[PATCH v4 3/5] drm/xe/eustall: Implement EU stall sampling APIs

Matt Roper matthew.d.roper at intel.com
Fri Oct 18 23:31:04 UTC 2024


On Sun, Oct 13, 2024 at 11:00:34PM -0700, Harish Chegondi wrote:
> Add support to the EU stall sampling APIs introduced in
> the previous patch. Add register definitions and the code
> that accesses these registers to the APIs.

You should make it clear in the subject / commit message that this is
only enabling stall sampling for Xe_HPC, Xe2, and beyond.

It might actually be best if this patch only enabled Xe_HPC, and then a
separate follow-up patch added the deltas necessary for Xe2.  That would
make it easier to ensure we don't miss anything.

> 
> A timer thread periodically polls the EU stall data buffer write pointer
> registers to look for any new data and caches the write pointer. The read
> function compares the cached read and write pointers and copies any new
> data to the user space. If the user space doesn't read the EU stall data
> fast enough, it is possible that the EU stall data buffer can get filled,
> and if the hardware wants to write more data, it simply drops data due to
> unavailable buffer space. In that case hardware sets a bit in a register.
> The driver read() returns -EIO error to let the user space know that
> the hardware has dropped data. A subsequent read by the user space returns
> the remaining EU stall data.
> 
> Signed-off-by: Harish Chegondi <harish.chegondi at intel.com>
> ---
>  drivers/gpu/drm/xe/regs/xe_eu_stall_regs.h |  29 +
>  drivers/gpu/drm/xe/xe_eu_stall.c           | 710 ++++++++++++++++++++-
>  drivers/gpu/drm/xe/xe_eu_stall.h           |  46 ++
>  drivers/gpu/drm/xe/xe_gt.c                 |   6 +
>  drivers/gpu/drm/xe/xe_gt_types.h           |   3 +
>  drivers/gpu/drm/xe/xe_trace.h              |  35 +
>  6 files changed, 804 insertions(+), 25 deletions(-)
>  create mode 100644 drivers/gpu/drm/xe/regs/xe_eu_stall_regs.h
> 
> diff --git a/drivers/gpu/drm/xe/regs/xe_eu_stall_regs.h b/drivers/gpu/drm/xe/regs/xe_eu_stall_regs.h
> new file mode 100644
> index 000000000000..2b742890b283
> --- /dev/null
> +++ b/drivers/gpu/drm/xe/regs/xe_eu_stall_regs.h
> @@ -0,0 +1,29 @@
> +/* SPDX-License-Identifier: MIT */
> +/*
> + * Copyright © 2024 Intel Corporation
> + */
> +
> +#ifndef _XE_EU_STALL_REGS_H_
> +#define _XE_EU_STALL_REGS_H_
> +
> +#include "regs/xe_reg_defs.h"
> +
> +#define XEHPC_EUSTALL_BASE			XE_REG_MCR(0xe520)
> +#define   XEHPC_EUSTALL_BASE_BUF_ADDR		REG_GENMASK(31, 6)
> +#define   XEHPC_EUSTALL_BASE_XECORE_BUF_SZ	REG_GENMASK(5, 3)
> +#define   XEHPC_EUSTALL_BASE_ENABLE_SAMPLING	REG_BIT(1)
> +
> +#define XEHPC_EUSTALL_BASE_UPPER		XE_REG_MCR(0xe524)
> +
> +#define XEHPC_EUSTALL_REPORT			XE_REG_MCR(0xe528, XE_REG_OPTION_MASKED)
> +#define   XEHPC_EUSTALL_REPORT_WRITE_PTR_MASK	REG_GENMASK(15, 2)
> +#define   XEHPC_EUSTALL_REPORT_OVERFLOW_DROP	REG_BIT(1)
> +
> +#define XEHPC_EUSTALL_REPORT1			XE_REG_MCR(0xe52c, XE_REG_OPTION_MASKED)
> +#define   XEHPC_EUSTALL_REPORT1_READ_PTR_MASK	REG_GENMASK(15, 2)
> +
> +#define XEHPC_EUSTALL_CTRL			XE_REG_MCR(0xe53c, XE_REG_OPTION_MASKED)
> +#define   EUSTALL_MOCS				REG_GENMASK(9, 3)
> +#define   EUSTALL_SAMPLE_RATE			REG_GENMASK(2, 0)
> +
> +#endif
> diff --git a/drivers/gpu/drm/xe/xe_eu_stall.c b/drivers/gpu/drm/xe/xe_eu_stall.c
> index 5e4c90f9614d..6a730ffa6d2f 100644
> --- a/drivers/gpu/drm/xe/xe_eu_stall.c
> +++ b/drivers/gpu/drm/xe/xe_eu_stall.c
> @@ -8,15 +8,27 @@
>  #include <linux/poll.h>
>  #include <linux/fs.h>
>  
> +#include <drm/drm_drv.h>
>  #include <uapi/drm/xe_drm.h>
>  
> +#include "xe_bo.h"
> +#include "xe_pm.h"
> +#include "xe_trace.h"
>  #include "xe_macros.h"
>  #include "xe_device.h"
> +#include "xe_gt_mcr.h"
>  #include "xe_eu_stall.h"
>  #include "xe_gt_printk.h"
> +#include "xe_force_wake.h"
>  #include "xe_gt_topology.h"
>  #include "xe_observation.h"
>  
> +#include "regs/xe_gt_regs.h"
> +#include "regs/xe_eu_stall_regs.h"
> +
> +#define DEFAULT_POLL_FREQUENCY_HZ 100
> +#define DEFAULT_POLL_PERIOD_NS (NSEC_PER_SEC / DEFAULT_POLL_FREQUENCY_HZ)
> +
>  /**
>   * struct eu_stall_open_properties
>   *
> @@ -45,6 +57,22 @@ num_data_rows(u32 data_size)
>  	return (data_size >> 6);
>  }
>  
> +int xe_eu_stall_init(struct xe_gt *gt)
> +{
> +	gt->eu_stall_cntr = kzalloc(sizeof(*gt->eu_stall_cntr), GFP_KERNEL);

Nitpick; like Jani mentioned on a different series recently,
abbreviating words in ways that aren't already widely-used makes things
a bit harder to read.  Every time I see 'cntr' I read it as "control"
before I remember that it's supposed to be "counter."  I don't think
we're really saving enough characters to make it worth abbreviating.
Writing this out as "counter" would probably be best.


