[PATCH v9 00/11] Enable i915 perf stream for Haswell OA unit

[PATCH v9 00/11] Enable i915 perf stream for Haswell OA unit

[Robert Bragg]

Rebased and updated with more feedback from Sourab and Matt.

In particular the patch that added the oa_min_timer_exponent sysctl parameter
has now been replaced with one adding an oa_max_sample_rate parameter in Hz.

This way userspace policy won't need to be tailored to different systems when
gen9 OA is enabled where the exponents don't represent the same periods as for
Haswell.

- Robert

Robert Bragg (11):
  drm/i915: Add i915 perf infrastructure
  drm/i915: rename OACONTROL GEN7_OACONTROL
  drm/i915: return EACCES for check_cmd() failures
  drm/i915: don't whitelist oacontrol in cmd parser
  drm/i915: Add 'render basic' Haswell OA unit config
  drm/i915: Enable i915 perf stream for Haswell OA unit
  drm/i915: advertise available metrics via sysfs
  drm/i915: Add dev.i915.perf_stream_paranoid sysctl option
  drm/i915: add dev.i915.oa_max_sample_rate sysctl
  drm/i915: Add more Haswell OA metric sets
  drm/i915: Add a kerneldoc summary for i915_perf.c

 drivers/gpu/drm/i915/Makefile          |    4 +
 drivers/gpu/drm/i915/gvt/handlers.c    |    2 +-
 drivers/gpu/drm/i915/i915_cmd_parser.c |   45 +-
 drivers/gpu/drm/i915/i915_drv.c        |    9 +
 drivers/gpu/drm/i915/i915_drv.h        |  156 +++
 drivers/gpu/drm/i915/i915_oa_hsw.c     |  752 ++++++++++++++
 drivers/gpu/drm/i915/i915_oa_hsw.h     |   38 +
 drivers/gpu/drm/i915/i915_perf.c       | 1753 ++++++++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_reg.h        |  340 ++++++-
 include/uapi/drm/i915_drm.h            |  134 +++
 10 files changed, 3193 insertions(+), 40 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

-- 
2.10.1

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v9 01/11] drm/i915: Add i915 perf infrastructure

[Robert Bragg]

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_PROP_CTX_HANDLE,        ctx_handle,

      /* Include OA reports in samples */
      DRM_I915_PERF_PROP_SAMPLE_OA,         true,

      /* OA unit configuration */
      DRM_I915_PERF_PROP_OA_METRICS_SET,    metrics_set_id,
      DRM_I915_PERF_PROP_OA_FORMAT,         report_format,
      DRM_I915_PERF_PROP_OA_EXPONENT,       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_ptr = (uint64_t)properties,
      .num_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_PROP_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.

v2:
    use i915_gem_context_get() - Chris Wilson
v3:
    update read() interface to avoid passing state struct - Chris Wilson
    fix some rebase fallout, with i915-perf init/deinit
v4:
    s/DRM_IORW/DRM_IOW/ - Emil Velikov

Signed-off-by: Robert Bragg <robert@sixbynine.org>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>
---
 drivers/gpu/drm/i915/Makefile    |   3 +
 drivers/gpu/drm/i915/i915_drv.c  |   4 +
 drivers/gpu/drm/i915/i915_drv.h  |  91 ++++++++
 drivers/gpu/drm/i915/i915_perf.c | 449 +++++++++++++++++++++++++++++++++++++++
 include/uapi/drm/i915_drm.h      |  67 ++++++
 5 files changed, 614 insertions(+)
 create mode 100644 drivers/gpu/drm/i915/i915_perf.c

diff --git a/drivers/gpu/drm/i915/Makefile b/drivers/gpu/drm/i915/Makefile
index 0857e50..08b43af 100644
--- a/drivers/gpu/drm/i915/Makefile
+++ b/drivers/gpu/drm/i915/Makefile
@@ -116,6 +116,9 @@ i915-$(CONFIG_DRM_I915_CAPTURE_ERROR) += i915_gpu_error.o
 # virtual gpu code
 i915-y += i915_vgpu.o
 
+# perf code
+i915-y += i915_perf.o
+
 ifeq ($(CONFIG_DRM_I915_GVT),y)
 i915-y += intel_gvt.o
 include $(src)/gvt/Makefile
diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c
index 0213a30..22b4166 100644
--- a/drivers/gpu/drm/i915/i915_drv.c
+++ b/drivers/gpu/drm/i915/i915_drv.c
@@ -844,6 +844,8 @@ static int i915_driver_init_early(struct drm_i915_private *dev_priv,
 
 	intel_detect_preproduction_hw(dev_priv);
 
+	i915_perf_init(dev_priv);
+
 	return 0;
 
 err_gvt:
@@ -859,6 +861,7 @@ static int i915_driver_init_early(struct drm_i915_private *dev_priv,
  */
 static void i915_driver_cleanup_early(struct drm_i915_private *dev_priv)
 {
+	i915_perf_fini(dev_priv);
 	i915_gem_load_cleanup(&dev_priv->drm);
 	i915_workqueues_cleanup(dev_priv);
 }
@@ -2561,6 +2564,7 @@ static 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),
 };
 
 static struct drm_driver driver = {
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 4735b417..e4cd322 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1785,6 +1785,84 @@ struct intel_wm_config {
 	bool sprites_scaled;
 };
 
+struct i915_perf_stream;
+
+struct i915_perf_stream_ops {
+	/* Enables the collection of HW samples, either in response to
+	 * I915_PERF_IOCTL_ENABLE or implicitly called when stream is
+	 * opened without I915_PERF_FLAG_DISABLED.
+	 */
+	void (*enable)(struct i915_perf_stream *stream);
+
+	/* Disables the collection of HW samples, either in response to
+	 * I915_PERF_IOCTL_DISABLE or implicitly called before
+	 * destroying the stream.
+	 */
+	void (*disable)(struct i915_perf_stream *stream);
+
+	/* Return: true if any i915 perf records are ready to read()
+	 * for this stream.
+	 */
+	bool (*can_read)(struct i915_perf_stream *stream);
+
+	/* Call poll_wait, passing a wait queue that will be woken
+	 * once there is something ready to read() for the stream
+	 */
+	void (*poll_wait)(struct i915_perf_stream *stream,
+			  struct file *file,
+			  poll_table *wait);
+
+	/* For handling a blocking read, wait until there is something
+	 * to ready to read() for the stream. E.g. wait on the same
+	 * wait queue that would be passed to poll_wait() until
+	 * ->can_read() returns true (if its safe to call ->can_read()
+	 * without the i915 perf lock held).
+	 */
+	int (*wait_unlocked)(struct i915_perf_stream *stream);
+
+	/* read - Copy buffered metrics as records to userspace
+	 * @buf: the userspace, destination buffer
+	 * @count: the number of bytes to copy, requested by userspace
+	 * @offset: zero at the start of the read, updated as the read
+	 *          proceeds, it represents how many bytes have been
+	 *          copied so far and the buffer offset for copying the
+	 *          next record.
+	 *
+	 * Copy as many buffered i915 perf samples and records for
+	 * this stream to userspace as will fit in the given buffer.
+	 *
+	 * Only write complete records; returning -ENOSPC if there
+	 * isn't room for a complete record.
+	 *
+	 * Return any error condition that results in a short read
+	 * such as -ENOSPC or -EFAULT, even though these may be
+	 * squashed before returning to userspace.
+	 */
+	int (*read)(struct i915_perf_stream *stream,
+		    char __user *buf,
+		    size_t count,
+		    size_t *offset);
+
+	/* Cleanup any stream specific resources.
+	 *
+	 * The stream will always be disabled before this is called.
+	 */
+	void (*destroy)(struct i915_perf_stream *stream);
+};
+
+struct i915_perf_stream {
+	struct drm_i915_private *dev_priv;
+
+	struct list_head link;
+
+	u32 sample_flags;
+
+	struct i915_gem_context *ctx;
+	bool enabled;
+
+	struct i915_perf_stream_ops *ops;
+};
+
 struct drm_i915_private {
 	struct drm_device drm;
 
@@ -2086,6 +2164,12 @@ struct drm_i915_private {
 
 	struct i915_runtime_pm pm;
 
+	struct {
+		bool initialized;
+		struct mutex lock;
+		struct list_head streams;
+	} perf;
+
 	/* Abstract the submission mechanism (legacy ringbuffer or execlists) away */
 	struct {
 		void (*resume)(struct drm_i915_private *);
@@ -3604,6 +3688,9 @@ int i915_gem_context_setparam_ioctl(struct drm_device *dev, void *data,
 int i915_gem_context_reset_stats_ioctl(struct drm_device *dev, void *data,
 				       struct drm_file *file);
 
+int i915_perf_open_ioctl(struct drm_device *dev, void *data,
+			 struct drm_file *file);
+
 /* i915_gem_evict.c */
 int __must_check i915_gem_evict_something(struct i915_address_space *vm,
 					  u64 min_size, u64 alignment,
@@ -3734,6 +3821,10 @@ int intel_engine_cmd_parser(struct intel_engine_cs *engine,
 			    u32 batch_len,
 			    bool is_master);
 
+/* i915_perf.c */
+extern void i915_perf_init(struct drm_i915_private *dev_priv);
+extern void i915_perf_fini(struct drm_i915_private *dev_priv);
+
 /* i915_suspend.c */
 extern int i915_save_state(struct drm_device *dev);
 extern int i915_restore_state(struct drm_device *dev);
diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c
new file mode 100644
index 0000000..777ce65
--- /dev/null
+++ b/drivers/gpu/drm/i915/i915_perf.c
@@ -0,0 +1,449 @@
+/*
+ * Copyright © 2015-2016 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.
+ *
+ * Authors:
+ *   Robert Bragg <robert@sixbynine.org>
+ */
+
+#include <linux/anon_inodes.h>
+
+#include "i915_drv.h"
+
+struct perf_open_properties {
+	u32 sample_flags;
+
+	u64 single_context:1;
+	u64 ctx_handle;
+};
+
+static ssize_t i915_perf_read_locked(struct i915_perf_stream *stream,
+				     struct file *file,
+				     char __user *buf,
+				     size_t count,
+				     loff_t *ppos)
+{
+	/* Note we keep the offset (aka bytes read) separate from any
+	 * error status so that the final check for whether we return
+	 * the bytes read with a higher precedence than any error (see
+	 * comment below) doesn't need to be handled/duplicated in
+	 * stream->ops->read() implementations.
+	 */
+	size_t offset = 0;
+	int ret = stream->ops->read(stream, buf, count, &offset);
+
+	/* If we've successfully copied any data then reporting that
+	 * takes precedence over any internal error status, so the
+	 * data isn't lost.
+	 *
+	 * For example ret will be -ENOSPC whenever there is more
+	 * buffered data than can be copied to userspace, but that's
+	 * only interesting if we weren't able to copy some data
+	 * because it implies the userspace buffer is too small to
+	 * receive a single record (and we never split records).
+	 *
+	 * Another case with ret == -EFAULT is more of a grey area
+	 * since it would seem like bad form for userspace to ask us
+	 * to overrun its buffer, but the user knows best:
+	 *
+	 *   http://yarchive.net/comp/linux/partial_reads_writes.html
+	 */
+	return offset ?: (ret ?: -EAGAIN);
+}
+
+static ssize_t i915_perf_read(struct file *file,
+			      char __user *buf,
+			      size_t count,
+			      loff_t *ppos)
+{
+	struct i915_perf_stream *stream = file->private_data;
+	struct drm_i915_private *dev_priv = stream->dev_priv;
+	ssize_t ret;
+
+	if (!(file->f_flags & O_NONBLOCK)) {
+		/* Allow false positives from stream->ops->wait_unlocked.
+		 */
+		do {
+			ret = stream->ops->wait_unlocked(stream);
+			if (ret)
+				return ret;
+
+			mutex_lock(&dev_priv->perf.lock);
+			ret = i915_perf_read_locked(stream, file,
+						    buf, count, ppos);
+			mutex_unlock(&dev_priv->perf.lock);
+		} while (ret == -EAGAIN);
+	} else {
+		mutex_lock(&dev_priv->perf.lock);
+		ret = i915_perf_read_locked(stream, file, buf, count, ppos);
+		mutex_unlock(&dev_priv->perf.lock);
+	}
+
+	return ret;
+}
+
+static unsigned int i915_perf_poll_locked(struct i915_perf_stream *stream,
+					  struct file *file,
+					  poll_table *wait)
+{
+	unsigned int streams = 0;
+
+	stream->ops->poll_wait(stream, file, wait);
+
+	if (stream->ops->can_read(stream))
+		streams |= POLLIN;
+
+	return streams;
+}
+
+static unsigned int i915_perf_poll(struct file *file, poll_table *wait)
+{
+	struct i915_perf_stream *stream = file->private_data;
+	struct drm_i915_private *dev_priv = stream->dev_priv;
+	int ret;
+
+	mutex_lock(&dev_priv->perf.lock);
+	ret = i915_perf_poll_locked(stream, file, wait);
+	mutex_unlock(&dev_priv->perf.lock);
+
+	return ret;
+}
+
+static void i915_perf_enable_locked(struct i915_perf_stream *stream)
+{
+	if (stream->enabled)
+		return;
+
+	/* Allow stream->ops->enable() to refer to this */
+	stream->enabled = true;
+
+	if (stream->ops->enable)
+		stream->ops->enable(stream);
+}
+
+static void i915_perf_disable_locked(struct i915_perf_stream *stream)
+{
+	if (!stream->enabled)
+		return;
+
+	/* Allow stream->ops->disable() to refer to this */
+	stream->enabled = false;
+
+	if (stream->ops->disable)
+		stream->ops->disable(stream);
+}
+
+static long i915_perf_ioctl_locked(struct i915_perf_stream *stream,
+				   unsigned int cmd,
+				   unsigned long arg)
+{
+	switch (cmd) {
+	case I915_PERF_IOCTL_ENABLE:
+		i915_perf_enable_locked(stream);
+		return 0;
+	case I915_PERF_IOCTL_DISABLE:
+		i915_perf_disable_locked(stream);
+		return 0;
+	}
+
+	return -EINVAL;
+}
+
+static long i915_perf_ioctl(struct file *file,
+			    unsigned int cmd,
+			    unsigned long arg)
+{
+	struct i915_perf_stream *stream = file->private_data;
+	struct drm_i915_private *dev_priv = stream->dev_priv;
+	long ret;
+
+	mutex_lock(&dev_priv->perf.lock);
+	ret = i915_perf_ioctl_locked(stream, cmd, arg);
+	mutex_unlock(&dev_priv->perf.lock);
+
+	return ret;
+}
+
+static void i915_perf_destroy_locked(struct i915_perf_stream *stream)
+{
+	struct drm_i915_private *dev_priv = stream->dev_priv;
+
+	if (stream->enabled)
+		i915_perf_disable_locked(stream);
+
+	if (stream->ops->destroy)
+		stream->ops->destroy(stream);
+
+	list_del(&stream->link);
+
+	if (stream->ctx) {
+		mutex_lock(&dev_priv->drm.struct_mutex);
+		i915_gem_context_put(stream->ctx);
+		mutex_unlock(&dev_priv->drm.struct_mutex);
+	}
+
+	kfree(stream);
+}
+
+static int i915_perf_release(struct inode *inode, struct file *file)
+{
+	struct i915_perf_stream *stream = file->private_data;
+	struct drm_i915_private *dev_priv = stream->dev_priv;
+
+	mutex_lock(&dev_priv->perf.lock);
+	i915_perf_destroy_locked(stream);
+	mutex_unlock(&dev_priv->perf.lock);
+
+	return 0;
+}
+
+
+static const struct file_operations fops = {
+	.owner		= THIS_MODULE,
+	.llseek		= no_llseek,
+	.release	= i915_perf_release,
+	.poll		= i915_perf_poll,
+	.read		= i915_perf_read,
+	.unlocked_ioctl	= i915_perf_ioctl,
+};
+
+
+static struct i915_gem_context *
+lookup_context(struct drm_i915_private *dev_priv,
+	       struct drm_i915_file_private *file_priv,
+	       u32 ctx_user_handle)
+{
+	struct i915_gem_context *ctx;
+	int ret;
+
+	ret = i915_mutex_lock_interruptible(&dev_priv->drm);
+	if (ret)
+		return ERR_PTR(ret);
+
+	ctx = i915_gem_context_lookup(file_priv, ctx_user_handle);
+	if (!IS_ERR(ctx))
+		i915_gem_context_get(ctx);
+
+	mutex_unlock(&dev_priv->drm.struct_mutex);
+
+	return ctx;
+}
+
+static int
+i915_perf_open_ioctl_locked(struct drm_i915_private *dev_priv,
+			    struct drm_i915_perf_open_param *param,
+			    struct perf_open_properties *props,
+			    struct drm_file *file)
+{
+	struct i915_gem_context *specific_ctx = NULL;
+	struct i915_perf_stream *stream = NULL;
+	unsigned long f_flags = 0;
+	int stream_fd;
+	int ret;
+
+	if (props->single_context) {
+		u32 ctx_handle = props->ctx_handle;
+		struct drm_i915_file_private *file_priv = file->driver_priv;
+
+		specific_ctx = lookup_context(dev_priv, file_priv, ctx_handle);
+		if (IS_ERR(specific_ctx)) {
+			ret = PTR_ERR(specific_ctx);
+			if (ret != -EINTR)
+				DRM_ERROR("Failed to look up context with ID %u for opening perf stream\n",
+					  ctx_handle);
+			goto err;
+		}
+	}
+
+	if (!specific_ctx && !capable(CAP_SYS_ADMIN)) {
+		DRM_ERROR("Insufficient privileges to open system-wide i915 perf stream\n");
+		ret = -EACCES;
+		goto err_ctx;
+	}
+
+	stream = kzalloc(sizeof(*stream), GFP_KERNEL);
+	if (!stream) {
+		ret = -ENOMEM;
+		goto err_ctx;
+	}
+
+	stream->sample_flags = props->sample_flags;
+	stream->dev_priv = dev_priv;
+	stream->ctx = specific_ctx;
+
+	/*
+	 * TODO: support sampling something
+	 *
+	 * For now this is as far as we can go.
+	 */
+	DRM_ERROR("Unsupported i915 perf stream configuration\n");
+	ret = -EINVAL;
+	goto err_alloc;
+
+	list_add(&stream->link, &dev_priv->perf.streams);
+
+	if (param->flags & I915_PERF_FLAG_FD_CLOEXEC)
+		f_flags |= O_CLOEXEC;
+	if (param->flags & I915_PERF_FLAG_FD_NONBLOCK)
+		f_flags |= O_NONBLOCK;
+
+	stream_fd = anon_inode_getfd("[i915_perf]", &fops, stream, f_flags);
+	if (stream_fd < 0) {
+		ret = stream_fd;
+		goto err_open;
+	}
+
+	if (!(param->flags & I915_PERF_FLAG_DISABLED))
+		i915_perf_enable_locked(stream);
+
+	return stream_fd;
+
+err_open:
+	list_del(&stream->link);
+	if (stream->ops->destroy)
+		stream->ops->destroy(stream);
+err_alloc:
+	kfree(stream);
+err_ctx:
+	if (specific_ctx) {
+		mutex_lock(&dev_priv->drm.struct_mutex);
+		i915_gem_context_put(specific_ctx);
+		mutex_unlock(&dev_priv->drm.struct_mutex);
+	}
+err:
+	return ret;
+}
+
+/* Note we copy the properties from userspace outside of the i915 perf
+ * mutex to avoid an awkward lockdep with mmap_sem.
+ *
+ * Note this function only validates properties in isolation it doesn't
+ * validate that the combination of properties makes sense or that all
+ * properties necessary for a particular kind of stream have been set.
+ */
+static int read_properties_unlocked(struct drm_i915_private *dev_priv,
+				    u64 __user *uprops,
+				    u32 n_props,
+				    struct perf_open_properties *props)
+{
+	u64 __user *uprop = uprops;
+	int i;
+
+	memset(props, 0, sizeof(struct perf_open_properties));
+
+	if (!n_props) {
+		DRM_ERROR("No i915 perf properties given");
+		return -EINVAL;
+	}
+
+	/* Considering that ID = 0 is reserved and assuming that we don't
+	 * (currently) expect any configurations to ever specify duplicate
+	 * values for a particular property ID then the last _PROP_MAX value is
+	 * one greater than the maximum number of properties we expect to get
+	 * from userspace.
+	 */
+	if (n_props >= DRM_I915_PERF_PROP_MAX) {
+		DRM_ERROR("More i915 perf properties specified than exist");
+		return -EINVAL;
+	}
+
+	for (i = 0; i < n_props; i++) {
+		u64 id, value;
+		int ret;
+
+		ret = get_user(id, uprop);
+		if (ret)
+			return ret;
+
+		ret = get_user(value, uprop + 1);
+		if (ret)
+			return ret;
+
+		switch ((enum drm_i915_perf_property_id)id) {
+		case DRM_I915_PERF_PROP_CTX_HANDLE:
+			props->single_context = 1;
+			props->ctx_handle = value;
+			break;
+		default:
+			MISSING_CASE(id);
+			DRM_ERROR("Unknown i915 perf property ID");
+			return -EINVAL;
+		}
+
+		uprop += 2;
+	}
+
+	return 0;
+}
+
+int i915_perf_open_ioctl(struct drm_device *dev, void *data,
+			 struct drm_file *file)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	struct drm_i915_perf_open_param *param = data;
+	struct perf_open_properties props;
+	u32 known_open_flags;
+	int ret;
+
+	if (!dev_priv->perf.initialized) {
+		DRM_ERROR("i915 perf interface not available for this system");
+		return -ENOTSUPP;
+	}
+
+	known_open_flags = I915_PERF_FLAG_FD_CLOEXEC |
+			   I915_PERF_FLAG_FD_NONBLOCK |
+			   I915_PERF_FLAG_DISABLED;
+	if (param->flags & ~known_open_flags) {
+		DRM_ERROR("Unknown drm_i915_perf_open_param flag\n");
+		return -EINVAL;
+	}
+
+	ret = read_properties_unlocked(dev_priv,
+				       u64_to_user_ptr(param->properties_ptr),
+				       param->num_properties,
+				       &props);
+	if (ret)
+		return ret;
+
+	mutex_lock(&dev_priv->perf.lock);
+	ret = i915_perf_open_ioctl_locked(dev_priv, param, &props, file);
+	mutex_unlock(&dev_priv->perf.lock);
+
+	return ret;
+}
+
+void i915_perf_init(struct drm_i915_private *dev_priv)
+{
+	INIT_LIST_HEAD(&dev_priv->perf.streams);
+	mutex_init(&dev_priv->perf.lock);
+
+	dev_priv->perf.initialized = true;
+}
+
+void i915_perf_fini(struct drm_i915_private *dev_priv)
+{
+	if (!dev_priv->perf.initialized)
+		return;
+
+	/* Currently nothing to clean up */
+
+	dev_priv->perf.initialized = false;
+}
diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h
index 03725fe..98cd493 100644
--- a/include/uapi/drm/i915_drm.h
+++ b/include/uapi/drm/i915_drm.h
@@ -258,6 +258,7 @@ typedef struct _drm_i915_sarea {
 #define DRM_I915_GEM_USERPTR		0x33
 #define DRM_I915_GEM_CONTEXT_GETPARAM	0x34
 #define DRM_I915_GEM_CONTEXT_SETPARAM	0x35
+#define DRM_I915_PERF_OPEN		0x36
 
 #define DRM_IOCTL_I915_INIT		DRM_IOW( DRM_COMMAND_BASE + DRM_I915_INIT, drm_i915_init_t)
 #define DRM_IOCTL_I915_FLUSH		DRM_IO ( DRM_COMMAND_BASE + DRM_I915_FLUSH)
@@ -311,6 +312,7 @@ typedef struct _drm_i915_sarea {
 #define DRM_IOCTL_I915_GEM_USERPTR			DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_USERPTR, struct drm_i915_gem_userptr)
 #define DRM_IOCTL_I915_GEM_CONTEXT_GETPARAM	DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_CONTEXT_GETPARAM, struct drm_i915_gem_context_param)
 #define DRM_IOCTL_I915_GEM_CONTEXT_SETPARAM	DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_CONTEXT_SETPARAM, struct drm_i915_gem_context_param)
+#define DRM_IOCTL_I915_PERF_OPEN	DRM_IOW(DRM_COMMAND_BASE + DRM_I915_PERF_OPEN, struct drm_i915_perf_open_param)
 
 /* Allow drivers to submit batchbuffers directly to hardware, relying
  * on the security mechanisms provided by hardware.
@@ -1222,6 +1224,71 @@ struct drm_i915_gem_context_param {
 	__u64 value;
 };
 
+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_PROP_CTX_HANDLE = 1,
+
+	DRM_I915_PERF_PROP_MAX /* non-ABI */
+};
+
+struct drm_i915_perf_open_param {
+	__u32 flags;
+#define I915_PERF_FLAG_FD_CLOEXEC	(1<<0)
+#define I915_PERF_FLAG_FD_NONBLOCK	(1<<1)
+#define I915_PERF_FLAG_DISABLED		(1<<2)
+
+	/** The number of u64 (id, value) pairs */
+	__u32 num_properties;
+
+	/**
+	 * Pointer to array of u64 (id, value) pairs configuring the stream
+	 * to open.
+	 */
+	__u64 __user properties_ptr;
+};
+
+#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;
+	__u16 size;
+};
+
+enum drm_i915_perf_record_type {
+
+	/**
+	 * Samples are the work horse record type whose contents are extensible
+	 * and defined when opening an i915 perf stream based on the given
+	 * properties.
+	 *
+	 * Boolean properties following the naming convention
+	 * DRM_I915_PERF_SAMPLE_xyz_PROP request the inclusion of 'xyz' data in
+	 * every sample.
+	 *
+	 * The order of these sample properties given by userspace has no
+	 * affect on the ordering of data within a sample. The order will be
+	 * documented here.
+	 *
+	 * struct {
+	 *     struct drm_i915_perf_record_header header;
+	 *
+	 *     TODO: itemize extensible sample data here
+	 * };
+	 */
+	DRM_I915_PERF_RECORD_SAMPLE = 1,
+
+	DRM_I915_PERF_RECORD_MAX /* non-ABI */
+};
+
 #if defined(__cplusplus)
 }
 #endif
-- 
2.10.1

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v9 02/11] drm/i915: rename OACONTROL GEN7_OACONTROL

[Robert Bragg]

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

Signed-off-by: Robert Bragg <robert@sixbynine.org>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>
---
 drivers/gpu/drm/i915/gvt/handlers.c    | 2 +-
 drivers/gpu/drm/i915/i915_cmd_parser.c | 4 ++--
 drivers/gpu/drm/i915/i915_reg.h        | 2 +-
 3 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/gvt/handlers.c b/drivers/gpu/drm/i915/gvt/handlers.c
index 9ab1f95..4527cb7 100644
--- a/drivers/gpu/drm/i915/gvt/handlers.c
+++ b/drivers/gpu/drm/i915/gvt/handlers.c
@@ -2177,7 +2177,7 @@ static int init_generic_mmio_info(struct intel_gvt *gvt)
 	MMIO_DFH(0x1217c, D_ALL, F_CMD_ACCESS, NULL, NULL);
 
 	MMIO_F(0x2290, 8, 0, 0, 0, D_HSW_PLUS, NULL, NULL);
-	MMIO_D(OACONTROL, D_HSW);
+	MMIO_D(GEN7_OACONTROL, D_HSW);
 	MMIO_D(0x2b00, D_BDW_PLUS);
 	MMIO_D(0x2360, D_BDW_PLUS);
 	MMIO_F(0x5200, 32, 0, 0, 0, D_ALL, NULL, NULL);
diff --git a/drivers/gpu/drm/i915/i915_cmd_parser.c b/drivers/gpu/drm/i915/i915_cmd_parser.c
index f5039f4..7719aed 100644
--- a/drivers/gpu/drm/i915/i915_cmd_parser.c
+++ b/drivers/gpu/drm/i915/i915_cmd_parser.c
@@ -450,7 +450,7 @@ static const struct drm_i915_reg_descriptor gen7_render_regs[] = {
 	REG64(PS_INVOCATION_COUNT),
 	REG64(PS_DEPTH_COUNT),
 	REG64_IDX(RING_TIMESTAMP, RENDER_RING_BASE),
-	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),
@@ -1108,7 +1108,7 @@ static bool check_cmd(const struct intel_engine_cs *engine,
 			 * 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;
diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h
index 3361d7f..4623dbb 100644
--- a/drivers/gpu/drm/i915/i915_reg.h
+++ b/drivers/gpu/drm/i915/i915_reg.h
@@ -615,7 +615,7 @@ static inline bool i915_mmio_reg_valid(i915_reg_t reg)
 #define HSW_CS_GPR(n)                   _MMIO(0x2600 + (n) * 8)
 #define HSW_CS_GPR_UDW(n)               _MMIO(0x2600 + (n) * 8 + 4)
 
-#define OACONTROL _MMIO(0x2360)
+#define GEN7_OACONTROL _MMIO(0x2360)
 
 #define _GEN7_PIPEA_DE_LOAD_SL	0x70068
 #define _GEN7_PIPEB_DE_LOAD_SL	0x71068
-- 
2.10.1

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v9 03/11] drm/i915: return EACCES for check_cmd() failures

[Robert Bragg]

check_cmd() is checking whether a command adheres to certain
restrictions that ensure it's safe to execute within a privileged batch
buffer. Returning false implies a privilege problem, not that the
command is invalid.

The distinction makes the difference between allowing the buffer to be
executed as an unprivileged batch buffer or returning an EINVAL error to
userspace without executing anything.

In a case where userspace may want to test whether it can successfully
write to a register that needs privileges the distinction may be
important and an EINVAL error may be considered fatal.

In particular this is currently true for Mesa, which includes a test for
whether OACONTROL can be written too, but Mesa treats any error when
flushing a batch buffer as fatal, calling exit(1).

As it is currently Mesa can gracefully handle a failure to write to
OACONTROL if the command parser is disabled, but if we were to remove
OACONTROL from the parser's whitelist then the returned EINVAL would
break Mesa applications as they attempt an OACONTROL write.

This bumps the command parser version from 7 to 8, as the change is
visible to userspace.

Signed-off-by: Robert Bragg <robert@sixbynine.org>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>
---
 drivers/gpu/drm/i915/i915_cmd_parser.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_cmd_parser.c b/drivers/gpu/drm/i915/i915_cmd_parser.c
index 7719aed..c9d2ecd 100644
--- a/drivers/gpu/drm/i915/i915_cmd_parser.c
+++ b/drivers/gpu/drm/i915/i915_cmd_parser.c
@@ -1272,7 +1272,7 @@ int intel_engine_cmd_parser(struct intel_engine_cs *engine,
 
 		if (!check_cmd(engine, desc, cmd, length, is_master,
 			       &oacontrol_set)) {
-			ret = -EINVAL;
+			ret = -EACCES;
 			break;
 		}
 
@@ -1333,6 +1333,9 @@ int i915_cmd_parser_get_version(struct drm_i915_private *dev_priv)
 	 * 5. GPGPU dispatch compute indirect registers.
 	 * 6. TIMESTAMP register and Haswell CS GPR registers
 	 * 7. Allow MI_LOAD_REGISTER_REG between whitelisted registers.
+	 * 8. Don't report cmd_check() failures as EINVAL errors to userspace;
+	 *    rely on the HW to NOOP disallowed commands as it would without
+	 *    the parser enabled.
 	 */
-	return 7;
+	return 8;
 }
-- 
2.10.1

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v9 04/11] drm/i915: don't whitelist oacontrol in cmd parser

[Robert Bragg]

Being able to program OACONTROL from a non-privileged batch buffer is
not sufficient to be able to configure the OA unit. This was originally
allowed to help enable Mesa to expose OA counters via the
INTEL_performance_query extension, but the current implementation based
on programming OACONTROL via a batch buffer isn't able to report useable
data without a more complete OA unit configuration. Mesa handles the
possibility that writes to OACONTROL may not be allowed and so only
advertises the extension after explicitly testing that a write to
OACONTROL succeeds. Based on this; removing OACONTROL from the whitelist
should be ok for userspace.

Removing this simplifies adding a new kernel api for configuring the OA
unit without needing to consider the possibility that userspace might
trample on OACONTROL state which we'd like to start managing within
the kernel instead. In particular running any Mesa based GL application
currently results in clearing OACONTROL when initializing which would
disable the capturing of metrics.

Signed-off-by: Robert Bragg <robert@sixbynine.org>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>
---
 drivers/gpu/drm/i915/i915_cmd_parser.c | 38 ++--------------------------------
 1 file changed, 2 insertions(+), 36 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_cmd_parser.c b/drivers/gpu/drm/i915/i915_cmd_parser.c
index c9d2ecd..f3453dc 100644
--- a/drivers/gpu/drm/i915/i915_cmd_parser.c
+++ b/drivers/gpu/drm/i915/i915_cmd_parser.c
@@ -450,7 +450,6 @@ static const struct drm_i915_reg_descriptor gen7_render_regs[] = {
 	REG64(PS_INVOCATION_COUNT),
 	REG64(PS_DEPTH_COUNT),
 	REG64_IDX(RING_TIMESTAMP, RENDER_RING_BASE),
-	REG32(GEN7_OACONTROL), /* Only allowed for LRI and SRM. See below. */
 	REG64(MI_PREDICATE_SRC0),
 	REG64(MI_PREDICATE_SRC1),
 	REG32(GEN7_3DPRIM_END_OFFSET),
@@ -1060,8 +1059,7 @@ bool intel_engine_needs_cmd_parser(struct intel_engine_cs *engine)
 static bool check_cmd(const struct intel_engine_cs *engine,
 		      const struct drm_i915_cmd_descriptor *desc,
 		      const u32 *cmd, u32 length,
-		      const bool is_master,
-		      bool *oacontrol_set)
+		      const bool is_master)
 {
 	if (desc->flags & CMD_DESC_SKIP)
 		return true;
@@ -1099,31 +1097,6 @@ static bool check_cmd(const struct intel_engine_cs *engine,
 			}
 
 			/*
-			 * OACONTROL requires some special handling for
-			 * writes. We want to make sure that any batch which
-			 * enables OA also disables it before the end of the
-			 * batch. The goal is to prevent one process from
-			 * snooping on the perf data from another process. To do
-			 * that, we need to check the value that will be written
-			 * to the register. Hence, limit OACONTROL writes to
-			 * only MI_LOAD_REGISTER_IMM commands.
-			 */
-			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;
-				}
-
-				if (desc->cmd.value == MI_LOAD_REGISTER_REG) {
-					DRM_DEBUG_DRIVER("CMD: Rejected LRR to OACONTROL\n");
-					return false;
-				}
-
-				if (desc->cmd.value == MI_LOAD_REGISTER_IMM(1))
-					*oacontrol_set = (cmd[offset + 1] != 0);
-			}
-
-			/*
 			 * Check the value written to the register against the
 			 * allowed mask/value pair given in the whitelist entry.
 			 */
@@ -1214,7 +1187,6 @@ int intel_engine_cmd_parser(struct intel_engine_cs *engine,
 	u32 *cmd, *batch_end;
 	struct drm_i915_cmd_descriptor default_desc = noop_desc;
 	const struct drm_i915_cmd_descriptor *desc = &default_desc;
-	bool oacontrol_set = false; /* OACONTROL tracking. See check_cmd() */
 	bool needs_clflush_after = false;
 	int ret = 0;
 
@@ -1270,8 +1242,7 @@ int intel_engine_cmd_parser(struct intel_engine_cs *engine,
 			break;
 		}
 
-		if (!check_cmd(engine, desc, cmd, length, is_master,
-			       &oacontrol_set)) {
+		if (!check_cmd(engine, desc, cmd, length, is_master)) {
 			ret = -EACCES;
 			break;
 		}
@@ -1279,11 +1250,6 @@ int intel_engine_cmd_parser(struct intel_engine_cs *engine,
 		cmd += length;
 	}
 
-	if (oacontrol_set) {
-		DRM_DEBUG_DRIVER("CMD: batch set OACONTROL but did not clear it\n");
-		ret = -EINVAL;
-	}
-
 	if (cmd >= batch_end) {
 		DRM_DEBUG_DRIVER("CMD: Got to the end of the buffer w/o a BBE cmd!\n");
 		ret = -EINVAL;
-- 
2.10.1

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v9 05/11] drm/i915: Add 'render basic' Haswell OA unit config

[Robert Bragg]

Adds a static OA unit, MUX + B Counter configuration for basic render
metrics on Haswell. This is auto generated from an XML
description of metric sets, currently maintained in gputop, ref:

  https://github.com/rib/gputop
  > gputop-data/oa-*.xml
  > scripts/i915-perf-kernelgen.py

  $ make -C gputop-data -f Makefile.xml SYSFS=0 WHITELIST=RenderBasic

Signed-off-by: Robert Bragg <robert@sixbynine.org>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
---
 drivers/gpu/drm/i915/Makefile      |   3 +-
 drivers/gpu/drm/i915/i915_drv.h    |  14 ++++
 drivers/gpu/drm/i915/i915_oa_hsw.c | 144 +++++++++++++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_oa_hsw.h |  34 +++++++++
 4 files changed, 194 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

diff --git a/drivers/gpu/drm/i915/Makefile b/drivers/gpu/drm/i915/Makefile
index 08b43af..7e9e6d0 100644
--- a/drivers/gpu/drm/i915/Makefile
+++ b/drivers/gpu/drm/i915/Makefile
@@ -117,7 +117,8 @@ i915-$(CONFIG_DRM_I915_CAPTURE_ERROR) += i915_gpu_error.o
 i915-y += i915_vgpu.o
 
 # perf code
-i915-y += i915_perf.o
+i915-y += i915_perf.o \
+	  i915_oa_hsw.o
 
 ifeq ($(CONFIG_DRM_I915_GVT),y)
 i915-y += intel_gvt.o
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index e4cd322..bdebb66 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1785,6 +1785,11 @@ struct intel_wm_config {
 	bool sprites_scaled;
 };
 
+struct i915_oa_reg {
+	i915_reg_t addr;
+	u32 value;
+};
+
 struct i915_perf_stream;
 
 struct i915_perf_stream_ops {
@@ -2168,6 +2173,15 @@ 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 */
diff --git a/drivers/gpu/drm/i915/i915_oa_hsw.c b/drivers/gpu/drm/i915/i915_oa_hsw.c
new file mode 100644
index 0000000..8906380
--- /dev/null
+++ b/drivers/gpu/drm/i915/i915_oa_hsw.c
@@ -0,0 +1,144 @@
+/*
+ * 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.
+ *
+ */
+
+#include "i915_drv.h"
+#include "i915_oa_hsw.h"
+
+enum metric_set_id {
+	METRIC_SET_ID_RENDER_BASIC = 1,
+};
+
+int i915_oa_n_builtin_metric_sets_hsw = 1;
+
+static const struct i915_oa_reg b_counter_config_render_basic[] = {
+	{ _MMIO(0x2724), 0x00800000 },
+	{ _MMIO(0x2720), 0x00000000 },
+	{ _MMIO(0x2714), 0x00800000 },
+	{ _MMIO(0x2710), 0x00000000 },
+};
+
+static const struct i915_oa_reg mux_config_render_basic[] = {
+	{ _MMIO(0x253a4), 0x01600000 },
+	{ _MMIO(0x25440), 0x00100000 },
+	{ _MMIO(0x25128), 0x00000000 },
+	{ _MMIO(0x2691c), 0x00000800 },
+	{ _MMIO(0x26aa0), 0x01500000 },
+	{ _MMIO(0x26b9c), 0x00006000 },
+	{ _MMIO(0x2791c), 0x00000800 },
+	{ _MMIO(0x27aa0), 0x01500000 },
+	{ _MMIO(0x27b9c), 0x00006000 },
+	{ _MMIO(0x2641c), 0x00000400 },
+	{ _MMIO(0x25380), 0x00000010 },
+	{ _MMIO(0x2538c), 0x00000000 },
+	{ _MMIO(0x25384), 0x0800aaaa },
+	{ _MMIO(0x25400), 0x00000004 },
+	{ _MMIO(0x2540c), 0x06029000 },
+	{ _MMIO(0x25410), 0x00000002 },
+	{ _MMIO(0x25404), 0x5c30ffff },
+	{ _MMIO(0x25100), 0x00000016 },
+	{ _MMIO(0x25110), 0x00000400 },
+	{ _MMIO(0x25104), 0x00000000 },
+	{ _MMIO(0x26804), 0x00001211 },
+	{ _MMIO(0x26884), 0x00000100 },
+	{ _MMIO(0x26900), 0x00000002 },
+	{ _MMIO(0x26908), 0x00700000 },
+	{ _MMIO(0x26904), 0x00000000 },
+	{ _MMIO(0x26984), 0x00001022 },
+	{ _MMIO(0x26a04), 0x00000011 },
+	{ _MMIO(0x26a80), 0x00000006 },
+	{ _MMIO(0x26a88), 0x00000c02 },
+	{ _MMIO(0x26a84), 0x00000000 },
+	{ _MMIO(0x26b04), 0x00001000 },
+	{ _MMIO(0x26b80), 0x00000002 },
+	{ _MMIO(0x26b8c), 0x00000007 },
+	{ _MMIO(0x26b84), 0x00000000 },
+	{ _MMIO(0x27804), 0x00004844 },
+	{ _MMIO(0x27884), 0x00000400 },
+	{ _MMIO(0x27900), 0x00000002 },
+	{ _MMIO(0x27908), 0x0e000000 },
+	{ _MMIO(0x27904), 0x00000000 },
+	{ _MMIO(0x27984), 0x00004088 },
+	{ _MMIO(0x27a04), 0x00000044 },
+	{ _MMIO(0x27a80), 0x00000006 },
+	{ _MMIO(0x27a88), 0x00018040 },
+	{ _MMIO(0x27a84), 0x00000000 },
+	{ _MMIO(0x27b04), 0x00004000 },
+	{ _MMIO(0x27b80), 0x00000002 },
+	{ _MMIO(0x27b8c), 0x000000e0 },
+	{ _MMIO(0x27b84), 0x00000000 },
+	{ _MMIO(0x26104), 0x00002222 },
+	{ _MMIO(0x26184), 0x0c006666 },
+	{ _MMIO(0x26284), 0x04000000 },
+	{ _MMIO(0x26304), 0x04000000 },
+	{ _MMIO(0x26400), 0x00000002 },
+	{ _MMIO(0x26410), 0x000000a0 },
+	{ _MMIO(0x26404), 0x00000000 },
+	{ _MMIO(0x25420), 0x04108020 },
+	{ _MMIO(0x25424), 0x1284a420 },
+	{ _MMIO(0x2541c), 0x00000000 },
+	{ _MMIO(0x25428), 0x00042049 },
+};
+
+static const struct i915_oa_reg *
+get_render_basic_mux_config(struct drm_i915_private *dev_priv,
+			    int *len)
+{
+	*len = ARRAY_SIZE(mux_config_render_basic);
+	return mux_config_render_basic;
+}
+
+int i915_oa_select_metric_set_hsw(struct drm_i915_private *dev_priv)
+{
+	dev_priv->perf.oa.mux_regs = NULL;
+	dev_priv->perf.oa.mux_regs_len = 0;
+	dev_priv->perf.oa.b_counter_regs = NULL;
+	dev_priv->perf.oa.b_counter_regs_len = 0;
+
+	switch (dev_priv->perf.oa.metrics_set) {
+	case METRIC_SET_ID_RENDER_BASIC:
+		dev_priv->perf.oa.mux_regs =
+			get_render_basic_mux_config(dev_priv,
+						    &dev_priv->perf.oa.mux_regs_len);
+		if (!dev_priv->perf.oa.mux_regs) {
+			DRM_DEBUG_DRIVER("No suitable MUX config for \"RENDER_BASIC\" metric set");
+
+			/* EINVAL because *_register_sysfs already checked this
+			 * and so it wouldn't have been advertised so userspace and
+			 * so shouldn't have been requested
+			 */
+			return -EINVAL;
+		}
+
+		dev_priv->perf.oa.b_counter_regs =
+			b_counter_config_render_basic;
+		dev_priv->perf.oa.b_counter_regs_len =
+			ARRAY_SIZE(b_counter_config_render_basic);
+
+		return 0;
+	default:
+		return -ENODEV;
+	}
+}
diff --git a/drivers/gpu/drm/i915/i915_oa_hsw.h b/drivers/gpu/drm/i915/i915_oa_hsw.h
new file mode 100644
index 0000000..b618a1f
--- /dev/null
+++ b/drivers/gpu/drm/i915/i915_oa_hsw.h
@@ -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
-- 
2.10.1

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v9 06/11] drm/i915: Enable i915 perf stream for Haswell OA unit

[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.

v2:
   Make sure to initialize ->specific_ctx_id when opening, without
   relying on _pin_notify hook, in case ctx already pinned.
v3:
   Revert back to pinning ctx upfront when opening stream, removing
   need to hook in to pinning and to update OACONTROL on the fly.

Signed-off-by: Robert Bragg <robert@sixbynine.org>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
---
 drivers/gpu/drm/i915/i915_drv.h  |   66 ++-
 drivers/gpu/drm/i915/i915_perf.c | 1036 +++++++++++++++++++++++++++++++++++++-
 drivers/gpu/drm/i915/i915_reg.h  |  338 +++++++++++++
 include/uapi/drm/i915_drm.h      |   71 ++-
 4 files changed, 1482 insertions(+), 29 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index bdebb66..8003120 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1785,6 +1785,11 @@ struct intel_wm_config {
 	bool sprites_scaled;
 };
 
+struct i915_oa_format {
+	u32 format;
+	int size;
+};
+
 struct i915_oa_reg {
 	i915_reg_t addr;
 	u32 value;
@@ -1805,11 +1810,6 @@ struct i915_perf_stream_ops {
 	 */
 	void (*disable)(struct i915_perf_stream *stream);
 
-	/* Return: true if any i915 perf records are ready to read()
-	 * for this stream.
-	 */
-	bool (*can_read)(struct i915_perf_stream *stream);
-
 	/* Call poll_wait, passing a wait queue that will be woken
 	 * once there is something ready to read() for the stream
 	 */
@@ -1819,9 +1819,7 @@ struct i915_perf_stream_ops {
 
 	/* For handling a blocking read, wait until there is something
 	 * to ready to read() for the stream. E.g. wait on the same
-	 * wait queue that would be passed to poll_wait() until
-	 * ->can_read() returns true (if its safe to call ->can_read()
-	 * without the i915 perf lock held).
+	 * wait queue that would be passed to poll_wait().
 	 */
 	int (*wait_unlocked)(struct i915_perf_stream *stream);
 
@@ -1861,11 +1859,28 @@ struct i915_perf_stream {
 	struct list_head link;
 
 	u32 sample_flags;
+	int sample_size;
 
 	struct i915_gem_context *ctx;
 	bool enabled;
 
-	struct i915_perf_stream_ops *ops;
+	const struct i915_perf_stream_ops *ops;
+};
+
+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,
+		    char __user *buf,
+		    size_t count,
+		    size_t *offset);
+	bool (*oa_buffer_is_empty)(struct drm_i915_private *dev_priv);
 };
 
 struct drm_i915_private {
@@ -2171,16 +2186,47 @@ struct drm_i915_private {
 
 	struct {
 		bool initialized;
+
 		struct mutex lock;
 		struct list_head streams;
 
+		spinlock_t hook_lock;
+
 		struct {
-			u32 metrics_set;
+			struct i915_perf_stream *exclusive_stream;
+
+			u32 specific_ctx_id;
+			struct i915_vma *pinned_rcs_vma;
+
+			struct hrtimer poll_check_timer;
+			wait_queue_head_t poll_wq;
+			bool pollin;
+
+			bool periodic;
+			int period_exponent;
+			int timestamp_frequency;
+
+			int tail_margin;
+
+			int 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 i915_vma *vma;
+				u8 *vaddr;
+				int format;
+				int format_size;
+			} oa_buffer;
+
+			u32 gen7_latched_oastatus1;
+
+			struct i915_oa_ops ops;
+			const struct i915_oa_format *oa_formats;
+			int n_builtin_sets;
 		} oa;
 	} perf;
 
diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c
index 777ce65..54653bc 100644
--- a/drivers/gpu/drm/i915/i915_perf.c
+++ b/drivers/gpu/drm/i915/i915_perf.c
@@ -25,16 +25,899 @@
  */
 
 #include <linux/anon_inodes.h>
+#include <linux/sizes.h>
 
 #include "i915_drv.h"
+#include "i915_oa_hsw.h"
+
+/* HW requires this to be a power of two, between 128k and 16M, though driver
+ * is currently generally designed assuming the largest 16M size is used such
+ * that the overflow cases are unlikely in normal operation.
+ */
+#define OA_BUFFER_SIZE		SZ_16M
+
+#define OA_TAKEN(tail, head)	((tail - head) & (OA_BUFFER_SIZE - 1))
+
+/* There's a HW race condition between OA unit tail pointer register updates and
+ * writes to memory whereby the tail pointer can sometimes get ahead of what's
+ * been written out to the OA buffer so far.
+ *
+ * Although this can be observed explicitly by checking for a zeroed report-id
+ * field in tail reports, it seems preferable to account for this earlier e.g.
+ * as part of the _oa_buffer_is_empty checks to minimize -EAGAIN polling cycles
+ * in this situation.
+ *
+ * To give time for the most recent reports to land before they may be copied to
+ * userspace, the driver operates as if the tail pointer effectively lags behind
+ * the HW tail pointer by 'tail_margin' bytes. The margin in bytes is calculated
+ * based on this constant in nanoseconds, the current OA sampling exponent
+ * and current report size.
+ *
+ * There is also a fallback check while reading to simply skip over reports with
+ * a zeroed report-id.
+ */
+#define OA_TAIL_MARGIN_NSEC	100000ULL
+
+/* frequency for checking whether the OA unit has written new reports to the
+ * circular OA buffer...
+ */
+#define POLL_FREQUENCY 200
+#define POLL_PERIOD (NSEC_PER_SEC / POLL_FREQUENCY)
+
+/* The maximum exponent the hardware accepts is 63 (essentially it selects one
+ * of the 64bit timestamp bits to trigger reports from) but there's currently
+ * no known use case for sampling as infrequently as once per 47 thousand years.
+ *
+ * Since the timestamps included in OA reports are only 32bits it seems
+ * reasonable to limit the OA exponent where it's still possible to account for
+ * overflow in OA report timestamps.
+ */
+#define OA_EXPONENT_MAX 31
+
+#define INVALID_CTX_ID 0xffffffff
+
+
+/* XXX: beware if future OA HW adds new report formats that the current
+ * code assumes all reports have a power-of-two size and ~(size - 1) can
+ * be used as a mask to align the OA tail pointer.
+ */
+static struct i915_oa_format hsw_oa_formats[I915_OA_FORMAT_MAX] = {
+	[I915_OA_FORMAT_A13]	    = { 0, 64 },
+	[I915_OA_FORMAT_A29]	    = { 1, 128 },
+	[I915_OA_FORMAT_A13_B8_C8]  = { 2, 128 },
+	/* A29_B8_C8 Disallowed as 192 bytes doesn't factor into buffer size */
+	[I915_OA_FORMAT_B4_C8]	    = { 4, 64 },
+	[I915_OA_FORMAT_A45_B8_C8]  = { 5, 256 },
+	[I915_OA_FORMAT_B4_C8_A16]  = { 6, 128 },
+	[I915_OA_FORMAT_C4_B8]	    = { 7, 64 },
+};
+
+#define SAMPLE_OA_REPORT      (1<<0)
 
 struct perf_open_properties {
 	u32 sample_flags;
 
 	u64 single_context:1;
 	u64 ctx_handle;
+
+	/* OA sampling state */
+	int metrics_set;
+	int oa_format;
+	bool oa_periodic;
+	int oa_period_exponent;
+};
+
+/* NB: This is either called via fops or the poll check hrtimer (atomic ctx)
+ *
+ * It's safe to read OA config state here unlocked, assuming that this is only
+ * called while the stream is enabled, while the global OA configuration can't
+ * be modified.
+ *
+ * Note: we don't lock around the head/tail reads even though there's the slim
+ * possibility of read() fop errors forcing a re-init of the OA buffer
+ * pointers.  A race here could result in a false positive !empty status which
+ * is acceptable.
+ */
+static bool gen7_oa_buffer_is_empty_fop_unlocked(struct drm_i915_private *dev_priv)
+{
+	int report_size = dev_priv->perf.oa.oa_buffer.format_size;
+	u32 oastatus2 = I915_READ(GEN7_OASTATUS2);
+	u32 oastatus1 = I915_READ(GEN7_OASTATUS1);
+	u32 head = oastatus2 & GEN7_OASTATUS2_HEAD_MASK;
+	u32 tail = oastatus1 & GEN7_OASTATUS1_TAIL_MASK;
+
+	return OA_TAKEN(tail, head) <
+		dev_priv->perf.oa.tail_margin + report_size;
+}
+
+/**
+ * Appends a status record to a userspace read() buffer.
+ */
+static int append_oa_status(struct i915_perf_stream *stream,
+			    char __user *buf,
+			    size_t count,
+			    size_t *offset,
+			    enum drm_i915_perf_record_type type)
+{
+	struct drm_i915_perf_record_header header = { type, 0, sizeof(header) };
+
+	if ((count - *offset) < header.size)
+		return -ENOSPC;
+
+	if (copy_to_user(buf + *offset, &header, sizeof(header)))
+		return -EFAULT;
+
+	(*offset) += header.size;
+
+	return 0;
+}
+
+/**
+ * Copies single OA report into userspace read() buffer.
+ */
+static int append_oa_sample(struct i915_perf_stream *stream,
+			    char __user *buf,
+			    size_t count,
+			    size_t *offset,
+			    const u8 *report)
+{
+	struct drm_i915_private *dev_priv = stream->dev_priv;
+	int report_size = dev_priv->perf.oa.oa_buffer.format_size;
+	struct drm_i915_perf_record_header header;
+	u32 sample_flags = stream->sample_flags;
+
+	header.type = DRM_I915_PERF_RECORD_SAMPLE;
+	header.pad = 0;
+	header.size = stream->sample_size;
+
+	if ((count - *offset) < header.size)
+		return -ENOSPC;
+
+	buf += *offset;
+	if (copy_to_user(buf, &header, sizeof(header)))
+		return -EFAULT;
+	buf += sizeof(header);
+
+	if (sample_flags & SAMPLE_OA_REPORT) {
+		if (copy_to_user(buf, report, report_size))
+			return -EFAULT;
+	}
+
+	(*offset) += header.size;
+
+	return 0;
+}
+
+/**
+ * Copies all buffered OA reports into userspace read() buffer.
+ * @stream: An i915-perf stream opened for OA metrics
+ * @buf: destination buffer given by userspace
+ * @count: the number of bytes userspace wants to read
+ * @offset: (inout): the current position for writing into @buf
+ * @head_ptr: (inout): the current oa buffer cpu read position
+ * @tail: the current oa buffer gpu write position
+ *
+ * Returns 0 on success, negative error code on failure.
+ *
+ * Notably any error condition resulting in a short read (-ENOSPC or
+ * -EFAULT) will be returned even though one or more records may
+ * have been successfully copied. In this case it's up to the caller
+ * to decide if the error should be squashed before returning to
+ * userspace.
+ *
+ * Note: reports are consumed from the head, and appended to the
+ * tail, so the head chases the tail?... If you think that's mad
+ * and back-to-front you're not alone, but this follows the
+ * Gen PRM naming convention.
+ */
+static int gen7_append_oa_reports(struct i915_perf_stream *stream,
+				  char __user *buf,
+				  size_t count,
+				  size_t *offset,
+				  u32 *head_ptr,
+				  u32 tail)
+{
+	struct drm_i915_private *dev_priv = stream->dev_priv;
+	int report_size = dev_priv->perf.oa.oa_buffer.format_size;
+	u8 *oa_buf_base = dev_priv->perf.oa.oa_buffer.vaddr;
+	int tail_margin = dev_priv->perf.oa.tail_margin;
+	u32 gtt_offset = i915_ggtt_offset(dev_priv->perf.oa.oa_buffer.vma);
+	u32 mask = (OA_BUFFER_SIZE - 1);
+	u32 head;
+	u32 taken;
+	int ret = 0;
+
+	if (WARN_ON(!stream->enabled))
+		return -EIO;
+
+	head = *head_ptr - gtt_offset;
+	tail -= gtt_offset;
+
+	/* The OA unit is expected to wrap the tail pointer according to the OA
+	 * buffer size and since we should never write a misaligned head
+	 * pointer we don't expect to read one back either...
+	 */
+	if (tail > OA_BUFFER_SIZE || head > OA_BUFFER_SIZE ||
+	    head % report_size) {
+		DRM_ERROR("Inconsistent OA buffer pointer (head = %u, tail = %u): force restart\n",
+			  head, tail);
+		dev_priv->perf.oa.ops.oa_disable(dev_priv);
+		dev_priv->perf.oa.ops.oa_enable(dev_priv);
+		*head_ptr = I915_READ(GEN7_OASTATUS2) &
+			GEN7_OASTATUS2_HEAD_MASK;
+		return -EIO;
+	}
+
+
+	/* The tail pointer increases in 64 byte increments, not in report_size
+	 * steps...
+	 */
+	tail &= ~(report_size - 1);
+
+	/* Move the tail pointer back by the current tail_margin to account for
+	 * the possibility that the latest reports may not have really landed
+	 * in memory yet...
+	 */
+
+	if (OA_TAKEN(tail, head) < report_size + tail_margin)
+		return -EAGAIN;
+
+	tail -= tail_margin;
+	tail &= mask;
+
+	for (/* none */;
+	     (taken = OA_TAKEN(tail, head));
+	     head = (head + report_size) & mask) {
+		u8 *report = oa_buf_base + head;
+		u32 *report32 = (void *)report;
+
+		/* All the report sizes factor neatly into the buffer
+		 * size so we never expect to see a report split
+		 * between the beginning and end of the buffer.
+		 *
+		 * Given the initial alignment check a misalignment
+		 * here would imply a driver bug that would result
+		 * in an overrun.
+		 */
+		if (WARN_ON((OA_BUFFER_SIZE - head) < report_size)) {
+			DRM_ERROR("Spurious OA head ptr: non-integral report offset\n");
+			break;
+		}
+
+		/* The report-ID field for periodic samples includes
+		 * some undocumented flags related to what triggered
+		 * the report and is never expected to be zero so we
+		 * can check that the report isn't invalid before
+		 * copying it to userspace...
+		 */
+		if (report32[0] == 0) {
+			DRM_ERROR("Skipping spurious, invalid OA report\n");
+			continue;
+		}
+
+		ret = append_oa_sample(stream, buf, count, offset, report);
+		if (ret)
+			break;
+
+		/* The above report-id field sanity check is based on
+		 * the assumption that the OA buffer is initially
+		 * zeroed and we reset the field after copying so the
+		 * check is still meaningful once old reports start
+		 * being overwritten.
+		 */
+		report32[0] = 0;
+	}
+
+	*head_ptr = gtt_offset + head;
+
+	return ret;
+}
+
+static int gen7_oa_read(struct i915_perf_stream *stream,
+			char __user *buf,
+			size_t count,
+			size_t *offset)
+{
+	struct drm_i915_private *dev_priv = stream->dev_priv;
+	int report_size = dev_priv->perf.oa.oa_buffer.format_size;
+	u32 oastatus2;
+	u32 oastatus1;
+	u32 head;
+	u32 tail;
+	int ret;
+
+	if (WARN_ON(!dev_priv->perf.oa.oa_buffer.vaddr))
+		return -EIO;
+
+	oastatus2 = I915_READ(GEN7_OASTATUS2);
+	oastatus1 = I915_READ(GEN7_OASTATUS1);
+
+	head = oastatus2 & GEN7_OASTATUS2_HEAD_MASK;
+	tail = oastatus1 & GEN7_OASTATUS1_TAIL_MASK;
+
+	/* XXX: On Haswell we don't have a safe way to clear oastatus1
+	 * bits while the OA unit is enabled (while the tail pointer
+	 * may be updated asynchronously) so we ignore status bits
+	 * that have already been reported to userspace.
+	 */
+	oastatus1 &= ~dev_priv->perf.oa.gen7_latched_oastatus1;
+
+	/* We treat OABUFFER_OVERFLOW as a significant error:
+	 *
+	 * - The status can be interpreted to mean that the buffer is
+	 *   currently full (with a higher precedence than OA_TAKEN()
+	 *   which will start to report a near-empty buffer after an
+	 *   overflow) but it's awkward that we can't clear the status
+	 *   on Haswell, so without a reset we won't be able to catch
+	 *   the state again.
+	 *
+	 * - Since it also implies the HW has started overwriting old
+	 *   reports it may also affect our sanity checks for invalid
+	 *   reports when copying to userspace that assume new reports
+	 *   are being written to cleared memory.
+	 *
+	 * - In the future we may want to introduce a flight recorder
+	 *   mode where the driver will automatically maintain a safe
+	 *   guard band between head/tail, avoiding this overflow
+	 *   condition, but we avoid the added driver complexity for
+	 *   now.
+	 */
+	if (unlikely(oastatus1 & GEN7_OASTATUS1_OABUFFER_OVERFLOW)) {
+		ret = append_oa_status(stream, buf, count, offset,
+				       DRM_I915_PERF_RECORD_OA_BUFFER_LOST);
+		if (ret)
+			return ret;
+
+		DRM_ERROR("OA buffer overflow: force restart\n");
+
+		dev_priv->perf.oa.ops.oa_disable(dev_priv);
+		dev_priv->perf.oa.ops.oa_enable(dev_priv);
+
+		oastatus2 = I915_READ(GEN7_OASTATUS2);
+		oastatus1 = I915_READ(GEN7_OASTATUS1);
+
+		head = oastatus2 & GEN7_OASTATUS2_HEAD_MASK;
+		tail = oastatus1 & GEN7_OASTATUS1_TAIL_MASK;
+	}
+
+	if (unlikely(oastatus1 & GEN7_OASTATUS1_REPORT_LOST)) {
+		ret = append_oa_status(stream, buf, count, offset,
+				       DRM_I915_PERF_RECORD_OA_REPORT_LOST);
+		if (ret)
+			return ret;
+		dev_priv->perf.oa.gen7_latched_oastatus1 |=
+			GEN7_OASTATUS1_REPORT_LOST;
+	}
+
+	ret = gen7_append_oa_reports(stream, buf, count, offset,
+				     &head, tail);
+
+	/* All the report sizes are a power of two and the
+	 * head should always be incremented by some multiple
+	 * of the report size.
+	 *
+	 * A warning here, but notably if we later read back a
+	 * misaligned pointer we will treat that as a bug since
+	 * it could lead to a buffer overrun.
+	 */
+	WARN_ONCE(head & (report_size - 1),
+		  "i915: Writing misaligned OA head pointer");
+
+	/* Note: we update the head pointer here even if an error
+	 * was returned since the error may represent a short read
+	 * where some some reports were successfully copied.
+	 */
+	I915_WRITE(GEN7_OASTATUS2,
+		   ((head & GEN7_OASTATUS2_HEAD_MASK) |
+		    OA_MEM_SELECT_GGTT));
+
+	return ret;
+}
+
+static int i915_oa_wait_unlocked(struct i915_perf_stream *stream)
+{
+	struct drm_i915_private *dev_priv = stream->dev_priv;
+
+	/* We would wait indefinitely if periodic sampling is not enabled */
+	if (!dev_priv->perf.oa.periodic)
+		return -EIO;
+
+	/* Note: the oa_buffer_is_empty() condition is ok to run unlocked as it
+	 * just performs mmio reads of the OA buffer head + tail pointers and
+	 * it's assumed we're handling some operation that implies the stream
+	 * can't be destroyed until completion (such as a read()) that ensures
+	 * the device + OA buffer can't disappear
+	 */
+	return wait_event_interruptible(dev_priv->perf.oa.poll_wq,
+					!dev_priv->perf.oa.ops.oa_buffer_is_empty(dev_priv));
+}
+
+static void i915_oa_poll_wait(struct i915_perf_stream *stream,
+			      struct file *file,
+			      poll_table *wait)
+{
+	struct drm_i915_private *dev_priv = stream->dev_priv;
+
+	poll_wait(file, &dev_priv->perf.oa.poll_wq, wait);
+}
+
+static int i915_oa_read(struct i915_perf_stream *stream,
+			char __user *buf,
+			size_t count,
+			size_t *offset)
+{
+	struct drm_i915_private *dev_priv = stream->dev_priv;
+
+	return dev_priv->perf.oa.ops.read(stream, buf, count, offset);
+}
+
+/* Determine the render context hw id, and ensure it remains fixed for the
+ * lifetime of the stream. This ensures that we don't have to worry about
+ * updating the context ID in OACONTROL on the fly.
+ */
+static int oa_get_render_ctx_id(struct i915_perf_stream *stream)
+{
+	struct drm_i915_private *dev_priv = stream->dev_priv;
+	struct i915_vma *vma;
+	int ret;
+
+	ret = i915_mutex_lock_interruptible(&dev_priv->drm);
+	if (ret)
+		return ret;
+
+	/* As the ID is the gtt offset of the context's vma we pin
+	 * the vma to ensure the ID remains fixed.
+	 *
+	 * NB: implied RCS engine...
+	 */
+	vma = i915_gem_context_pin_legacy(stream->ctx, 0);
+	if (IS_ERR(vma)) {
+		ret = PTR_ERR(vma);
+		goto unlock;
+	}
+
+	dev_priv->perf.oa.pinned_rcs_vma = vma;
+
+	/* Explicitly track the ID (instead of calling i915_ggtt_offset()
+	 * on the fly) considering the difference with gen8+ and
+	 * execlists
+	 */
+	dev_priv->perf.oa.specific_ctx_id = i915_ggtt_offset(vma);
+
+unlock:
+	mutex_unlock(&dev_priv->drm.struct_mutex);
+
+	return ret;
+}
+
+static void oa_put_render_ctx_id(struct i915_perf_stream *stream)
+{
+	struct drm_i915_private *dev_priv = stream->dev_priv;
+
+	mutex_lock(&dev_priv->drm.struct_mutex);
+
+	i915_vma_unpin(dev_priv->perf.oa.pinned_rcs_vma);
+	dev_priv->perf.oa.pinned_rcs_vma = NULL;
+
+	dev_priv->perf.oa.specific_ctx_id = INVALID_CTX_ID;
+
+	mutex_unlock(&dev_priv->drm.struct_mutex);
+}
+
+static void
+free_oa_buffer(struct drm_i915_private *i915)
+{
+	mutex_lock(&i915->drm.struct_mutex);
+
+	i915_gem_object_unpin_map(i915->perf.oa.oa_buffer.vma->obj);
+	i915_vma_unpin(i915->perf.oa.oa_buffer.vma);
+	i915_gem_object_put(i915->perf.oa.oa_buffer.vma->obj);
+
+	i915->perf.oa.oa_buffer.vma = NULL;
+	i915->perf.oa.oa_buffer.vaddr = NULL;
+
+	mutex_unlock(&i915->drm.struct_mutex);
+}
+
+static void i915_oa_stream_destroy(struct i915_perf_stream *stream)
+{
+	struct drm_i915_private *dev_priv = stream->dev_priv;
+
+	BUG_ON(stream != dev_priv->perf.oa.exclusive_stream);
+
+	dev_priv->perf.oa.ops.disable_metric_set(dev_priv);
+
+	free_oa_buffer(dev_priv);
+
+	intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
+	intel_runtime_pm_put(dev_priv);
+
+	if (stream->ctx)
+		oa_put_render_ctx_id(stream);
+
+	dev_priv->perf.oa.exclusive_stream = NULL;
+}
+
+static void gen7_init_oa_buffer(struct drm_i915_private *dev_priv)
+{
+	u32 gtt_offset = i915_ggtt_offset(dev_priv->perf.oa.oa_buffer.vma);
+
+	/* Pre-DevBDW: OABUFFER must be set with counters off,
+	 * before OASTATUS1, but after OASTATUS2
+	 */
+	I915_WRITE(GEN7_OASTATUS2, gtt_offset | OA_MEM_SELECT_GGTT); /* head */
+	I915_WRITE(GEN7_OABUFFER, gtt_offset);
+	I915_WRITE(GEN7_OASTATUS1, gtt_offset | OABUFFER_SIZE_16M); /* tail */
+
+	/* On Haswell we have to track which OASTATUS1 flags we've
+	 * already seen since they can't be cleared while periodic
+	 * sampling is enabled.
+	 */
+	dev_priv->perf.oa.gen7_latched_oastatus1 = 0;
+
+	/* NB: although the OA buffer will initially be allocated
+	 * zeroed via shmfs (and so this memset is redundant when
+	 * first allocating), we may re-init the OA buffer, either
+	 * when re-enabling a stream or in error/reset paths.
+	 *
+	 * The reason we clear the buffer for each re-init is for the
+	 * sanity check in gen7_append_oa_reports() that looks at the
+	 * report-id field to make sure it's non-zero which relies on
+	 * the assumption that new reports are being written to zeroed
+	 * memory...
+	 */
+	memset(dev_priv->perf.oa.oa_buffer.vaddr, 0, OA_BUFFER_SIZE);
+
+	/* Maybe make ->pollin per-stream state if we support multiple
+	 * concurrent streams in the future.
+	 */
+	dev_priv->perf.oa.pollin = false;
+}
+
+static int alloc_oa_buffer(struct drm_i915_private *dev_priv)
+{
+	struct drm_i915_gem_object *bo;
+	struct i915_vma *vma;
+	int ret;
+
+	if (WARN_ON(dev_priv->perf.oa.oa_buffer.vma))
+		return -ENODEV;
+
+	ret = i915_mutex_lock_interruptible(&dev_priv->drm);
+	if (ret)
+		return ret;
+
+	BUILD_BUG_ON_NOT_POWER_OF_2(OA_BUFFER_SIZE);
+	BUILD_BUG_ON(OA_BUFFER_SIZE < SZ_128K || OA_BUFFER_SIZE > SZ_16M);
+
+	bo = i915_gem_object_create(&dev_priv->drm, OA_BUFFER_SIZE);
+	if (IS_ERR(bo)) {
+		DRM_ERROR("Failed to allocate OA buffer\n");
+		ret = PTR_ERR(bo);
+		goto unlock;
+	}
+
+	ret = i915_gem_object_set_cache_level(bo, I915_CACHE_LLC);
+	if (ret)
+		goto err_unref;
+
+	/* PreHSW required 512K alignment, HSW requires 16M */
+	vma = i915_gem_object_ggtt_pin(bo, NULL, 0, SZ_16M, 0);
+	if (IS_ERR(vma)) {
+		ret = PTR_ERR(vma);
+		goto err_unref;
+	}
+	dev_priv->perf.oa.oa_buffer.vma = vma;
+
+	dev_priv->perf.oa.oa_buffer.vaddr =
+		i915_gem_object_pin_map(bo, I915_MAP_WB);
+	if (IS_ERR(dev_priv->perf.oa.oa_buffer.vaddr)) {
+		ret = PTR_ERR(dev_priv->perf.oa.oa_buffer.vaddr);
+		goto err_unpin;
+	}
+
+	dev_priv->perf.oa.ops.init_oa_buffer(dev_priv);
+
+	DRM_DEBUG_DRIVER("OA Buffer initialized, gtt offset = 0x%x, vaddr = %p\n",
+			 i915_ggtt_offset(dev_priv->perf.oa.oa_buffer.vma),
+			 dev_priv->perf.oa.oa_buffer.vaddr);
+
+	goto unlock;
+
+err_unpin:
+	__i915_vma_unpin(vma);
+
+err_unref:
+	i915_gem_object_put(bo);
+
+	dev_priv->perf.oa.oa_buffer.vaddr = NULL;
+	dev_priv->perf.oa.oa_buffer.vma = NULL;
+
+unlock:
+	mutex_unlock(&dev_priv->drm.struct_mutex);
+	return ret;
+}
+
+static void config_oa_regs(struct drm_i915_private *dev_priv,
+			   const struct i915_oa_reg *regs,
+			   int n_regs)
+{
+	int i;
+
+	for (i = 0; i < n_regs; i++) {
+		const struct i915_oa_reg *reg = regs + i;
+
+		I915_WRITE(reg->addr, reg->value);
+	}
+}
+
+static int hsw_enable_metric_set(struct drm_i915_private *dev_priv)
+{
+	int ret = i915_oa_select_metric_set_hsw(dev_priv);
+
+	if (ret)
+		return ret;
+
+	I915_WRITE(GDT_CHICKEN_BITS, (I915_READ(GDT_CHICKEN_BITS) |
+				      GT_NOA_ENABLE));
+
+	/* PRM:
+	 *
+	 * OA unit is using “crclk” for its functionality. When trunk
+	 * level clock gating takes place, OA clock would be gated,
+	 * unable to count the events from non-render clock domain.
+	 * Render clock gating must be disabled when OA is enabled to
+	 * count the events from non-render domain. Unit level clock
+	 * gating for RCS should also be disabled.
+	 */
+	I915_WRITE(GEN7_MISCCPCTL, (I915_READ(GEN7_MISCCPCTL) &
+				    ~GEN7_DOP_CLOCK_GATE_ENABLE));
+	I915_WRITE(GEN6_UCGCTL1, (I915_READ(GEN6_UCGCTL1) |
+				  GEN6_CSUNIT_CLOCK_GATE_DISABLE));
+
+	config_oa_regs(dev_priv, dev_priv->perf.oa.mux_regs,
+		       dev_priv->perf.oa.mux_regs_len);
+
+	/* It apparently takes a fairly long time for a new MUX
+	 * configuration to be be applied after these register writes.
+	 * This delay duration was derived empirically based on the
+	 * render_basic config but hopefully it covers the maximum
+	 * configuration latency.
+	 *
+	 * As a fallback, the checks in _append_oa_reports() to skip
+	 * invalid OA reports do also seem to work to discard reports
+	 * generated before this config has completed - albeit not
+	 * silently.
+	 *
+	 * Unfortunately this is essentially a magic number, since we
+	 * don't currently know of a reliable mechanism for predicting
+	 * how long the MUX config will take to apply and besides
+	 * seeing invalid reports we don't know of a reliable way to
+	 * explicitly check that the MUX config has landed.
+	 *
+	 * It's even possible we've miss characterized the underlying
+	 * problem - it just seems like the simplest explanation why
+	 * a delay at this location would mitigate any invalid reports.
+	 */
+	usleep_range(15000, 20000);
+
+	config_oa_regs(dev_priv, dev_priv->perf.oa.b_counter_regs,
+		       dev_priv->perf.oa.b_counter_regs_len);
+
+	return 0;
+}
+
+static void hsw_disable_metric_set(struct drm_i915_private *dev_priv)
+{
+	I915_WRITE(GEN6_UCGCTL1, (I915_READ(GEN6_UCGCTL1) &
+				  ~GEN6_CSUNIT_CLOCK_GATE_DISABLE));
+	I915_WRITE(GEN7_MISCCPCTL, (I915_READ(GEN7_MISCCPCTL) |
+				    GEN7_DOP_CLOCK_GATE_ENABLE));
+
+	I915_WRITE(GDT_CHICKEN_BITS, (I915_READ(GDT_CHICKEN_BITS) &
+				      ~GT_NOA_ENABLE));
+}
+
+static void gen7_update_oacontrol_locked(struct drm_i915_private *dev_priv)
+{
+	assert_spin_locked(&dev_priv->perf.hook_lock);
+
+	if (dev_priv->perf.oa.exclusive_stream->enabled) {
+		struct i915_gem_context *ctx =
+			dev_priv->perf.oa.exclusive_stream->ctx;
+		u32 ctx_id = dev_priv->perf.oa.specific_ctx_id;
+
+		bool periodic = dev_priv->perf.oa.periodic;
+		u32 period_exponent = dev_priv->perf.oa.period_exponent;
+		u32 report_format = dev_priv->perf.oa.oa_buffer.format;
+
+		I915_WRITE(GEN7_OACONTROL,
+			   (ctx_id & GEN7_OACONTROL_CTX_MASK) |
+			   (period_exponent <<
+			    GEN7_OACONTROL_TIMER_PERIOD_SHIFT) |
+			   (periodic ? GEN7_OACONTROL_TIMER_ENABLE : 0) |
+			   (report_format << GEN7_OACONTROL_FORMAT_SHIFT) |
+			   (ctx ? GEN7_OACONTROL_PER_CTX_ENABLE : 0) |
+			   GEN7_OACONTROL_ENABLE);
+	} else
+		I915_WRITE(GEN7_OACONTROL, 0);
+}
+
+static void gen7_oa_enable(struct drm_i915_private *dev_priv)
+{
+	unsigned long flags;
+
+	/* Reset buf pointers so we don't forward reports from before now.
+	 *
+	 * Think carefully if considering trying to avoid this, since it
+	 * also ensures status flags and the buffer itself are cleared
+	 * in error paths, and we have checks for invalid reports based
+	 * on the assumption that certain fields are written to zeroed
+	 * memory which this helps maintains.
+	 */
+	gen7_init_oa_buffer(dev_priv);
+
+	spin_lock_irqsave(&dev_priv->perf.hook_lock, flags);
+	gen7_update_oacontrol_locked(dev_priv);
+	spin_unlock_irqrestore(&dev_priv->perf.hook_lock, flags);
+}
+
+static void i915_oa_stream_enable(struct i915_perf_stream *stream)
+{
+	struct drm_i915_private *dev_priv = stream->dev_priv;
+
+	dev_priv->perf.oa.ops.oa_enable(dev_priv);
+
+	if (dev_priv->perf.oa.periodic)
+		hrtimer_start(&dev_priv->perf.oa.poll_check_timer,
+			      ns_to_ktime(POLL_PERIOD),
+			      HRTIMER_MODE_REL_PINNED);
+}
+
+static void gen7_oa_disable(struct drm_i915_private *dev_priv)
+{
+	I915_WRITE(GEN7_OACONTROL, 0);
+}
+
+static void i915_oa_stream_disable(struct i915_perf_stream *stream)
+{
+	struct drm_i915_private *dev_priv = stream->dev_priv;
+
+	dev_priv->perf.oa.ops.oa_disable(dev_priv);
+
+	if (dev_priv->perf.oa.periodic)
+		hrtimer_cancel(&dev_priv->perf.oa.poll_check_timer);
+}
+
+static u64 oa_exponent_to_ns(struct drm_i915_private *dev_priv, int exponent)
+{
+	return 1000000000ULL * (2ULL << exponent) /
+		dev_priv->perf.oa.timestamp_frequency;
+}
+
+static const struct i915_perf_stream_ops i915_oa_stream_ops = {
+	.destroy = i915_oa_stream_destroy,
+	.enable = i915_oa_stream_enable,
+	.disable = i915_oa_stream_disable,
+	.wait_unlocked = i915_oa_wait_unlocked,
+	.poll_wait = i915_oa_poll_wait,
+	.read = i915_oa_read,
 };
 
+static int i915_oa_stream_init(struct i915_perf_stream *stream,
+			       struct drm_i915_perf_open_param *param,
+			       struct perf_open_properties *props)
+{
+	struct drm_i915_private *dev_priv = stream->dev_priv;
+	int format_size;
+	int ret;
+
+	if (!(props->sample_flags & SAMPLE_OA_REPORT)) {
+		DRM_ERROR("Only OA report sampling supported\n");
+		return -EINVAL;
+	}
+
+	if (!dev_priv->perf.oa.ops.init_oa_buffer) {
+		DRM_ERROR("OA unit not supported\n");
+		return -ENODEV;
+	}
+
+	/* To avoid the complexity of having to accurately filter
+	 * counter reports and marshal to the appropriate client
+	 * we currently only allow exclusive access
+	 */
+	if (dev_priv->perf.oa.exclusive_stream) {
+		DRM_ERROR("OA unit already in use\n");
+		return -EBUSY;
+	}
+
+	if (!props->metrics_set) {
+		DRM_ERROR("OA metric set not specified\n");
+		return -EINVAL;
+	}
+
+	if (!props->oa_format) {
+		DRM_ERROR("OA report format not specified\n");
+		return -EINVAL;
+	}
+
+	stream->sample_size = sizeof(struct drm_i915_perf_record_header);
+
+	format_size = dev_priv->perf.oa.oa_formats[props->oa_format].size;
+
+	stream->sample_flags |= SAMPLE_OA_REPORT;
+	stream->sample_size += format_size;
+
+	dev_priv->perf.oa.oa_buffer.format_size = format_size;
+	if (WARN_ON(dev_priv->perf.oa.oa_buffer.format_size == 0))
+		return -EINVAL;
+
+	dev_priv->perf.oa.oa_buffer.format =
+		dev_priv->perf.oa.oa_formats[props->oa_format].format;
+
+	dev_priv->perf.oa.metrics_set = props->metrics_set;
+
+	dev_priv->perf.oa.periodic = props->oa_periodic;
+	if (dev_priv->perf.oa.periodic) {
+		u64 period_ns = oa_exponent_to_ns(dev_priv,
+						  props->oa_period_exponent);
+
+		dev_priv->perf.oa.period_exponent = props->oa_period_exponent;
+
+		/* See comment for OA_TAIL_MARGIN_NSEC for details
+		 * about this tail_margin...
+		 */
+		dev_priv->perf.oa.tail_margin =
+			((OA_TAIL_MARGIN_NSEC / period_ns) + 1) * format_size;
+	}
+
+	if (stream->ctx) {
+		ret = oa_get_render_ctx_id(stream);
+		if (ret)
+			return ret;
+	}
+
+	ret = alloc_oa_buffer(dev_priv);
+	if (ret)
+		goto err_oa_buf_alloc;
+
+	/* PRM - observability performance counters:
+	 *
+	 *   OACONTROL, performance counter enable, note:
+	 *
+	 *   "When this bit is set, in order to have coherent counts,
+	 *   RC6 power state and trunk clock gating must be disabled.
+	 *   This can be achieved by programming MMIO registers as
+	 *   0xA094=0 and 0xA090[31]=1"
+	 *
+	 *   In our case we are expecting that taking pm + FORCEWAKE
+	 *   references will effectively disable RC6.
+	 */
+	intel_runtime_pm_get(dev_priv);
+	intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
+
+	ret = dev_priv->perf.oa.ops.enable_metric_set(dev_priv);
+	if (ret)
+		goto err_enable;
+
+	stream->ops = &i915_oa_stream_ops;
+
+	dev_priv->perf.oa.exclusive_stream = stream;
+
+	return 0;
+
+err_enable:
+	intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
+	intel_runtime_pm_put(dev_priv);
+	free_oa_buffer(dev_priv);
+
+err_oa_buf_alloc:
+	if (stream->ctx)
+		oa_put_render_ctx_id(stream);
+
+	return ret;
+}
+
 static ssize_t i915_perf_read_locked(struct i915_perf_stream *stream,
 				     struct file *file,
 				     char __user *buf,
@@ -78,8 +961,20 @@ static ssize_t i915_perf_read(struct file *file,
 	struct drm_i915_private *dev_priv = stream->dev_priv;
 	ssize_t ret;
 
+	/* To ensure it's handled consistently we simply treat all reads of a
+	 * disabled stream as an error. In particular it might otherwise lead
+	 * to a deadlock for blocking file descriptors...
+	 */
+	if (!stream->enabled)
+		return -EIO;
+
 	if (!(file->f_flags & O_NONBLOCK)) {
-		/* Allow false positives from stream->ops->wait_unlocked.
+		/* There's the small chance of false positives from
+		 * stream->ops->wait_unlocked.
+		 *
+		 * E.g. with single context filtering since we only wait until
+		 * oabuffer has >= 1 report we don't immediately know whether
+		 * any reports really belong to the current context
 		 */
 		do {
 			ret = stream->ops->wait_unlocked(stream);
@@ -97,21 +992,51 @@ static ssize_t i915_perf_read(struct file *file,
 		mutex_unlock(&dev_priv->perf.lock);
 	}
 
+	if (ret >= 0) {
+		/* Maybe make ->pollin per-stream state if we support multiple
+		 * concurrent streams in the future.
+		 */
+		dev_priv->perf.oa.pollin = false;
+	}
+
 	return ret;
 }
 
-static unsigned int i915_perf_poll_locked(struct i915_perf_stream *stream,
+static enum hrtimer_restart oa_poll_check_timer_cb(struct hrtimer *hrtimer)
+{
+	struct drm_i915_private *dev_priv =
+		container_of(hrtimer, typeof(*dev_priv),
+			     perf.oa.poll_check_timer);
+
+	if (!dev_priv->perf.oa.ops.oa_buffer_is_empty(dev_priv)) {
+		dev_priv->perf.oa.pollin = true;
+		wake_up(&dev_priv->perf.oa.poll_wq);
+	}
+
+	hrtimer_forward_now(hrtimer, ns_to_ktime(POLL_PERIOD));
+
+	return HRTIMER_RESTART;
+}
+
+static unsigned int i915_perf_poll_locked(struct drm_i915_private *dev_priv,
+					  struct i915_perf_stream *stream,
 					  struct file *file,
 					  poll_table *wait)
 {
-	unsigned int streams = 0;
+	unsigned int events = 0;
 
 	stream->ops->poll_wait(stream, file, wait);
 
-	if (stream->ops->can_read(stream))
-		streams |= POLLIN;
+	/* Note: we don't explicitly check whether there's something to read
+	 * here since this path may be very hot depending on what else
+	 * userspace is polling, or on the timeout in use. We rely solely on
+	 * the hrtimer/oa_poll_check_timer_cb to notify us when there are
+	 * samples to read.
+	 */
+	if (dev_priv->perf.oa.pollin)
+		events |= POLLIN;
 
-	return streams;
+	return events;
 }
 
 static unsigned int i915_perf_poll(struct file *file, poll_table *wait)
@@ -121,7 +1046,7 @@ static unsigned int i915_perf_poll(struct file *file, poll_table *wait)
 	int ret;
 
 	mutex_lock(&dev_priv->perf.lock);
-	ret = i915_perf_poll_locked(stream, file, wait);
+	ret = i915_perf_poll_locked(dev_priv, stream, file, wait);
 	mutex_unlock(&dev_priv->perf.lock);
 
 	return ret;
@@ -285,18 +1210,21 @@ i915_perf_open_ioctl_locked(struct drm_i915_private *dev_priv,
 		goto err_ctx;
 	}
 
-	stream->sample_flags = props->sample_flags;
 	stream->dev_priv = dev_priv;
 	stream->ctx = specific_ctx;
 
-	/*
-	 * TODO: support sampling something
-	 *
-	 * For now this is as far as we can go.
+	ret = i915_oa_stream_init(stream, param, props);
+	if (ret)
+		goto err_alloc;
+
+	/* we avoid simply assigning stream->sample_flags = props->sample_flags
+	 * to have _stream_init check the combination of sample flags more
+	 * thoroughly, but still this is the expected result at this point.
 	 */
-	DRM_ERROR("Unsupported i915 perf stream configuration\n");
-	ret = -EINVAL;
-	goto err_alloc;
+	if (WARN_ON(stream->sample_flags != props->sample_flags)) {
+		ret = -ENODEV;
+		goto err_alloc;
+	}
 
 	list_add(&stream->link, &dev_priv->perf.streams);
 
@@ -382,6 +1310,56 @@ static int read_properties_unlocked(struct drm_i915_private *dev_priv,
 			props->single_context = 1;
 			props->ctx_handle = value;
 			break;
+		case DRM_I915_PERF_PROP_SAMPLE_OA:
+			props->sample_flags |= SAMPLE_OA_REPORT;
+			break;
+		case DRM_I915_PERF_PROP_OA_METRICS_SET:
+			if (value == 0 ||
+			    value > dev_priv->perf.oa.n_builtin_sets) {
+				DRM_ERROR("Unknown OA metric set ID");
+				return -EINVAL;
+			}
+			props->metrics_set = value;
+			break;
+		case DRM_I915_PERF_PROP_OA_FORMAT:
+			if (value == 0 || value >= I915_OA_FORMAT_MAX) {
+				DRM_ERROR("Invalid OA report format\n");
+				return -EINVAL;
+			}
+			if (!dev_priv->perf.oa.oa_formats[value].size) {
+				DRM_ERROR("Invalid OA report format\n");
+				return -EINVAL;
+			}
+			props->oa_format = value;
+			break;
+		case DRM_I915_PERF_PROP_OA_EXPONENT:
+			if (value > OA_EXPONENT_MAX) {
+				DRM_ERROR("OA timer exponent too high (> %u)\n",
+					  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
+			 * 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)) {
+				DRM_ERROR("Minimum OA sampling exponent is 6 without root privileges\n");
+				return -EACCES;
+			}
+
+			props->oa_periodic = true;
+			props->oa_period_exponent = value;
+			break;
 		default:
 			MISSING_CASE(id);
 			DRM_ERROR("Unknown i915 perf property ID");
@@ -432,8 +1410,33 @@ int i915_perf_open_ioctl(struct drm_device *dev, void *data,
 
 void i915_perf_init(struct drm_i915_private *dev_priv)
 {
+	if (!IS_HASWELL(dev_priv))
+		return;
+
+	hrtimer_init(&dev_priv->perf.oa.poll_check_timer,
+		     CLOCK_MONOTONIC, HRTIMER_MODE_REL);
+	dev_priv->perf.oa.poll_check_timer.function = oa_poll_check_timer_cb;
+	init_waitqueue_head(&dev_priv->perf.oa.poll_wq);
+
 	INIT_LIST_HEAD(&dev_priv->perf.streams);
 	mutex_init(&dev_priv->perf.lock);
+	spin_lock_init(&dev_priv->perf.hook_lock);
+
+	dev_priv->perf.oa.ops.init_oa_buffer = gen7_init_oa_buffer;
+	dev_priv->perf.oa.ops.enable_metric_set = hsw_enable_metric_set;
+	dev_priv->perf.oa.ops.disable_metric_set = hsw_disable_metric_set;
+	dev_priv->perf.oa.ops.oa_enable = gen7_oa_enable;
+	dev_priv->perf.oa.ops.oa_disable = gen7_oa_disable;
+	dev_priv->perf.oa.ops.read = gen7_oa_read;
+	dev_priv->perf.oa.ops.oa_buffer_is_empty =
+		gen7_oa_buffer_is_empty_fop_unlocked;
+
+	dev_priv->perf.oa.timestamp_frequency = 12500000;
+
+	dev_priv->perf.oa.oa_formats = hsw_oa_formats;
+
+	dev_priv->perf.oa.n_builtin_sets =
+		i915_oa_n_builtin_metric_sets_hsw;
 
 	dev_priv->perf.initialized = true;
 }
@@ -443,7 +1446,6 @@ void i915_perf_fini(struct drm_i915_private *dev_priv)
 	if (!dev_priv->perf.initialized)
 		return;
 
-	/* Currently nothing to clean up */
-
+	memset(&dev_priv->perf.oa.ops, 0, sizeof(dev_priv->perf.oa.ops));
 	dev_priv->perf.initialized = false;
 }
diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h
index 4623dbb..df5f88d 100644
--- a/drivers/gpu/drm/i915/i915_reg.h
+++ b/drivers/gpu/drm/i915/i915_reg.h
@@ -616,6 +616,343 @@ static inline bool i915_mmio_reg_valid(i915_reg_t reg)
 #define HSW_CS_GPR_UDW(n)               _MMIO(0x2600 + (n) * 8 + 4)
 
 #define GEN7_OACONTROL _MMIO(0x2360)
+#define  GEN7_OACONTROL_CTX_MASK	    0xFFFFF000
+#define  GEN7_OACONTROL_TIMER_PERIOD_MASK   0x3F
+#define  GEN7_OACONTROL_TIMER_PERIOD_SHIFT  6
+#define  GEN7_OACONTROL_TIMER_ENABLE	    (1<<5)
+#define  GEN7_OACONTROL_FORMAT_A13	    (0<<2)
+#define  GEN7_OACONTROL_FORMAT_A29	    (1<<2)
+#define  GEN7_OACONTROL_FORMAT_A13_B8_C8    (2<<2)
+#define  GEN7_OACONTROL_FORMAT_A29_B8_C8    (3<<2)
+#define  GEN7_OACONTROL_FORMAT_B4_C8	    (4<<2)
+#define  GEN7_OACONTROL_FORMAT_A45_B8_C8    (5<<2)
+#define  GEN7_OACONTROL_FORMAT_B4_C8_A16    (6<<2)
+#define  GEN7_OACONTROL_FORMAT_C4_B8	    (7<<2)
+#define  GEN7_OACONTROL_FORMAT_SHIFT	    2
+#define  GEN7_OACONTROL_PER_CTX_ENABLE	    (1<<1)
+#define  GEN7_OACONTROL_ENABLE		    (1<<0)
+
+#define GEN8_OACTXID _MMIO(0x2364)
+
+#define GEN8_OACONTROL _MMIO(0x2B00)
+#define  GEN8_OA_REPORT_FORMAT_A12	    (0<<2)
+#define  GEN8_OA_REPORT_FORMAT_A12_B8_C8    (2<<2)
+#define  GEN8_OA_REPORT_FORMAT_A36_B8_C8    (5<<2)
+#define  GEN8_OA_REPORT_FORMAT_C4_B8	    (7<<2)
+#define  GEN8_OA_REPORT_FORMAT_SHIFT	    2
+#define  GEN8_OA_SPECIFIC_CONTEXT_ENABLE    (1<<1)
+#define  GEN8_OA_COUNTER_ENABLE             (1<<0)
+
+#define GEN8_OACTXCONTROL _MMIO(0x2360)
+#define  GEN8_OA_TIMER_PERIOD_MASK	    0x3F
+#define  GEN8_OA_TIMER_PERIOD_SHIFT	    2
+#define  GEN8_OA_TIMER_ENABLE		    (1<<1)
+#define  GEN8_OA_COUNTER_RESUME		    (1<<0)
+
+#define GEN7_OABUFFER _MMIO(0x23B0) /* R/W */
+#define  GEN7_OABUFFER_OVERRUN_DISABLE	    (1<<3)
+#define  GEN7_OABUFFER_EDGE_TRIGGER	    (1<<2)
+#define  GEN7_OABUFFER_STOP_RESUME_ENABLE   (1<<1)
+#define  GEN7_OABUFFER_RESUME		    (1<<0)
+
+#define GEN8_OABUFFER _MMIO(0x2b14)
+
+#define GEN7_OASTATUS1 _MMIO(0x2364)
+#define  GEN7_OASTATUS1_TAIL_MASK	    0xffffffc0
+#define  GEN7_OASTATUS1_COUNTER_OVERFLOW    (1<<2)
+#define  GEN7_OASTATUS1_OABUFFER_OVERFLOW   (1<<1)
+#define  GEN7_OASTATUS1_REPORT_LOST	    (1<<0)
+
+#define GEN7_OASTATUS2 _MMIO(0x2368)
+#define GEN7_OASTATUS2_HEAD_MASK    0xffffffc0
+
+#define GEN8_OASTATUS _MMIO(0x2b08)
+#define  GEN8_OASTATUS_OVERRUN_STATUS	    (1<<3)
+#define  GEN8_OASTATUS_COUNTER_OVERFLOW     (1<<2)
+#define  GEN8_OASTATUS_OABUFFER_OVERFLOW    (1<<1)
+#define  GEN8_OASTATUS_REPORT_LOST	    (1<<0)
+
+#define GEN8_OAHEADPTR _MMIO(0x2B0C)
+#define GEN8_OATAILPTR _MMIO(0x2B10)
+
+#define OABUFFER_SIZE_128K  (0<<3)
+#define OABUFFER_SIZE_256K  (1<<3)
+#define OABUFFER_SIZE_512K  (2<<3)
+#define OABUFFER_SIZE_1M    (3<<3)
+#define OABUFFER_SIZE_2M    (4<<3)
+#define OABUFFER_SIZE_4M    (5<<3)
+#define OABUFFER_SIZE_8M    (6<<3)
+#define OABUFFER_SIZE_16M   (7<<3)
+
+#define OA_MEM_SELECT_GGTT  (1<<0)
+
+#define EU_PERF_CNTL0	    _MMIO(0xe458)
+
+#define GDT_CHICKEN_BITS    _MMIO(0x9840)
+#define GT_NOA_ENABLE	    0x00000080
+
+/*
+ * OA Boolean state
+ */
+
+#define OAREPORTTRIG1 _MMIO(0x2740)
+#define OAREPORTTRIG1_THRESHOLD_MASK 0xffff
+#define OAREPORTTRIG1_EDGE_LEVEL_TRIGER_SELECT_MASK 0xffff0000 /* 0=level */
+
+#define OAREPORTTRIG2 _MMIO(0x2744)
+#define OAREPORTTRIG2_INVERT_A_0  (1<<0)
+#define OAREPORTTRIG2_INVERT_A_1  (1<<1)
+#define OAREPORTTRIG2_INVERT_A_2  (1<<2)
+#define OAREPORTTRIG2_INVERT_A_3  (1<<3)
+#define OAREPORTTRIG2_INVERT_A_4  (1<<4)
+#define OAREPORTTRIG2_INVERT_A_5  (1<<5)
+#define OAREPORTTRIG2_INVERT_A_6  (1<<6)
+#define OAREPORTTRIG2_INVERT_A_7  (1<<7)
+#define OAREPORTTRIG2_INVERT_A_8  (1<<8)
+#define OAREPORTTRIG2_INVERT_A_9  (1<<9)
+#define OAREPORTTRIG2_INVERT_A_10 (1<<10)
+#define OAREPORTTRIG2_INVERT_A_11 (1<<11)
+#define OAREPORTTRIG2_INVERT_A_12 (1<<12)
+#define OAREPORTTRIG2_INVERT_A_13 (1<<13)
+#define OAREPORTTRIG2_INVERT_A_14 (1<<14)
+#define OAREPORTTRIG2_INVERT_A_15 (1<<15)
+#define OAREPORTTRIG2_INVERT_B_0  (1<<16)
+#define OAREPORTTRIG2_INVERT_B_1  (1<<17)
+#define OAREPORTTRIG2_INVERT_B_2  (1<<18)
+#define OAREPORTTRIG2_INVERT_B_3  (1<<19)
+#define OAREPORTTRIG2_INVERT_C_0  (1<<20)
+#define OAREPORTTRIG2_INVERT_C_1  (1<<21)
+#define OAREPORTTRIG2_INVERT_D_0  (1<<22)
+#define OAREPORTTRIG2_THRESHOLD_ENABLE	    (1<<23)
+#define OAREPORTTRIG2_REPORT_TRIGGER_ENABLE (1<<31)
+
+#define OAREPORTTRIG3 _MMIO(0x2748)
+#define OAREPORTTRIG3_NOA_SELECT_MASK	    0xf
+#define OAREPORTTRIG3_NOA_SELECT_8_SHIFT    0
+#define OAREPORTTRIG3_NOA_SELECT_9_SHIFT    4
+#define OAREPORTTRIG3_NOA_SELECT_10_SHIFT   8
+#define OAREPORTTRIG3_NOA_SELECT_11_SHIFT   12
+#define OAREPORTTRIG3_NOA_SELECT_12_SHIFT   16
+#define OAREPORTTRIG3_NOA_SELECT_13_SHIFT   20
+#define OAREPORTTRIG3_NOA_SELECT_14_SHIFT   24
+#define OAREPORTTRIG3_NOA_SELECT_15_SHIFT   28
+
+#define OAREPORTTRIG4 _MMIO(0x274c)
+#define OAREPORTTRIG4_NOA_SELECT_MASK	    0xf
+#define OAREPORTTRIG4_NOA_SELECT_0_SHIFT    0
+#define OAREPORTTRIG4_NOA_SELECT_1_SHIFT    4
+#define OAREPORTTRIG4_NOA_SELECT_2_SHIFT    8
+#define OAREPORTTRIG4_NOA_SELECT_3_SHIFT    12
+#define OAREPORTTRIG4_NOA_SELECT_4_SHIFT    16
+#define OAREPORTTRIG4_NOA_SELECT_5_SHIFT    20
+#define OAREPORTTRIG4_NOA_SELECT_6_SHIFT    24
+#define OAREPORTTRIG4_NOA_SELECT_7_SHIFT    28
+
+#define OAREPORTTRIG5 _MMIO(0x2750)
+#define OAREPORTTRIG5_THRESHOLD_MASK 0xffff
+#define OAREPORTTRIG5_EDGE_LEVEL_TRIGER_SELECT_MASK 0xffff0000 /* 0=level */
+
+#define OAREPORTTRIG6 _MMIO(0x2754)
+#define OAREPORTTRIG6_INVERT_A_0  (1<<0)
+#define OAREPORTTRIG6_INVERT_A_1  (1<<1)
+#define OAREPORTTRIG6_INVERT_A_2  (1<<2)
+#define OAREPORTTRIG6_INVERT_A_3  (1<<3)
+#define OAREPORTTRIG6_INVERT_A_4  (1<<4)
+#define OAREPORTTRIG6_INVERT_A_5  (1<<5)
+#define OAREPORTTRIG6_INVERT_A_6  (1<<6)
+#define OAREPORTTRIG6_INVERT_A_7  (1<<7)
+#define OAREPORTTRIG6_INVERT_A_8  (1<<8)
+#define OAREPORTTRIG6_INVERT_A_9  (1<<9)
+#define OAREPORTTRIG6_INVERT_A_10 (1<<10)
+#define OAREPORTTRIG6_INVERT_A_11 (1<<11)
+#define OAREPORTTRIG6_INVERT_A_12 (1<<12)
+#define OAREPORTTRIG6_INVERT_A_13 (1<<13)
+#define OAREPORTTRIG6_INVERT_A_14 (1<<14)
+#define OAREPORTTRIG6_INVERT_A_15 (1<<15)
+#define OAREPORTTRIG6_INVERT_B_0  (1<<16)
+#define OAREPORTTRIG6_INVERT_B_1  (1<<17)
+#define OAREPORTTRIG6_INVERT_B_2  (1<<18)
+#define OAREPORTTRIG6_INVERT_B_3  (1<<19)
+#define OAREPORTTRIG6_INVERT_C_0  (1<<20)
+#define OAREPORTTRIG6_INVERT_C_1  (1<<21)
+#define OAREPORTTRIG6_INVERT_D_0  (1<<22)
+#define OAREPORTTRIG6_THRESHOLD_ENABLE	    (1<<23)
+#define OAREPORTTRIG6_REPORT_TRIGGER_ENABLE (1<<31)
+
+#define OAREPORTTRIG7 _MMIO(0x2758)
+#define OAREPORTTRIG7_NOA_SELECT_MASK	    0xf
+#define OAREPORTTRIG7_NOA_SELECT_8_SHIFT    0
+#define OAREPORTTRIG7_NOA_SELECT_9_SHIFT    4
+#define OAREPORTTRIG7_NOA_SELECT_10_SHIFT   8
+#define OAREPORTTRIG7_NOA_SELECT_11_SHIFT   12
+#define OAREPORTTRIG7_NOA_SELECT_12_SHIFT   16
+#define OAREPORTTRIG7_NOA_SELECT_13_SHIFT   20
+#define OAREPORTTRIG7_NOA_SELECT_14_SHIFT   24
+#define OAREPORTTRIG7_NOA_SELECT_15_SHIFT   28
+
+#define OAREPORTTRIG8 _MMIO(0x275c)
+#define OAREPORTTRIG8_NOA_SELECT_MASK	    0xf
+#define OAREPORTTRIG8_NOA_SELECT_0_SHIFT    0
+#define OAREPORTTRIG8_NOA_SELECT_1_SHIFT    4
+#define OAREPORTTRIG8_NOA_SELECT_2_SHIFT    8
+#define OAREPORTTRIG8_NOA_SELECT_3_SHIFT    12
+#define OAREPORTTRIG8_NOA_SELECT_4_SHIFT    16
+#define OAREPORTTRIG8_NOA_SELECT_5_SHIFT    20
+#define OAREPORTTRIG8_NOA_SELECT_6_SHIFT    24
+#define OAREPORTTRIG8_NOA_SELECT_7_SHIFT    28
+
+#define OASTARTTRIG1 _MMIO(0x2710)
+#define OASTARTTRIG1_THRESHOLD_COUNT_MASK_MBZ 0xffff0000
+#define OASTARTTRIG1_THRESHOLD_MASK	      0xffff
+
+#define OASTARTTRIG2 _MMIO(0x2714)
+#define OASTARTTRIG2_INVERT_A_0 (1<<0)
+#define OASTARTTRIG2_INVERT_A_1 (1<<1)
+#define OASTARTTRIG2_INVERT_A_2 (1<<2)
+#define OASTARTTRIG2_INVERT_A_3 (1<<3)
+#define OASTARTTRIG2_INVERT_A_4 (1<<4)
+#define OASTARTTRIG2_INVERT_A_5 (1<<5)
+#define OASTARTTRIG2_INVERT_A_6 (1<<6)
+#define OASTARTTRIG2_INVERT_A_7 (1<<7)
+#define OASTARTTRIG2_INVERT_A_8 (1<<8)
+#define OASTARTTRIG2_INVERT_A_9 (1<<9)
+#define OASTARTTRIG2_INVERT_A_10 (1<<10)
+#define OASTARTTRIG2_INVERT_A_11 (1<<11)
+#define OASTARTTRIG2_INVERT_A_12 (1<<12)
+#define OASTARTTRIG2_INVERT_A_13 (1<<13)
+#define OASTARTTRIG2_INVERT_A_14 (1<<14)
+#define OASTARTTRIG2_INVERT_A_15 (1<<15)
+#define OASTARTTRIG2_INVERT_B_0 (1<<16)
+#define OASTARTTRIG2_INVERT_B_1 (1<<17)
+#define OASTARTTRIG2_INVERT_B_2 (1<<18)
+#define OASTARTTRIG2_INVERT_B_3 (1<<19)
+#define OASTARTTRIG2_INVERT_C_0 (1<<20)
+#define OASTARTTRIG2_INVERT_C_1 (1<<21)
+#define OASTARTTRIG2_INVERT_D_0 (1<<22)
+#define OASTARTTRIG2_THRESHOLD_ENABLE	    (1<<23)
+#define OASTARTTRIG2_START_TRIG_FLAG_MBZ    (1<<24)
+#define OASTARTTRIG2_EVENT_SELECT_0  (1<<28)
+#define OASTARTTRIG2_EVENT_SELECT_1  (1<<29)
+#define OASTARTTRIG2_EVENT_SELECT_2  (1<<30)
+#define OASTARTTRIG2_EVENT_SELECT_3  (1<<31)
+
+#define OASTARTTRIG3 _MMIO(0x2718)
+#define OASTARTTRIG3_NOA_SELECT_MASK	   0xf
+#define OASTARTTRIG3_NOA_SELECT_8_SHIFT    0
+#define OASTARTTRIG3_NOA_SELECT_9_SHIFT    4
+#define OASTARTTRIG3_NOA_SELECT_10_SHIFT   8
+#define OASTARTTRIG3_NOA_SELECT_11_SHIFT   12
+#define OASTARTTRIG3_NOA_SELECT_12_SHIFT   16
+#define OASTARTTRIG3_NOA_SELECT_13_SHIFT   20
+#define OASTARTTRIG3_NOA_SELECT_14_SHIFT   24
+#define OASTARTTRIG3_NOA_SELECT_15_SHIFT   28
+
+#define OASTARTTRIG4 _MMIO(0x271c)
+#define OASTARTTRIG4_NOA_SELECT_MASK	    0xf
+#define OASTARTTRIG4_NOA_SELECT_0_SHIFT    0
+#define OASTARTTRIG4_NOA_SELECT_1_SHIFT    4
+#define OASTARTTRIG4_NOA_SELECT_2_SHIFT    8
+#define OASTARTTRIG4_NOA_SELECT_3_SHIFT    12
+#define OASTARTTRIG4_NOA_SELECT_4_SHIFT    16
+#define OASTARTTRIG4_NOA_SELECT_5_SHIFT    20
+#define OASTARTTRIG4_NOA_SELECT_6_SHIFT    24
+#define OASTARTTRIG4_NOA_SELECT_7_SHIFT    28
+
+#define OASTARTTRIG5 _MMIO(0x2720)
+#define OASTARTTRIG5_THRESHOLD_COUNT_MASK_MBZ 0xffff0000
+#define OASTARTTRIG5_THRESHOLD_MASK	      0xffff
+
+#define OASTARTTRIG6 _MMIO(0x2724)
+#define OASTARTTRIG6_INVERT_A_0 (1<<0)
+#define OASTARTTRIG6_INVERT_A_1 (1<<1)
+#define OASTARTTRIG6_INVERT_A_2 (1<<2)
+#define OASTARTTRIG6_INVERT_A_3 (1<<3)
+#define OASTARTTRIG6_INVERT_A_4 (1<<4)
+#define OASTARTTRIG6_INVERT_A_5 (1<<5)
+#define OASTARTTRIG6_INVERT_A_6 (1<<6)
+#define OASTARTTRIG6_INVERT_A_7 (1<<7)
+#define OASTARTTRIG6_INVERT_A_8 (1<<8)
+#define OASTARTTRIG6_INVERT_A_9 (1<<9)
+#define OASTARTTRIG6_INVERT_A_10 (1<<10)
+#define OASTARTTRIG6_INVERT_A_11 (1<<11)
+#define OASTARTTRIG6_INVERT_A_12 (1<<12)
+#define OASTARTTRIG6_INVERT_A_13 (1<<13)
+#define OASTARTTRIG6_INVERT_A_14 (1<<14)
+#define OASTARTTRIG6_INVERT_A_15 (1<<15)
+#define OASTARTTRIG6_INVERT_B_0 (1<<16)
+#define OASTARTTRIG6_INVERT_B_1 (1<<17)
+#define OASTARTTRIG6_INVERT_B_2 (1<<18)
+#define OASTARTTRIG6_INVERT_B_3 (1<<19)
+#define OASTARTTRIG6_INVERT_C_0 (1<<20)
+#define OASTARTTRIG6_INVERT_C_1 (1<<21)
+#define OASTARTTRIG6_INVERT_D_0 (1<<22)
+#define OASTARTTRIG6_THRESHOLD_ENABLE	    (1<<23)
+#define OASTARTTRIG6_START_TRIG_FLAG_MBZ    (1<<24)
+#define OASTARTTRIG6_EVENT_SELECT_4  (1<<28)
+#define OASTARTTRIG6_EVENT_SELECT_5  (1<<29)
+#define OASTARTTRIG6_EVENT_SELECT_6  (1<<30)
+#define OASTARTTRIG6_EVENT_SELECT_7  (1<<31)
+
+#define OASTARTTRIG7 _MMIO(0x2728)
+#define OASTARTTRIG7_NOA_SELECT_MASK	   0xf
+#define OASTARTTRIG7_NOA_SELECT_8_SHIFT    0
+#define OASTARTTRIG7_NOA_SELECT_9_SHIFT    4
+#define OASTARTTRIG7_NOA_SELECT_10_SHIFT   8
+#define OASTARTTRIG7_NOA_SELECT_11_SHIFT   12
+#define OASTARTTRIG7_NOA_SELECT_12_SHIFT   16
+#define OASTARTTRIG7_NOA_SELECT_13_SHIFT   20
+#define OASTARTTRIG7_NOA_SELECT_14_SHIFT   24
+#define OASTARTTRIG7_NOA_SELECT_15_SHIFT   28
+
+#define OASTARTTRIG8 _MMIO(0x272c)
+#define OASTARTTRIG8_NOA_SELECT_MASK	   0xf
+#define OASTARTTRIG8_NOA_SELECT_0_SHIFT    0
+#define OASTARTTRIG8_NOA_SELECT_1_SHIFT    4
+#define OASTARTTRIG8_NOA_SELECT_2_SHIFT    8
+#define OASTARTTRIG8_NOA_SELECT_3_SHIFT    12
+#define OASTARTTRIG8_NOA_SELECT_4_SHIFT    16
+#define OASTARTTRIG8_NOA_SELECT_5_SHIFT    20
+#define OASTARTTRIG8_NOA_SELECT_6_SHIFT    24
+#define OASTARTTRIG8_NOA_SELECT_7_SHIFT    28
+
+/* CECX_0 */
+#define OACEC_COMPARE_LESS_OR_EQUAL	6
+#define OACEC_COMPARE_NOT_EQUAL		5
+#define OACEC_COMPARE_LESS_THAN		4
+#define OACEC_COMPARE_GREATER_OR_EQUAL	3
+#define OACEC_COMPARE_EQUAL		2
+#define OACEC_COMPARE_GREATER_THAN	1
+#define OACEC_COMPARE_ANY_EQUAL		0
+
+#define OACEC_COMPARE_VALUE_MASK    0xffff
+#define OACEC_COMPARE_VALUE_SHIFT   3
+
+#define OACEC_SELECT_NOA	(0<<19)
+#define OACEC_SELECT_PREV	(1<<19)
+#define OACEC_SELECT_BOOLEAN	(2<<19)
+
+/* CECX_1 */
+#define OACEC_MASK_MASK		    0xffff
+#define OACEC_CONSIDERATIONS_MASK   0xffff
+#define OACEC_CONSIDERATIONS_SHIFT  16
+
+#define OACEC0_0 _MMIO(0x2770)
+#define OACEC0_1 _MMIO(0x2774)
+#define OACEC1_0 _MMIO(0x2778)
+#define OACEC1_1 _MMIO(0x277c)
+#define OACEC2_0 _MMIO(0x2780)
+#define OACEC2_1 _MMIO(0x2784)
+#define OACEC3_0 _MMIO(0x2788)
+#define OACEC3_1 _MMIO(0x278c)
+#define OACEC4_0 _MMIO(0x2790)
+#define OACEC4_1 _MMIO(0x2794)
+#define OACEC5_0 _MMIO(0x2798)
+#define OACEC5_1 _MMIO(0x279c)
+#define OACEC6_0 _MMIO(0x27a0)
+#define OACEC6_1 _MMIO(0x27a4)
+#define OACEC7_0 _MMIO(0x27a8)
+#define OACEC7_1 _MMIO(0x27ac)
+
 
 #define _GEN7_PIPEA_DE_LOAD_SL	0x70068
 #define _GEN7_PIPEB_DE_LOAD_SL	0x71068
@@ -6914,6 +7251,7 @@ enum {
 # define GEN6_RCCUNIT_CLOCK_GATE_DISABLE		(1 << 11)
 
 #define GEN6_UCGCTL3				_MMIO(0x9408)
+# define GEN6_OACSUNIT_CLOCK_GATE_DISABLE		(1 << 20)
 
 #define GEN7_UCGCTL4				_MMIO(0x940c)
 #define  GEN7_L3BANK2X_CLOCK_GATE_DISABLE	(1<<25)
diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h
index 98cd493..82d6239 100644
--- a/include/uapi/drm/i915_drm.h
+++ b/include/uapi/drm/i915_drm.h
@@ -1224,6 +1224,18 @@ struct drm_i915_gem_context_param {
 	__u64 value;
 };
 
+enum drm_i915_oa_format {
+	I915_OA_FORMAT_A13 = 1,
+	I915_OA_FORMAT_A29,
+	I915_OA_FORMAT_A13_B8_C8,
+	I915_OA_FORMAT_B4_C8,
+	I915_OA_FORMAT_A45_B8_C8,
+	I915_OA_FORMAT_B4_C8_A16,
+	I915_OA_FORMAT_C4_B8,
+
+	I915_OA_FORMAT_MAX	    /* non-ABI */
+};
+
 enum drm_i915_perf_property_id {
 	/**
 	 * Open the stream for a specific context handle (as used with
@@ -1232,6 +1244,32 @@ enum drm_i915_perf_property_id {
 	 */
 	DRM_I915_PERF_PROP_CTX_HANDLE = 1,
 
+	/**
+	 * A value of 1 requests the inclusion of raw OA unit reports as
+	 * part of stream samples.
+	 */
+	DRM_I915_PERF_PROP_SAMPLE_OA,
+
+	/**
+	 * The value specifies which set of OA unit metrics should be
+	 * be configured, defining the contents of any OA unit reports.
+	 */
+	DRM_I915_PERF_PROP_OA_METRICS_SET,
+
+	/**
+	 * The value specifies the size and layout of OA unit reports.
+	 */
+	DRM_I915_PERF_PROP_OA_FORMAT,
+
+	/**
+	 * Specifying this property implicitly requests periodic OA unit
+	 * sampling and (at least on Haswell) the sampling frequency is derived
+	 * from this exponent as follows:
+	 *
+	 *   80ns * 2^(period_exponent + 1)
+	 */
+	DRM_I915_PERF_PROP_OA_EXPONENT,
+
 	DRM_I915_PERF_PROP_MAX /* non-ABI */
 };
 
@@ -1251,7 +1289,23 @@ struct drm_i915_perf_open_param {
 	__u64 __user properties_ptr;
 };
 
+/**
+ * Enable data capture for a stream that was either opened in a disabled state
+ * via I915_PERF_FLAG_DISABLED or was later disabled via
+ * I915_PERF_IOCTL_DISABLE.
+ *
+ * It is intended to be cheaper to disable and enable a stream than it may be
+ * to close and re-open a stream with the same configuration.
+ *
+ * It's undefined whether any pending data for the stream will be lost.
+ */
 #define I915_PERF_IOCTL_ENABLE	_IO('i', 0x0)
+
+/**
+ * Disable data capture for a stream.
+ *
+ * It is an error to try and read a stream that is disabled.
+ */
 #define I915_PERF_IOCTL_DISABLE	_IO('i', 0x1)
 
 /**
@@ -1275,17 +1329,30 @@ enum drm_i915_perf_record_type {
 	 * every sample.
 	 *
 	 * The order of these sample properties given by userspace has no
-	 * affect on the ordering of data within a sample. The order will be
+	 * affect on the ordering of data within a sample. The order is
 	 * documented here.
 	 *
 	 * struct {
 	 *     struct drm_i915_perf_record_header header;
 	 *
-	 *     TODO: itemize extensible sample data here
+	 *     { u32 oa_report[]; } && DRM_I915_PERF_PROP_SAMPLE_OA
 	 * };
 	 */
 	DRM_I915_PERF_RECORD_SAMPLE = 1,
 
+	/*
+	 * Indicates that one or more OA reports were not written by the
+	 * hardware. This can happen for example if an MI_REPORT_PERF_COUNT
+	 * command collides with periodic sampling - which would be more likely
+	 * at higher sampling frequencies.
+	 */
+	DRM_I915_PERF_RECORD_OA_REPORT_LOST = 2,
+
+	/**
+	 * An error occurred that resulted in all pending OA reports being lost.
+	 */
+	DRM_I915_PERF_RECORD_OA_BUFFER_LOST = 3,
+
 	DRM_I915_PERF_RECORD_MAX /* non-ABI */
 };
 
-- 
2.10.1

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v9 07/11] drm/i915: advertise available metrics via sysfs

[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 register configuration that can be reliably used by
userspace as a key to lookup corresponding counter meta data and
normalization equations.

The guid registry is currently maintained as part of gputop along with
the XML metric set descriptions and code generation scripts, ref:

 https://github.com/rib/gputop
 > gputop-data/guids.xml
 > scripts/update-guids.py
 > gputop-data/oa-*.xml
 > scripts/i915-perf-kernelgen.py

 $ make -C gputop-data -f Makefile.xml SYSFS=1 WHITELIST=RenderBasic

Signed-off-by: Robert Bragg <robert@sixbynine.org>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>
---
 drivers/gpu/drm/i915/i915_drv.c    |  5 ++++
 drivers/gpu/drm/i915/i915_drv.h    |  4 +++
 drivers/gpu/drm/i915/i915_oa_hsw.c | 51 +++++++++++++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_oa_hsw.h |  4 +++
 drivers/gpu/drm/i915/i915_perf.c   | 52 ++++++++++++++++++++++++++++++++++++++
 5 files changed, 116 insertions(+)

diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c
index 22b4166..4fd8650 100644
--- a/drivers/gpu/drm/i915/i915_drv.c
+++ b/drivers/gpu/drm/i915/i915_drv.c
@@ -1125,6 +1125,9 @@ static void i915_driver_register(struct drm_i915_private *dev_priv)
 		i915_debugfs_register(dev_priv);
 		i915_guc_register(dev_priv);
 		i915_setup_sysfs(dev_priv);
+
+		/* Depends on sysfs having been initialized */
+		i915_perf_register(dev_priv);
 	} else
 		DRM_ERROR("Failed to register driver for userspace access!\n");
 
@@ -1161,6 +1164,8 @@ static void i915_driver_unregister(struct drm_i915_private *dev_priv)
 	acpi_video_unregister();
 	intel_opregion_unregister(dev_priv);
 
+	i915_perf_unregister(dev_priv);
+
 	i915_teardown_sysfs(dev_priv);
 	i915_guc_unregister(dev_priv);
 	i915_debugfs_unregister(dev_priv);
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 8003120..15fba6b 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -2187,6 +2187,8 @@ struct drm_i915_private {
 	struct {
 		bool initialized;
 
+		struct kobject *metrics_kobj;
+
 		struct mutex lock;
 		struct list_head streams;
 
@@ -3884,6 +3886,8 @@ int intel_engine_cmd_parser(struct intel_engine_cs *engine,
 /* i915_perf.c */
 extern void i915_perf_init(struct drm_i915_private *dev_priv);
 extern void i915_perf_fini(struct drm_i915_private *dev_priv);
+extern void i915_perf_register(struct drm_i915_private *dev_priv);
+extern void i915_perf_unregister(struct drm_i915_private *dev_priv);
 
 /* i915_suspend.c */
 extern int i915_save_state(struct drm_device *dev);
diff --git a/drivers/gpu/drm/i915/i915_oa_hsw.c b/drivers/gpu/drm/i915/i915_oa_hsw.c
index 8906380..6af25cf 100644
--- a/drivers/gpu/drm/i915/i915_oa_hsw.c
+++ b/drivers/gpu/drm/i915/i915_oa_hsw.c
@@ -24,6 +24,8 @@
  *
  */
 
+#include <linux/sysfs.h>
+
 #include "i915_drv.h"
 #include "i915_oa_hsw.h"
 
@@ -142,3 +144,52 @@ int i915_oa_select_metric_set_hsw(struct drm_i915_private *dev_priv)
 		return -ENODEV;
 	}
 }
+
+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);
+}
+
+static struct device_attribute dev_attr_render_basic_id = {
+	.attr = { .name = "id", .mode = 0444 },
+	.show = show_render_basic_id,
+	.store = NULL,
+};
+
+static struct attribute *attrs_render_basic[] = {
+	&dev_attr_render_basic_id.attr,
+	NULL,
+};
+
+static struct attribute_group group_render_basic = {
+	.name = "403d8832-1a27-4aa6-a64e-f5389ce7b212",
+	.attrs =  attrs_render_basic,
+};
+
+int
+i915_perf_register_sysfs_hsw(struct drm_i915_private *dev_priv)
+{
+	int mux_len;
+	int ret = 0;
+
+	if (get_render_basic_mux_config(dev_priv, &mux_len)) {
+		ret = sysfs_create_group(dev_priv->perf.metrics_kobj, &group_render_basic);
+		if (ret)
+			goto error_render_basic;
+	}
+
+	return 0;
+
+error_render_basic:
+	return ret;
+}
+
+void
+i915_perf_unregister_sysfs_hsw(struct drm_i915_private *dev_priv)
+{
+	int mux_len;
+
+	if (get_render_basic_mux_config(dev_priv, &mux_len))
+		sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_render_basic);
+}
diff --git a/drivers/gpu/drm/i915/i915_oa_hsw.h b/drivers/gpu/drm/i915/i915_oa_hsw.h
index b618a1f..429a229 100644
--- a/drivers/gpu/drm/i915/i915_oa_hsw.h
+++ b/drivers/gpu/drm/i915/i915_oa_hsw.h
@@ -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_register_sysfs_hsw(struct drm_i915_private *dev_priv);
+
+extern void i915_perf_unregister_sysfs_hsw(struct drm_i915_private *dev_priv);
+
 #endif
diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c
index 54653bc..c427cd8c 100644
--- a/drivers/gpu/drm/i915/i915_perf.c
+++ b/drivers/gpu/drm/i915/i915_perf.c
@@ -812,6 +812,15 @@ static int i915_oa_stream_init(struct i915_perf_stream *stream,
 	int format_size;
 	int ret;
 
+	/* If the sysfs metrics/ directory wasn't registered for some
+	 * reason then don't let userspace try their luck with config
+	 * IDs
+	 */
+	if (!dev_priv->perf.metrics_kobj) {
+		DRM_ERROR("OA metrics weren't advertised via sysfs\n");
+		return -EINVAL;
+	}
+
 	if (!(props->sample_flags & SAMPLE_OA_REPORT)) {
 		DRM_ERROR("Only OA report sampling supported\n");
 		return -EINVAL;
@@ -1408,6 +1417,49 @@ int i915_perf_open_ioctl(struct drm_device *dev, void *data,
 	return ret;
 }
 
+void i915_perf_register(struct drm_i915_private *dev_priv)
+{
+	if (!IS_HASWELL(dev_priv))
+		return;
+
+	if (!dev_priv->perf.initialized)
+		return;
+
+	/* To be sure we're synchronized with an attempted
+	 * i915_perf_open_ioctl(); considering that we register after
+	 * being exposed to userspace.
+	 */
+	mutex_lock(&dev_priv->perf.lock);
+
+	dev_priv->perf.metrics_kobj =
+		kobject_create_and_add("metrics",
+				       &dev_priv->drm.primary->kdev->kobj);
+	if (!dev_priv->perf.metrics_kobj)
+		goto exit;
+
+	if (i915_perf_register_sysfs_hsw(dev_priv)) {
+		kobject_put(dev_priv->perf.metrics_kobj);
+		dev_priv->perf.metrics_kobj = NULL;
+	}
+
+exit:
+	mutex_unlock(&dev_priv->perf.lock);
+}
+
+void i915_perf_unregister(struct drm_i915_private *dev_priv)
+{
+	if (!IS_HASWELL(dev_priv))
+		return;
+
+	if (!dev_priv->perf.metrics_kobj)
+		return;
+
+	i915_perf_unregister_sysfs_hsw(dev_priv);
+
+	kobject_put(dev_priv->perf.metrics_kobj);
+	dev_priv->perf.metrics_kobj = NULL;
+}
+
 void i915_perf_init(struct drm_i915_private *dev_priv)
 {
 	if (!IS_HASWELL(dev_priv))
-- 
2.10.1

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v9 08/11] drm/i915: Add dev.i915.perf_stream_paranoid sysctl option

[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.

Signed-off-by: Robert Bragg <robert@sixbynine.org>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>
---
 drivers/gpu/drm/i915/i915_drv.h  |  1 +
 drivers/gpu/drm/i915/i915_perf.c | 50 +++++++++++++++++++++++++++++++++++++++-
 2 files changed, 50 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 15fba6b..8962bfd 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -2188,6 +2188,7 @@ struct drm_i915_private {
 		bool initialized;
 
 		struct kobject *metrics_kobj;
+		struct ctl_table_header *sysctl_header;
 
 		struct mutex lock;
 		struct list_head streams;
diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c
index c427cd8c..e51c1d8 100644
--- a/drivers/gpu/drm/i915/i915_perf.c
+++ b/drivers/gpu/drm/i915/i915_perf.c
@@ -64,6 +64,11 @@
 #define POLL_FREQUENCY 200
 #define POLL_PERIOD (NSEC_PER_SEC / POLL_FREQUENCY)
 
+/* for sysctl proc_dointvec_minmax of dev.i915.perf_stream_paranoid */
+static int zero;
+static int one = 1;
+static u32 i915_perf_stream_paranoid = true;
+
 /* The maximum exponent the hardware accepts is 63 (essentially it selects one
  * of the 64bit timestamp bits to trigger reports from) but there's currently
  * no known use case for sampling as infrequently as once per 47 thousand years.
@@ -1207,7 +1212,13 @@ i915_perf_open_ioctl_locked(struct drm_i915_private *dev_priv,
 		}
 	}
 
-	if (!specific_ctx && !capable(CAP_SYS_ADMIN)) {
+	/* Similar to perf's kernel.perf_paranoid_cpu sysctl option
+	 * we check a dev.i915.perf_stream_paranoid sysctl option
+	 * to determine if it's ok to access system wide OA counters
+	 * without CAP_SYS_ADMIN privileges.
+	 */
+	if (!specific_ctx &&
+	    i915_perf_stream_paranoid && !capable(CAP_SYS_ADMIN)) {
 		DRM_ERROR("Insufficient privileges to open system-wide i915 perf stream\n");
 		ret = -EACCES;
 		goto err_ctx;
@@ -1460,6 +1471,39 @@ void i915_perf_unregister(struct drm_i915_private *dev_priv)
 	dev_priv->perf.metrics_kobj = NULL;
 }
 
+static struct ctl_table oa_table[] = {
+	{
+	 .procname = "perf_stream_paranoid",
+	 .data = &i915_perf_stream_paranoid,
+	 .maxlen = sizeof(i915_perf_stream_paranoid),
+	 .mode = 0644,
+	 .proc_handler = proc_dointvec_minmax,
+	 .extra1 = &zero,
+	 .extra2 = &one,
+	 },
+	{}
+};
+
+static struct ctl_table i915_root[] = {
+	{
+	 .procname = "i915",
+	 .maxlen = 0,
+	 .mode = 0555,
+	 .child = oa_table,
+	 },
+	{}
+};
+
+static struct ctl_table dev_root[] = {
+	{
+	 .procname = "dev",
+	 .maxlen = 0,
+	 .mode = 0555,
+	 .child = i915_root,
+	 },
+	{}
+};
+
 void i915_perf_init(struct drm_i915_private *dev_priv)
 {
 	if (!IS_HASWELL(dev_priv))
@@ -1490,6 +1534,8 @@ void i915_perf_init(struct drm_i915_private *dev_priv)
 	dev_priv->perf.oa.n_builtin_sets =
 		i915_oa_n_builtin_metric_sets_hsw;
 
+	dev_priv->perf.sysctl_header = register_sysctl_table(dev_root);
+
 	dev_priv->perf.initialized = true;
 }
 
@@ -1498,6 +1544,8 @@ void i915_perf_fini(struct drm_i915_private *dev_priv)
 	if (!dev_priv->perf.initialized)
 		return;
 
+	unregister_sysctl_table(dev_priv->perf.sysctl_header);
+
 	memset(&dev_priv->perf.oa.ops, 0, sizeof(dev_priv->perf.oa.ops));
 	dev_priv->perf.initialized = false;
 }
-- 
2.10.1

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v9 09/11] drm/i915: add dev.i915.oa_max_sample_rate sysctl

[Robert Bragg]

The maximum OA sampling frequency is now configurable via a
dev.i915.oa_max_sample_rate sysctl parameter.

Following the precedent set by perf's similar
kernel.perf_event_max_sample_rate the default maximum rate is 100000Hz

Signed-off-by: Robert Bragg <robert@sixbynine.org>
---
 drivers/gpu/drm/i915/i915_perf.c | 61 ++++++++++++++++++++++++++++++++--------
 1 file changed, 50 insertions(+), 11 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c
index e51c1d8..1a87fe9 100644
--- a/drivers/gpu/drm/i915/i915_perf.c
+++ b/drivers/gpu/drm/i915/i915_perf.c
@@ -82,6 +82,21 @@ static u32 i915_perf_stream_paranoid = true;
 #define INVALID_CTX_ID 0xffffffff
 
 
+/* For sysctl proc_dointvec_minmax of i915_oa_max_sample_rate
+ *
+ * 160ns is the smallest sampling period we can theoretically program the OA
+ * unit with on Haswell, corresponding to 6.25MHz.
+ */
+static int oa_sample_rate_hard_limit = 6250000;
+
+/* Theoretically we can program the OA unit to sample every 160ns but don't
+ * allow that by default unless root...
+ *
+ * The default threshold of 100000Hz is based on perf's similar
+ * kernel.perf_event_max_sample_rate sysctl parameter.
+ */
+static u32 i915_oa_max_sample_rate = 100000;
+
 /* XXX: beware if future OA HW adds new report formats that the current
  * code assumes all reports have a power-of-two size and ~(size - 1) can
  * be used as a mask to align the OA tail pointer.
@@ -1314,6 +1329,7 @@ static int read_properties_unlocked(struct drm_i915_private *dev_priv,
 	}
 
 	for (i = 0; i < n_props; i++) {
+		u64 oa_period, oa_freq_hz;
 		u64 id, value;
 		int ret;
 
@@ -1359,21 +1375,35 @@ static int read_properties_unlocked(struct drm_i915_private *dev_priv,
 				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
+			 * On Haswell the period is derived from the exponent
+			 * as:
+			 *
+			 *   period = 80ns * 2^(exponent + 1)
+			 */
+			BUILD_BUG_ON(sizeof(oa_period) != 8);
+			oa_period = 80ull * (2ull << value);
+
+			/* This check is primarily to ensure that oa_period <=
+			 * UINT32_MAX (before passing to do_div which only
+			 * accepts a u32 denominator), but we can also skip
+			 * checking anything < 1Hz which implicitly can't be
+			 * limited via an integer oa_max_sample_rate.
 			 */
-			if (value < 6 && !capable(CAP_SYS_ADMIN)) {
-				DRM_ERROR("Minimum OA sampling exponent is 6 without root privileges\n");
+			if (oa_period <= NSEC_PER_SEC) {
+				u64 tmp = NSEC_PER_SEC;
+				do_div(tmp, oa_period);
+				oa_freq_hz = tmp;
+			} else
+				oa_freq_hz = 0;
+
+			if (oa_freq_hz > i915_oa_max_sample_rate &&
+			    !capable(CAP_SYS_ADMIN)) {
+				DRM_ERROR("OA exponent would exceed the max sampling frequency (sysctl dev.i915.oa_max_sample_rate) %uHz without root privileges\n",
+					  i915_oa_max_sample_rate);
 				return -EACCES;
 			}
 
@@ -1481,6 +1511,15 @@ static struct ctl_table oa_table[] = {
 	 .extra1 = &zero,
 	 .extra2 = &one,
 	 },
+	{
+	 .procname = "oa_max_sample_rate",
+	 .data = &i915_oa_max_sample_rate,
+	 .maxlen = sizeof(i915_oa_max_sample_rate),
+	 .mode = 0644,
+	 .proc_handler = proc_dointvec_minmax,
+	 .extra1 = &zero,
+	 .extra2 = &oa_sample_rate_hard_limit,
+	 },
 	{}
 };
 
-- 
2.10.1

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v9 10/11] drm/i915: Add more Haswell OA metric sets

[Robert Bragg]

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

The code is auto generated from an XML description of metric sets,
currently maintained in gputop, ref:

 https://github.com/rib/gputop
 > gputop-data/oa-*.xml
 > scripts/i915-perf-kernelgen.py

 $ make -C gputop-data -f Makefile.xml

Signed-off-by: Robert Bragg <robert@sixbynine.org>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
---
 drivers/gpu/drm/i915/i915_oa_hsw.c | 559 ++++++++++++++++++++++++++++++++++++-
 1 file changed, 558 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/i915_oa_hsw.c b/drivers/gpu/drm/i915/i915_oa_hsw.c
index 6af25cf..4ddf756 100644
--- a/drivers/gpu/drm/i915/i915_oa_hsw.c
+++ b/drivers/gpu/drm/i915/i915_oa_hsw.c
@@ -31,9 +31,14 @@
 
 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 },
@@ -112,6 +117,298 @@ get_render_basic_mux_config(struct drm_i915_private *dev_priv,
 	return mux_config_render_basic;
 }
 
+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 },
+	{ _MMIO(0x236c), 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 const struct i915_oa_reg *
+get_compute_basic_mux_config(struct drm_i915_private *dev_priv,
+			     int *len)
+{
+	*len = ARRAY_SIZE(mux_config_compute_basic);
+	return mux_config_compute_basic;
+}
+
+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 const struct i915_oa_reg *
+get_compute_extended_mux_config(struct drm_i915_private *dev_priv,
+				int *len)
+{
+	*len = ARRAY_SIZE(mux_config_compute_extended);
+	return mux_config_compute_extended;
+}
+
+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 const struct i915_oa_reg *
+get_memory_reads_mux_config(struct drm_i915_private *dev_priv,
+			    int *len)
+{
+	*len = ARRAY_SIZE(mux_config_memory_reads);
+	return mux_config_memory_reads;
+}
+
+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 const struct i915_oa_reg *
+get_memory_writes_mux_config(struct drm_i915_private *dev_priv,
+			     int *len)
+{
+	*len = ARRAY_SIZE(mux_config_memory_writes);
+	return mux_config_memory_writes;
+}
+
+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 const struct i915_oa_reg *
+get_sampler_balance_mux_config(struct drm_i915_private *dev_priv,
+			       int *len)
+{
+	*len = ARRAY_SIZE(mux_config_sampler_balance);
+	return mux_config_sampler_balance;
+}
+
 int i915_oa_select_metric_set_hsw(struct drm_i915_private *dev_priv)
 {
 	dev_priv->perf.oa.mux_regs = NULL;
@@ -140,6 +437,106 @@ int i915_oa_select_metric_set_hsw(struct drm_i915_private *dev_priv)
 			ARRAY_SIZE(b_counter_config_render_basic);
 
 		return 0;
+	case METRIC_SET_ID_COMPUTE_BASIC:
+		dev_priv->perf.oa.mux_regs =
+			get_compute_basic_mux_config(dev_priv,
+						     &dev_priv->perf.oa.mux_regs_len);
+		if (!dev_priv->perf.oa.mux_regs) {
+			DRM_DEBUG_DRIVER("No suitable MUX config for \"COMPUTE_BASIC\" metric set");
+
+			/* EINVAL because *_register_sysfs already checked this
+			 * and so it wouldn't have been advertised so userspace and
+			 * so shouldn't have been requested
+			 */
+			return -EINVAL;
+		}
+
+		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;
+	case METRIC_SET_ID_COMPUTE_EXTENDED:
+		dev_priv->perf.oa.mux_regs =
+			get_compute_extended_mux_config(dev_priv,
+							&dev_priv->perf.oa.mux_regs_len);
+		if (!dev_priv->perf.oa.mux_regs) {
+			DRM_DEBUG_DRIVER("No suitable MUX config for \"COMPUTE_EXTENDED\" metric set");
+
+			/* EINVAL because *_register_sysfs already checked this
+			 * and so it wouldn't have been advertised so userspace and
+			 * so shouldn't have been requested
+			 */
+			return -EINVAL;
+		}
+
+		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;
+	case METRIC_SET_ID_MEMORY_READS:
+		dev_priv->perf.oa.mux_regs =
+			get_memory_reads_mux_config(dev_priv,
+						    &dev_priv->perf.oa.mux_regs_len);
+		if (!dev_priv->perf.oa.mux_regs) {
+			DRM_DEBUG_DRIVER("No suitable MUX config for \"MEMORY_READS\" metric set");
+
+			/* EINVAL because *_register_sysfs already checked this
+			 * and so it wouldn't have been advertised so userspace and
+			 * so shouldn't have been requested
+			 */
+			return -EINVAL;
+		}
+
+		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;
+	case METRIC_SET_ID_MEMORY_WRITES:
+		dev_priv->perf.oa.mux_regs =
+			get_memory_writes_mux_config(dev_priv,
+						     &dev_priv->perf.oa.mux_regs_len);
+		if (!dev_priv->perf.oa.mux_regs) {
+			DRM_DEBUG_DRIVER("No suitable MUX config for \"MEMORY_WRITES\" metric set");
+
+			/* EINVAL because *_register_sysfs already checked this
+			 * and so it wouldn't have been advertised so userspace and
+			 * so shouldn't have been requested
+			 */
+			return -EINVAL;
+		}
+
+		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;
+	case METRIC_SET_ID_SAMPLER_BALANCE:
+		dev_priv->perf.oa.mux_regs =
+			get_sampler_balance_mux_config(dev_priv,
+						       &dev_priv->perf.oa.mux_regs_len);
+		if (!dev_priv->perf.oa.mux_regs) {
+			DRM_DEBUG_DRIVER("No suitable MUX config for \"SAMPLER_BALANCE\" metric set");
+
+			/* EINVAL because *_register_sysfs already checked this
+			 * and so it wouldn't have been advertised so userspace and
+			 * so shouldn't have been requested
+			 */
+			return -EINVAL;
+		}
+
+		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;
 	default:
 		return -ENODEV;
 	}
@@ -167,6 +564,116 @@ static struct attribute_group group_render_basic = {
 	.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 = 0444 },
+	.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 = 0444 },
+	.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 = 0444 },
+	.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 = 0444 },
+	.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 = 0444 },
+	.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_register_sysfs_hsw(struct drm_i915_private *dev_priv)
 {
@@ -178,9 +685,49 @@ i915_perf_register_sysfs_hsw(struct drm_i915_private *dev_priv)
 		if (ret)
 			goto error_render_basic;
 	}
+	if (get_compute_basic_mux_config(dev_priv, &mux_len)) {
+		ret = sysfs_create_group(dev_priv->perf.metrics_kobj, &group_compute_basic);
+		if (ret)
+			goto error_compute_basic;
+	}
+	if (get_compute_extended_mux_config(dev_priv, &mux_len)) {
+		ret = sysfs_create_group(dev_priv->perf.metrics_kobj, &group_compute_extended);
+		if (ret)
+			goto error_compute_extended;
+	}
+	if (get_memory_reads_mux_config(dev_priv, &mux_len)) {
+		ret = sysfs_create_group(dev_priv->perf.metrics_kobj, &group_memory_reads);
+		if (ret)
+			goto error_memory_reads;
+	}
+	if (get_memory_writes_mux_config(dev_priv, &mux_len)) {
+		ret = sysfs_create_group(dev_priv->perf.metrics_kobj, &group_memory_writes);
+		if (ret)
+			goto error_memory_writes;
+	}
+	if (get_sampler_balance_mux_config(dev_priv, &mux_len)) {
+		ret = sysfs_create_group(dev_priv->perf.metrics_kobj, &group_sampler_balance);
+		if (ret)
+			goto error_sampler_balance;
+	}
 
 	return 0;
 
+error_sampler_balance:
+	if (get_sampler_balance_mux_config(dev_priv, &mux_len))
+		sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_memory_writes);
+error_memory_writes:
+	if (get_sampler_balance_mux_config(dev_priv, &mux_len))
+		sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_memory_reads);
+error_memory_reads:
+	if (get_sampler_balance_mux_config(dev_priv, &mux_len))
+		sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_compute_extended);
+error_compute_extended:
+	if (get_sampler_balance_mux_config(dev_priv, &mux_len))
+		sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_compute_basic);
+error_compute_basic:
+	if (get_sampler_balance_mux_config(dev_priv, &mux_len))
+		sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_render_basic);
 error_render_basic:
 	return ret;
 }
@@ -192,4 +739,14 @@ i915_perf_unregister_sysfs_hsw(struct drm_i915_private *dev_priv)
 
 	if (get_render_basic_mux_config(dev_priv, &mux_len))
 		sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_render_basic);
+	if (get_compute_basic_mux_config(dev_priv, &mux_len))
+		sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_compute_basic);
+	if (get_compute_extended_mux_config(dev_priv, &mux_len))
+		sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_compute_extended);
+	if (get_memory_reads_mux_config(dev_priv, &mux_len))
+		sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_memory_reads);
+	if (get_memory_writes_mux_config(dev_priv, &mux_len))
+		sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_memory_writes);
+	if (get_sampler_balance_mux_config(dev_priv, &mux_len))
+		sysfs_remove_group(dev_priv->perf.metrics_kobj, &group_sampler_balance);
 }
-- 
2.10.1

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v9 11/11] drm/i915: Add a kerneldoc summary for i915_perf.c

[Robert Bragg]

In particular this tries to capture for posterity some of the early
challenges we had with using the core perf infrastructure in case we
ever want to revisit adapting perf for device metrics.

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Robert Bragg <robert@sixbynine.org>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
---
 drivers/gpu/drm/i915/i915_perf.c | 163 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 163 insertions(+)

diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c
index 1a87fe9..9551282 100644
--- a/drivers/gpu/drm/i915/i915_perf.c
+++ b/drivers/gpu/drm/i915/i915_perf.c
@@ -24,6 +24,169 @@
  *   Robert Bragg <robert@sixbynine.org>
  */
 
+
+/**
+ * DOC: i915 Perf, streaming API for GPU metrics
+ *
+ * Gen graphics supports a large number of performance counters that can help
+ * driver and application developers understand and optimize their use of the
+ * GPU.
+ *
+ * This i915 perf interface enables userspace to configure and open a file
+ * descriptor representing a stream of GPU metrics which can then be read() as
+ * a stream of sample records.
+ *
+ * The interface is particularly suited to exposing buffered metrics that are
+ * captured by DMA from the GPU, unsynchronized with and unrelated to the CPU.
+ *
+ * Streams representing a single context are accessible to applications with a
+ * corresponding drm file descriptor, such that OpenGL can use the interface
+ * without special privileges. Access to system-wide metrics requires root
+ * privileges by default, unless changed via the dev.i915.perf_event_paranoid
+ * sysctl option.
+ *
+ *
+ * The interface was initially inspired by the core Perf infrastructure but
+ * some notable differences are:
+ *
+ * i915 perf file descriptors represent a "stream" instead of an "event"; where
+ * a perf event primarily corresponds to a single 64bit value, while a stream
+ * might sample sets of tightly-coupled counters, depending on the
+ * configuration.  For example the Gen OA unit isn't designed to support
+ * orthogonal configurations of individual counters; it's configured for a set
+ * of related counters. Samples for an i915 perf stream capturing OA metrics
+ * will include a set of counter values packed in a compact HW specific format.
+ * The OA unit supports a number of different packing formats which can be
+ * selected by the user opening the stream. Perf has support for grouping
+ * events, but each event in the group is configured, validated and
+ * authenticated individually with separate system calls.
+ *
+ * i915 perf stream configurations are provided as an array of u64 (key,value)
+ * pairs, instead of a fixed struct with multiple miscellaneous config members,
+ * interleaved with event-type specific members.
+ *
+ * i915 perf doesn't support exposing metrics via an mmap'd circular buffer.
+ * The supported metrics are being written to memory by the GPU unsynchronized
+ * with the CPU, using HW specific packing formats for counter sets. Sometimes
+ * the constraints on HW configuration require reports to be filtered before it
+ * would be acceptable to expose them to unprivileged applications - to hide
+ * the metrics of other processes/contexts. For these use cases a read() based
+ * interface is a good fit, and provides an opportunity to filter data as it
+ * gets copied from the GPU mapped buffers to userspace buffers.
+ *
+ *
+ * Some notes regarding Linux Perf:
+ * --------------------------------
+ *
+ * The first prototype of this driver was based on the core perf
+ * infrastructure, and while we did make that mostly work, with some changes to
+ * perf, we found we were breaking or working around too many assumptions baked
+ * into perf's currently cpu centric design.
+ *
+ * In the end we didn't see a clear benefit to making perf's implementation and
+ * interface more complex by changing design assumptions while we knew we still
+ * wouldn't be able to use any existing perf based userspace tools.
+ *
+ * Also considering the Gen specific nature of the Observability hardware and
+ * how userspace will sometimes need to combine i915 perf OA metrics with
+ * side-band OA data captured via MI_REPORT_PERF_COUNT commands; we're
+ * expecting the interface to be used by a platform specific userspace such as
+ * OpenGL or tools. This is to say; we aren't inherently missing out on having
+ * a standard vendor/architecture agnostic interface by not using perf.
+ *
+ *
+ * For posterity, in case we might re-visit trying to adapt core perf to be
+ * better suited to exposing i915 metrics these were the main pain points we
+ * hit:
+ *
+ * - The perf based OA PMU driver broke some significant design assumptions:
+ *
+ *   Existing perf pmus are used for profiling work on a cpu and we were
+ *   introducing the idea of _IS_DEVICE pmus with different security
+ *   implications, the need to fake cpu-related data (such as user/kernel
+ *   registers) to fit with perf's current design, and adding _DEVICE records
+ *   as a way to forward device-specific status records.
+ *
+ *   The OA unit writes reports of counters into a circular buffer, without
+ *   involvement from the CPU, making our PMU driver the first of a kind.
+ *
+ *   Given the way we were periodically forward data from the GPU-mapped, OA
+ *   buffer to perf's buffer, those bursts of sample writes looked to perf like
+ *   we were sampling too fast and so we had to subvert its throttling checks.
+ *
+ *   Perf supports groups of counters and allows those to be read via
+ *   transactions internally but transactions currently seem designed to be
+ *   explicitly initiated from the cpu (say in response to a userspace read())
+ *   and while we could pull a report out of the OA buffer we can't
+ *   trigger a report from the cpu on demand.
+ *
+ *   Related to being report based; the OA counters are configured in HW as a
+ *   set while perf generally expects counter configurations to be orthogonal.
+ *   Although counters can be associated with a group leader as they are
+ *   opened, there's no clear precedent for being able to provide group-wide
+ *   configuration attributes (for example we want to let userspace choose the
+ *   OA unit report format used to capture all counters in a set, or specify a
+ *   GPU context to filter metrics on). We avoided using perf's grouping
+ *   feature and forwarded OA reports to userspace via perf's 'raw' sample
+ *   field. This suited our userspace well considering how coupled the counters
+ *   are when dealing with normalizing. It would be inconvenient to split
+ *   counters up into separate events, only to require userspace to recombine
+ *   them. For Mesa it's also convenient to be forwarded raw, periodic reports
+ *   for combining with the side-band raw reports it captures using
+ *   MI_REPORT_PERF_COUNT commands.
+ *
+ *   _ As a side note on perf's grouping feature; there was also some concern
+ *     that using PERF_FORMAT_GROUP as a way to pack together counter values
+ *     would quite drastically inflate our sample sizes, which would likely
+ *     lower the effective sampling resolutions we could use when the available
+ *     memory bandwidth is limited.
+ *
+ *     With the OA unit's report formats, counters are packed together as 32
+ *     or 40bit values, with the largest report size being 256 bytes.
+ *
+ *     PERF_FORMAT_GROUP values are 64bit, but there doesn't appear to be a
+ *     documented ordering to the values, implying PERF_FORMAT_ID must also be
+ *     used to add a 64bit ID before each value; giving 16 bytes per counter.
+ *
+ *   Related to counter orthogonality; we can't time share the OA unit, while
+ *   event scheduling is a central design idea within perf for allowing
+ *   userspace to open + enable more events than can be configured in HW at any
+ *   one time.  The OA unit is not designed to allow re-configuration while in
+ *   use. We can't reconfigure the OA unit without losing internal OA unit
+ *   state which we can't access explicitly to save and restore. Reconfiguring
+ *   the OA unit is also relatively slow, involving ~100 register writes. From
+ *   userspace Mesa also depends on a stable OA configuration when emitting
+ *   MI_REPORT_PERF_COUNT commands and importantly the OA unit can't be
+ *   disabled while there are outstanding MI_RPC commands lest we hang the
+ *   command streamer.
+ *
+ *   The contents of sample records aren't extensible by device drivers (i.e.
+ *   the sample_type bits). As an example; Sourab Gupta had been looking to
+ *   attach GPU timestamps to our OA samples. We were shoehorning OA reports
+ *   into sample records by using the 'raw' field, but it's tricky to pack more
+ *   than one thing into this field because events/core.c currently only lets a
+ *   pmu give a single raw data pointer plus len which will be copied into the
+ *   ring buffer. To include more than the OA report we'd have to copy the
+ *   report into an intermediate larger buffer. I'd been considering allowing a
+ *   vector of data+len values to be specified for copying the raw data, but
+ *   it felt like a kludge to being using the raw field for this purpose.
+ *
+ * - It felt like our perf based PMU was making some technical compromises
+ *   just for the sake of using perf:
+ *
+ *   perf_event_open() requires events to either relate to a pid or a specific
+ *   cpu core, while our device pmu related to neither.  Events opened with a
+ *   pid will be automatically enabled/disabled according to the scheduling of
+ *   that process - so not appropriate for us. When an event is related to a
+ *   cpu id, perf ensures pmu methods will be invoked via an inter process
+ *   interrupt on that core. To avoid invasive changes our userspace opened OA
+ *   perf events for a specific cpu. This was workable but it meant the
+ *   majority of the OA driver ran in atomic context, including all OA report
+ *   forwarding, which wasn't really necessary in our case and seems to make
+ *   our locking requirements somewhat complex as we handled the interaction
+ *   with the rest of the i915 driver.
+ */
+
 #include <linux/anon_inodes.h>
 #include <linux/sizes.h>
 
-- 
2.10.1

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

✗ Fi.CI.BAT: failure for Enable i915 perf stream for Haswell OA unit

[Patchwork]

== Series Details ==

Series: Enable i915 perf stream for Haswell OA unit
URL   : https://patchwork.freedesktop.org/series/14935/
State : failure

== Summary ==

Series 14935v1 Enable i915 perf stream for Haswell OA unit
https://patchwork.freedesktop.org/api/1.0/series/14935/revisions/1/mbox/

Test gem_exec_parse:
        Subgroup basic-rejected:
                pass       -> FAIL       (fi-hsw-4770r)
                pass       -> FAIL       (fi-byt-n2820)
                pass       -> FAIL       (fi-hsw-4770)
                pass       -> FAIL       (fi-ivb-3520m)
                pass       -> FAIL       (fi-ivb-3770)
                pass       -> FAIL       (fi-byt-j1900)

fi-bdw-5557u     total:241  pass:226  dwarn:0   dfail:0   fail:0   skip:15 
fi-bxt-t5700     total:241  pass:213  dwarn:0   dfail:0   fail:0   skip:28 
fi-byt-j1900     total:241  pass:212  dwarn:0   dfail:0   fail:1   skip:28 
fi-byt-n2820     total:241  pass:208  dwarn:0   dfail:0   fail:1   skip:32 
fi-hsw-4770      total:241  pass:220  dwarn:0   dfail:0   fail:1   skip:20 
fi-hsw-4770r     total:241  pass:220  dwarn:0   dfail:0   fail:1   skip:20 
fi-ilk-650       total:241  pass:188  dwarn:0   dfail:0   fail:0   skip:53 
fi-ivb-3520m     total:241  pass:218  dwarn:0   dfail:0   fail:1   skip:22 
fi-ivb-3770      total:241  pass:218  dwarn:0   dfail:0   fail:1   skip:22 
fi-skl-6260u     total:241  pass:227  dwarn:0   dfail:0   fail:0   skip:14 
fi-skl-6700hq    total:241  pass:220  dwarn:0   dfail:0   fail:0   skip:21 
fi-skl-6700k     total:241  pass:219  dwarn:1   dfail:0   fail:0   skip:21 
fi-skl-6770hq    total:241  pass:227  dwarn:0   dfail:0   fail:0   skip:14 
fi-snb-2520m     total:241  pass:209  dwarn:0   dfail:0   fail:0   skip:32 
fi-snb-2600      total:241  pass:208  dwarn:0   dfail:0   fail:0   skip:33 

fb7fc6fcfc660044f50b98e63ab608afdf4eb1ef drm-intel-nightly: 2016y-11m-07d-17h-05m-46s UTC integration manifest
9fb1bf7 drm/i915: Add a kerneldoc summary for i915_perf.c
78a09b4 drm/i915: Add more Haswell OA metric sets
1167139 drm/i915: add dev.i915.oa_max_sample_rate sysctl
2046027 drm/i915: Add dev.i915.perf_stream_paranoid sysctl option
c3f7e74 drm/i915: advertise available metrics via sysfs
8a66445 drm/i915: Enable i915 perf stream for Haswell OA unit
64a5c5e drm/i915: Add 'render basic' Haswell OA unit config
2f6d0a1 drm/i915: don't whitelist oacontrol in cmd parser
735a8a5 drm/i915: return EACCES for check_cmd() failures
5d5a243 drm/i915: rename OACONTROL GEN7_OACONTROL
19883ca0 drm/i915: Add i915 perf infrastructure

== Logs ==

For more details see: https://intel-gfx-ci.01.org/CI/Patchwork_2925/
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v9 05/11] drm/i915: Add 'render basic' Haswell OA unit config

[sourab gupta]

On Mon, 2016-11-07 at 11:49 -0800, Robert Bragg wrote:
> Adds a static OA unit, MUX + B Counter configuration for basic render
> metrics on Haswell. This is auto generated from an XML
> description of metric sets, currently maintained in gputop, ref:
> 
>   https://github.com/rib/gputop
>   > gputop-data/oa-*.xml
>   > scripts/i915-perf-kernelgen.py
> 
>   $ make -C gputop-data -f Makefile.xml SYSFS=0 WHITELIST=RenderBasic
> 
> Signed-off-by: Robert Bragg <robert@sixbynine.org>
> Reviewed-by: Matthew Auld <matthew.auld@intel.com>
> ---
Looks good.
Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v9 09/11] drm/i915: add dev.i915.oa_max_sample_rate sysctl

[sourab gupta]

On Mon, 2016-11-07 at 11:49 -0800, Robert Bragg wrote:
> The maximum OA sampling frequency is now configurable via a
> dev.i915.oa_max_sample_rate sysctl parameter.
> 
> Following the precedent set by perf's similar
> kernel.perf_event_max_sample_rate the default maximum rate is 100000Hz
> 
> Signed-off-by: Robert Bragg <robert@sixbynine.org>
> ---
>  drivers/gpu/drm/i915/i915_perf.c | 61 ++++++++++++++++++++++++++++++++--------
>  1 file changed, 50 insertions(+), 11 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c
> index e51c1d8..1a87fe9 100644
> --- a/drivers/gpu/drm/i915/i915_perf.c
> +++ b/drivers/gpu/drm/i915/i915_perf.c
> @@ -82,6 +82,21 @@ static u32 i915_perf_stream_paranoid = true;
>  #define INVALID_CTX_ID 0xffffffff
>  
> 
> +/* For sysctl proc_dointvec_minmax of i915_oa_max_sample_rate
> + *
> + * 160ns is the smallest sampling period we can theoretically program the OA
> + * unit with on Haswell, corresponding to 6.25MHz.
> + */
> +static int oa_sample_rate_hard_limit = 6250000;
There's no check for 'oa_sample_rate_hard_limit' anywhere below.

> +
> +/* Theoretically we can program the OA unit to sample every 160ns but don't
> + * allow that by default unless root...
> + *
> + * The default threshold of 100000Hz is based on perf's similar
> + * kernel.perf_event_max_sample_rate sysctl parameter.
> + */
> +static u32 i915_oa_max_sample_rate = 100000;
> +
>  /* XXX: beware if future OA HW adds new report formats that the current
>   * code assumes all reports have a power-of-two size and ~(size - 1) can
>   * be used as a mask to align the OA tail pointer.
> @@ -1314,6 +1329,7 @@ static int read_properties_unlocked(struct drm_i915_private *dev_priv,
>  	}
>  
>  	for (i = 0; i < n_props; i++) {
> +		u64 oa_period, oa_freq_hz;
>  		u64 id, value;
>  		int ret;
>  
> @@ -1359,21 +1375,35 @@ static int read_properties_unlocked(struct drm_i915_private *dev_priv,
>  				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
> +			 * On Haswell the period is derived from the exponent
> +			 * as:
> +			 *
> +			 *   period = 80ns * 2^(exponent + 1)
> +			 */
> +			BUILD_BUG_ON(sizeof(oa_period) != 8);
> +			oa_period = 80ull * (2ull << value);
I assume now that there'll be a platform specific check for 80ull, while
programming oa_period, for subquent Gen8+ platforms, which should be
fine.

> +
> +			/* This check is primarily to ensure that oa_period <=
> +			 * UINT32_MAX (before passing to do_div which only
> +			 * accepts a u32 denominator), but we can also skip
> +			 * checking anything < 1Hz which implicitly can't be
> +			 * limited via an integer oa_max_sample_rate.
>  			 */
> -			if (value < 6 && !capable(CAP_SYS_ADMIN)) {
> -				DRM_ERROR("Minimum OA sampling exponent is 6 without root privileges\n");
> +			if (oa_period <= NSEC_PER_SEC) {
> +				u64 tmp = NSEC_PER_SEC;
> +				do_div(tmp, oa_period);
> +				oa_freq_hz = tmp;
> +			} else
> +				oa_freq_hz = 0;
> +
> +			if (oa_freq_hz > i915_oa_max_sample_rate &&
> +			    !capable(CAP_SYS_ADMIN)) {
> +				DRM_ERROR("OA exponent would exceed the max sampling frequency (sysctl dev.i915.oa_max_sample_rate) %uHz without root privileges\n",
> +					  i915_oa_max_sample_rate);
>  				return -EACCES;
>  			}
>  
> @@ -1481,6 +1511,15 @@ static struct ctl_table oa_table[] = {
>  	 .extra1 = &zero,
>  	 .extra2 = &one,
>  	 },
> +	{
> +	 .procname = "oa_max_sample_rate",
> +	 .data = &i915_oa_max_sample_rate,
> +	 .maxlen = sizeof(i915_oa_max_sample_rate),
> +	 .mode = 0644,
> +	 .proc_handler = proc_dointvec_minmax,
> +	 .extra1 = &zero,
> +	 .extra2 = &oa_sample_rate_hard_limit,
> +	 },
>  	{}
>  };
>  


_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v9 11/11] drm/i915: Add a kerneldoc summary for i915_perf.c

[sourab gupta]

On Mon, 2016-11-07 at 11:49 -0800, Robert Bragg wrote:
> In particular this tries to capture for posterity some of the early
> challenges we had with using the core perf infrastructure in case we
> ever want to revisit adapting perf for device metrics.
> 
> Cc: Chris Wilson <chris@chris-wilson.co.uk>
> Signed-off-by: Robert Bragg <robert@sixbynine.org>
> Reviewed-by: Matthew Auld <matthew.auld@intel.com>
> ---
Good summary of early challenges faced while adapting core perf.

Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v9 09/11] drm/i915: add dev.i915.oa_max_sample_rate sysctl

[Robert Bragg]

[-- Attachment #1.1: Type: text/plain, Size: 5788 bytes --]

On Tue, Nov 8, 2016 at 6:19 AM, sourab gupta <sourab.gupta@intel.com> wrote:

> On Mon, 2016-11-07 at 11:49 -0800, Robert Bragg wrote:
> > The maximum OA sampling frequency is now configurable via a
> > dev.i915.oa_max_sample_rate sysctl parameter.
> >
> > Following the precedent set by perf's similar
> > kernel.perf_event_max_sample_rate the default maximum rate is 100000Hz
> >
> > Signed-off-by: Robert Bragg <robert@sixbynine.org>
> > ---
> >  drivers/gpu/drm/i915/i915_perf.c | 61 ++++++++++++++++++++++++++++++
> ++--------
> >  1 file changed, 50 insertions(+), 11 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/i915/i915_perf.c
> b/drivers/gpu/drm/i915/i915_perf.c
> > index e51c1d8..1a87fe9 100644
> > --- a/drivers/gpu/drm/i915/i915_perf.c
> > +++ b/drivers/gpu/drm/i915/i915_perf.c
> > @@ -82,6 +82,21 @@ static u32 i915_perf_stream_paranoid = true;
> >  #define INVALID_CTX_ID 0xffffffff
> >
> >
> > +/* For sysctl proc_dointvec_minmax of i915_oa_max_sample_rate
> > + *
> > + * 160ns is the smallest sampling period we can theoretically program
> the OA
> > + * unit with on Haswell, corresponding to 6.25MHz.
> > + */
> > +static int oa_sample_rate_hard_limit = 6250000;
> There's no check for 'oa_sample_rate_hard_limit' anywhere below.
>

It's in the struct ctl_table oa_table[] declaration of the
"oa_max_sample_rate" paramater, assigned to .extra2 which is referenced by
the proc_dointvec_minmax validation handler for the parameter.


>
> > +
> > +/* Theoretically we can program the OA unit to sample every 160ns but
> don't
> > + * allow that by default unless root...
> > + *
> > + * The default threshold of 100000Hz is based on perf's similar
> > + * kernel.perf_event_max_sample_rate sysctl parameter.
> > + */
> > +static u32 i915_oa_max_sample_rate = 100000;
> > +
> >  /* XXX: beware if future OA HW adds new report formats that the current
> >   * code assumes all reports have a power-of-two size and ~(size - 1) can
> >   * be used as a mask to align the OA tail pointer.
> > @@ -1314,6 +1329,7 @@ static int read_properties_unlocked(struct
> drm_i915_private *dev_priv,
> >       }
> >
> >       for (i = 0; i < n_props; i++) {
> > +             u64 oa_period, oa_freq_hz;
> >               u64 id, value;
> >               int ret;
> >
> > @@ -1359,21 +1375,35 @@ static int read_properties_unlocked(struct
> drm_i915_private *dev_priv,
> >                               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
> > +                      * On Haswell the period is derived from the
> exponent
> > +                      * as:
> > +                      *
> > +                      *   period = 80ns * 2^(exponent + 1)
> > +                      */
> > +                     BUILD_BUG_ON(sizeof(oa_period) != 8);
> > +                     oa_period = 80ull * (2ull << value);
> I assume now that there'll be a platform specific check for 80ull, while
> programming oa_period, for subquent Gen8+ platforms, which should be
> fine.
>

Yeah, this code will need adapting for gen9+. I guess we'll change it to
work in terms of ((2<<exp) * NSEC_PER_SEC) / timestamp_frequency.


>
> > +
> > +                     /* This check is primarily to ensure that
> oa_period <=
> > +                      * UINT32_MAX (before passing to do_div which only
> > +                      * accepts a u32 denominator), but we can also skip
> > +                      * checking anything < 1Hz which implicitly can't
> be
> > +                      * limited via an integer oa_max_sample_rate.
> >                        */
> > -                     if (value < 6 && !capable(CAP_SYS_ADMIN)) {
> > -                             DRM_ERROR("Minimum OA sampling exponent is
> 6 without root privileges\n");
> > +                     if (oa_period <= NSEC_PER_SEC) {
> > +                             u64 tmp = NSEC_PER_SEC;
> > +                             do_div(tmp, oa_period);
> > +                             oa_freq_hz = tmp;
> > +                     } else
> > +                             oa_freq_hz = 0;
> > +
> > +                     if (oa_freq_hz > i915_oa_max_sample_rate &&
> > +                         !capable(CAP_SYS_ADMIN)) {
> > +                             DRM_ERROR("OA exponent would exceed the
> max sampling frequency (sysctl dev.i915.oa_max_sample_rate) %uHz without
> root privileges\n",
> > +                                       i915_oa_max_sample_rate);
> >                               return -EACCES;
> >                       }
> >
> > @@ -1481,6 +1511,15 @@ static struct ctl_table oa_table[] = {
> >        .extra1 = &zero,
> >        .extra2 = &one,
> >        },
> > +     {
> > +      .procname = "oa_max_sample_rate",
> > +      .data = &i915_oa_max_sample_rate,
> > +      .maxlen = sizeof(i915_oa_max_sample_rate),
> > +      .mode = 0644,
> > +      .proc_handler = proc_dointvec_minmax,
> > +      .extra1 = &zero,
> > +      .extra2 = &oa_sample_rate_hard_limit,
> > +      },
> >       {}
> >  };
> >
>
>
>

[-- Attachment #1.2: Type: text/html, Size: 8000 bytes --]

[-- Attachment #2: Type: text/plain, Size: 160 bytes --]

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH v9 09/11] drm/i915: add dev.i915.oa_max_sample_rate sysctl

[sourab gupta]

On Tue, 2016-11-08 at 03:47 -0800, Robert Bragg wrote:
> 
> 
> On Tue, Nov 8, 2016 at 6:19 AM, sourab gupta <sourab.gupta@intel.com>
> wrote:
>         On Mon, 2016-11-07 at 11:49 -0800, Robert Bragg wrote:
>         > The maximum OA sampling frequency is now configurable via a
>         > dev.i915.oa_max_sample_rate sysctl parameter.
>         >
>         > Following the precedent set by perf's similar
>         > kernel.perf_event_max_sample_rate the default maximum rate
>         is 100000Hz
>         >
>         > Signed-off-by: Robert Bragg <robert@sixbynine.org>
>         > ---
>         >  drivers/gpu/drm/i915/i915_perf.c | 61
>         ++++++++++++++++++++++++++++++++--------
>         >  1 file changed, 50 insertions(+), 11 deletions(-)
>         >
>         > diff --git a/drivers/gpu/drm/i915/i915_perf.c
>         b/drivers/gpu/drm/i915/i915_perf.c
>         > index e51c1d8..1a87fe9 100644
>         > --- a/drivers/gpu/drm/i915/i915_perf.c
>         > +++ b/drivers/gpu/drm/i915/i915_perf.c
>         > @@ -82,6 +82,21 @@ static u32 i915_perf_stream_paranoid =
>         true;
>         >  #define INVALID_CTX_ID 0xffffffff
>         >
>         >
>         > +/* For sysctl proc_dointvec_minmax of
>         i915_oa_max_sample_rate
>         > + *
>         > + * 160ns is the smallest sampling period we can
>         theoretically program the OA
>         > + * unit with on Haswell, corresponding to 6.25MHz.
>         > + */
>         > +static int oa_sample_rate_hard_limit = 6250000;
>         There's no check for 'oa_sample_rate_hard_limit' anywhere
>         below.
> 
> 
> It's in the struct ctl_table oa_table[] declaration of the
> "oa_max_sample_rate" paramater, assigned to .extra2 which is
> referenced by the proc_dointvec_minmax validation handler for the
> parameter.
> 
Ok. Seems fine then.
>  
>         
>         > +
>         > +/* Theoretically we can program the OA unit to sample every
>         160ns but don't
>         > + * allow that by default unless root...
>         > + *
>         > + * The default threshold of 100000Hz is based on perf's
>         similar
>         > + * kernel.perf_event_max_sample_rate sysctl parameter.
>         > + */
>         > +static u32 i915_oa_max_sample_rate = 100000;
>         > +
>         >  /* XXX: beware if future OA HW adds new report formats that
>         the current
>         >   * code assumes all reports have a power-of-two size and
>         ~(size - 1) can
>         >   * be used as a mask to align the OA tail pointer.
>         > @@ -1314,6 +1329,7 @@ static int
>         read_properties_unlocked(struct drm_i915_private *dev_priv,
>         >       }
>         >
>         >       for (i = 0; i < n_props; i++) {
>         > +             u64 oa_period, oa_freq_hz;
>         >               u64 id, value;
>         >               int ret;
>         >
>         > @@ -1359,21 +1375,35 @@ static int
>         read_properties_unlocked(struct drm_i915_private *dev_priv,
>         >                               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
>         > +                      * On Haswell the period is derived
>         from the exponent
>         > +                      * as:
>         > +                      *
>         > +                      *   period = 80ns * 2^(exponent + 1)
>         > +                      */
>         > +                     BUILD_BUG_ON(sizeof(oa_period) != 8);
>         > +                     oa_period = 80ull * (2ull << value);
>         
>         I assume now that there'll be a platform specific check for
>         80ull, while
>         programming oa_period, for subquent Gen8+ platforms, which
>         should be
>         fine.
> 
> 
> Yeah, this code will need adapting for gen9+. I guess we'll change it
> to work in terms of ((2<<exp) * NSEC_PER_SEC) / timestamp_frequency.
> 
Seems reasonable.
>  
>         
>         > +
>         > +                     /* This check is primarily to ensure
>         that oa_period <=
>         > +                      * UINT32_MAX (before passing to
>         do_div which only
>         > +                      * accepts a u32 denominator), but we
>         can also skip
>         > +                      * checking anything < 1Hz which
>         implicitly can't be
>         > +                      * limited via an integer
>         oa_max_sample_rate.
>         >                        */
>         > -                     if (value < 6 && !
>         capable(CAP_SYS_ADMIN)) {
>         > -                             DRM_ERROR("Minimum OA sampling
>         exponent is 6 without root privileges\n");
>         > +                     if (oa_period <= NSEC_PER_SEC) {
>         > +                             u64 tmp = NSEC_PER_SEC;
>         > +                             do_div(tmp, oa_period);
>         > +                             oa_freq_hz = tmp;
>         > +                     } else
>         > +                             oa_freq_hz = 0;
>         > +
>         > +                     if (oa_freq_hz >
>         i915_oa_max_sample_rate &&
>         > +                         !capable(CAP_SYS_ADMIN)) {
>         > +                             DRM_ERROR("OA exponent would
>         exceed the max sampling frequency (sysctl
>         dev.i915.oa_max_sample_rate) %uHz without root privileges\n",
>         > +
>          i915_oa_max_sample_rate);
>         >                               return -EACCES;
>         >                       }
>         >
>         > @@ -1481,6 +1511,15 @@ static struct ctl_table oa_table[] =
>         {
>         >        .extra1 = &zero,
>         >        .extra2 = &one,
>         >        },
>         > +     {
>         > +      .procname = "oa_max_sample_rate",
>         > +      .data = &i915_oa_max_sample_rate,
>         > +      .maxlen = sizeof(i915_oa_max_sample_rate),
>         > +      .mode = 0644,
>         > +      .proc_handler = proc_dointvec_minmax,
>         > +      .extra1 = &zero,
>         > +      .extra2 = &oa_sample_rate_hard_limit,
>         > +      },
>         >       {}
>         >  };
>         >
>         
The patch looks good to me.
Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>
>         
>         
> 
> 


_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v2] drm/i915: don't whitelist oacontrol in cmd parser

[Robert Bragg]

This v2 patch bumps the command parser version so it can be referenced in
corresponding i-g-t gem_exec_parse changes.

--- >8 ---

Being able to program OACONTROL from a non-privileged batch buffer is
not sufficient to be able to configure the OA unit. This was originally
allowed to help enable Mesa to expose OA counters via the
INTEL_performance_query extension, but the current implementation based
on programming OACONTROL via a batch buffer isn't able to report useable
data without a more complete OA unit configuration. Mesa handles the
possibility that writes to OACONTROL may not be allowed and so only
advertises the extension after explicitly testing that a write to
OACONTROL succeeds. Based on this; removing OACONTROL from the whitelist
should be ok for userspace.

Removing this simplifies adding a new kernel api for configuring the OA
unit without needing to consider the possibility that userspace might
trample on OACONTROL state which we'd like to start managing within
the kernel instead. In particular running any Mesa based GL application
currently results in clearing OACONTROL when initializing which would
disable the capturing of metrics.

v2:
    This bumps the command parser version from 8 to 9, as the change is
    visible to userspace.

Signed-off-by: Robert Bragg <robert@sixbynine.org>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>
---
 drivers/gpu/drm/i915/i915_cmd_parser.c | 42 ++++------------------------------
 1 file changed, 5 insertions(+), 37 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_cmd_parser.c b/drivers/gpu/drm/i915/i915_cmd_parser.c
index c9d2ecd..f5762cd 100644
--- a/drivers/gpu/drm/i915/i915_cmd_parser.c
+++ b/drivers/gpu/drm/i915/i915_cmd_parser.c
@@ -450,7 +450,6 @@ static const struct drm_i915_reg_descriptor gen7_render_regs[] = {
 	REG64(PS_INVOCATION_COUNT),
 	REG64(PS_DEPTH_COUNT),
 	REG64_IDX(RING_TIMESTAMP, RENDER_RING_BASE),
-	REG32(GEN7_OACONTROL), /* Only allowed for LRI and SRM. See below. */
 	REG64(MI_PREDICATE_SRC0),
 	REG64(MI_PREDICATE_SRC1),
 	REG32(GEN7_3DPRIM_END_OFFSET),
@@ -1060,8 +1059,7 @@ bool intel_engine_needs_cmd_parser(struct intel_engine_cs *engine)
 static bool check_cmd(const struct intel_engine_cs *engine,
 		      const struct drm_i915_cmd_descriptor *desc,
 		      const u32 *cmd, u32 length,
-		      const bool is_master,
-		      bool *oacontrol_set)
+		      const bool is_master)
 {
 	if (desc->flags & CMD_DESC_SKIP)
 		return true;
@@ -1099,31 +1097,6 @@ static bool check_cmd(const struct intel_engine_cs *engine,
 			}
 
 			/*
-			 * OACONTROL requires some special handling for
-			 * writes. We want to make sure that any batch which
-			 * enables OA also disables it before the end of the
-			 * batch. The goal is to prevent one process from
-			 * snooping on the perf data from another process. To do
-			 * that, we need to check the value that will be written
-			 * to the register. Hence, limit OACONTROL writes to
-			 * only MI_LOAD_REGISTER_IMM commands.
-			 */
-			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;
-				}
-
-				if (desc->cmd.value == MI_LOAD_REGISTER_REG) {
-					DRM_DEBUG_DRIVER("CMD: Rejected LRR to OACONTROL\n");
-					return false;
-				}
-
-				if (desc->cmd.value == MI_LOAD_REGISTER_IMM(1))
-					*oacontrol_set = (cmd[offset + 1] != 0);
-			}
-
-			/*
 			 * Check the value written to the register against the
 			 * allowed mask/value pair given in the whitelist entry.
 			 */
@@ -1214,7 +1187,6 @@ int intel_engine_cmd_parser(struct intel_engine_cs *engine,
 	u32 *cmd, *batch_end;
 	struct drm_i915_cmd_descriptor default_desc = noop_desc;
 	const struct drm_i915_cmd_descriptor *desc = &default_desc;
-	bool oacontrol_set = false; /* OACONTROL tracking. See check_cmd() */
 	bool needs_clflush_after = false;
 	int ret = 0;
 
@@ -1270,8 +1242,7 @@ int intel_engine_cmd_parser(struct intel_engine_cs *engine,
 			break;
 		}
 
-		if (!check_cmd(engine, desc, cmd, length, is_master,
-			       &oacontrol_set)) {
+		if (!check_cmd(engine, desc, cmd, length, is_master)) {
 			ret = -EACCES;
 			break;
 		}
@@ -1279,11 +1250,6 @@ int intel_engine_cmd_parser(struct intel_engine_cs *engine,
 		cmd += length;
 	}
 
-	if (oacontrol_set) {
-		DRM_DEBUG_DRIVER("CMD: batch set OACONTROL but did not clear it\n");
-		ret = -EINVAL;
-	}
-
 	if (cmd >= batch_end) {
 		DRM_DEBUG_DRIVER("CMD: Got to the end of the buffer w/o a BBE cmd!\n");
 		ret = -EINVAL;
@@ -1336,6 +1302,8 @@ int i915_cmd_parser_get_version(struct drm_i915_private *dev_priv)
 	 * 8. Don't report cmd_check() failures as EINVAL errors to userspace;
 	 *    rely on the HW to NOOP disallowed commands as it would without
 	 *    the parser enabled.
+	 * 9. Don't whitelist or handle oacontrol specially, as ownership
+	 *    for oacontrol state is moving to i915-perf.
 	 */
-	return 8;
+	return 9;
 }
-- 
2.10.1

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

✗ Fi.CI.BAT: failure for Enable i915 perf stream for Haswell OA unit (rev2)

[Patchwork]

== Series Details ==

Series: Enable i915 perf stream for Haswell OA unit (rev2)
URL   : https://patchwork.freedesktop.org/series/14935/
State : failure

== Summary ==

Series 14935v2 Enable i915 perf stream for Haswell OA unit
https://patchwork.freedesktop.org/api/1.0/series/14935/revisions/2/mbox/

Test gem_exec_parse:
        Subgroup basic-rejected:
                pass       -> FAIL       (fi-ivb-3520m)
                pass       -> FAIL       (fi-byt-j1900)
                pass       -> FAIL       (fi-ivb-3770)
                pass       -> FAIL       (fi-byt-n2820)
                pass       -> FAIL       (fi-hsw-4770r)
                pass       -> FAIL       (fi-hsw-4770)
Test kms_pipe_crc_basic:
        Subgroup suspend-read-crc-pipe-b:
                dmesg-warn -> PASS       (fi-skl-6770hq)

fi-bdw-5557u     total:244  pass:229  dwarn:0   dfail:0   fail:0   skip:15 
fi-bsw-n3050     total:244  pass:204  dwarn:0   dfail:0   fail:0   skip:40 
fi-bxt-t5700     total:244  pass:216  dwarn:0   dfail:0   fail:0   skip:28 
fi-byt-j1900     total:244  pass:215  dwarn:0   dfail:0   fail:1   skip:28 
fi-byt-n2820     total:244  pass:211  dwarn:0   dfail:0   fail:1   skip:32 
fi-hsw-4770      total:244  pass:223  dwarn:0   dfail:0   fail:1   skip:20 
fi-hsw-4770r     total:244  pass:223  dwarn:0   dfail:0   fail:1   skip:20 
fi-ilk-650       total:244  pass:191  dwarn:0   dfail:0   fail:0   skip:53 
fi-ivb-3520m     total:244  pass:221  dwarn:0   dfail:0   fail:1   skip:22 
fi-ivb-3770      total:244  pass:221  dwarn:0   dfail:0   fail:1   skip:22 
fi-kbl-7200u     total:244  pass:222  dwarn:0   dfail:0   fail:0   skip:22 
fi-skl-6260u     total:244  pass:230  dwarn:0   dfail:0   fail:0   skip:14 
fi-skl-6700hq    total:244  pass:223  dwarn:0   dfail:0   fail:0   skip:21 
fi-skl-6700k     total:244  pass:222  dwarn:1   dfail:0   fail:0   skip:21 
fi-skl-6770hq    total:244  pass:230  dwarn:0   dfail:0   fail:0   skip:14 
fi-snb-2520m     total:244  pass:212  dwarn:0   dfail:0   fail:0   skip:32 
fi-snb-2600      total:244  pass:211  dwarn:0   dfail:0   fail:0   skip:33 

aa74f4d7a58d239a7ff25c02eebf00c9ceb42c95 drm-intel-nightly: 2016y-11m-08d-11h-05m-41s UTC integration manifest
6bbae39 drm/i915: Add a kerneldoc summary for i915_perf.c
8b05137 drm/i915: Add more Haswell OA metric sets
f4c6eec drm/i915: add dev.i915.oa_max_sample_rate sysctl
9f486e7 drm/i915: Add dev.i915.perf_stream_paranoid sysctl option
576e12b drm/i915: advertise available metrics via sysfs
2fe39d2 drm/i915: Enable i915 perf stream for Haswell OA unit
b478f8e drm/i915: Add 'render basic' Haswell OA unit config
bcf0893 drm/i915: don't whitelist oacontrol in cmd parser
cf8aecb drm/i915: return EACCES for check_cmd() failures
676b8a2 drm/i915: rename OACONTROL GEN7_OACONTROL
289c71d drm/i915: Add i915 perf infrastructure

== Logs ==

For more details see: https://intel-gfx-ci.01.org/CI/Patchwork_2930/
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v9 09/11] drm/i915: add dev.i915.oa_max_sample_rate sysctl

[Matthew Auld]

On 7 November 2016 at 19:49, Robert Bragg <robert@sixbynine.org> wrote:
> The maximum OA sampling frequency is now configurable via a
> dev.i915.oa_max_sample_rate sysctl parameter.
>
> Following the precedent set by perf's similar
> kernel.perf_event_max_sample_rate the default maximum rate is 100000Hz
>
> Signed-off-by: Robert Bragg <robert@sixbynine.org>
> ---
>  drivers/gpu/drm/i915/i915_perf.c | 61 ++++++++++++++++++++++++++++++++--------
>  1 file changed, 50 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c
> index e51c1d8..1a87fe9 100644
> --- a/drivers/gpu/drm/i915/i915_perf.c
> +++ b/drivers/gpu/drm/i915/i915_perf.c
> @@ -82,6 +82,21 @@ static u32 i915_perf_stream_paranoid = true;
>  #define INVALID_CTX_ID 0xffffffff
>
>
> +/* For sysctl proc_dointvec_minmax of i915_oa_max_sample_rate
> + *
> + * 160ns is the smallest sampling period we can theoretically program the OA
> + * unit with on Haswell, corresponding to 6.25MHz.
> + */
> +static int oa_sample_rate_hard_limit = 6250000;
> +
> +/* Theoretically we can program the OA unit to sample every 160ns but don't
> + * allow that by default unless root...
> + *
> + * The default threshold of 100000Hz is based on perf's similar
> + * kernel.perf_event_max_sample_rate sysctl parameter.
> + */
> +static u32 i915_oa_max_sample_rate = 100000;
> +
>  /* XXX: beware if future OA HW adds new report formats that the current
>   * code assumes all reports have a power-of-two size and ~(size - 1) can
>   * be used as a mask to align the OA tail pointer.
> @@ -1314,6 +1329,7 @@ static int read_properties_unlocked(struct drm_i915_private *dev_priv,
>         }
>
>         for (i = 0; i < n_props; i++) {
> +               u64 oa_period, oa_freq_hz;
>                 u64 id, value;
>                 int ret;
>
> @@ -1359,21 +1375,35 @@ static int read_properties_unlocked(struct drm_i915_private *dev_priv,
>                                 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
> +                        * On Haswell the period is derived from the exponent
> +                        * as:
> +                        *
> +                        *   period = 80ns * 2^(exponent + 1)
> +                        */
> +                       BUILD_BUG_ON(sizeof(oa_period) != 8);
> +                       oa_period = 80ull * (2ull << value);
> +
> +                       /* This check is primarily to ensure that oa_period <=
> +                        * UINT32_MAX (before passing to do_div which only
> +                        * accepts a u32 denominator), but we can also skip
> +                        * checking anything < 1Hz which implicitly can't be
> +                        * limited via an integer oa_max_sample_rate.
>                          */
> -                       if (value < 6 && !capable(CAP_SYS_ADMIN)) {
> -                               DRM_ERROR("Minimum OA sampling exponent is 6 without root privileges\n");
> +                       if (oa_period <= NSEC_PER_SEC) {
> +                               u64 tmp = NSEC_PER_SEC;
> +                               do_div(tmp, oa_period);
> +                               oa_freq_hz = tmp;
> +                       } else
> +                               oa_freq_hz = 0;
> +
> +                       if (oa_freq_hz > i915_oa_max_sample_rate &&
> +                           !capable(CAP_SYS_ADMIN)) {
> +                               DRM_ERROR("OA exponent would exceed the max sampling frequency (sysctl dev.i915.oa_max_sample_rate) %uHz without root privileges\n",
This line is getting a little too long.

Reviewed-by: Matthew Auld <matthew.auld@intel.com>
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v9 01/11] drm/i915: Add i915 perf infrastructure

[Matthew Auld]

On 7 November 2016 at 19:49, Robert Bragg <robert@sixbynine.org> wrote:
> 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_PROP_CTX_HANDLE,        ctx_handle,
>
>       /* Include OA reports in samples */
>       DRM_I915_PERF_PROP_SAMPLE_OA,         true,
>
>       /* OA unit configuration */
>       DRM_I915_PERF_PROP_OA_METRICS_SET,    metrics_set_id,
>       DRM_I915_PERF_PROP_OA_FORMAT,         report_format,
>       DRM_I915_PERF_PROP_OA_EXPONENT,       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_ptr = (uint64_t)properties,
>       .num_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_PROP_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.
>
> v2:
>     use i915_gem_context_get() - Chris Wilson
> v3:
>     update read() interface to avoid passing state struct - Chris Wilson
>     fix some rebase fallout, with i915-perf init/deinit
> v4:
>     s/DRM_IORW/DRM_IOW/ - Emil Velikov
>
> Signed-off-by: Robert Bragg <robert@sixbynine.org>
> Reviewed-by: Matthew Auld <matthew.auld@intel.com>
> Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>
Minor nit, there are a fair few DRM_ERROR's missing a new line.
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v9 09/11] drm/i915: add dev.i915.oa_max_sample_rate sysctl

[Robert Bragg]

[-- Attachment #1.1: Type: text/plain, Size: 5090 bytes --]

On Wed, Nov 9, 2016 at 7:52 PM, Matthew Auld <matthew.william.auld@gmail.com
> wrote:

> On 7 November 2016 at 19:49, Robert Bragg <robert@sixbynine.org> wrote:
> > The maximum OA sampling frequency is now configurable via a
> > dev.i915.oa_max_sample_rate sysctl parameter.
> >
> > Following the precedent set by perf's similar
> > kernel.perf_event_max_sample_rate the default maximum rate is 100000Hz
> >
> > Signed-off-by: Robert Bragg <robert@sixbynine.org>
> > ---
> >  drivers/gpu/drm/i915/i915_perf.c | 61 ++++++++++++++++++++++++++++++
> ++--------
> >  1 file changed, 50 insertions(+), 11 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/i915/i915_perf.c
> b/drivers/gpu/drm/i915/i915_perf.c
> > index e51c1d8..1a87fe9 100644
> > --- a/drivers/gpu/drm/i915/i915_perf.c
> > +++ b/drivers/gpu/drm/i915/i915_perf.c
> > @@ -82,6 +82,21 @@ static u32 i915_perf_stream_paranoid = true;
> >  #define INVALID_CTX_ID 0xffffffff
> >
> >
> > +/* For sysctl proc_dointvec_minmax of i915_oa_max_sample_rate
> > + *
> > + * 160ns is the smallest sampling period we can theoretically program
> the OA
> > + * unit with on Haswell, corresponding to 6.25MHz.
> > + */
> > +static int oa_sample_rate_hard_limit = 6250000;
> > +
> > +/* Theoretically we can program the OA unit to sample every 160ns but
> don't
> > + * allow that by default unless root...
> > + *
> > + * The default threshold of 100000Hz is based on perf's similar
> > + * kernel.perf_event_max_sample_rate sysctl parameter.
> > + */
> > +static u32 i915_oa_max_sample_rate = 100000;
> > +
> >  /* XXX: beware if future OA HW adds new report formats that the current
> >   * code assumes all reports have a power-of-two size and ~(size - 1) can
> >   * be used as a mask to align the OA tail pointer.
> > @@ -1314,6 +1329,7 @@ static int read_properties_unlocked(struct
> drm_i915_private *dev_priv,
> >         }
> >
> >         for (i = 0; i < n_props; i++) {
> > +               u64 oa_period, oa_freq_hz;
> >                 u64 id, value;
> >                 int ret;
> >
> > @@ -1359,21 +1375,35 @@ static int read_properties_unlocked(struct
> drm_i915_private *dev_priv,
> >                                 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
> > +                        * On Haswell the period is derived from the
> exponent
> > +                        * as:
> > +                        *
> > +                        *   period = 80ns * 2^(exponent + 1)
> > +                        */
> > +                       BUILD_BUG_ON(sizeof(oa_period) != 8);
> > +                       oa_period = 80ull * (2ull << value);
> > +
> > +                       /* This check is primarily to ensure that
> oa_period <=
> > +                        * UINT32_MAX (before passing to do_div which
> only
> > +                        * accepts a u32 denominator), but we can also
> skip
> > +                        * checking anything < 1Hz which implicitly
> can't be
> > +                        * limited via an integer oa_max_sample_rate.
> >                          */
> > -                       if (value < 6 && !capable(CAP_SYS_ADMIN)) {
> > -                               DRM_ERROR("Minimum OA sampling exponent
> is 6 without root privileges\n");
> > +                       if (oa_period <= NSEC_PER_SEC) {
> > +                               u64 tmp = NSEC_PER_SEC;
> > +                               do_div(tmp, oa_period);
> > +                               oa_freq_hz = tmp;
> > +                       } else
> > +                               oa_freq_hz = 0;
> > +
> > +                       if (oa_freq_hz > i915_oa_max_sample_rate &&
> > +                           !capable(CAP_SYS_ADMIN)) {
> > +                               DRM_ERROR("OA exponent would exceed the
> max sampling frequency (sysctl dev.i915.oa_max_sample_rate) %uHz without
> root privileges\n",
> This line is getting a little too long.
>

yeah, though the coding style has the exception for string constants so
they remain grepable which can be useful too. Could maybe remove the
bracketed naming of the corresponding sysctl parameter, though it does seem
like pertinent/useful info if someone were to hit this limit.



>
> Reviewed-by: Matthew Auld <matthew.auld@intel.com>
>

thanks

[-- Attachment #1.2: Type: text/html, Size: 7038 bytes --]

[-- Attachment #2: Type: text/plain, Size: 160 bytes --]

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v9 06/11] drm/i915: Enable i915 perf stream for Haswell OA unit

[sourab gupta]

On Mon, 2016-11-07 at 11:49 -0800, Robert Bragg wrote:
> 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.
> 
> v2:
>    Make sure to initialize ->specific_ctx_id when opening, without
>    relying on _pin_notify hook, in case ctx already pinned.
> v3:
>    Revert back to pinning ctx upfront when opening stream, removing
>    need to hook in to pinning and to update OACONTROL on the fly.
> 
> Signed-off-by: Robert Bragg <robert@sixbynine.org>
> Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
> Cc: Chris Wilson <chris@chris-wilson.co.uk>
> Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Have been working for quite some time on extending the interfaces per
the usecase of multiple concurrent streams (on different engines), and
infrastructure fits quite well for these usecases. With Chris' comments
addressed, the patch can have my r-b.
Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [Intel-gfx] [PATCH v9 01/11] drm/i915: Add i915 perf infrastructure

[Daniel Vetter]

On Wed, Nov 09, 2016 at 08:00:06PM +0000, Matthew Auld wrote:
> On 7 November 2016 at 19:49, Robert Bragg <robert@sixbynine.org> wrote:
> > 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_PROP_CTX_HANDLE,        ctx_handle,
> >
> >       /* Include OA reports in samples */
> >       DRM_I915_PERF_PROP_SAMPLE_OA,         true,
> >
> >       /* OA unit configuration */
> >       DRM_I915_PERF_PROP_OA_METRICS_SET,    metrics_set_id,
> >       DRM_I915_PERF_PROP_OA_FORMAT,         report_format,
> >       DRM_I915_PERF_PROP_OA_EXPONENT,       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_ptr = (uint64_t)properties,
> >       .num_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_PROP_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.
> >
> > v2:
> >     use i915_gem_context_get() - Chris Wilson
> > v3:
> >     update read() interface to avoid passing state struct - Chris Wilson
> >     fix some rebase fallout, with i915-perf init/deinit
> > v4:
> >     s/DRM_IORW/DRM_IOW/ - Emil Velikov
> >
> > Signed-off-by: Robert Bragg <robert@sixbynine.org>
> > Reviewed-by: Matthew Auld <matthew.auld@intel.com>
> > Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>
> Minor nit, there are a fair few DRM_ERROR's missing a new line.

Also, DRM_ERROR for userspace-triggerable failures is no good. igt
testcase are supposed to exercise all the invalid stuff, and would then
fail if you spam dmesg. Why was this not caught?

Fixup patch totally fine, but if this wasn't caught due to missing igt
that needs to be fixed, too.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH v9 01/11] drm/i915: Add i915 perf infrastructure

[Daniel Vetter]

On Tue, Nov 22, 2016 at 02:29:18PM +0100, Daniel Vetter wrote:
> On Wed, Nov 09, 2016 at 08:00:06PM +0000, Matthew Auld wrote:
> > On 7 November 2016 at 19:49, Robert Bragg <robert@sixbynine.org> wrote:
> > > 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_PROP_CTX_HANDLE,        ctx_handle,
> > >
> > >       /* Include OA reports in samples */
> > >       DRM_I915_PERF_PROP_SAMPLE_OA,         true,
> > >
> > >       /* OA unit configuration */
> > >       DRM_I915_PERF_PROP_OA_METRICS_SET,    metrics_set_id,
> > >       DRM_I915_PERF_PROP_OA_FORMAT,         report_format,
> > >       DRM_I915_PERF_PROP_OA_EXPONENT,       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_ptr = (uint64_t)properties,
> > >       .num_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_PROP_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.
> > >
> > > v2:
> > >     use i915_gem_context_get() - Chris Wilson
> > > v3:
> > >     update read() interface to avoid passing state struct - Chris Wilson
> > >     fix some rebase fallout, with i915-perf init/deinit
> > > v4:
> > >     s/DRM_IORW/DRM_IOW/ - Emil Velikov
> > >
> > > Signed-off-by: Robert Bragg <robert@sixbynine.org>
> > > Reviewed-by: Matthew Auld <matthew.auld@intel.com>
> > > Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>
> > Minor nit, there are a fair few DRM_ERROR's missing a new line.
> 
> Also, DRM_ERROR for userspace-triggerable failures is no good. igt
> testcase are supposed to exercise all the invalid stuff, and would then
> fail if you spam dmesg. Why was this not caught?
> 
> Fixup patch totally fine, but if this wasn't caught due to missing igt
> that needs to be fixed, too.

Another nitpick for the future: Enabling new features first and then
fixing up the fallout is the wrong way round, if someone bisects over this
range mesa might blow up in really bad ways.

Oh well, this has been out there for way too long, so meh.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [Intel-gfx] [PATCH v2] drm/i915: don't whitelist oacontrol in cmd parser

[Daniel Vetter]

On Tue, Nov 08, 2016 at 12:51:48PM +0000, Robert Bragg wrote:
> This v2 patch bumps the command parser version so it can be referenced in
> corresponding i-g-t gem_exec_parse changes.
> 
> --- >8 ---

Scissors cut everything below, not everything above, hence next time
around pls switch around your comment and the commit message, as-is not
much left ;-)

Fixed up while applying.
-Daniel

> 
> Being able to program OACONTROL from a non-privileged batch buffer is
> not sufficient to be able to configure the OA unit. This was originally
> allowed to help enable Mesa to expose OA counters via the
> INTEL_performance_query extension, but the current implementation based
> on programming OACONTROL via a batch buffer isn't able to report useable
> data without a more complete OA unit configuration. Mesa handles the
> possibility that writes to OACONTROL may not be allowed and so only
> advertises the extension after explicitly testing that a write to
> OACONTROL succeeds. Based on this; removing OACONTROL from the whitelist
> should be ok for userspace.
> 
> Removing this simplifies adding a new kernel api for configuring the OA
> unit without needing to consider the possibility that userspace might
> trample on OACONTROL state which we'd like to start managing within
> the kernel instead. In particular running any Mesa based GL application
> currently results in clearing OACONTROL when initializing which would
> disable the capturing of metrics.
> 
> v2:
>     This bumps the command parser version from 8 to 9, as the change is
>     visible to userspace.
> 
> Signed-off-by: Robert Bragg <robert@sixbynine.org>
> Reviewed-by: Matthew Auld <matthew.auld@intel.com>
> Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>
> ---
>  drivers/gpu/drm/i915/i915_cmd_parser.c | 42 ++++------------------------------
>  1 file changed, 5 insertions(+), 37 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_cmd_parser.c b/drivers/gpu/drm/i915/i915_cmd_parser.c
> index c9d2ecd..f5762cd 100644
> --- a/drivers/gpu/drm/i915/i915_cmd_parser.c
> +++ b/drivers/gpu/drm/i915/i915_cmd_parser.c
> @@ -450,7 +450,6 @@ static const struct drm_i915_reg_descriptor gen7_render_regs[] = {
>  	REG64(PS_INVOCATION_COUNT),
>  	REG64(PS_DEPTH_COUNT),
>  	REG64_IDX(RING_TIMESTAMP, RENDER_RING_BASE),
> -	REG32(GEN7_OACONTROL), /* Only allowed for LRI and SRM. See below. */
>  	REG64(MI_PREDICATE_SRC0),
>  	REG64(MI_PREDICATE_SRC1),
>  	REG32(GEN7_3DPRIM_END_OFFSET),
> @@ -1060,8 +1059,7 @@ bool intel_engine_needs_cmd_parser(struct intel_engine_cs *engine)
>  static bool check_cmd(const struct intel_engine_cs *engine,
>  		      const struct drm_i915_cmd_descriptor *desc,
>  		      const u32 *cmd, u32 length,
> -		      const bool is_master,
> -		      bool *oacontrol_set)
> +		      const bool is_master)
>  {
>  	if (desc->flags & CMD_DESC_SKIP)
>  		return true;
> @@ -1099,31 +1097,6 @@ static bool check_cmd(const struct intel_engine_cs *engine,
>  			}
>  
>  			/*
> -			 * OACONTROL requires some special handling for
> -			 * writes. We want to make sure that any batch which
> -			 * enables OA also disables it before the end of the
> -			 * batch. The goal is to prevent one process from
> -			 * snooping on the perf data from another process. To do
> -			 * that, we need to check the value that will be written
> -			 * to the register. Hence, limit OACONTROL writes to
> -			 * only MI_LOAD_REGISTER_IMM commands.
> -			 */
> -			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;
> -				}
> -
> -				if (desc->cmd.value == MI_LOAD_REGISTER_REG) {
> -					DRM_DEBUG_DRIVER("CMD: Rejected LRR to OACONTROL\n");
> -					return false;
> -				}
> -
> -				if (desc->cmd.value == MI_LOAD_REGISTER_IMM(1))
> -					*oacontrol_set = (cmd[offset + 1] != 0);
> -			}
> -
> -			/*
>  			 * Check the value written to the register against the
>  			 * allowed mask/value pair given in the whitelist entry.
>  			 */
> @@ -1214,7 +1187,6 @@ int intel_engine_cmd_parser(struct intel_engine_cs *engine,
>  	u32 *cmd, *batch_end;
>  	struct drm_i915_cmd_descriptor default_desc = noop_desc;
>  	const struct drm_i915_cmd_descriptor *desc = &default_desc;
> -	bool oacontrol_set = false; /* OACONTROL tracking. See check_cmd() */
>  	bool needs_clflush_after = false;
>  	int ret = 0;
>  
> @@ -1270,8 +1242,7 @@ int intel_engine_cmd_parser(struct intel_engine_cs *engine,
>  			break;
>  		}
>  
> -		if (!check_cmd(engine, desc, cmd, length, is_master,
> -			       &oacontrol_set)) {
> +		if (!check_cmd(engine, desc, cmd, length, is_master)) {
>  			ret = -EACCES;
>  			break;
>  		}
> @@ -1279,11 +1250,6 @@ int intel_engine_cmd_parser(struct intel_engine_cs *engine,
>  		cmd += length;
>  	}
>  
> -	if (oacontrol_set) {
> -		DRM_DEBUG_DRIVER("CMD: batch set OACONTROL but did not clear it\n");
> -		ret = -EINVAL;
> -	}
> -
>  	if (cmd >= batch_end) {
>  		DRM_DEBUG_DRIVER("CMD: Got to the end of the buffer w/o a BBE cmd!\n");
>  		ret = -EINVAL;
> @@ -1336,6 +1302,8 @@ int i915_cmd_parser_get_version(struct drm_i915_private *dev_priv)
>  	 * 8. Don't report cmd_check() failures as EINVAL errors to userspace;
>  	 *    rely on the HW to NOOP disallowed commands as it would without
>  	 *    the parser enabled.
> +	 * 9. Don't whitelist or handle oacontrol specially, as ownership
> +	 *    for oacontrol state is moving to i915-perf.
>  	 */
> -	return 8;
> +	return 9;
>  }
> -- 
> 2.10.1
> 
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH v9 11/11] drm/i915: Add a kerneldoc summary for i915_perf.c

[Daniel Vetter]

On Mon, Nov 07, 2016 at 07:49:57PM +0000, Robert Bragg wrote:
> In particular this tries to capture for posterity some of the early
> challenges we had with using the core perf infrastructure in case we
> ever want to revisit adapting perf for device metrics.
> 
> Cc: Chris Wilson <chris@chris-wilson.co.uk>
> Signed-off-by: Robert Bragg <robert@sixbynine.org>
> Reviewed-by: Matthew Auld <matthew.auld@intel.com>
> ---
>  drivers/gpu/drm/i915/i915_perf.c | 163 +++++++++++++++++++++++++++++++++++++++

Missing the include stanza in Documentation/gpu/i915.rst, which means this
won't show up in the docs build. Also function/struct docs are not
included either it seems.

Please check out Documentation/kernel-documentation.rst and build it all
with

$ make DOCBOOKS="" htmldocs

Again fixup patch is fine.
-Daniel

>  1 file changed, 163 insertions(+)
> 
> diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c
> index 1a87fe9..9551282 100644
> --- a/drivers/gpu/drm/i915/i915_perf.c
> +++ b/drivers/gpu/drm/i915/i915_perf.c
> @@ -24,6 +24,169 @@
>   *   Robert Bragg <robert@sixbynine.org>
>   */
>  
> +
> +/**
> + * DOC: i915 Perf, streaming API for GPU metrics
> + *
> + * Gen graphics supports a large number of performance counters that can help
> + * driver and application developers understand and optimize their use of the
> + * GPU.
> + *
> + * This i915 perf interface enables userspace to configure and open a file
> + * descriptor representing a stream of GPU metrics which can then be read() as
> + * a stream of sample records.
> + *
> + * The interface is particularly suited to exposing buffered metrics that are
> + * captured by DMA from the GPU, unsynchronized with and unrelated to the CPU.
> + *
> + * Streams representing a single context are accessible to applications with a
> + * corresponding drm file descriptor, such that OpenGL can use the interface
> + * without special privileges. Access to system-wide metrics requires root
> + * privileges by default, unless changed via the dev.i915.perf_event_paranoid
> + * sysctl option.
> + *
> + *
> + * The interface was initially inspired by the core Perf infrastructure but
> + * some notable differences are:
> + *
> + * i915 perf file descriptors represent a "stream" instead of an "event"; where
> + * a perf event primarily corresponds to a single 64bit value, while a stream
> + * might sample sets of tightly-coupled counters, depending on the
> + * configuration.  For example the Gen OA unit isn't designed to support
> + * orthogonal configurations of individual counters; it's configured for a set
> + * of related counters. Samples for an i915 perf stream capturing OA metrics
> + * will include a set of counter values packed in a compact HW specific format.
> + * The OA unit supports a number of different packing formats which can be
> + * selected by the user opening the stream. Perf has support for grouping
> + * events, but each event in the group is configured, validated and
> + * authenticated individually with separate system calls.
> + *
> + * i915 perf stream configurations are provided as an array of u64 (key,value)
> + * pairs, instead of a fixed struct with multiple miscellaneous config members,
> + * interleaved with event-type specific members.
> + *
> + * i915 perf doesn't support exposing metrics via an mmap'd circular buffer.
> + * The supported metrics are being written to memory by the GPU unsynchronized
> + * with the CPU, using HW specific packing formats for counter sets. Sometimes
> + * the constraints on HW configuration require reports to be filtered before it
> + * would be acceptable to expose them to unprivileged applications - to hide
> + * the metrics of other processes/contexts. For these use cases a read() based
> + * interface is a good fit, and provides an opportunity to filter data as it
> + * gets copied from the GPU mapped buffers to userspace buffers.
> + *
> + *
> + * Some notes regarding Linux Perf:
> + * --------------------------------
> + *
> + * The first prototype of this driver was based on the core perf
> + * infrastructure, and while we did make that mostly work, with some changes to
> + * perf, we found we were breaking or working around too many assumptions baked
> + * into perf's currently cpu centric design.
> + *
> + * In the end we didn't see a clear benefit to making perf's implementation and
> + * interface more complex by changing design assumptions while we knew we still
> + * wouldn't be able to use any existing perf based userspace tools.
> + *
> + * Also considering the Gen specific nature of the Observability hardware and
> + * how userspace will sometimes need to combine i915 perf OA metrics with
> + * side-band OA data captured via MI_REPORT_PERF_COUNT commands; we're
> + * expecting the interface to be used by a platform specific userspace such as
> + * OpenGL or tools. This is to say; we aren't inherently missing out on having
> + * a standard vendor/architecture agnostic interface by not using perf.
> + *
> + *
> + * For posterity, in case we might re-visit trying to adapt core perf to be
> + * better suited to exposing i915 metrics these were the main pain points we
> + * hit:
> + *
> + * - The perf based OA PMU driver broke some significant design assumptions:
> + *
> + *   Existing perf pmus are used for profiling work on a cpu and we were
> + *   introducing the idea of _IS_DEVICE pmus with different security
> + *   implications, the need to fake cpu-related data (such as user/kernel
> + *   registers) to fit with perf's current design, and adding _DEVICE records
> + *   as a way to forward device-specific status records.
> + *
> + *   The OA unit writes reports of counters into a circular buffer, without
> + *   involvement from the CPU, making our PMU driver the first of a kind.
> + *
> + *   Given the way we were periodically forward data from the GPU-mapped, OA
> + *   buffer to perf's buffer, those bursts of sample writes looked to perf like
> + *   we were sampling too fast and so we had to subvert its throttling checks.
> + *
> + *   Perf supports groups of counters and allows those to be read via
> + *   transactions internally but transactions currently seem designed to be
> + *   explicitly initiated from the cpu (say in response to a userspace read())
> + *   and while we could pull a report out of the OA buffer we can't
> + *   trigger a report from the cpu on demand.
> + *
> + *   Related to being report based; the OA counters are configured in HW as a
> + *   set while perf generally expects counter configurations to be orthogonal.
> + *   Although counters can be associated with a group leader as they are
> + *   opened, there's no clear precedent for being able to provide group-wide
> + *   configuration attributes (for example we want to let userspace choose the
> + *   OA unit report format used to capture all counters in a set, or specify a
> + *   GPU context to filter metrics on). We avoided using perf's grouping
> + *   feature and forwarded OA reports to userspace via perf's 'raw' sample
> + *   field. This suited our userspace well considering how coupled the counters
> + *   are when dealing with normalizing. It would be inconvenient to split
> + *   counters up into separate events, only to require userspace to recombine
> + *   them. For Mesa it's also convenient to be forwarded raw, periodic reports
> + *   for combining with the side-band raw reports it captures using
> + *   MI_REPORT_PERF_COUNT commands.
> + *
> + *   _ As a side note on perf's grouping feature; there was also some concern
> + *     that using PERF_FORMAT_GROUP as a way to pack together counter values
> + *     would quite drastically inflate our sample sizes, which would likely
> + *     lower the effective sampling resolutions we could use when the available
> + *     memory bandwidth is limited.
> + *
> + *     With the OA unit's report formats, counters are packed together as 32
> + *     or 40bit values, with the largest report size being 256 bytes.
> + *
> + *     PERF_FORMAT_GROUP values are 64bit, but there doesn't appear to be a
> + *     documented ordering to the values, implying PERF_FORMAT_ID must also be
> + *     used to add a 64bit ID before each value; giving 16 bytes per counter.
> + *
> + *   Related to counter orthogonality; we can't time share the OA unit, while
> + *   event scheduling is a central design idea within perf for allowing
> + *   userspace to open + enable more events than can be configured in HW at any
> + *   one time.  The OA unit is not designed to allow re-configuration while in
> + *   use. We can't reconfigure the OA unit without losing internal OA unit
> + *   state which we can't access explicitly to save and restore. Reconfiguring
> + *   the OA unit is also relatively slow, involving ~100 register writes. From
> + *   userspace Mesa also depends on a stable OA configuration when emitting
> + *   MI_REPORT_PERF_COUNT commands and importantly the OA unit can't be
> + *   disabled while there are outstanding MI_RPC commands lest we hang the
> + *   command streamer.
> + *
> + *   The contents of sample records aren't extensible by device drivers (i.e.
> + *   the sample_type bits). As an example; Sourab Gupta had been looking to
> + *   attach GPU timestamps to our OA samples. We were shoehorning OA reports
> + *   into sample records by using the 'raw' field, but it's tricky to pack more
> + *   than one thing into this field because events/core.c currently only lets a
> + *   pmu give a single raw data pointer plus len which will be copied into the
> + *   ring buffer. To include more than the OA report we'd have to copy the
> + *   report into an intermediate larger buffer. I'd been considering allowing a
> + *   vector of data+len values to be specified for copying the raw data, but
> + *   it felt like a kludge to being using the raw field for this purpose.
> + *
> + * - It felt like our perf based PMU was making some technical compromises
> + *   just for the sake of using perf:
> + *
> + *   perf_event_open() requires events to either relate to a pid or a specific
> + *   cpu core, while our device pmu related to neither.  Events opened with a
> + *   pid will be automatically enabled/disabled according to the scheduling of
> + *   that process - so not appropriate for us. When an event is related to a
> + *   cpu id, perf ensures pmu methods will be invoked via an inter process
> + *   interrupt on that core. To avoid invasive changes our userspace opened OA
> + *   perf events for a specific cpu. This was workable but it meant the
> + *   majority of the OA driver ran in atomic context, including all OA report
> + *   forwarding, which wasn't really necessary in our case and seems to make
> + *   our locking requirements somewhat complex as we handled the interaction
> + *   with the rest of the i915 driver.
> + */
> +
>  #include <linux/anon_inodes.h>
>  #include <linux/sizes.h>
>  
> -- 
> 2.10.1
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [Intel-gfx] [PATCH v9 01/11] drm/i915: Add i915 perf infrastructure

[Robert Bragg]

[-- Attachment #1.1: Type: text/plain, Size: 3825 bytes --]

On Tue, Nov 22, 2016 at 1:31 PM, Daniel Vetter <daniel@ffwll.ch> wrote:

> On Tue, Nov 22, 2016 at 02:29:18PM +0100, Daniel Vetter wrote:
> > On Wed, Nov 09, 2016 at 08:00:06PM +0000, Matthew Auld wrote:
> > > On 7 November 2016 at 19:49, Robert Bragg <robert@sixbynine.org>
> wrote:
> > > > 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_PROP_CTX_HANDLE,        ctx_handle,
> > > >
> > > >       /* Include OA reports in samples */
> > > >       DRM_I915_PERF_PROP_SAMPLE_OA,         true,
> > > >
> > > >       /* OA unit configuration */
> > > >       DRM_I915_PERF_PROP_OA_METRICS_SET,    metrics_set_id,
> > > >       DRM_I915_PERF_PROP_OA_FORMAT,         report_format,
> > > >       DRM_I915_PERF_PROP_OA_EXPONENT,       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_ptr = (uint64_t)properties,
> > > >       .num_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_PROP_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.
> > > >
> > > > v2:
> > > >     use i915_gem_context_get() - Chris Wilson
> > > > v3:
> > > >     update read() interface to avoid passing state struct - Chris
> Wilson
> > > >     fix some rebase fallout, with i915-perf init/deinit
> > > > v4:
> > > >     s/DRM_IORW/DRM_IOW/ - Emil Velikov
> > > >
> > > > Signed-off-by: Robert Bragg <robert@sixbynine.org>
> > > > Reviewed-by: Matthew Auld <matthew.auld@intel.com>
> > > > Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>
> > > Minor nit, there are a fair few DRM_ERROR's missing a new line.
> >
> > Also, DRM_ERROR for userspace-triggerable failures is no good. igt
> > testcase are supposed to exercise all the invalid stuff, and would then
> > fail if you spam dmesg. Why was this not caught?
> >
> > Fixup patch totally fine, but if this wasn't caught due to missing igt
> > that needs to be fixed, too.
>
> Another nitpick for the future: Enabling new features first and then
> fixing up the fallout is the wrong way round, if someone bisects over this
> range mesa might blow up in really bad ways.
>
> Oh well, this has been out there for way too long, so meh.
>

Fwiw I'm aware of this, and think I've ordered the patches correctly to
avoid bisect problems in Mesa / userspace. This infrastructure patch should
have no fallout to fix for userspace. The command parser changes that
affect userspace were done before adding oacontrol usage to i915-perf and
the cmd parser's EINVAL reporting for access failures was changed *before*
removing oacontrol from the whitelist.

Did I overlook something in particular?

- Robert



> -Daniel
> --
> Daniel Vetter
> Software Engineer, Intel Corporation
> http://blog.ffwll.ch
>

[-- Attachment #1.2: Type: text/html, Size: 5541 bytes --]

[-- Attachment #2: Type: text/plain, Size: 160 bytes --]

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH v2] drm/i915: don't whitelist oacontrol in cmd parser

[Robert Bragg]

[-- Attachment #1.1: Type: text/plain, Size: 7333 bytes --]

On Tue, Nov 22, 2016 at 1:34 PM, Daniel Vetter <daniel@ffwll.ch> wrote:

> On Tue, Nov 08, 2016 at 12:51:48PM +0000, Robert Bragg wrote:
> > This v2 patch bumps the command parser version so it can be referenced in
> > corresponding i-g-t gem_exec_parse changes.
> >
> > --- >8 ---
>
> Scissors cut everything below, not everything above, hence next time
> around pls switch around your comment and the commit message, as-is not
> much left ;-)
>

Hmm, they cut away what's above and keep what's below in my experience -
what command are you seeing the opposite with?

I just double checked this with git am --scissors

- Robert



>
> Fixed up while applying.
> -Daniel
>
> >
> > Being able to program OACONTROL from a non-privileged batch buffer is
> > not sufficient to be able to configure the OA unit. This was originally
> > allowed to help enable Mesa to expose OA counters via the
> > INTEL_performance_query extension, but the current implementation based
> > on programming OACONTROL via a batch buffer isn't able to report useable
> > data without a more complete OA unit configuration. Mesa handles the
> > possibility that writes to OACONTROL may not be allowed and so only
> > advertises the extension after explicitly testing that a write to
> > OACONTROL succeeds. Based on this; removing OACONTROL from the whitelist
> > should be ok for userspace.
> >
> > Removing this simplifies adding a new kernel api for configuring the OA
> > unit without needing to consider the possibility that userspace might
> > trample on OACONTROL state which we'd like to start managing within
> > the kernel instead. In particular running any Mesa based GL application
> > currently results in clearing OACONTROL when initializing which would
> > disable the capturing of metrics.
> >
> > v2:
> >     This bumps the command parser version from 8 to 9, as the change is
> >     visible to userspace.
> >
> > Signed-off-by: Robert Bragg <robert@sixbynine.org>
> > Reviewed-by: Matthew Auld <matthew.auld@intel.com>
> > Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>
> > ---
> >  drivers/gpu/drm/i915/i915_cmd_parser.c | 42
> ++++------------------------------
> >  1 file changed, 5 insertions(+), 37 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/i915/i915_cmd_parser.c
> b/drivers/gpu/drm/i915/i915_cmd_parser.c
> > index c9d2ecd..f5762cd 100644
> > --- a/drivers/gpu/drm/i915/i915_cmd_parser.c
> > +++ b/drivers/gpu/drm/i915/i915_cmd_parser.c
> > @@ -450,7 +450,6 @@ static const struct drm_i915_reg_descriptor
> gen7_render_regs[] = {
> >       REG64(PS_INVOCATION_COUNT),
> >       REG64(PS_DEPTH_COUNT),
> >       REG64_IDX(RING_TIMESTAMP, RENDER_RING_BASE),
> > -     REG32(GEN7_OACONTROL), /* Only allowed for LRI and SRM. See below.
> */
> >       REG64(MI_PREDICATE_SRC0),
> >       REG64(MI_PREDICATE_SRC1),
> >       REG32(GEN7_3DPRIM_END_OFFSET),
> > @@ -1060,8 +1059,7 @@ bool intel_engine_needs_cmd_parser(struct
> intel_engine_cs *engine)
> >  static bool check_cmd(const struct intel_engine_cs *engine,
> >                     const struct drm_i915_cmd_descriptor *desc,
> >                     const u32 *cmd, u32 length,
> > -                   const bool is_master,
> > -                   bool *oacontrol_set)
> > +                   const bool is_master)
> >  {
> >       if (desc->flags & CMD_DESC_SKIP)
> >               return true;
> > @@ -1099,31 +1097,6 @@ static bool check_cmd(const struct
> intel_engine_cs *engine,
> >                       }
> >
> >                       /*
> > -                      * OACONTROL requires some special handling for
> > -                      * writes. We want to make sure that any batch
> which
> > -                      * enables OA also disables it before the end of
> the
> > -                      * batch. The goal is to prevent one process from
> > -                      * snooping on the perf data from another process.
> To do
> > -                      * that, we need to check the value that will be
> written
> > -                      * to the register. Hence, limit OACONTROL writes
> to
> > -                      * only MI_LOAD_REGISTER_IMM commands.
> > -                      */
> > -                     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;
> > -                             }
> > -
> > -                             if (desc->cmd.value ==
> MI_LOAD_REGISTER_REG) {
> > -                                     DRM_DEBUG_DRIVER("CMD: Rejected
> LRR to OACONTROL\n");
> > -                                     return false;
> > -                             }
> > -
> > -                             if (desc->cmd.value ==
> MI_LOAD_REGISTER_IMM(1))
> > -                                     *oacontrol_set = (cmd[offset + 1]
> != 0);
> > -                     }
> > -
> > -                     /*
> >                        * Check the value written to the register against
> the
> >                        * allowed mask/value pair given in the whitelist
> entry.
> >                        */
> > @@ -1214,7 +1187,6 @@ int intel_engine_cmd_parser(struct intel_engine_cs
> *engine,
> >       u32 *cmd, *batch_end;
> >       struct drm_i915_cmd_descriptor default_desc = noop_desc;
> >       const struct drm_i915_cmd_descriptor *desc = &default_desc;
> > -     bool oacontrol_set = false; /* OACONTROL tracking. See check_cmd()
> */
> >       bool needs_clflush_after = false;
> >       int ret = 0;
> >
> > @@ -1270,8 +1242,7 @@ int intel_engine_cmd_parser(struct intel_engine_cs
> *engine,
> >                       break;
> >               }
> >
> > -             if (!check_cmd(engine, desc, cmd, length, is_master,
> > -                            &oacontrol_set)) {
> > +             if (!check_cmd(engine, desc, cmd, length, is_master)) {
> >                       ret = -EACCES;
> >                       break;
> >               }
> > @@ -1279,11 +1250,6 @@ int intel_engine_cmd_parser(struct
> intel_engine_cs *engine,
> >               cmd += length;
> >       }
> >
> > -     if (oacontrol_set) {
> > -             DRM_DEBUG_DRIVER("CMD: batch set OACONTROL but did not
> clear it\n");
> > -             ret = -EINVAL;
> > -     }
> > -
> >       if (cmd >= batch_end) {
> >               DRM_DEBUG_DRIVER("CMD: Got to the end of the buffer w/o a
> BBE cmd!\n");
> >               ret = -EINVAL;
> > @@ -1336,6 +1302,8 @@ int i915_cmd_parser_get_version(struct
> drm_i915_private *dev_priv)
> >        * 8. Don't report cmd_check() failures as EINVAL errors to
> userspace;
> >        *    rely on the HW to NOOP disallowed commands as it would
> without
> >        *    the parser enabled.
> > +      * 9. Don't whitelist or handle oacontrol specially, as ownership
> > +      *    for oacontrol state is moving to i915-perf.
> >        */
> > -     return 8;
> > +     return 9;
> >  }
> > --
> > 2.10.1
> >
> > _______________________________________________
> > Intel-gfx mailing list
> > Intel-gfx@lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/intel-gfx
>
> --
> Daniel Vetter
> Software Engineer, Intel Corporation
> http://blog.ffwll.ch
>

[-- Attachment #1.2: Type: text/html, Size: 9917 bytes --]

[-- Attachment #2: Type: text/plain, Size: 160 bytes --]

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v9 01/11] drm/i915: Add i915 perf infrastructure

[Daniel Vetter]

On Tue, Nov 22, 2016 at 02:02:38PM +0000, Robert Bragg wrote:
> On Tue, Nov 22, 2016 at 1:31 PM, Daniel Vetter <daniel@ffwll.ch> wrote:
> 
> > On Tue, Nov 22, 2016 at 02:29:18PM +0100, Daniel Vetter wrote:
> > > On Wed, Nov 09, 2016 at 08:00:06PM +0000, Matthew Auld wrote:
> > > > On 7 November 2016 at 19:49, Robert Bragg <robert@sixbynine.org>
> > wrote:
> > > > > 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_PROP_CTX_HANDLE,        ctx_handle,
> > > > >
> > > > >       /* Include OA reports in samples */
> > > > >       DRM_I915_PERF_PROP_SAMPLE_OA,         true,
> > > > >
> > > > >       /* OA unit configuration */
> > > > >       DRM_I915_PERF_PROP_OA_METRICS_SET,    metrics_set_id,
> > > > >       DRM_I915_PERF_PROP_OA_FORMAT,         report_format,
> > > > >       DRM_I915_PERF_PROP_OA_EXPONENT,       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_ptr = (uint64_t)properties,
> > > > >       .num_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_PROP_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.
> > > > >
> > > > > v2:
> > > > >     use i915_gem_context_get() - Chris Wilson
> > > > > v3:
> > > > >     update read() interface to avoid passing state struct - Chris
> > Wilson
> > > > >     fix some rebase fallout, with i915-perf init/deinit
> > > > > v4:
> > > > >     s/DRM_IORW/DRM_IOW/ - Emil Velikov
> > > > >
> > > > > Signed-off-by: Robert Bragg <robert@sixbynine.org>
> > > > > Reviewed-by: Matthew Auld <matthew.auld@intel.com>
> > > > > Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>
> > > > Minor nit, there are a fair few DRM_ERROR's missing a new line.
> > >
> > > Also, DRM_ERROR for userspace-triggerable failures is no good. igt
> > > testcase are supposed to exercise all the invalid stuff, and would then
> > > fail if you spam dmesg. Why was this not caught?
> > >
> > > Fixup patch totally fine, but if this wasn't caught due to missing igt
> > > that needs to be fixed, too.
> >
> > Another nitpick for the future: Enabling new features first and then
> > fixing up the fallout is the wrong way round, if someone bisects over this
> > range mesa might blow up in really bad ways.
> >
> > Oh well, this has been out there for way too long, so meh.
> >
> 
> Fwiw I'm aware of this, and think I've ordered the patches correctly to
> avoid bisect problems in Mesa / userspace. This infrastructure patch should
> have no fallout to fix for userspace. The command parser changes that
> affect userspace were done before adding oacontrol usage to i915-perf and
> the cmd parser's EINVAL reporting for access failures was changed *before*
> removing oacontrol from the whitelist.
> 
> Did I overlook something in particular?

Ah, if you key the userspace side off the cmd parser change then I think
it should be all fine. I only looked at this ioctl here, and that alone
looked like it was unsafe. Ordering of the patches later on seemed correct
indeed.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v2] drm/i915: don't whitelist oacontrol in cmd parser

[Daniel Vetter]

On Tue, Nov 22, 2016 at 02:09:08PM +0000, Robert Bragg wrote:
> On Tue, Nov 22, 2016 at 1:34 PM, Daniel Vetter <daniel@ffwll.ch> wrote:
> 
> > On Tue, Nov 08, 2016 at 12:51:48PM +0000, Robert Bragg wrote:
> > > This v2 patch bumps the command parser version so it can be referenced in
> > > corresponding i-g-t gem_exec_parse changes.
> > >
> > > --- >8 ---
> >
> > Scissors cut everything below, not everything above, hence next time
> > around pls switch around your comment and the commit message, as-is not
> > much left ;-)
> >
> 
> Hmm, they cut away what's above and keep what's below in my experience -
> what command are you seeing the opposite with?
> 
> I just double checked this with git am --scissors

Plain git am works the other way round, that's why I never noticed there's
a special option. TIL ;-)
-Daniel

> 
> - Robert
> 
> 
> 
> >
> > Fixed up while applying.
> > -Daniel
> >
> > >
> > > Being able to program OACONTROL from a non-privileged batch buffer is
> > > not sufficient to be able to configure the OA unit. This was originally
> > > allowed to help enable Mesa to expose OA counters via the
> > > INTEL_performance_query extension, but the current implementation based
> > > on programming OACONTROL via a batch buffer isn't able to report useable
> > > data without a more complete OA unit configuration. Mesa handles the
> > > possibility that writes to OACONTROL may not be allowed and so only
> > > advertises the extension after explicitly testing that a write to
> > > OACONTROL succeeds. Based on this; removing OACONTROL from the whitelist
> > > should be ok for userspace.
> > >
> > > Removing this simplifies adding a new kernel api for configuring the OA
> > > unit without needing to consider the possibility that userspace might
> > > trample on OACONTROL state which we'd like to start managing within
> > > the kernel instead. In particular running any Mesa based GL application
> > > currently results in clearing OACONTROL when initializing which would
> > > disable the capturing of metrics.
> > >
> > > v2:
> > >     This bumps the command parser version from 8 to 9, as the change is
> > >     visible to userspace.
> > >
> > > Signed-off-by: Robert Bragg <robert@sixbynine.org>
> > > Reviewed-by: Matthew Auld <matthew.auld@intel.com>
> > > Reviewed-by: Sourab Gupta <sourab.gupta@intel.com>
> > > ---
> > >  drivers/gpu/drm/i915/i915_cmd_parser.c | 42
> > ++++------------------------------
> > >  1 file changed, 5 insertions(+), 37 deletions(-)
> > >
> > > diff --git a/drivers/gpu/drm/i915/i915_cmd_parser.c
> > b/drivers/gpu/drm/i915/i915_cmd_parser.c
> > > index c9d2ecd..f5762cd 100644
> > > --- a/drivers/gpu/drm/i915/i915_cmd_parser.c
> > > +++ b/drivers/gpu/drm/i915/i915_cmd_parser.c
> > > @@ -450,7 +450,6 @@ static const struct drm_i915_reg_descriptor
> > gen7_render_regs[] = {
> > >       REG64(PS_INVOCATION_COUNT),
> > >       REG64(PS_DEPTH_COUNT),
> > >       REG64_IDX(RING_TIMESTAMP, RENDER_RING_BASE),
> > > -     REG32(GEN7_OACONTROL), /* Only allowed for LRI and SRM. See below.
> > */
> > >       REG64(MI_PREDICATE_SRC0),
> > >       REG64(MI_PREDICATE_SRC1),
> > >       REG32(GEN7_3DPRIM_END_OFFSET),
> > > @@ -1060,8 +1059,7 @@ bool intel_engine_needs_cmd_parser(struct
> > intel_engine_cs *engine)
> > >  static bool check_cmd(const struct intel_engine_cs *engine,
> > >                     const struct drm_i915_cmd_descriptor *desc,
> > >                     const u32 *cmd, u32 length,
> > > -                   const bool is_master,
> > > -                   bool *oacontrol_set)
> > > +                   const bool is_master)
> > >  {
> > >       if (desc->flags & CMD_DESC_SKIP)
> > >               return true;
> > > @@ -1099,31 +1097,6 @@ static bool check_cmd(const struct
> > intel_engine_cs *engine,
> > >                       }
> > >
> > >                       /*
> > > -                      * OACONTROL requires some special handling for
> > > -                      * writes. We want to make sure that any batch
> > which
> > > -                      * enables OA also disables it before the end of
> > the
> > > -                      * batch. The goal is to prevent one process from
> > > -                      * snooping on the perf data from another process.
> > To do
> > > -                      * that, we need to check the value that will be
> > written
> > > -                      * to the register. Hence, limit OACONTROL writes
> > to
> > > -                      * only MI_LOAD_REGISTER_IMM commands.
> > > -                      */
> > > -                     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;
> > > -                             }
> > > -
> > > -                             if (desc->cmd.value ==
> > MI_LOAD_REGISTER_REG) {
> > > -                                     DRM_DEBUG_DRIVER("CMD: Rejected
> > LRR to OACONTROL\n");
> > > -                                     return false;
> > > -                             }
> > > -
> > > -                             if (desc->cmd.value ==
> > MI_LOAD_REGISTER_IMM(1))
> > > -                                     *oacontrol_set = (cmd[offset + 1]
> > != 0);
> > > -                     }
> > > -
> > > -                     /*
> > >                        * Check the value written to the register against
> > the
> > >                        * allowed mask/value pair given in the whitelist
> > entry.
> > >                        */
> > > @@ -1214,7 +1187,6 @@ int intel_engine_cmd_parser(struct intel_engine_cs
> > *engine,
> > >       u32 *cmd, *batch_end;
> > >       struct drm_i915_cmd_descriptor default_desc = noop_desc;
> > >       const struct drm_i915_cmd_descriptor *desc = &default_desc;
> > > -     bool oacontrol_set = false; /* OACONTROL tracking. See check_cmd()
> > */
> > >       bool needs_clflush_after = false;
> > >       int ret = 0;
> > >
> > > @@ -1270,8 +1242,7 @@ int intel_engine_cmd_parser(struct intel_engine_cs
> > *engine,
> > >                       break;
> > >               }
> > >
> > > -             if (!check_cmd(engine, desc, cmd, length, is_master,
> > > -                            &oacontrol_set)) {
> > > +             if (!check_cmd(engine, desc, cmd, length, is_master)) {
> > >                       ret = -EACCES;
> > >                       break;
> > >               }
> > > @@ -1279,11 +1250,6 @@ int intel_engine_cmd_parser(struct
> > intel_engine_cs *engine,
> > >               cmd += length;
> > >       }
> > >
> > > -     if (oacontrol_set) {
> > > -             DRM_DEBUG_DRIVER("CMD: batch set OACONTROL but did not
> > clear it\n");
> > > -             ret = -EINVAL;
> > > -     }
> > > -
> > >       if (cmd >= batch_end) {
> > >               DRM_DEBUG_DRIVER("CMD: Got to the end of the buffer w/o a
> > BBE cmd!\n");
> > >               ret = -EINVAL;
> > > @@ -1336,6 +1302,8 @@ int i915_cmd_parser_get_version(struct
> > drm_i915_private *dev_priv)
> > >        * 8. Don't report cmd_check() failures as EINVAL errors to
> > userspace;
> > >        *    rely on the HW to NOOP disallowed commands as it would
> > without
> > >        *    the parser enabled.
> > > +      * 9. Don't whitelist or handle oacontrol specially, as ownership
> > > +      *    for oacontrol state is moving to i915-perf.
> > >        */
> > > -     return 8;
> > > +     return 9;
> > >  }
> > > --
> > > 2.10.1
> > >
> > > _______________________________________________
> > > Intel-gfx mailing list
> > > Intel-gfx@lists.freedesktop.org
> > > https://lists.freedesktop.org/mailman/listinfo/intel-gfx
> >
> > --
> > Daniel Vetter
> > Software Engineer, Intel Corporation
> > http://blog.ffwll.ch
> >

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx