2015 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __I915_OA_HSW_H__
+#define __I915_OA_HSW_H__
+
+extern int i915_oa_n_builtin_metric_sets_hsw;
+
+extern int i915_oa_select_metric_set_hsw(struct drm_i915_private *dev_priv);
+
+#endif
Signature · 6 lines
2.7.0

_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel
#7 lore raw

PATCH 4/8 drm/i915: Add i915 perf event for Haswell OA unit

In reply to Robert Bragg

Gen graphics hardware can be set up to periodically write snapshots of performance counters into a circular buffer via its Observation Architecture and this patch exposes that capability to userspace via the i915 perf interface.

---
 drivers/gpu/drm/i915/i915_drv.h         |  51 ++-
 drivers/gpu/drm/i915/i915_gem_context.c |  23 +-
 drivers/gpu/drm/i915/i915_oa_hsw.c      | 166 +++----
 drivers/gpu/drm/i915/i915_perf.c        | 783 +++++++++++++++++++++++++++++++-
 drivers/gpu/drm/i915/i915_reg.h         | 338 ++++++++++++++
 include/uapi/drm/i915_drm.h             |  56 ++-
 6 files changed, 1317 insertions(+), 100 deletions(-)
drivers/gpu/drm/i915/i915_drv.h +50 -1
index 86da0ae..e4b9399 100644
@@ -1732,8 +1732,13 @@ struct intel_wm_config {
 	bool sprites_scaled;
 };
 
+struct i915_oa_format {
+	u32 format;
+	int size;
+};
+
 struct i915_oa_reg {
-	u32 addr;
+	i915_reg_t addr;
 	u32 value;
 };
 
@@ -1749,6 +1754,7 @@ struct i915_perf_stream {
 	struct list_head link;
 
 	u32 sample_flags;
+	int sample_size;
 
 	struct intel_context *ctx;
 	bool enabled;
@@ -1812,6 +1818,20 @@ struct i915_perf_stream {
 	void (*destroy)(struct i915_perf_stream *stream);
 };
 
+struct i915_oa_ops {
+	void (*init_oa_buffer)(struct drm_i915_private *dev_priv);
+	int (*enable_metric_set)(struct drm_i915_private *dev_priv);
+	void (*disable_metric_set)(struct drm_i915_private *dev_priv);
+	void (*oa_enable)(struct drm_i915_private *dev_priv);
+	void (*oa_disable)(struct drm_i915_private *dev_priv);
+	void (*update_oacontrol)(struct drm_i915_private *dev_priv);
+	void (*update_hw_ctx_id_locked)(struct drm_i915_private *dev_priv,
+					u32 ctx_id);
+	int (*read)(struct i915_perf_stream *stream,
+		    struct i915_perf_read_state *read_state);
+	bool (*oa_buffer_is_empty)(struct drm_i915_private *dev_priv);
+};
+
 struct drm_i915_private {
 	struct drm_device *dev;
 	struct kmem_cache *objects;
@@ -2065,16 +2085,43 @@ struct drm_i915_private {
 
 	struct {
 		bool initialized;
+
 		struct mutex lock;
 		struct list_head streams;
 
+		spinlock_t hook_lock;
+
 		struct {
+			struct i915_perf_stream *exclusive_stream;
+
+			u32 specific_ctx_id;
+
+			struct hrtimer poll_check_timer;
+			wait_queue_head_t poll_wq;
+
+			bool periodic;
+			u32 period_exponent;
+
 			u32 metrics_set;
 
 			const struct i915_oa_reg *mux_regs;
 			int mux_regs_len;
 			const struct i915_oa_reg *b_counter_regs;
 			int b_counter_regs_len;
+
+			struct {
+				struct drm_i915_gem_object *obj;
+				u32 gtt_offset;
+				u8 *addr;
+				u32 head;
+				u32 tail;
+				int format;
+				int format_size;
+			} oa_buffer;
+
+			struct i915_oa_ops ops;
+			const struct i915_oa_format *oa_formats;
+			int n_builtin_sets;
 		} oa;
 	} perf;
 
@@ -3348,6 +3395,8 @@ int i915_gem_context_setparam_ioctl(struct drm_device *dev, void *data,
 
 int i915_perf_open_ioctl(struct drm_device *dev, void *data,
 			 struct drm_file *file);
+void i915_oa_context_pin_notify(struct drm_i915_private *dev_priv,
+				struct intel_context *context);
 
 /* i915_gem_evict.c */
 int __must_check i915_gem_evict_something(struct drm_device *dev,
drivers/gpu/drm/i915/i915_gem_context.c +19 -4
index 83a097c..071cfa8 100644
@@ -133,6 +133,23 @@ static int get_context_size(struct drm_device *dev)
 	return ret;
 }
 
+static int i915_gem_context_pin_state(struct drm_device *dev,
+				      struct intel_context *ctx)
+{
+	int ret;
+
+	BUG_ON(!mutex_is_locked(&dev->struct_mutex));
+
+	ret = i915_gem_obj_ggtt_pin(ctx->legacy_hw_ctx.rcs_state,
+				    get_context_alignment(dev), 0);
+	if (ret)
+		return ret;
+
+	i915_oa_context_pin_notify(dev->dev_private, ctx);
+
+	return 0;
+}
+
 static void i915_gem_context_clean(struct intel_context *ctx)
 {
 	struct i915_hw_ppgtt *ppgtt = ctx->ppgtt;
@@ -287,8 +304,7 @@ i915_gem_create_context(struct drm_device *dev,
 		 * be available. To avoid this we always pin the default
 		 * context.
 		 */
-		ret = i915_gem_obj_ggtt_pin(ctx->legacy_hw_ctx.rcs_state,
-					    get_context_alignment(dev), 0);
+		ret = i915_gem_context_pin_state(dev, ctx);
 		if (ret) {
 			DRM_DEBUG_DRIVER("Couldn't pin %d\n", ret);
 			goto err_destroy;
@@ -665,8 +681,7 @@ static int do_switch(struct drm_i915_gem_request *req)
 
 	/* Trying to pin first makes error handling easier. */
 	if (ring == &dev_priv->ring[RCS]) {
-		ret = i915_gem_obj_ggtt_pin(to->legacy_hw_ctx.rcs_state,
-					    get_context_alignment(ring->dev), 0);
+		ret = i915_gem_context_pin_state(ring->dev, to);
 		if (ret)
 			return ret;
 	}
drivers/gpu/drm/i915/i915_oa_hsw.c +83 -83

189 lines not on this page · raw message

drivers/gpu/drm/i915/i915_perf.c +773 -10

833 lines not on this page · raw message

drivers/gpu/drm/i915/i915_reg.h +338 -0

350 lines not on this page · raw message

include/uapi/drm/i915_drm.h +54 -2

83 lines not on this page · raw message

Signature · 6 lines
2.7.0

_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel
#8 lore raw

PATCH 5/8 drm/i915: advertise available metrics via sysfs

In reply to Robert Bragg

Each metric set is given a sysfs entry like:

/sys/class/drm/card0/metrics/<guid>/id

This allows userspace to enumerate the specific sets that are available for the current system. The 'id' file contains an unsigned integer that can be used to open the associated metric set via DRM_IOCTL_I915_PERF_OPEN. The <guid> is a globally unique ID for a specific OA unit configuration that can be reliably used as a key to lookup corresponding counter meta data and normalization equations.

---
 drivers/gpu/drm/i915/i915_drv.h    |  2 ++
 drivers/gpu/drm/i915/i915_oa_hsw.c | 45 ++++++++++++++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_oa_hsw.h |  4 ++++
 drivers/gpu/drm/i915/i915_perf.c   | 18 ++++++++++++++-
 4 files changed, 68 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/i915_drv.h +2 -0
index e4b9399..b1c60bb 100644
@@ -2086,6 +2086,8 @@ struct drm_i915_private {
 	struct {
 		bool initialized;
 
+		struct kobject *metrics_kobj;
+
 		struct mutex lock;
 		struct list_head streams;
 
drivers/gpu/drm/i915/i915_oa_hsw.c +45 -0

54 lines not on this page · raw message

drivers/gpu/drm/i915/i915_oa_hsw.h +4 -0
index b618a1f..e4ba89d 100644
@@ -31,4 +31,8 @@ extern int i915_oa_n_builtin_metric_sets_hsw;
 
 extern int i915_oa_select_metric_set_hsw(struct drm_i915_private *dev_priv);
 
+extern int i915_perf_init_sysfs_hsw(struct drm_i915_private *dev_priv);
+
+extern void i915_perf_deinit_sysfs_hsw(struct drm_i915_private *dev_priv);
+
 #endif
drivers/gpu/drm/i915/i915_perf.c +17 -1
index b54f719..5c08a16 100644
@@ -1164,6 +1164,11 @@ void i915_perf_init(struct drm_device *dev)
 	if (!IS_HASWELL(dev))
 		return;
 
+	dev_priv->perf.metrics_kobj =
+		kobject_create_and_add("metrics", &dev->primary->kdev->kobj);
+	if (!dev_priv->perf.metrics_kobj)
+		return;
+
 	hrtimer_init(&dev_priv->perf.oa.poll_check_timer,
 		     CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 	dev_priv->perf.oa.poll_check_timer.function = poll_check_timer_cb;
@@ -1187,9 +1192,15 @@ void i915_perf_init(struct drm_device *dev)
 	dev_priv->perf.oa.n_builtin_sets =
 		i915_oa_n_builtin_metric_sets_hsw;
 
-	dev_priv->perf.oa.oa_formats = hsw_oa_formats;
+	if (i915_perf_init_sysfs_hsw(dev_priv)) {
+		kobject_put(dev_priv->perf.metrics_kobj);
+		dev_priv->perf.metrics_kobj = NULL;
+		return;
+	}
 
 	dev_priv->perf.initialized = true;
+
+	return;
 }
 
 void i915_perf_fini(struct drm_device *dev)
@@ -1199,6 +1210,11 @@ void i915_perf_fini(struct drm_device *dev)
 	if (!dev_priv->perf.initialized)
 		return;
 
+	i915_perf_deinit_sysfs_hsw(dev_priv);
+
+	kobject_put(dev_priv->perf.metrics_kobj);
+	dev_priv->perf.metrics_kobj = NULL;
+
 	dev_priv->perf.oa.ops.init_oa_buffer = NULL;
 
 	dev_priv->perf.initialized = false;
Signature · 6 lines
2.7.0

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#9 lore raw

PATCH 6/8 drm/i915: Add dev.i915.perf_event_paranoid sysctl option

In reply to Robert Bragg

Consistent with the kernel.perf_event_paranoid sysctl option that can allow non-root users to access system wide cpu metrics, this can optionally allow non-root users to access system wide OA counter metrics from Gen graphics hardware.