> +	if (!gt->eu_stall_cntr)
> +		return -ENOMEM;
> +
> +	mutex_init(&gt->eu_stall_cntr->lock);
> +	return 0;
> +}
> +
> +void xe_eu_stall_fini(struct xe_gt *gt)
> +{
> +	mutex_destroy(&gt->eu_stall_cntr->lock);
> +	kfree(gt->eu_stall_cntr);
> +}
> +
>  static int set_prop_eu_stall_sampling_rate(struct xe_device *xe, u64 value,
>  					   struct eu_stall_open_properties *props)
>  {
> @@ -152,6 +180,229 @@ static int xe_eu_stall_user_extensions(struct xe_device *xe, u64 extension,
>  	return 0;
>  }
>  
> +/**
> + * buf_data_size - Calculate the number of bytes in a circular buffer
> + *		   of size buf_size given the read and write pointers
> + *		   into the buffer.
> + *
> + * @read_ptr: Read pointer. Uses an additional overflow bit
> + * @write_ptr: Write pointer. Uses an additional overflow bit

@buf_size is missing from the kerneldoc.  Presumably this function only
works if the size is a power of 2?

The "uses an additional overflow bit" doesn't seem to be explained or
used in the code below.

> + *
> + * Returns: number of bytes of data in the buffer
> + */
> +static u32
> +buf_data_size(size_t buf_size, u32 read_ptr, u32 write_ptr)
> +{
> +	u32 read_offset, write_offset, size = 0;
> +
> +	read_offset = read_ptr & (buf_size - 1);
> +	write_offset = write_ptr & (buf_size - 1);
> +
> +	if (write_offset > read_offset)
> +		size = write_offset - read_offset;
> +	else
> +		size = buf_size - read_offset + write_offset;
> +
> +	return size;
> +}
> +
> +/**
> + * eu_stall_data_buf_check - check for EU stall data in the buffer
> + *
> + * @stream: xe EU stall data stream instance
> + *
> + * Returns: true if the EU stall buffer contains minimum stall data as
> + *	    specified by the event report count, else false.
> + */
> +static bool
> +eu_stall_data_buf_check(struct xe_eu_stall_data_stream *stream)
> +{
> +	u32 read_ptr, write_ptr_reg, write_ptr, total_data = 0;
> +	u32 buf_size = stream->per_xecore_buf_size;
> +	struct xe_gt *gt = stream->gt;
> +	struct per_xecore_buf *xecore_buf;
> +	bool min_data_present;
> +	u16 group, instance;
> +	unsigned int xecore;
> +
> +	min_data_present = false;
> +	for_each_dss_steering(xecore, gt, group, instance) {
> +		xecore_buf = &stream->xecore_buf[xecore];
> +		mutex_lock(&xecore_buf->lock);
> +		read_ptr = xecore_buf->read;
> +		write_ptr_reg = xe_gt_mcr_unicast_read(gt, XEHPC_EUSTALL_REPORT,
> +						       group, instance);
> +		write_ptr = REG_FIELD_GET(XEHPC_EUSTALL_REPORT_WRITE_PTR_MASK, write_ptr_reg);
> +		write_ptr <<= 6;
> +		write_ptr &= ((buf_size << 1) - 1);
> +		if ((write_ptr != read_ptr) && !min_data_present) {
> +			total_data += buf_data_size(buf_size, read_ptr, write_ptr);
> +			/*
> +			 * Check if there are at least minimum number of stall data
> +			 * rows for poll() to indicate that the data is present.
> +			 * Each stall data row is 64B (cacheline size).
> +			 */
> +			if (num_data_rows(total_data) >= stream->event_report_count)
> +				min_data_present = true;
> +		}
> +		if (write_ptr_reg & XEHPC_EUSTALL_REPORT_OVERFLOW_DROP) {
> +			spin_lock(&stream->data_drop.lock);
> +			set_bit(xecore, stream->data_drop.mask);
> +			spin_unlock(&stream->data_drop.lock);

Do we need the spinlock for this mask?  I believe set_bit and clear_bit
are supposed to be atomic already.


I haven't fully reviewed this patch yet; there's a lot going on here, so
I'll need to come back and look at it more closely later.


Matt

> +		}
> +		xecore_buf->write = write_ptr;
> +		mutex_unlock(&xecore_buf->lock);
> +	}
> +	return min_data_present;
> +}
> +
> +static void
> +clear_dropped_eviction_line_bit(struct xe_gt *gt, u16 group, u16 instance)
> +{
> +	struct xe_device *xe = gt_to_xe(gt);
> +	u32 write_ptr_reg;
> +
> +	/* On PVC, the overflow bit has to be cleared by writing 1 to it.
> +	 * On other GPUs, the bit has to be cleared by writing 0 to it.
> +	 */
> +	if (GRAPHICS_VER(xe) >= 20)
> +		write_ptr_reg = _MASKED_BIT_DISABLE(XEHPC_EUSTALL_REPORT_OVERFLOW_DROP);
> +	else
> +		write_ptr_reg = _MASKED_BIT_ENABLE(XEHPC_EUSTALL_REPORT_OVERFLOW_DROP);
> +
> +	xe_gt_mcr_unicast_write(gt, XEHPC_EUSTALL_REPORT, write_ptr_reg, group, instance);
> +}
> +
> +static int
> +xe_eu_stall_data_buf_read(struct xe_eu_stall_data_stream *stream,
> +			  char __user *buf, size_t count,
> +			  size_t *total_size, struct xe_gt *gt,
> +			  u16 group, u16 instance, unsigned int xecore)
> +{
> +	u32 read_ptr_reg, read_ptr, write_ptr;
> +	u8 *xecore_start_vaddr, *read_vaddr;
> +	u32 read_offset, write_offset;
> +	struct per_xecore_buf *xecore_buf;
> +	size_t size, buf_size;
> +
> +	/* Hardware increments the read and write pointers such that they can
> +	 * overflow into one additional bit. For example, a 256KB size buffer
> +	 * offset pointer needs 18 bits. But HW uses 19 bits for the read and
> +	 * write pointers. This technique avoids wasting a slot in the buffer.
> +	 * Read and write offsets are calculated from the pointers in order to
> +	 * check if the write pointer has wrapped around the array.
> +	 */
> +	xecore_buf = &stream->xecore_buf[xecore];
> +	mutex_lock(&xecore_buf->lock);
> +	xecore_start_vaddr = xecore_buf->vaddr;
> +	read_ptr = xecore_buf->read;
> +	write_ptr = xecore_buf->write;
> +	buf_size = stream->per_xecore_buf_size;
> +	read_offset = read_ptr & (buf_size - 1);
> +	write_offset = write_ptr & (buf_size - 1);
> +
> +	if (write_ptr == read_ptr) {
> +		mutex_unlock(&xecore_buf->lock);
> +		return 0;
> +	}
> +
> +	trace_xe_eu_stall_cntr_read(group, instance, read_ptr, write_ptr,
> +				    read_offset, write_offset, *total_size);
> +	/* If write pointer offset is less than the read pointer offset,
> +	 * it means, write pointer has wrapped around the array.
> +	 */
> +	if (write_offset > read_offset)
> +		size = write_offset - read_offset;
> +	else
> +		size = buf_size - read_offset + write_offset;
> +
> +	/* Read only the data that the user space buffer can accommodate */
> +	if ((*total_size + size) > count) {
> +		mutex_unlock(&xecore_buf->lock);
> +		return 0;
> +	}
> +
> +	read_vaddr = xecore_start_vaddr + read_offset;
> +
> +	if (write_offset > read_offset) {
> +		if (copy_to_user((buf + *total_size), read_vaddr, size)) {
> +			mutex_unlock(&xecore_buf->lock);
> +			return -EFAULT;
> +		}
> +	} else {
> +		if (copy_to_user((buf + *total_size), read_vaddr, (buf_size - read_offset))) {
> +			mutex_unlock(&xecore_buf->lock);
> +			return -EFAULT;
> +		}
> +		if (copy_to_user((buf + *total_size), xecore_start_vaddr, write_offset)) {
> +			mutex_unlock(&xecore_buf->lock);
> +			return -EFAULT;
> +		}
> +	}
> +
> +	*total_size += size;
> +	read_ptr += size;
> +
> +	/* Read pointer can overflow into one additional bit */
> +	read_ptr &= ((buf_size << 1) - 1);
> +	read_ptr_reg = REG_FIELD_PREP(XEHPC_EUSTALL_REPORT1_READ_PTR_MASK, (read_ptr >> 6));
> +	read_ptr_reg &= XEHPC_EUSTALL_REPORT1_READ_PTR_MASK;
> +	read_ptr_reg = _MASKED_FIELD(XEHPC_EUSTALL_REPORT1_READ_PTR_MASK, read_ptr_reg);
> +	xe_gt_mcr_unicast_write(gt, XEHPC_EUSTALL_REPORT1, read_ptr_reg, group, instance);
> +	if (test_bit(xecore, stream->data_drop.mask)) {
> +		clear_dropped_eviction_line_bit(gt, group, instance);
> +		spin_lock(&stream->data_drop.lock);
> +		clear_bit(xecore, stream->data_drop.mask);
> +		spin_unlock(&stream->data_drop.lock);
> +	}
> +	xecore_buf->read = read_ptr;
> +	mutex_unlock(&xecore_buf->lock);
> +	trace_xe_eu_stall_cntr_read(group, instance, read_ptr, write_ptr,
> +				    read_offset, write_offset, *total_size);
> +	return 0;
> +}
> +
> +/**
> + * xe_eu_stall_stream_read_locked - copy EU stall counters data from the
> + *				    per xecore buffers to the userspace buffer
> + * @stream: A stream opened for EU stall count metrics
> + * @buf: destination buffer given by userspace
> + * @count: the number of bytes userspace wants to read
> + * @ppos: (inout) file seek position (unused)
> + *
> + * Returns: Number of bytes copied or a negative error code
> + * If we've successfully copied any data then reporting that takes
> + * precedence over any internal error status, so the data isn't lost.
> + */
> +static ssize_t
> +xe_eu_stall_stream_read_locked(struct xe_eu_stall_data_stream *stream,
> +			       struct file *file, char __user *buf,
> +			       size_t count, loff_t *ppos)
> +{
> +	struct xe_gt *gt = stream->gt;
> +	size_t total_size = 0;
> +	u16 group, instance;
> +	unsigned int xecore;
> +	int ret = 0;
> +
> +	if (count == 0)
> +		return -EINVAL;
> +
> +	for_each_dss_steering(xecore, gt, group, instance) {
> +		ret = xe_eu_stall_data_buf_read(stream, buf, count, &total_size,
> +						gt, group, instance, xecore);
> +		if (ret || count == total_size)
> +			goto exit;
> +	}
> +exit:
> +	if (total_size)
> +		return total_size;
> +	else if (ret)
> +		return ret;
> +	else
> +		return -EAGAIN;
> +}
> +
>  /**
>   * xe_eu_stall_stream_read - handles userspace read() of a EU stall data stream fd.
>   *
> @@ -165,11 +416,263 @@ static int xe_eu_stall_user_extensions(struct xe_device *xe, u64 extension,
>  static ssize_t xe_eu_stall_stream_read(struct file *file, char __user *buf,
>  				       size_t count, loff_t *ppos)
>  {
> -	ssize_t ret = 0;
> +	struct xe_eu_stall_data_stream *stream = file->private_data;
> +	struct xe_gt *gt = stream->gt;
> +	ssize_t ret;
> +
> +	if (!stream->enabled) {
> +		xe_gt_dbg(gt, "EU stall data stream not enabled to read\n");
> +		return -EINVAL;
> +	}
> +
> +	if (bitmap_weight(stream->data_drop.mask, XE_MAX_DSS_FUSE_BITS)) {
> +		if (!stream->data_drop.reported_to_user) {
> +			stream->data_drop.reported_to_user = true;
> +			xe_gt_dbg(gt, "EU stall data dropped in XeCores: %*pb\n",
> +				  XE_MAX_DSS_FUSE_BITS, stream->data_drop.mask);
> +			return -EIO;
> +		}
> +		stream->data_drop.reported_to_user = false;
> +	}
> +
> +	if (!(file->f_flags & O_NONBLOCK)) {
> +		do {
> +			if (!stream->pollin) {
> +				ret = wait_event_interruptible(stream->poll_wq, stream->pollin);
> +				if (ret)
> +					return -EINTR;
> +			}
> +
> +			mutex_lock(&gt->eu_stall_cntr->lock);
> +			ret = xe_eu_stall_stream_read_locked(stream, file, buf, count, ppos);
> +			mutex_unlock(&gt->eu_stall_cntr->lock);
> +		} while (ret == -EAGAIN);
> +	} else {
> +		mutex_lock(&gt->eu_stall_cntr->lock);
> +		ret = xe_eu_stall_stream_read_locked(stream, file, buf, count, ppos);
> +		mutex_unlock(&gt->eu_stall_cntr->lock);
> +	}
> +
> +	stream->pollin = false;
>  
>  	return ret;
>  }
>  
> +static void
> +free_eu_stall_cntr_buf(struct xe_eu_stall_data_stream *stream)
> +{
> +	if (stream->bo) {
> +		xe_bo_unpin_map_no_vm(stream->bo);
> +		stream->bo = NULL;
> +	}
> +	destroy_workqueue(stream->buf_check_wq);
> +}
> +
> +static int alloc_eu_stall_data_buf(struct xe_eu_stall_data_stream *stream,
> +				   u16 num_xecore)
> +{
> +	struct xe_tile *tile = stream->gt->tile;
> +	struct xe_bo *bo;
> +	u32 size;
> +
> +	size = stream->per_xecore_buf_size * num_xecore;
> +
> +	bo = xe_bo_create_pin_map(tile->xe, tile, NULL,
> +				  size, ttm_bo_type_kernel,
> +				  XE_BO_FLAG_SYSTEM |
> +				  XE_BO_FLAG_GGTT);
> +	if (IS_ERR(bo))
> +		return PTR_ERR(bo);
> +
> +	stream->bo = bo;
> +
> +	return 0;
> +}
> +
> +static u32
> +gen_eustall_base(struct xe_eu_stall_data_stream *stream, bool enable)
> +{
> +	u32 val = xe_bo_ggtt_addr(stream->bo);
> +	u32 sz;
> +
> +	XE_WARN_ON(!IS_ALIGNED(val, 64));
> +
> +	switch (stream->per_xecore_buf_size) {
> +	case SZ_128K:
> +		sz = 0;
> +		break;
> +	case SZ_256K:
> +		sz = 1;
> +		break;
> +	case SZ_512K:
> +		sz = 2;
> +		break;
> +	default:
> +		xe_gt_warn(stream->gt, "Missing case per XeCore buffer size == %lu)\n",
> +			   (long)(stream->per_xecore_buf_size));
> +		sz = 2;
> +	}
> +
> +	val |= REG_FIELD_PREP(XEHPC_EUSTALL_BASE_XECORE_BUF_SZ, sz);
> +	if (enable)
> +		val |= XEHPC_EUSTALL_BASE_ENABLE_SAMPLING;
> +
> +	return val;
> +}
> +
> +static void
> +xe_eu_stall_stream_enable(struct xe_eu_stall_data_stream *stream)
> +{
> +	struct xe_gt *gt = stream->gt;
> +	u32 reg_value;
> +
> +	/* Take runtime pm ref and forcewake to disable RC6 */
> +	xe_pm_runtime_get(gt_to_xe(gt));
> +	XE_WARN_ON(xe_force_wake_get(gt_to_fw(gt), XE_FW_RENDER));
> +
> +	reg_value = gen_eustall_base(stream, true);
> +	xe_gt_mcr_multicast_write(gt, XEHPC_EUSTALL_BASE, reg_value);
> +}
> +
> +static void
> +xe_eu_stall_stream_disable(struct xe_eu_stall_data_stream *stream)
> +{
> +	struct xe_gt *gt = stream->gt;
> +	u16 group, instance;
> +	unsigned int xecore;
> +	u32 reg_value;
> +
> +	/*
> +	 * Before disabling EU stall sampling, check if any of the
> +	 * XEHPC_EUSTALL_REPORT registers have the drop bit set. If set,
> +	 * clear the bit. If the user space application reads all the
> +	 * stall data, the drop bit would be cleared during the read.
> +	 * But if there is any unread data and the drop bit is set for
> +	 * any subslice, the drop bit would continue to be set even
> +	 * after disabling EU stall sampling and may cause erroneous
> +	 * stall data in the subsequent stall data sampling run.
> +	 */
> +	for_each_dss_steering(xecore, gt, group, instance) {
> +		reg_value = xe_gt_mcr_unicast_read(gt, XEHPC_EUSTALL_REPORT,
> +						   group, instance);
> +		if (reg_value & XEHPC_EUSTALL_REPORT_OVERFLOW_DROP)
> +			clear_dropped_eviction_line_bit(gt, group, instance);
> +	}
> +	reg_value = gen_eustall_base(stream, false);
> +	xe_gt_mcr_multicast_write(gt, XEHPC_EUSTALL_BASE, reg_value);
> +
> +	XE_WARN_ON(xe_force_wake_put(gt_to_fw(gt), XE_FW_RENDER));
> +	xe_pm_runtime_put(gt_to_xe(gt));
> +}
> +
> +static void eu_stall_buf_check_work_fn(struct work_struct *work)
> +{
> +	struct xe_eu_stall_data_stream *stream =
> +		container_of(work, typeof(*stream), buf_check_work);
> +
> +	if (eu_stall_data_buf_check(stream)) {
> +		stream->pollin = true;
> +		wake_up(&stream->poll_wq);
> +	}
> +}
> +
> +static enum
> +hrtimer_restart eu_stall_poll_check_timer_cb(struct hrtimer *hrtimer)
> +{
> +	struct xe_eu_stall_data_stream *stream =
> +		container_of(hrtimer, typeof(*stream), poll_check_timer);
> +
> +	queue_work(stream->buf_check_wq, &stream->buf_check_work);
> +	hrtimer_forward_now(hrtimer, ns_to_ktime(stream->poll_period));
> +
> +	return HRTIMER_RESTART;
> +}
> +
> +static int xe_eu_stall_stream_init(struct xe_eu_stall_data_stream *stream,
> +				   struct eu_stall_open_properties *props,
> +				   u16 num_xecore)
> +{
> +	u32 write_ptr_reg, write_ptr, read_ptr_reg;
> +	u32 vaddr_offset, base_reg_value;
> +	struct xe_gt *gt = stream->gt;
> +	struct per_xecore_buf *xecore_buf;
> +	u16 group, instance;
> +	int ret, xecore;
> +
> +	init_waitqueue_head(&stream->poll_wq);
> +	INIT_WORK(&stream->buf_check_work, eu_stall_buf_check_work_fn);
> +	stream->buf_check_wq = alloc_ordered_workqueue("xe_eustall_cntr", 0);
> +	if (!stream->buf_check_wq)
> +		return -ENOMEM;
> +	hrtimer_init(&stream->poll_check_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
> +	stream->poll_check_timer.function = eu_stall_poll_check_timer_cb;
> +	stream->event_report_count = props->event_report_count;
> +	stream->per_xecore_buf_size = SZ_512K;
> +	stream->poll_period = props->poll_period;
> +
> +	ret = alloc_eu_stall_data_buf(stream, num_xecore);
> +	if (ret)
> +		return ret;
> +
> +	stream->xecore_buf = kcalloc(num_xecore, sizeof(*stream->xecore_buf), GFP_KERNEL);
> +	if (!stream->xecore_buf)
> +		return -ENOMEM;
> +
> +	spin_lock_init(&stream->data_drop.lock);
> +	stream->data_drop.reported_to_user = false;
> +	bitmap_zero(stream->data_drop.mask, XE_MAX_DSS_FUSE_BITS);
> +
> +	xe_pm_runtime_get(gt_to_xe(gt));
> +	XE_WARN_ON(xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL));
> +
> +	base_reg_value = gen_eustall_base(stream, false);
> +	xe_gt_mcr_multicast_write(gt, XEHPC_EUSTALL_BASE, base_reg_value);
> +	/* GGTT addresses can never be > 32 bits */
> +	xe_gt_mcr_multicast_write(gt, XEHPC_EUSTALL_BASE_UPPER, 0);
> +	base_reg_value = _MASKED_FIELD(EUSTALL_MOCS | EUSTALL_SAMPLE_RATE,
> +				       REG_FIELD_PREP(EUSTALL_MOCS, gt->mocs.uc_index << 1) |
> +				       REG_FIELD_PREP(EUSTALL_SAMPLE_RATE,
> +						      props->eu_stall_sampling_rate));
> +	xe_gt_mcr_multicast_write(gt, XEHPC_EUSTALL_CTRL, base_reg_value);
> +
> +	for_each_dss_steering(xecore, gt, group, instance) {
> +		write_ptr_reg = xe_gt_mcr_unicast_read(gt, XEHPC_EUSTALL_REPORT,
> +						       group, instance);
> +		write_ptr = REG_FIELD_GET(XEHPC_EUSTALL_REPORT_WRITE_PTR_MASK, write_ptr_reg);
> +		write_ptr <<= 6;
> +		write_ptr &= ((stream->per_xecore_buf_size << 1) - 1);
> +		read_ptr_reg = write_ptr >> 6;
> +		read_ptr_reg = REG_FIELD_PREP(XEHPC_EUSTALL_REPORT1_READ_PTR_MASK, write_ptr);
> +		read_ptr_reg &= XEHPC_EUSTALL_REPORT1_READ_PTR_MASK;
> +		read_ptr_reg = _MASKED_FIELD(XEHPC_EUSTALL_REPORT1_READ_PTR_MASK, read_ptr_reg);
> +		xe_gt_mcr_unicast_write(gt, XEHPC_EUSTALL_REPORT1,
> +					read_ptr_reg, group, instance);
> +		xecore_buf = &stream->xecore_buf[xecore];
> +		vaddr_offset = xecore * stream->per_xecore_buf_size;
> +		xecore_buf->vaddr = stream->bo->vmap.vaddr + vaddr_offset;
> +		xecore_buf->write = write_ptr;
> +		xecore_buf->read = write_ptr;
> +		mutex_init(&xecore_buf->lock);
> +	}
> +	XE_WARN_ON(xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL));
> +	xe_pm_runtime_put(gt_to_xe(gt));
> +	return 0;
> +}
> +
> +static __poll_t
> +xe_eu_stall_stream_poll_locked(struct xe_eu_stall_data_stream *stream,
> +			       struct file *file, poll_table *wait)
> +{
> +	__poll_t events = 0;
> +
> +	poll_wait(file, &stream->poll_wq, wait);
> +
> +	if (stream->pollin)
> +		events |= EPOLLIN;
> +
> +	return events;
> +}
> +
>  /**
>   * xe_eu_stall_stream_poll - handles userspace poll() of a EU stall data stream fd.
>   *
> @@ -181,11 +684,60 @@ static ssize_t xe_eu_stall_stream_read(struct file *file, char __user *buf,
>  static __poll_t
>  xe_eu_stall_stream_poll(struct file *file, poll_table *wait)
>  {
> -	__poll_t ret = 0;
> +	struct xe_eu_stall_data_stream *stream = file->private_data;
> +	struct xe_gt *gt = stream->gt;
> +	__poll_t ret;
> +
> +	mutex_lock(&gt->eu_stall_cntr->lock);
> +	ret = xe_eu_stall_stream_poll_locked(stream, file, wait);
> +	mutex_unlock(&gt->eu_stall_cntr->lock);
>  
>  	return ret;
>  }
>  
> +static void
> +xe_eu_stall_cntr_enable_locked(struct xe_eu_stall_data_stream *stream)
> +{
> +	if (stream->enabled)
> +		return;
> +
> +	stream->enabled = true;
> +
> +	xe_eu_stall_stream_enable(stream);
> +	hrtimer_start(&stream->poll_check_timer,
> +		      ns_to_ktime(stream->poll_period),
> +		      HRTIMER_MODE_REL);
> +}
> +
> +static void
> +xe_eu_stall_cntr_disable_locked(struct xe_eu_stall_data_stream *stream)
> +{
> +	if (!stream->enabled)
> +		return;
> +
> +	stream->enabled = false;
> +
> +	hrtimer_cancel(&stream->poll_check_timer);
> +	flush_workqueue(stream->buf_check_wq);
> +	xe_eu_stall_stream_disable(stream);
> +}
> +
> +static long
> +xe_eu_stall_stream_ioctl_locked(struct xe_eu_stall_data_stream *stream,
> +				unsigned int cmd, unsigned long arg)
> +{
> +	switch (cmd) {
> +	case DRM_XE_OBSERVATION_IOCTL_ENABLE:
> +		xe_eu_stall_cntr_enable_locked(stream);
> +		return 0;
> +	case DRM_XE_OBSERVATION_IOCTL_DISABLE:
> +		xe_eu_stall_cntr_disable_locked(stream);
> +		return 0;
> +	}
> +
> +	return -EINVAL;
> +}
> +
>  /**
>   * xe_eu_stall_stream_ioctl - support ioctl() usage with xe EU stall data
>   *								stream fd
> @@ -200,14 +752,22 @@ static long xe_eu_stall_stream_ioctl(struct file *file,
>  				     unsigned int cmd,
>  				     unsigned long arg)
>  {
> -	switch (cmd) {
> -	case DRM_XE_OBSERVATION_IOCTL_ENABLE:
> -		return 0;
> -	case DRM_XE_OBSERVATION_IOCTL_DISABLE:
> -		return 0;
> -	}
> +	struct xe_eu_stall_data_stream *stream = file->private_data;
> +	struct xe_gt *gt = stream->gt;
> +	long ret;
>  
> -	return -EINVAL;
> +	mutex_lock(&gt->eu_stall_cntr->lock);
> +	ret = xe_eu_stall_stream_ioctl_locked(stream, cmd, arg);
> +	mutex_unlock(&gt->eu_stall_cntr->lock);
> +
> +	return ret;
> +}
> +
> +static void
> +xe_eu_stall_stream_close_locked(struct xe_eu_stall_data_stream *stream)
> +{
> +	xe_eu_stall_cntr_disable_locked(stream);
> +	free_eu_stall_cntr_buf(stream);
>  }
>  
>  /**
> @@ -220,6 +780,19 @@ static long xe_eu_stall_stream_ioctl(struct file *file,
>   */
>  static int xe_eu_stall_stream_close(struct inode *inode, struct file *file)
>  {
> +	struct xe_eu_stall_data_stream *stream = file->private_data;
> +	struct xe_gt *gt = stream->gt;
> +
> +	mutex_lock(&gt->eu_stall_cntr->lock);
> +	xe_eu_stall_stream_close_locked(stream);
> +	kfree(stream->xecore_buf);
> +	kfree(stream);
> +	gt->eu_stall_cntr->stream = NULL;
> +	mutex_unlock(&gt->eu_stall_cntr->lock);
> +
> +	/* Release the reference the EU stall stream kept on the driver */
> +	drm_dev_put(&gt->tile->xe->drm);
> +
>  	return 0;
>  }
>  
> @@ -235,7 +808,98 @@ static const struct file_operations fops_eu_stall = {
>  
>  static inline bool has_eu_stall_sampling_support(struct xe_device *xe)
>  {
> -	return false;
> +	return ((xe->info.platform == XE_PVC || GRAPHICS_VER(xe) >= 20) ? true : false);
> +}
> +
> +/**
> + * xe_eu_stall_stream_open_locked - Open a EU stall data stream FD.
> + * @dev: drm device instance
> + * @props: individually validated u64 property value pairs
> + * @file: drm file
> + *
> + * Returns: zero on success or a negative error code.
> + */
> +static int
> +xe_eu_stall_stream_open_locked(struct drm_device *dev,
> +			       struct eu_stall_open_properties *props,
> +			       struct drm_file *file)
> +{
> +	struct xe_device *xe = to_xe_device(dev);
> +	struct xe_eu_stall_data_stream *stream;
> +	struct xe_gt *gt = props->gt;
> +	unsigned long f_flags = 0;
> +	xe_dss_mask_t all_xecore;
> +	int ret, stream_fd;
> +	u32 gt_buf_size;
> +	u16 num_xecore;
> +
> +	if (!has_eu_stall_sampling_support(xe)) {
> +		xe_gt_dbg(gt, "EU stall monitoring is not supported on this platform\n");
> +		return -EPERM;
> +	}
> +
> +	if (xe_observation_paranoid && !perfmon_capable()) {
> +		xe_gt_dbg(gt, "Insufficient privileges for EU stall monitoring\n");
> +		return -EACCES;
> +	}
> +
> +	/* Only one session can be active at any time */
> +	if (gt->eu_stall_cntr->stream) {
> +		xe_gt_dbg(gt, "EU stall cntr session already active\n");
> +		return -EBUSY;
> +	}
> +
> +	bitmap_or(all_xecore, gt->fuse_topo.g_dss_mask, gt->fuse_topo.c_dss_mask,
> +		  XE_MAX_DSS_FUSE_BITS);
> +	/*
> +	 * Enabled subslices can be discontiguous. Find the last subslice
> +	 * and calculate total buffer size based on that.
> +	 */
> +	num_xecore = xe_dss_mask_last_dss(all_xecore) + 1;
> +	gt_buf_size = SZ_512K * num_xecore;
> +	if (props->event_report_count > num_data_rows(gt_buf_size)) {
> +		xe_gt_dbg(gt, "Invalid EU stall data poll event report count %u\n",
> +			  props->event_report_count);
> +		xe_gt_dbg(gt, "Maximum event report count for the given buffer size is %u\n",
> +			  num_data_rows(gt_buf_size));
> +		return -EINVAL;
> +	}
> +
> +	stream = kzalloc(sizeof(*stream), GFP_KERNEL);
> +	if (!stream)
> +		return -ENOMEM;
> +
> +	gt->eu_stall_cntr->stream = stream;
> +	stream->gt = gt;
> +
> +	ret = xe_eu_stall_stream_init(stream, props, num_xecore);
> +	if (ret) {
> +		xe_gt_dbg(gt, "EU stall stream init failed : %d\n", ret);
> +		goto err_alloc;
> +	}
> +
> +	stream_fd = anon_inode_getfd("[xe_eu_stall]", &fops_eu_stall,
> +				     stream, f_flags);
> +	if (stream_fd < 0) {
> +		ret = stream_fd;
> +		xe_gt_dbg(gt, "EU stall inode get fd failed : %d\n", ret);
> +		goto err_open;
> +	}
> +
> +	/* Take a reference on the driver that will be kept with stream_fd
> +	 * until its release.
> +	 */
> +	drm_dev_get(&gt->tile->xe->drm);
> +
> +	return stream_fd;
> +
> +err_open:
> +	free_eu_stall_cntr_buf(stream);
> +err_alloc:
> +	gt->eu_stall_cntr->stream = NULL;
> +	kfree(stream->xecore_buf);
> +	kfree(stream);
> +	return ret;
>  }
>  
>  int xe_eu_stall_stream_open(struct drm_device *dev,
> @@ -244,10 +908,16 @@ int xe_eu_stall_stream_open(struct drm_device *dev,
>  {
>  	struct xe_device *xe = to_xe_device(dev);
>  	struct eu_stall_open_properties props;
> -	int ret, stream_fd;
> +	int ret;
>  
>  	memset(&props, 0, sizeof(struct eu_stall_open_properties));
>  
> +	/* Set default values */
> +	props.gt = NULL;
> +	props.eu_stall_sampling_rate = 4;
> +	props.poll_period = DEFAULT_POLL_PERIOD_NS;
> +	props.event_report_count = 1;
> +
>  	ret = xe_eu_stall_user_extensions(xe, data, &props);
>  	if (ret)
>  		return ret;
> @@ -257,19 +927,9 @@ int xe_eu_stall_stream_open(struct drm_device *dev,
>  		return -EINVAL;
>  	}
>  
> -	if (xe_observation_paranoid && !perfmon_capable()) {
> -		xe_gt_dbg(props.gt, "Insufficient privileges for EU stall monitoring\n");
> -		return -EACCES;
> -	}
> -
> -	if (!has_eu_stall_sampling_support(xe)) {
> -		xe_gt_dbg(props.gt, "EU stall monitoring is not supported on this platform\n");
> -		return -EPERM;
> -	}
> -	stream_fd = anon_inode_getfd("[xe_eu_stall]", &fops_eu_stall,
> -				     NULL, 0);
> -	if (stream_fd < 0)
> -		xe_gt_dbg(props.gt, "EU stall inode get fd failed : %d\n", stream_fd);
> +	mutex_lock(&props.gt->eu_stall_cntr->lock);
> +	ret = xe_eu_stall_stream_open_locked(dev, &props, file);
> +	mutex_unlock(&props.gt->eu_stall_cntr->lock);
>  
> -	return stream_fd;
> +	return ret;
>  }
> diff --git a/drivers/gpu/drm/xe/xe_eu_stall.h b/drivers/gpu/drm/xe/xe_eu_stall.h
> index 70fc89480df2..ca83fe487278 100644
> --- a/drivers/gpu/drm/xe/xe_eu_stall.h
> +++ b/drivers/gpu/drm/xe/xe_eu_stall.h
> @@ -6,6 +6,52 @@
>  #ifndef __XE_EU_STALL_H__
>  #define __XE_EU_STALL_H__
>  
> +#include "xe_gt_types.h"
> +
> +struct per_xecore_buf {
> +	u8 *vaddr;
> +	u32 write;
> +	u32 read;
> +	/* lock to protect read and write pointers */
> +	struct mutex lock;
> +};
> +
> +/**
> + * struct xe_eu_stall_data_stream - state of EU stall data stream FD
> + */
> +struct xe_eu_stall_data_stream {
> +	bool pollin;
> +	bool enabled;
> +	u64 poll_period;
> +	u32 event_report_count;
> +	size_t per_xecore_buf_size;
> +	wait_queue_head_t poll_wq;
> +
> +	struct xe_gt *gt;
> +	struct xe_bo *bo;
> +	struct per_xecore_buf *xecore_buf;
> +	struct {
> +		bool reported_to_user;
> +		xe_dss_mask_t mask;
> +		/* lock to protect mask */
> +		spinlock_t lock;
> +	} data_drop;
> +	struct hrtimer poll_check_timer;
> +	struct work_struct buf_check_work;
> +	struct workqueue_struct *buf_check_wq;
> +};
> +
> +struct xe_eu_stall_cntr_gt {
> +	/* Lock to protect stream */
> +	struct mutex lock;
> +
> +	/* Execution Unit (EU) stall data stream */
> +	struct xe_eu_stall_data_stream *stream;
> +};
> +
> +int xe_eu_stall_init(struct xe_gt *gt);
> +void xe_eu_stall_fini(struct xe_gt *gt);
> +
>  int xe_eu_stall_stream_open(struct drm_device *dev,
>  			    u64 data,
>  			    struct drm_file *file);
> diff --git a/drivers/gpu/drm/xe/xe_gt.c b/drivers/gpu/drm/xe/xe_gt.c
> index 1c79660fb086..968ad733f49d 100644
> --- a/drivers/gpu/drm/xe/xe_gt.c
> +++ b/drivers/gpu/drm/xe/xe_gt.c
> @@ -59,6 +59,7 @@
>  #include "xe_vm.h"
>  #include "xe_wa.h"
>  #include "xe_wopcm.h"
> +#include "xe_eu_stall.h"
>  
>  static void gt_fini(struct drm_device *drm, void *arg)
>  {
> @@ -157,6 +158,7 @@ void xe_gt_remove(struct xe_gt *gt)
>  		xe_hw_fence_irq_finish(&gt->fence_irq[i]);
>  
>  	xe_gt_disable_host_l2_vram(gt);
> +	xe_eu_stall_fini(gt);
>  }
>  
>  static void gt_reset_worker(struct work_struct *w);
> @@ -619,6 +621,10 @@ int xe_gt_init(struct xe_gt *gt)
>  
>  	xe_gt_record_user_engines(gt);
>  
> +	err = xe_eu_stall_init(gt);
> +	if (err)
> +		return err;
> +
>  	return 0;
>  }
>  
> diff --git a/drivers/gpu/drm/xe/xe_gt_types.h b/drivers/gpu/drm/xe/xe_gt_types.h
> index a287b98ee70b..8160fa894409 100644
> --- a/drivers/gpu/drm/xe/xe_gt_types.h
> +++ b/drivers/gpu/drm/xe/xe_gt_types.h
> @@ -430,6 +430,9 @@ struct xe_gt {
>  
>  	/** @oa: oa observation subsystem per gt info */
>  	struct xe_oa_gt oa;
> +
> +	/** @eu_stall_cntr: EU stall counters subsystem per gt info */
> +	struct xe_eu_stall_cntr_gt *eu_stall_cntr;
>  };
>  
>  #endif
> diff --git a/drivers/gpu/drm/xe/xe_trace.h b/drivers/gpu/drm/xe/xe_trace.h
> index 91130ad8999c..e65bdd916d00 100644
> --- a/drivers/gpu/drm/xe/xe_trace.h
> +++ b/drivers/gpu/drm/xe/xe_trace.h
> @@ -422,6 +422,41 @@ DEFINE_EVENT(xe_pm_runtime, xe_pm_runtime_get_ioctl,
>  	     TP_ARGS(xe, caller)
>  );
>  
> +TRACE_EVENT(xe_eu_stall_cntr_read,
> +	    TP_PROTO(u8 slice, u8 subslice,
> +		     u32 read_ptr, u32 write_ptr,
> +		     u32 read_offset, u32 write_offset,
> +		     size_t total_size),
> +	    TP_ARGS(slice, subslice, read_ptr, write_ptr,
> +		    read_offset, write_offset, total_size),
> +
> +	    TP_STRUCT__entry(
> +			     __field(u8, slice)
> +			     __field(u8, subslice)
> +			     __field(u32, read_ptr)
> +			     __field(u32, write_ptr)
> +			     __field(u32, read_offset)
> +			     __field(u32, write_offset)
> +			     __field(size_t, total_size)
> +			     ),
> +
> +	    TP_fast_assign(
> +			   __entry->slice = slice;
> +			   __entry->subslice = subslice;
> +			   __entry->read_ptr = read_ptr;
> +			   __entry->write_ptr = write_ptr;
> +			   __entry->read_offset = read_offset;
> +			   __entry->write_offset = write_offset;
> +			   __entry->total_size = total_size;
> +			   ),
> +
> +	    TP_printk("slice:%u subslice:%u readptr:0x%x writeptr:0x%x read off:%u write off:%u size:%zu ",
> +		      __entry->slice, __entry->subslice,
> +		      __entry->read_ptr, __entry->write_ptr,
> +		      __entry->read_offset, __entry->write_offset,
> +		      __entry->total_size)
> +);
> +
>  #endif
>  
>  /* This part must be outside protection */
> -- 
> 2.45.1
> 

-- 
Matt Roper
Graphics Software Engineer
Linux GPU Platform Enablement
Intel Corporation


More information about the Intel-xe mailing list