xe_query.c

// SPDX-License-Identifier: MIT
/*
 * Copyright © 2022 Intel Corporation
 */

#include "xe_query.h"

#include <linux/nospec.h>
#include <linux/sched/clock.h>

#include <drm/ttm/ttm_placement.h>
#include <drm/xe_drm.h>

#include "regs/xe_engine_regs.h"
#include "xe_bo.h"
#include "xe_device.h"
#include "xe_exec_queue.h"
#include "xe_ggtt.h"
#include "xe_gt.h"
#include "xe_guc_hwconfig.h"
#include "xe_macros.h"
#include "xe_mmio.h"
#include "xe_ttm_vram_mgr.h"

static const u16 xe_to_user_engine_class[] = {
	[XE_ENGINE_CLASS_RENDER] = DRM_XE_ENGINE_CLASS_RENDER,
	[XE_ENGINE_CLASS_COPY] = DRM_XE_ENGINE_CLASS_COPY,
	[XE_ENGINE_CLASS_VIDEO_DECODE] = DRM_XE_ENGINE_CLASS_VIDEO_DECODE,
	[XE_ENGINE_CLASS_VIDEO_ENHANCE] = DRM_XE_ENGINE_CLASS_VIDEO_ENHANCE,
	[XE_ENGINE_CLASS_COMPUTE] = DRM_XE_ENGINE_CLASS_COMPUTE,
};

static const enum xe_engine_class user_to_xe_engine_class[] = {
	[DRM_XE_ENGINE_CLASS_RENDER] = XE_ENGINE_CLASS_RENDER,
	[DRM_XE_ENGINE_CLASS_COPY] = XE_ENGINE_CLASS_COPY,
	[DRM_XE_ENGINE_CLASS_VIDEO_DECODE] = XE_ENGINE_CLASS_VIDEO_DECODE,
	[DRM_XE_ENGINE_CLASS_VIDEO_ENHANCE] = XE_ENGINE_CLASS_VIDEO_ENHANCE,
	[DRM_XE_ENGINE_CLASS_COMPUTE] = XE_ENGINE_CLASS_COMPUTE,
};

static size_t calc_hw_engine_info_size(struct xe_device *xe)
{
	struct xe_hw_engine *hwe;
	enum xe_hw_engine_id id;
	struct xe_gt *gt;
	u8 gt_id;
	int i = 0;

	for_each_gt(gt, xe, gt_id)
		for_each_hw_engine(hwe, gt, id) {
			if (xe_hw_engine_is_reserved(hwe))
				continue;
			i++;
		}

	return i * sizeof(struct drm_xe_query_engine_info);
}

typedef u64 (*__ktime_func_t)(void);
static __ktime_func_t __clock_id_to_func(clockid_t clk_id)
{
	/*
	 * Use logic same as the perf subsystem to allow user to select the
	 * reference clock id to be used for timestamps.
	 */
	switch (clk_id) {
	case CLOCK_MONOTONIC:
		return &ktime_get_ns;
	case CLOCK_MONOTONIC_RAW:
		return &ktime_get_raw_ns;
	case CLOCK_REALTIME:
		return &ktime_get_real_ns;
	case CLOCK_BOOTTIME:
		return &ktime_get_boottime_ns;
	case CLOCK_TAI:
		return &ktime_get_clocktai_ns;
	default:
		return NULL;
	}
}

static void
__read_timestamps(struct xe_gt *gt,
		  struct xe_reg lower_reg,
		  struct xe_reg upper_reg,
		  u64 *engine_ts,
		  u64 *cpu_ts,
		  u64 *cpu_delta,
		  __ktime_func_t cpu_clock)
{
	u32 upper, lower, old_upper, loop = 0;

	upper = xe_mmio_read32(gt, upper_reg);
	do {
		*cpu_delta = local_clock();
		*cpu_ts = cpu_clock();
		lower = xe_mmio_read32(gt, lower_reg);
		*cpu_delta = local_clock() - *cpu_delta;
		old_upper = upper;
		upper = xe_mmio_read32(gt, upper_reg);
	} while (upper != old_upper && loop++ < 2);

	*engine_ts = (u64)upper << 32 | lower;
}

static int
query_engine_cycles(struct xe_device *xe,
		    struct drm_xe_device_query *query)
{
	struct drm_xe_query_engine_cycles __user *query_ptr;
	struct drm_xe_engine_class_instance *eci;
	struct drm_xe_query_engine_cycles resp;
	size_t size = sizeof(resp);
	__ktime_func_t cpu_clock;
	struct xe_hw_engine *hwe;
	struct xe_gt *gt;

	if (query->size == 0) {
		query->size = size;
		return 0;
	} else if (XE_IOCTL_DBG(xe, query->size != size)) {
		return -EINVAL;
	}

	query_ptr = u64_to_user_ptr(query->data);
	if (copy_from_user(&resp, query_ptr, size))
		return -EFAULT;

	cpu_clock = __clock_id_to_func(resp.clockid);
	if (!cpu_clock)
		return -EINVAL;

	eci = &resp.eci;
	if (eci->gt_id > XE_MAX_GT_PER_TILE)
		return -EINVAL;

	gt = xe_device_get_gt(xe, eci->gt_id);
	if (!gt)
		return -EINVAL;

	if (eci->engine_class >= ARRAY_SIZE(user_to_xe_engine_class))
		return -EINVAL;

	hwe = xe_gt_hw_engine(gt, user_to_xe_engine_class[eci->engine_class],
			      eci->engine_instance, true);
	if (!hwe)
		return -EINVAL;

	resp.engine_frequency = gt->info.clock_freq;

	xe_device_mem_access_get(xe);
	xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL);

	__read_timestamps(gt,
			  RING_TIMESTAMP(hwe->mmio_base),
			  RING_TIMESTAMP_UDW(hwe->mmio_base),
			  &resp.engine_cycles,
			  &resp.cpu_timestamp,
			  &resp.cpu_delta,
			  cpu_clock);

	xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL);
	xe_device_mem_access_put(xe);
	resp.width = 36;

	/* Only write to the output fields of user query */
	if (put_user(resp.engine_frequency, &query_ptr->engine_frequency))
		return -EFAULT;

	if (put_user(resp.cpu_timestamp, &query_ptr->cpu_timestamp))
		return -EFAULT;

	if (put_user(resp.cpu_delta, &query_ptr->cpu_delta))
		return -EFAULT;

	if (put_user(resp.engine_cycles, &query_ptr->engine_cycles))
		return -EFAULT;

	if (put_user(resp.width, &query_ptr->width))
		return -EFAULT;

	return 0;
}

static int query_engines(struct xe_device *xe,
			 struct drm_xe_device_query *query)
{
	size_t size = calc_hw_engine_info_size(xe);
	struct drm_xe_query_engine_info __user *query_ptr =
		u64_to_user_ptr(query->data);
	struct drm_xe_query_engine_info *hw_engine_info;
	struct xe_hw_engine *hwe;
	enum xe_hw_engine_id id;
	struct xe_gt *gt;
	u8 gt_id;
	int i = 0;

	if (query->size == 0) {
		query->size = size;
		return 0;
	} else if (XE_IOCTL_DBG(xe, query->size != size)) {
		return -EINVAL;
	}

	hw_engine_info = kmalloc(size, GFP_KERNEL);
	if (!hw_engine_info)
		return -ENOMEM;

	for_each_gt(gt, xe, gt_id)
		for_each_hw_engine(hwe, gt, id) {
			if (xe_hw_engine_is_reserved(hwe))
				continue;

			hw_engine_info[i].instance.engine_class =
				xe_to_user_engine_class[hwe->class];
			hw_engine_info[i].instance.engine_instance =
				hwe->logical_instance;
			hw_engine_info[i].instance.gt_id = gt->info.id;
			hw_engine_info[i].instance.pad = 0;
			memset(hw_engine_info->reserved, 0, sizeof(hw_engine_info->reserved));

			i++;
		}

	if (copy_to_user(query_ptr, hw_engine_info, size)) {
		kfree(hw_engine_info);
		return -EFAULT;
	}
	kfree(hw_engine_info);

	return 0;
}

static size_t calc_mem_regions_size(struct xe_device *xe)
{
	u32 num_managers = 1;
	int i;

	for (i = XE_PL_VRAM0; i <= XE_PL_VRAM1; ++i)
		if (ttm_manager_type(&xe->ttm, i))
			num_managers++;

	return offsetof(struct drm_xe_query_mem_regions, mem_regions[num_managers]);
}

static int query_mem_regions(struct xe_device *xe,
			    struct drm_xe_device_query *query)
{
	size_t size = calc_mem_regions_size(xe);
	struct drm_xe_query_mem_regions *mem_regions;
	struct drm_xe_query_mem_regions __user *query_ptr =
		u64_to_user_ptr(query->data);
	struct ttm_resource_manager *man;
	int ret, i;

	if (query->size == 0) {
		query->size = size;
		return 0;
	} else if (XE_IOCTL_DBG(xe, query->size != size)) {
		return -EINVAL;
	}

	mem_regions = kzalloc(size, GFP_KERNEL);
	if (XE_IOCTL_DBG(xe, !mem_regions))
		return -ENOMEM;

	man = ttm_manager_type(&xe->ttm, XE_PL_TT);
	mem_regions->mem_regions[0].mem_class = DRM_XE_MEM_REGION_CLASS_SYSMEM;
	mem_regions->mem_regions[0].instance = 0;
	mem_regions->mem_regions[0].min_page_size = PAGE_SIZE;
	mem_regions->mem_regions[0].total_size = man->size << PAGE_SHIFT;
	if (perfmon_capable())
		mem_regions->mem_regions[0].used = ttm_resource_manager_usage(man);
	mem_regions->num_mem_regions = 1;

	for (i = XE_PL_VRAM0; i <= XE_PL_VRAM1; ++i) {
		man = ttm_manager_type(&xe->ttm, i);
		if (man) {
			mem_regions->mem_regions[mem_regions->num_mem_regions].mem_class =
				DRM_XE_MEM_REGION_CLASS_VRAM;
			mem_regions->mem_regions[mem_regions->num_mem_regions].instance =
				mem_regions->num_mem_regions;
			mem_regions->mem_regions[mem_regions->num_mem_regions].min_page_size =
				xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K ?
				SZ_64K : PAGE_SIZE;
			mem_regions->mem_regions[mem_regions->num_mem_regions].total_size =
				man->size;

			if (perfmon_capable()) {
				xe_ttm_vram_get_used(man,
					&mem_regions->mem_regions
					[mem_regions->num_mem_regions].used,
					&mem_regions->mem_regions
					[mem_regions->num_mem_regions].cpu_visible_used);
			}

			mem_regions->mem_regions[mem_regions->num_mem_regions].cpu_visible_size =
				xe_ttm_vram_get_cpu_visible_size(man);
			mem_regions->num_mem_regions++;
		}
	}

	if (!copy_to_user(query_ptr, mem_regions, size))
		ret = 0;
	else
		ret = -ENOSPC;

	kfree(mem_regions);
	return ret;
}

static int query_config(struct xe_device *xe, struct drm_xe_device_query *query)
{
	const u32 num_params = DRM_XE_QUERY_CONFIG_MAX_EXEC_QUEUE_PRIORITY + 1;
	size_t size =
		sizeof(struct drm_xe_query_config) + num_params * sizeof(u64);
	struct drm_xe_query_config __user *query_ptr =
		u64_to_user_ptr(query->data);
	struct drm_xe_query_config *config;

	if (query->size == 0) {
		query->size = size;
		return 0;
	} else if (XE_IOCTL_DBG(xe, query->size != size)) {
		return -EINVAL;
	}

	config = kzalloc(size, GFP_KERNEL);
	if (!config)
		return -ENOMEM;

	config->num_params = num_params;
	config->info[DRM_XE_QUERY_CONFIG_REV_AND_DEVICE_ID] =
		xe->info.devid | (xe->info.revid << 16);
	if (xe_device_get_root_tile(xe)->mem.vram.usable_size)
		config->info[DRM_XE_QUERY_CONFIG_FLAGS] =
			DRM_XE_QUERY_CONFIG_FLAG_HAS_VRAM;
	config->info[DRM_XE_QUERY_CONFIG_MIN_ALIGNMENT] =
		xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K ? SZ_64K : SZ_4K;
	config->info[DRM_XE_QUERY_CONFIG_VA_BITS] = xe->info.va_bits;
	config->info[DRM_XE_QUERY_CONFIG_MAX_EXEC_QUEUE_PRIORITY] =
		xe_exec_queue_device_get_max_priority(xe);

	if (copy_to_user(query_ptr, config, size)) {
		kfree(config);
		return -EFAULT;
	}
	kfree(config);

	return 0;
}

static int query_gt_list(struct xe_device *xe, struct drm_xe_device_query *query)
{
	struct xe_gt *gt;
	size_t size = sizeof(struct drm_xe_query_gt_list) +
		xe->info.gt_count * sizeof(struct drm_xe_gt);
	struct drm_xe_query_gt_list __user *query_ptr =
		u64_to_user_ptr(query->data);
	struct drm_xe_query_gt_list *gt_list;
	u8 id;

	if (query->size == 0) {
		query->size = size;
		return 0;
	} else if (XE_IOCTL_DBG(xe, query->size != size)) {
		return -EINVAL;
	}

	gt_list = kzalloc(size, GFP_KERNEL);
	if (!gt_list)
		return -ENOMEM;

	gt_list->num_gt = xe->info.gt_count;
	for_each_gt(gt, xe, id) {
		if (xe_gt_is_media_type(gt))
			gt_list->gt_list[id].type = DRM_XE_QUERY_GT_TYPE_MEDIA;
		else
			gt_list->gt_list[id].type = DRM_XE_QUERY_GT_TYPE_MAIN;
		gt_list->gt_list[id].gt_id = gt->info.id;
		gt_list->gt_list[id].clock_freq = gt->info.clock_freq;
		if (!IS_DGFX(xe))
			gt_list->gt_list[id].near_mem_regions = 0x1;
		else
			gt_list->gt_list[id].near_mem_regions =
				BIT(gt_to_tile(gt)->id) << 1;
		gt_list->gt_list[id].far_mem_regions = xe->info.mem_region_mask ^
			gt_list->gt_list[id].near_mem_regions;
	}

	if (copy_to_user(query_ptr, gt_list, size)) {
		kfree(gt_list);
		return -EFAULT;
	}
	kfree(gt_list);

	return 0;
}

static int query_hwconfig(struct xe_device *xe,
			  struct drm_xe_device_query *query)
{
	struct xe_gt *gt = xe_root_mmio_gt(xe);
	size_t size = xe_guc_hwconfig_size(&gt->uc.guc);
	void __user *query_ptr = u64_to_user_ptr(query->data);
	void *hwconfig;

	if (query->size == 0) {
		query->size = size;
		return 0;
	} else if (XE_IOCTL_DBG(xe, query->size != size)) {
		return -EINVAL;
	}

	hwconfig = kzalloc(size, GFP_KERNEL);
	if (!hwconfig)
		return -ENOMEM;

	xe_device_mem_access_get(xe);
	xe_guc_hwconfig_copy(&gt->uc.guc, hwconfig);
	xe_device_mem_access_put(xe);

	if (copy_to_user(query_ptr, hwconfig, size)) {
		kfree(hwconfig);
		return -EFAULT;
	}
	kfree(hwconfig);

	return 0;
}

static size_t calc_topo_query_size(struct xe_device *xe)
{
	return xe->info.gt_count *
		(3 * sizeof(struct drm_xe_query_topology_mask) +
		 sizeof_field(struct xe_gt, fuse_topo.g_dss_mask) +
		 sizeof_field(struct xe_gt, fuse_topo.c_dss_mask) +
		 sizeof_field(struct xe_gt, fuse_topo.eu_mask_per_dss));
}

static void __user *copy_mask(void __user *ptr,
			      struct drm_xe_query_topology_mask *topo,
			      void *mask, size_t mask_size)
{
	topo->num_bytes = mask_size;

	if (copy_to_user(ptr, topo, sizeof(*topo)))
		return ERR_PTR(-EFAULT);
	ptr += sizeof(topo);

	if (copy_to_user(ptr, mask, mask_size))
		return ERR_PTR(-EFAULT);
	ptr += mask_size;

	return ptr;
}

static int query_gt_topology(struct xe_device *xe,
			     struct drm_xe_device_query *query)
{
	void __user *query_ptr = u64_to_user_ptr(query->data);
	size_t size = calc_topo_query_size(xe);
	struct drm_xe_query_topology_mask topo;
	struct xe_gt *gt;
	int id;

	if (query->size == 0) {
		query->size = size;
		return 0;
	} else if (XE_IOCTL_DBG(xe, query->size != size)) {
		return -EINVAL;
	}

	for_each_gt(gt, xe, id) {
		topo.gt_id = id;

		topo.type = DRM_XE_TOPO_DSS_GEOMETRY;
		query_ptr = copy_mask(query_ptr, &topo,
				      gt->fuse_topo.g_dss_mask,
				      sizeof(gt->fuse_topo.g_dss_mask));
		if (IS_ERR(query_ptr))
			return PTR_ERR(query_ptr);

		topo.type = DRM_XE_TOPO_DSS_COMPUTE;
		query_ptr = copy_mask(query_ptr, &topo,
				      gt->fuse_topo.c_dss_mask,
				      sizeof(gt->fuse_topo.c_dss_mask));
		if (IS_ERR(query_ptr))
			return PTR_ERR(query_ptr);

		topo.type = DRM_XE_TOPO_EU_PER_DSS;
		query_ptr = copy_mask(query_ptr, &topo,
				      gt->fuse_topo.eu_mask_per_dss,
				      sizeof(gt->fuse_topo.eu_mask_per_dss));
		if (IS_ERR(query_ptr))
			return PTR_ERR(query_ptr);
	}

	return 0;
}

static int (* const xe_query_funcs[])(struct xe_device *xe,
				      struct drm_xe_device_query *query) = {
	query_engines,
	query_mem_regions,
	query_config,
	query_gt_list,
	query_hwconfig,
	query_gt_topology,
	query_engine_cycles,
};

int xe_query_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
{
	struct xe_device *xe = to_xe_device(dev);
	struct drm_xe_device_query *query = data;
	u32 idx;

	if (XE_IOCTL_DBG(xe, query->extensions) ||
	    XE_IOCTL_DBG(xe, query->reserved[0] || query->reserved[1]))
		return -EINVAL;

	if (XE_IOCTL_DBG(xe, query->query >= ARRAY_SIZE(xe_query_funcs)))
		return -EINVAL;

	idx = array_index_nospec(query->query, ARRAY_SIZE(xe_query_funcs));
	if (XE_IOCTL_DBG(xe, !xe_query_funcs[idx]))
		return -EINVAL;

	return xe_query_funcs[idx](xe, query);
}