---
 drivers/gpu/drm/i915/i915_drv.h  |  1 +
 drivers/gpu/drm/i915/i915_perf.c | 46 +++++++++++++++++++++++++++++++++++++++-
 2 files changed, 46 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/i915_drv.h +1 -0
index b1c60bb..6776e18 100644
@@ -2087,6 +2087,7 @@ struct drm_i915_private {
 		bool initialized;
 
 		struct kobject *metrics_kobj;
+		struct ctl_table_header *sysctl_header;
 
 		struct mutex lock;
 		struct list_head streams;
drivers/gpu/drm/i915/i915_perf.c +45 -1

76 lines not on this page · raw message

Signature · 6 lines
2.7.0

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#10 lore raw

PATCH 7/8 drm/i915: add oa_event_min_timer_exponent sysctl

In reply to Robert Bragg

The minimal sampling period is now configurable via a dev.i915.oa_min_timer_exponent sysctl parameter.

Following the precedent set by perf, the default is the minimum that won't (on its own) exceed the default kernel.perf_event_max_sample_rate default of 100000 samples/s.

---
 drivers/gpu/drm/i915/i915_perf.c | 40 +++++++++++++++++++++++++++++-----------
 1 file changed, 29 insertions(+), 11 deletions(-)
drivers/gpu/drm/i915/i915_perf.c +29 -11
index 7c27d93..ca4860b 100644
@@ -41,6 +41,23 @@ static u32 i915_perf_stream_paranoid = true;
 
 #define OA_EXPONENT_MAX 0x3f
 
+/* for sysctl proc_dointvec_minmax of i915_oa_min_timer_exponent */
+static int zero;
+static int oa_exponent_max = OA_EXPONENT_MAX;
+
+/* Theoretically we can program the OA unit to sample every 160ns but don't
+ * allow that by default unless root...
+ *
+ * The period is derived from the exponent as:
+ *
+ *   period = 80ns * 2^(exponent + 1)
+ *
+ * Referring to perf's kernel.perf_event_max_sample_rate for a precedent
+ * (100000 by default); with an OA exponent of 6 we get a period of 10.240
+ * microseconds - just under 100000Hz
+ */
+static u32 i915_oa_min_timer_exponent = 6;
+
 static struct i915_oa_format hsw_oa_formats[I915_OA_FORMAT_MAX] = {
 	[I915_OA_FORMAT_A13]	    = { 0, 64 },
 	[I915_OA_FORMAT_A29]	    = { 1, 128 },
@@ -1098,20 +1115,12 @@ static int read_properties_unlocked(struct drm_i915_private *dev_priv,
 			if (value > OA_EXPONENT_MAX)
 				return -EINVAL;
 
-			/* NB: The exponent represents a period as follows:
-			 *
-			 *   80ns * 2^(period_exponent + 1)
-			 *
-			 * Theoretically we can program the OA unit to sample
+			/* Theoretically we can program the OA unit to sample
 			 * every 160ns but don't allow that by default unless
 			 * root.
-			 *
-			 * Referring to perf's
-			 * kernel.perf_event_max_sample_rate for a precedent
-			 * (100000 by default); with an OA exponent of 6 we get
-			 * a period of 10.240 microseconds -just under 100000Hz
 			 */
-			if (value < 6 && !capable(CAP_SYS_ADMIN)) {
+			if (value < i915_oa_min_timer_exponent &&
+			    !capable(CAP_SYS_ADMIN)) {
 				DRM_ERROR("Sampling period too high without root privileges\n");
 				return -EACCES;
 			}
@@ -1174,6 +1183,15 @@ static struct ctl_table oa_table[] = {
 	 .mode = 0644,
 	 .proc_handler = proc_dointvec,
 	 },
+	{
+	 .procname = "oa_min_timer_exponent",
+	 .data = &i915_oa_min_timer_exponent,
+	 .maxlen = sizeof(i915_oa_min_timer_exponent),
+	 .mode = 0644,
+	 .proc_handler = proc_dointvec_minmax,
+	 .extra1 = &zero,
+	 .extra2 = &oa_exponent_max,
+	 },
 	{}
 };
 
Signature · 6 lines
2.7.0

_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel
#11 lore raw

PATCH 8/8 drm/i915: Add more Haswell OA metric sets

In reply to Robert Bragg

This adds 'compute', 'compute extended', 'memory reads', 'memory writes' and 'sampler balance' metric sets for Haswell.

---
 drivers/gpu/drm/i915/i915_oa_hsw.c | 513 +++++++++++++++++++++++++++++++++++--
 1 file changed, 497 insertions(+), 16 deletions(-)
drivers/gpu/drm/i915/i915_oa_hsw.c +497 -16
index ecd7d2b..4a2de8a 100644
@@ -30,9 +30,14 @@
 
 enum metric_set_id {
 	METRIC_SET_ID_RENDER_BASIC = 1,
+	METRIC_SET_ID_COMPUTE_BASIC,
+	METRIC_SET_ID_COMPUTE_EXTENDED,
+	METRIC_SET_ID_MEMORY_READS,
+	METRIC_SET_ID_MEMORY_WRITES,
+	METRIC_SET_ID_SAMPLER_BALANCE,
 };
 
-int i915_oa_n_builtin_metric_sets_hsw = 1;
+int i915_oa_n_builtin_metric_sets_hsw = 6;
 
 static const struct i915_oa_reg b_counter_config_render_basic[] = {
 	{ _MMIO(0x2724), 0x00800000 },
@@ -118,6 +123,332 @@ static int select_render_basic_config(struct drm_i915_private *dev_priv)
 	return 0;
 }
 
+static const struct i915_oa_reg b_counter_config_compute_basic[] = {
+	{ _MMIO(0x2710), 0x00000000 },
+	{ _MMIO(0x2714), 0x00800000 },
+	{ _MMIO(0x2718), 0xAAAAAAAA },
+	{ _MMIO(0x271C), 0xAAAAAAAA },
+	{ _MMIO(0x2720), 0x00000000 },
+	{ _MMIO(0x2724), 0x00800000 },
+	{ _MMIO(0x2728), 0xAAAAAAAA },
+	{ _MMIO(0x272C), 0xAAAAAAAA },
+	{ _MMIO(0x2740), 0x00000000 },
+	{ _MMIO(0x2744), 0x00000000 },
+	{ _MMIO(0x2748), 0x00000000 },
+	{ _MMIO(0x274C), 0x00000000 },
+	{ _MMIO(0x2750), 0x00000000 },
+	{ _MMIO(0x2754), 0x00000000 },
+	{ _MMIO(0x2758), 0x00000000 },
+	{ _MMIO(0x275C), 0x00000000 },
+};
+
+static const struct i915_oa_reg mux_config_compute_basic[] = {
+	{ _MMIO(0x253A4), 0x00000000 },
+	{ _MMIO(0x2681C), 0x01F00800 },
+	{ _MMIO(0x26820), 0x00001000 },
+	{ _MMIO(0x2781C), 0x01F00800 },
+	{ _MMIO(0x26520), 0x00000007 },
+	{ _MMIO(0x265A0), 0x00000007 },
+	{ _MMIO(0x25380), 0x00000010 },
+	{ _MMIO(0x2538C), 0x00300000 },
+	{ _MMIO(0x25384), 0xAA8AAAAA },
+	{ _MMIO(0x25404), 0xFFFFFFFF },
+	{ _MMIO(0x26800), 0x00004202 },
+	{ _MMIO(0x26808), 0x00605817 },
+	{ _MMIO(0x2680C), 0x10001005 },
+	{ _MMIO(0x26804), 0x00000000 },
+	{ _MMIO(0x27800), 0x00000102 },
+	{ _MMIO(0x27808), 0x0C0701E0 },
+	{ _MMIO(0x2780C), 0x000200A0 },
+	{ _MMIO(0x27804), 0x00000000 },
+	{ _MMIO(0x26484), 0x44000000 },
+	{ _MMIO(0x26704), 0x44000000 },
+	{ _MMIO(0x26500), 0x00000006 },
+	{ _MMIO(0x26510), 0x00000001 },
+	{ _MMIO(0x26504), 0x88000000 },
+	{ _MMIO(0x26580), 0x00000006 },
+	{ _MMIO(0x26590), 0x00000020 },
+	{ _MMIO(0x26584), 0x00000000 },
+	{ _MMIO(0x26104), 0x55822222 },
+	{ _MMIO(0x26184), 0xAA866666 },
+	{ _MMIO(0x25420), 0x08320C83 },
+	{ _MMIO(0x25424), 0x06820C83 },
+	{ _MMIO(0x2541C), 0x00000000 },
+	{ _MMIO(0x25428), 0x00000C03 },
+};
+
+static int select_compute_basic_config(struct drm_i915_private *dev_priv)
+{
+	dev_priv->perf.oa.mux_regs =
+		mux_config_compute_basic;
+	dev_priv->perf.oa.mux_regs_len =
+		ARRAY_SIZE(mux_config_compute_basic);
+
+	dev_priv->perf.oa.b_counter_regs =
+		b_counter_config_compute_basic;
+	dev_priv->perf.oa.b_counter_regs_len =
+		ARRAY_SIZE(b_counter_config_compute_basic);
+
+	return 0;
+}
+
+static const struct i915_oa_reg b_counter_config_compute_extended[] = {
+	{ _MMIO(0x2724), 0xf0800000 },
+	{ _MMIO(0x2720), 0x00000000 },
+	{ _MMIO(0x2714), 0xf0800000 },
+	{ _MMIO(0x2710), 0x00000000 },
+	{ _MMIO(0x2770), 0x0007fe2a },
+	{ _MMIO(0x2774), 0x0000ff00 },
+	{ _MMIO(0x2778), 0x0007fe6a },
+	{ _MMIO(0x277c), 0x0000ff00 },
+	{ _MMIO(0x2780), 0x0007fe92 },
+	{ _MMIO(0x2784), 0x0000ff00 },
+	{ _MMIO(0x2788), 0x0007fea2 },
+	{ _MMIO(0x278c), 0x0000ff00 },
+	{ _MMIO(0x2790), 0x0007fe32 },
+	{ _MMIO(0x2794), 0x0000ff00 },
+	{ _MMIO(0x2798), 0x0007fe9a },
+	{ _MMIO(0x279c), 0x0000ff00 },
+	{ _MMIO(0x27a0), 0x0007ff23 },
+	{ _MMIO(0x27a4), 0x0000ff00 },
+	{ _MMIO(0x27a8), 0x0007fff3 },
+	{ _MMIO(0x27ac), 0x0000fffe },
+};
+
+static const struct i915_oa_reg mux_config_compute_extended[] = {
+	{ _MMIO(0x2681C), 0x3EB00800 },
+	{ _MMIO(0x26820), 0x00900000 },
+	{ _MMIO(0x25384), 0x02AAAAAA },
+	{ _MMIO(0x25404), 0x03FFFFFF },
+	{ _MMIO(0x26800), 0x00142284 },
+	{ _MMIO(0x26808), 0x0E629062 },
+	{ _MMIO(0x2680C), 0x3F6F55CB },
+	{ _MMIO(0x26810), 0x00000014 },
+	{ _MMIO(0x26804), 0x00000000 },
+	{ _MMIO(0x26104), 0x02AAAAAA },
+	{ _MMIO(0x26184), 0x02AAAAAA },
+	{ _MMIO(0x25420), 0x00000000 },
+	{ _MMIO(0x25424), 0x00000000 },
+	{ _MMIO(0x2541C), 0x00000000 },
+	{ _MMIO(0x25428), 0x00000000 },
+};
+
+static int select_compute_extended_config(struct drm_i915_private *dev_priv)
+{
+	dev_priv->perf.oa.mux_regs =
+		mux_config_compute_extended;
+	dev_priv->perf.oa.mux_regs_len =
+		ARRAY_SIZE(mux_config_compute_extended);
+
+	dev_priv->perf.oa.b_counter_regs =
+		b_counter_config_compute_extended;
+	dev_priv->perf.oa.b_counter_regs_len =
+		ARRAY_SIZE(b_counter_config_compute_extended);
+
+	return 0;
+}
+
+static const struct i915_oa_reg b_counter_config_memory_reads[] = {
+	{ _MMIO(0x2724), 0xf0800000 },
+	{ _MMIO(0x2720), 0x00000000 },
+	{ _MMIO(0x2714), 0xf0800000 },
+	{ _MMIO(0x2710), 0x00000000 },
+	{ _MMIO(0x274c), 0x76543298 },
+	{ _MMIO(0x2748), 0x98989898 },
+	{ _MMIO(0x2744), 0x000000e4 },
+	{ _MMIO(0x2740), 0x00000000 },
+	{ _MMIO(0x275c), 0x98a98a98 },
+	{ _MMIO(0x2758), 0x88888888 },
+	{ _MMIO(0x2754), 0x000c5500 },
+	{ _MMIO(0x2750), 0x00000000 },
+	{ _MMIO(0x2770), 0x0007f81a },
+	{ _MMIO(0x2774), 0x0000fc00 },
+	{ _MMIO(0x2778), 0x0007f82a },
+	{ _MMIO(0x277c), 0x0000fc00 },
+	{ _MMIO(0x2780), 0x0007f872 },
+	{ _MMIO(0x2784), 0x0000fc00 },
+	{ _MMIO(0x2788), 0x0007f8ba },
+	{ _MMIO(0x278c), 0x0000fc00 },
+	{ _MMIO(0x2790), 0x0007f87a },
+	{ _MMIO(0x2794), 0x0000fc00 },
+	{ _MMIO(0x2798), 0x0007f8ea },
+	{ _MMIO(0x279c), 0x0000fc00 },
+	{ _MMIO(0x27a0), 0x0007f8e2 },
+	{ _MMIO(0x27a4), 0x0000fc00 },
+	{ _MMIO(0x27a8), 0x0007f8f2 },
+	{ _MMIO(0x27ac), 0x0000fc00 },
+};
+
+static const struct i915_oa_reg mux_config_memory_reads[] = {
+	{ _MMIO(0x253A4), 0x34300000 },
+	{ _MMIO(0x25440), 0x2D800000 },
+	{ _MMIO(0x25444), 0x00000008 },
+	{ _MMIO(0x25128), 0x0E600000 },
+	{ _MMIO(0x25380), 0x00000450 },
+	{ _MMIO(0x25390), 0x00052C43 },
+	{ _MMIO(0x25384), 0x00000000 },
+	{ _MMIO(0x25400), 0x00006144 },
+	{ _MMIO(0x25408), 0x0A418820 },
+	{ _MMIO(0x2540C), 0x000820E6 },
+	{ _MMIO(0x25404), 0xFF500000 },
+	{ _MMIO(0x25100), 0x000005D6 },
+	{ _MMIO(0x2510C), 0x0EF00000 },
+	{ _MMIO(0x25104), 0x00000000 },
+	{ _MMIO(0x25420), 0x02108421 },
+	{ _MMIO(0x25424), 0x00008421 },
+	{ _MMIO(0x2541C), 0x00000000 },
+	{ _MMIO(0x25428), 0x00000000 },
+};
+
+static int select_memory_reads_config(struct drm_i915_private *dev_priv)
+{
+	dev_priv->perf.oa.mux_regs =
+		mux_config_memory_reads;
+	dev_priv->perf.oa.mux_regs_len =
+		ARRAY_SIZE(mux_config_memory_reads);
+
+	dev_priv->perf.oa.b_counter_regs =
+		b_counter_config_memory_reads;
+	dev_priv->perf.oa.b_counter_regs_len =
+		ARRAY_SIZE(b_counter_config_memory_reads);
+
+	return 0;
+}
+
+static const struct i915_oa_reg b_counter_config_memory_writes[] = {
+	{ _MMIO(0x2724), 0xf0800000 },
+	{ _MMIO(0x2720), 0x00000000 },
+	{ _MMIO(0x2714), 0xf0800000 },
+	{ _MMIO(0x2710), 0x00000000 },
+	{ _MMIO(0x274c), 0x76543298 },
+	{ _MMIO(0x2748), 0x98989898 },
+	{ _MMIO(0x2744), 0x000000e4 },
+	{ _MMIO(0x2740), 0x00000000 },
+	{ _MMIO(0x275c), 0xbabababa },
+	{ _MMIO(0x2758), 0x88888888 },
+	{ _MMIO(0x2754), 0x000c5500 },
+	{ _MMIO(0x2750), 0x00000000 },
+	{ _MMIO(0x2770), 0x0007f81a },
+	{ _MMIO(0x2774), 0x0000fc00 },
+	{ _MMIO(0x2778), 0x0007f82a },
+	{ _MMIO(0x277c), 0x0000fc00 },
+	{ _MMIO(0x2780), 0x0007f822 },
+	{ _MMIO(0x2784), 0x0000fc00 },
+	{ _MMIO(0x2788), 0x0007f8ba },
+	{ _MMIO(0x278c), 0x0000fc00 },
+	{ _MMIO(0x2790), 0x0007f87a },
+	{ _MMIO(0x2794), 0x0000fc00 },
+	{ _MMIO(0x2798), 0x0007f8ea },
+	{ _MMIO(0x279c), 0x0000fc00 },
+	{ _MMIO(0x27a0), 0x0007f8e2 },
+	{ _MMIO(0x27a4), 0x0000fc00 },
+	{ _MMIO(0x27a8), 0x0007f8f2 },
+	{ _MMIO(0x27ac), 0x0000fc00 },
+};
+
+static const struct i915_oa_reg mux_config_memory_writes[] = {
+	{ _MMIO(0x253A4), 0x34300000 },
+	{ _MMIO(0x25440), 0x01500000 },
+	{ _MMIO(0x25444), 0x00000120 },
+	{ _MMIO(0x25128), 0x0C200000 },
+	{ _MMIO(0x25380), 0x00000450 },
+	{ _MMIO(0x25390), 0x00052C43 },
+	{ _MMIO(0x25384), 0x00000000 },
+	{ _MMIO(0x25400), 0x00007184 },
+	{ _MMIO(0x25408), 0x0A418820 },
+	{ _MMIO(0x2540C), 0x000820E6 },
+	{ _MMIO(0x25404), 0xFF500000 },
+	{ _MMIO(0x25100), 0x000005D6 },
+	{ _MMIO(0x2510C), 0x1E700000 },
+	{ _MMIO(0x25104), 0x00000000 },
+	{ _MMIO(0x25420), 0x02108421 },
+	{ _MMIO(0x25424), 0x00008421 },
+	{ _MMIO(0x2541C), 0x00000000 },
+	{ _MMIO(0x25428), 0x00000000 },
+};
+
+static int select_memory_writes_config(struct drm_i915_private *dev_priv)
+{
+	dev_priv->perf.oa.mux_regs =
+		mux_config_memory_writes;
+	dev_priv->perf.oa.mux_regs_len =
+		ARRAY_SIZE(mux_config_memory_writes);
+
+	dev_priv->perf.oa.b_counter_regs =
+		b_counter_config_memory_writes;
+	dev_priv->perf.oa.b_counter_regs_len =
+		ARRAY_SIZE(b_counter_config_memory_writes);
+
+	return 0;
+}
+
+static const struct i915_oa_reg b_counter_config_sampler_balance[] = {
+	{ _MMIO(0x2740), 0x00000000 },
+	{ _MMIO(0x2744), 0x00800000 },
+	{ _MMIO(0x2710), 0x00000000 },
+	{ _MMIO(0x2714), 0x00800000 },
+	{ _MMIO(0x2720), 0x00000000 },
+	{ _MMIO(0x2724), 0x00800000 },
+};
+
+static const struct i915_oa_reg mux_config_sampler_balance[] = {
+	{ _MMIO(0x2eb9c), 0x01906400 },
+	{ _MMIO(0x2fb9c), 0x01906400 },
+	{ _MMIO(0x253a4), 0x00000000 },
+	{ _MMIO(0x26b9c), 0x01906400 },
+	{ _MMIO(0x27b9c), 0x01906400 },
+	{ _MMIO(0x27104), 0x00a00000 },
+	{ _MMIO(0x27184), 0x00a50000 },
+	{ _MMIO(0x2e804), 0x00500000 },
+	{ _MMIO(0x2e984), 0x00500000 },
+	{ _MMIO(0x2eb04), 0x00500000 },
+	{ _MMIO(0x2eb80), 0x00000084 },
+	{ _MMIO(0x2eb8c), 0x14200000 },
+	{ _MMIO(0x2eb84), 0x00000000 },
+	{ _MMIO(0x2f804), 0x00050000 },
+	{ _MMIO(0x2f984), 0x00050000 },
+	{ _MMIO(0x2fb04), 0x00050000 },
+	{ _MMIO(0x2fb80), 0x00000084 },
+	{ _MMIO(0x2fb8c), 0x00050800 },
+	{ _MMIO(0x2fb84), 0x00000000 },
+	{ _MMIO(0x25380), 0x00000010 },
+	{ _MMIO(0x2538c), 0x000000c0 },
+	{ _MMIO(0x25384), 0xaa550000 },
+	{ _MMIO(0x25404), 0xffffc000 },
+	{ _MMIO(0x26804), 0x50000000 },
+	{ _MMIO(0x26984), 0x50000000 },
+	{ _MMIO(0x26b04), 0x50000000 },
+	{ _MMIO(0x26b80), 0x00000084 },
+	{ _MMIO(0x26b90), 0x00050800 },
+	{ _MMIO(0x26b84), 0x00000000 },
+	{ _MMIO(0x27804), 0x05000000 },
+	{ _MMIO(0x27984), 0x05000000 },
+	{ _MMIO(0x27b04), 0x05000000 },
+	{ _MMIO(0x27b80), 0x00000084 },
+	{ _MMIO(0x27b90), 0x00000142 },
+	{ _MMIO(0x27b84), 0x00000000 },
+	{ _MMIO(0x26104), 0xa0000000 },
+	{ _MMIO(0x26184), 0xa5000000 },
+	{ _MMIO(0x25424), 0x00008620 },
+	{ _MMIO(0x2541c), 0x00000000 },
+	{ _MMIO(0x25428), 0x0004a54a },
+};
+
+static int select_sampler_balance_config(struct drm_i915_private *dev_priv)
+{
+	dev_priv->perf.oa.mux_regs =
+		mux_config_sampler_balance;
+	dev_priv->perf.oa.mux_regs_len =
+		ARRAY_SIZE(mux_config_sampler_balance);
+
+	dev_priv->perf.oa.b_counter_regs =
+		b_counter_config_sampler_balance;
+	dev_priv->perf.oa.b_counter_regs_len =
+		ARRAY_SIZE(b_counter_config_sampler_balance);
+
+	return 0;
+}
+
 int i915_oa_select_metric_set_hsw(struct drm_i915_private *dev_priv)
 {
 	dev_priv->perf.oa.mux_regs = NULL;
@@ -128,6 +459,16 @@ int i915_oa_select_metric_set_hsw(struct drm_i915_private *dev_priv)
 	switch (dev_priv->perf.oa.metrics_set) {
 	case METRIC_SET_ID_RENDER_BASIC:
 		return select_render_basic_config(dev_priv);
+	case METRIC_SET_ID_COMPUTE_BASIC:
+		return select_compute_basic_config(dev_priv);
+	case METRIC_SET_ID_COMPUTE_EXTENDED:
+		return select_compute_extended_config(dev_priv);
+	case METRIC_SET_ID_MEMORY_READS:
+		return select_memory_reads_config(dev_priv);
+	case METRIC_SET_ID_MEMORY_WRITES:
+		return select_memory_writes_config(dev_priv);
+	case METRIC_SET_ID_SAMPLER_BALANCE:
+		return select_sampler_balance_config(dev_priv);
 	default:
 		return -ENODEV;
 	}
@@ -136,42 +477,182 @@ int i915_oa_select_metric_set_hsw(struct drm_i915_private *dev_priv)
 static ssize_t
 show_render_basic_id(struct device *kdev, struct device_attribute *attr, char *buf)
 {
-        return sprintf(buf, "%d\n", METRIC_SET_ID_RENDER_BASIC);
+	return sprintf(buf, "%d\n", METRIC_SET_ID_RENDER_BASIC);
 }
 
 static struct device_attribute dev_attr_render_basic_id = {
-        .attr = { .name = "id", .mode = S_IRUGO },
-        .show = show_render_basic_id,
-        .store = NULL,
+	.attr = { .name = "id", .mode = S_IRUGO },
+	.show = show_render_basic_id,
+	.store = NULL,
 };
 
 static struct attribute *attrs_render_basic[] = {
-        &dev_attr_render_basic_id.attr,
-        NULL,
+	&dev_attr_render_basic_id.attr,
+	NULL,
 };
 
 static struct attribute_group group_render_basic = {
-        .name = "403d8832-1a27-4aa6-a64e-f5389ce7b212",
-        .attrs =  attrs_render_basic,
+	.name = "403d8832-1a27-4aa6-a64e-f5389ce7b212",
+	.attrs =  attrs_render_basic,
+};
+
+static ssize_t
+show_compute_basic_id(struct device *kdev, struct device_attribute *attr, char *buf)
+{
+	return sprintf(buf, "%d\n", METRIC_SET_ID_COMPUTE_BASIC);
+}
+
+static struct device_attribute dev_attr_compute_basic_id = {
+	.attr = { .name = "id", .mode = S_IRUGO },
+	.show = show_compute_basic_id,
+	.store = NULL,
+};
+
+static struct attribute *attrs_compute_basic[] = {
+	&dev_attr_compute_basic_id.attr,
+	NULL,
+};
+
+static struct attribute_group group_compute_basic = {
+	.name = "39ad14bc-2380-45c4-91eb-fbcb3aa7ae7b",
+	.attrs =  attrs_compute_basic,
+};
+
+static ssize_t
+show_compute_extended_id(struct device *kdev, struct device_attribute *attr, char *buf)
+{
+	return sprintf(buf, "%d\n", METRIC_SET_ID_COMPUTE_EXTENDED);
+}
+
+static struct device_attribute dev_attr_compute_extended_id = {
+	.attr = { .name = "id", .mode = S_IRUGO },
+	.show = show_compute_extended_id,
+	.store = NULL,
+};
+
+static struct attribute *attrs_compute_extended[] = {
+	&dev_attr_compute_extended_id.attr,
+	NULL,
+};
+
+static struct attribute_group group_compute_extended = {
+	.name = "3865be28-6982-49fe-9494-e4d1b4795413",
+	.attrs =  attrs_compute_extended,
+};
+
+static ssize_t
+show_memory_reads_id(struct device *kdev, struct device_attribute *attr, char *buf)
+{
+	return sprintf(buf, "%d\n", METRIC_SET_ID_MEMORY_READS);
+}
+
+static struct device_attribute dev_attr_memory_reads_id = {
+	.attr = { .name = "id", .mode = S_IRUGO },
+	.show = show_memory_reads_id,
+	.store = NULL,
+};
+
+static struct attribute *attrs_memory_reads[] = {
+	&dev_attr_memory_reads_id.attr,
+	NULL,
+};
+
+static struct attribute_group group_memory_reads = {
+	.name = "bb5ed49b-2497-4095-94f6-26ba294db88a",
+	.attrs =  attrs_memory_reads,
+};
+
+static ssize_t
+show_memory_writes_id(struct device *kdev, struct device_attribute *attr, char *buf)
+{
+	return sprintf(buf, "%d\n", METRIC_SET_ID_MEMORY_WRITES);
+}
+
+static struct device_attribute dev_attr_memory_writes_id = {
+	.attr = { .name = "id", .mode = S_IRUGO },
+	.show = show_memory_writes_id,
+	.store = NULL,
+};
+
+static struct attribute *attrs_memory_writes[] = {
+	&dev_attr_memory_writes_id.attr,
+	NULL,
+};
+
+static struct attribute_group group_memory_writes = {
+	.name = "3358d639-9b5f-45ab-976d-9b08cbfc6240",
+	.attrs =  attrs_memory_writes,
+};
+
+static ssize_t
+show_sampler_balance_id(struct device *kdev, struct device_attribute *attr, char *buf)
+{
+	return sprintf(buf, "%d\n", METRIC_SET_ID_SAMPLER_BALANCE);
+}
+
+static struct device_attribute dev_attr_sampler_balance_id = {
+	.attr = { .name = "id", .mode = S_IRUGO },
+	.show = show_sampler_balance_id,
+	.store = NULL,
+};
+
+static struct attribute *attrs_sampler_balance[] = {
+	&dev_attr_sampler_balance_id.attr,
+	NULL,
+};
+
+static struct attribute_group group_sampler_balance = {
+	.name = "bc274488-b4b6-40c7-90da-b77d7ad16189",
+	.attrs =  attrs_sampler_balance,
 };
 
 int
 i915_perf_init_sysfs_hsw(struct drm_i915_private *dev_priv)
 {
-        int ret;
+	int ret;
 
-        ret = sysfs_create_group(dev_priv->perf.metrics_kobj, &group_render_basic);
-        if (ret)
-                goto error_render_basic;
+	ret = sysfs_create_group(dev_priv->perf.metrics_kobj, &group_render_basic);
+	if (ret)
+		goto error_render_basic;
+	ret = sysfs_create_group(dev_priv->perf.metrics_kobj, &group_compute_basic);
+	if (ret)
+		goto error_compute_basic;
+	ret = sysfs_create_group(dev_priv->perf.metrics_kobj, &group_compute_extended);
+	if (ret)
+		goto error_compute_extended;
+	ret = sysfs_create_group(dev_priv->perf.metrics_kobj, &group_memory_reads);
+	if (ret)
+		goto error_memory_reads;
+	ret = sysfs_create_group(dev_priv->perf.metrics_kobj, &group_memory_writes);
+	if (ret)
+		goto error_memory_writes;
+	ret = sysfs_create_group(dev_priv->perf.metrics_kobj, &group_sampler_balance);
+	if (ret)
+		goto error_sampler_balance;
 
-        return 0;
+	return 0;
 
+error_sampler_balance:
+	sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_memory_writes);
+error_memory_writes:
+	sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_memory_reads);
+error_memory_reads:
+	sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_compute_extended);
+error_compute_extended:
+	sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_compute_basic);
+error_compute_basic:
+	sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_render_basic);
 error_render_basic:
-        return ret;
+	return ret;
 }
 
 void
 i915_perf_deinit_sysfs_hsw(struct drm_i915_private *dev_priv)
 {
-        sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_render_basic);
+	sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_render_basic);
+	sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_compute_basic);
+	sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_compute_extended);
+	sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_memory_reads);
+	sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_memory_writes);
+	sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_sampler_balance);
 }
Signature · 6 lines
2.7.0

_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel
#12 lore raw

✓ Fi.CI.BAT: success for Enable Gen 7 Observation Architecture (rev2)

In reply to Robert Bragg
== Summary ==

== Summary ==

Series 3024v2 Enable Gen 7 Observation Architecture
http://patchwork.freedesktop.org/api/1.0/series/3024/revisions/2/mbox/

bdw-nuci7        total:156  pass:147  dwarn:0   dfail:0   fail:0   skip:9  
bdw-ultra        total:159  pass:147  dwarn:0   dfail:0   fail:0   skip:12 
bsw-nuc-2        total:159  pass:131  dwarn:0   dfail:0   fail:0   skip:28 
byt-nuc          total:159  pass:136  dwarn:0   dfail:0   fail:0   skip:23 
hsw-brixbox      total:159  pass:146  dwarn:0   dfail:0   fail:0   skip:13 
hsw-gt2          total:159  pass:149  dwarn:0   dfail:0   fail:0   skip:10 
ilk-hp8440p      total:159  pass:111  dwarn:0   dfail:0   fail:0   skip:48 
ivb-t430s        total:159  pass:145  dwarn:0   dfail:0   fail:0   skip:14 
skl-i5k-2        total:159  pass:144  dwarn:1   dfail:0   fail:0   skip:14 
skl-i7k-2        total:159  pass:144  dwarn:1   dfail:0   fail:0   skip:14 
snb-dellxps      total:159  pass:137  dwarn:0   dfail:0   fail:0   skip:22 

Results at /archive/results/CI_IGT_test/Patchwork_1359/

1ff67c58aa3d7cacd37451397c740b8df27994e6 drm-intel-nightly: 2016y-02m-03d-18h-22m-38s UTC integration manifest 302f43ab1f73bfa02331cf48d09c0a4633a09a2a drm/i915: Add more Haswell OA metric sets 2d51257b0c7989ed1dae829ce4fe2b737342fbb3 drm/i915: add oa_event_min_timer_exponent sysctl e25850343facd37fa0db9e1f564fc4a1aa2d04a9 drm/i915: Add dev.i915.perf_event_paranoid sysctl option 4068269d6f53b90ef589684279714086887d1070 drm/i915: advertise available metrics via sysfs 9541e82d530e05442ce06e52a608545003edd8c2 drm/i915: Add i915 perf event for Haswell OA unit 730c581d81a0588abb81d846069d614f489548a9 drm/i915: Add 'render basic' Haswell OA unit config de7203e5abe978328f41dcad28f96b306aacf684 drm/i915: rename OACONTROL GEN7_OACONTROL d73e9b28435b554910bf1449f5261945b482bac7 drm/i915: Add i915 perf infrastructure

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx