Merge tag 'drm-for-v4.16' of git://people.freedesktop.org/~airlied/linux

Pull drm updates from Dave Airlie:
 "This seems to have been a comparatively quieter merge window, I assume
  due to holidays etc. The "biggest" change is AMD header cleanups, which
  merge/remove a bunch of them. The AMD gpu scheduler is now being made generic
  with the etnaviv driver wanting to reuse the code, hopefully other drivers
  can go in the same direction.

  Otherwise it's the usual lots of stuff in i915/amdgpu, not so much stuff
  elsewhere.

  Core:
   - Add .last_close and .output_poll_changed helpers to reduce driver footprints
   - Fix plane clipping
   - Improved debug printing support
   - Add panel orientation property
   - Update edid derived properties at edid setting
   - Reduction in fbdev driver footprint
   - Move amdgpu scheduler into core for other drivers to use.

  i915:
   - Selftest and IGT improvements
   - Fast boot prep work on IPS, pipe config
   - HW workarounds for Cannonlake, Geminilake
   - Cannonlake clock and HDMI2.0 fixes
   - GPU cache invalidation and context switch improvements
   - Display planes cleanup
   - New PMU interface for perf queries
   - New firmware support for KBL/SKL
   - Geminilake HW workaround for perforamce
   - Coffeelake stolen memory improvements
   - GPU reset robustness work
   - Cannonlake horizontal plane flipping
   - GVT work

  amdgpu/radeon:
   - RV and Vega header file cleanups (lots of lines gone!)
   - TTM operation context support
   - 48-bit GPUVM support for Vega/RV
   - ECC support for Vega
   - Resizeable BAR support
   - Multi-display sync support
   - Enable swapout for reserved BOs during allocation
   - S3 fixes on Raven
   - GPU reset cleanup and fixes
   - 2+1 level GPU page table

  amdkfd:
   - GFX7/8 SDMA user queues support
   - Hardware scheduling for multiple processes
   - dGPU prep work

  rcar:
   - Added R8A7743/5 support
   - System suspend/resume support

  sun4i:
   - Multi-plane support for YUV formats
   - A83T and LVDS support

  msm:
   - Devfreq support for GPU

  tegra:
   - Prep work for adding Tegra186 support
   - Tegra186 HDMI support
   - HDMI2.0 and zpos support by using generic helpers

  tilcdc:
   - Misc fixes

  omapdrm:
   - Support memory bandwidth limits
   - DSI command mode panel cleanups
   - DMM error handling

  exynos:
   - drop the old IPP subdriver.

  etnaviv:
   - Occlusion query fixes
   - Job handling fixes
   - Prep work for hooking in gpu scheduler

  armada:
   - Move closer to atomic modesetting
   - Allow disabling primary plane if overlay is full screen

  imx:
   - Format modifier support
   - Add tile prefetch to PRE
   - Runtime PM support for PRG

  ast:
   - fix LUT loading"

* tag 'drm-for-v4.16' of git://people.freedesktop.org/~airlied/linux: (1471 commits)
  drm/ast: Load lut in crtc_commit
  drm: Check for lessee in DROP_MASTER ioctl
  drm: fix gpu scheduler link order
  drm/amd/display: Demote error print to debug print when ATOM impl missing
  dma-buf: fix reservation_object_wait_timeout_rcu once more v2
  drm/amdgpu: Avoid leaking PM domain on driver unbind (v2)
  drm/amd/amdgpu: Add Polaris version check
  drm/amdgpu: Reenable manual GPU reset from sysfs
  drm/amdgpu: disable MMHUB power gating on raven
  drm/ttm: Don't unreserve swapped BOs that were previously reserved
  drm/ttm: Don't add swapped BOs to swap-LRU list
  drm/amdgpu: only check for ECC on Vega10
  drm/amd/powerplay: Fix smu_table_entry.handle type
  drm/ttm: add VADDR_FLAG_UPDATED_COUNT to correctly update dma_page global count
  drm: Fix PANEL_ORIENTATION_QUIRKS breaking the Kconfig DRM menuconfig
  drm/radeon: fill in rb backend map on evergreen/ni.
  drm/amdgpu/gfx9: fix ngg enablement to clear gds reserved memory (v2)
  drm/ttm: only free pages rather than update global memory count together
  drm/amdgpu: fix CPU based VM updates
  drm/amdgpu: fix typo in amdgpu_vce_validate_bo
  ...

1 files changed, 620 insertions, 441 deletions

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
index 19ce59028d6b..c6a76090a725 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
@@ -28,27 +28,32 @@
 #include <linux/hash.h>
 #include <linux/cpufreq.h>
 #include <linux/log2.h>
+#include <linux/dmi.h>
+#include <linux/atomic.h>
 
 #include "kfd_priv.h"
 #include "kfd_crat.h"
 #include "kfd_topology.h"
+#include "kfd_device_queue_manager.h"
 
+/* topology_device_list - Master list of all topology devices */
 static struct list_head topology_device_list;
-static int topology_crat_parsed;
 static struct kfd_system_properties sys_props;
 
 static DECLARE_RWSEM(topology_lock);
+static atomic_t topology_crat_proximity_domain;
 
-struct kfd_dev *kfd_device_by_id(uint32_t gpu_id)
+struct kfd_topology_device *kfd_topology_device_by_proximity_domain(
+						uint32_t proximity_domain)
 {
 	struct kfd_topology_device *top_dev;
-	struct kfd_dev *device = NULL;
+	struct kfd_topology_device *device = NULL;
 
 	down_read(&topology_lock);
 
 	list_for_each_entry(top_dev, &topology_device_list, list)
-		if (top_dev->gpu_id == gpu_id) {
-			device = top_dev->gpu;
+		if (top_dev->proximity_domain == proximity_domain) {
+			device = top_dev;
 			break;
 		}
 
@@ -57,7 +62,7 @@ struct kfd_dev *kfd_device_by_id(uint32_t gpu_id)
 	return device;
 }
 
-struct kfd_dev *kfd_device_by_pci_dev(const struct pci_dev *pdev)
+struct kfd_dev *kfd_device_by_id(uint32_t gpu_id)
 {
 	struct kfd_topology_device *top_dev;
 	struct kfd_dev *device = NULL;
@@ -65,7 +70,7 @@ struct kfd_dev *kfd_device_by_pci_dev(const struct pci_dev *pdev)
 	down_read(&topology_lock);
 
 	list_for_each_entry(top_dev, &topology_device_list, list)
-		if (top_dev->gpu->pdev == pdev) {
+		if (top_dev->gpu_id == gpu_id) {
 			device = top_dev->gpu;
 			break;
 		}
@@ -75,282 +80,31 @@ struct kfd_dev *kfd_device_by_pci_dev(const struct pci_dev *pdev)
 	return device;
 }
 
-static int kfd_topology_get_crat_acpi(void *crat_image, size_t *size)
-{
-	struct acpi_table_header *crat_table;
-	acpi_status status;
-
-	if (!size)
-		return -EINVAL;
-
-	/*
-	 * Fetch the CRAT table from ACPI
-	 */
-	status = acpi_get_table(CRAT_SIGNATURE, 0, &crat_table);
-	if (status == AE_NOT_FOUND) {
-		pr_warn("CRAT table not found\n");
-		return -ENODATA;
-	} else if (ACPI_FAILURE(status)) {
-		const char *err = acpi_format_exception(status);
-
-		pr_err("CRAT table error: %s\n", err);
-		return -EINVAL;
-	}
-
-	if (*size >= crat_table->length && crat_image != NULL)
-		memcpy(crat_image, crat_table, crat_table->length);
-
-	*size = crat_table->length;
-
-	return 0;
-}
-
-static void kfd_populated_cu_info_cpu(struct kfd_topology_device *dev,
-		struct crat_subtype_computeunit *cu)
-{
-	dev->node_props.cpu_cores_count = cu->num_cpu_cores;
-	dev->node_props.cpu_core_id_base = cu->processor_id_low;
-	if (cu->hsa_capability & CRAT_CU_FLAGS_IOMMU_PRESENT)
-		dev->node_props.capability |= HSA_CAP_ATS_PRESENT;
-
-	pr_info("CU CPU: cores=%d id_base=%d\n", cu->num_cpu_cores,
-			cu->processor_id_low);
-}
-
-static void kfd_populated_cu_info_gpu(struct kfd_topology_device *dev,
-		struct crat_subtype_computeunit *cu)
-{
-	dev->node_props.simd_id_base = cu->processor_id_low;
-	dev->node_props.simd_count = cu->num_simd_cores;
-	dev->node_props.lds_size_in_kb = cu->lds_size_in_kb;
-	dev->node_props.max_waves_per_simd = cu->max_waves_simd;
-	dev->node_props.wave_front_size = cu->wave_front_size;
-	dev->node_props.mem_banks_count = cu->num_banks;
-	dev->node_props.array_count = cu->num_arrays;
-	dev->node_props.cu_per_simd_array = cu->num_cu_per_array;
-	dev->node_props.simd_per_cu = cu->num_simd_per_cu;
-	dev->node_props.max_slots_scratch_cu = cu->max_slots_scatch_cu;
-	if (cu->hsa_capability & CRAT_CU_FLAGS_HOT_PLUGGABLE)
-		dev->node_props.capability |= HSA_CAP_HOT_PLUGGABLE;
-	pr_info("CU GPU: simds=%d id_base=%d\n", cu->num_simd_cores,
-				cu->processor_id_low);
-}
-
-/* kfd_parse_subtype_cu is called when the topology mutex is already acquired */
-static int kfd_parse_subtype_cu(struct crat_subtype_computeunit *cu)
-{
-	struct kfd_topology_device *dev;
-	int i = 0;
-
-	pr_info("Found CU entry in CRAT table with proximity_domain=%d caps=%x\n",
-			cu->proximity_domain, cu->hsa_capability);
-	list_for_each_entry(dev, &topology_device_list, list) {
-		if (cu->proximity_domain == i) {
-			if (cu->flags & CRAT_CU_FLAGS_CPU_PRESENT)
-				kfd_populated_cu_info_cpu(dev, cu);
-
-			if (cu->flags & CRAT_CU_FLAGS_GPU_PRESENT)
-				kfd_populated_cu_info_gpu(dev, cu);
-			break;
-		}
-		i++;
-	}
-
-	return 0;
-}
-
-/*
- * kfd_parse_subtype_mem is called when the topology mutex is
- * already acquired
- */
-static int kfd_parse_subtype_mem(struct crat_subtype_memory *mem)
-{
-	struct kfd_mem_properties *props;
-	struct kfd_topology_device *dev;
-	int i = 0;
-
-	pr_info("Found memory entry in CRAT table with proximity_domain=%d\n",
-			mem->promixity_domain);
-	list_for_each_entry(dev, &topology_device_list, list) {
-		if (mem->promixity_domain == i) {
-			props = kfd_alloc_struct(props);
-			if (props == NULL)
-				return -ENOMEM;
-
-			if (dev->node_props.cpu_cores_count == 0)
-				props->heap_type = HSA_MEM_HEAP_TYPE_FB_PRIVATE;
-			else
-				props->heap_type = HSA_MEM_HEAP_TYPE_SYSTEM;
-
-			if (mem->flags & CRAT_MEM_FLAGS_HOT_PLUGGABLE)
-				props->flags |= HSA_MEM_FLAGS_HOT_PLUGGABLE;
-			if (mem->flags & CRAT_MEM_FLAGS_NON_VOLATILE)
-				props->flags |= HSA_MEM_FLAGS_NON_VOLATILE;
-
-			props->size_in_bytes =
-				((uint64_t)mem->length_high << 32) +
-							mem->length_low;
-			props->width = mem->width;
-
-			dev->mem_bank_count++;
-			list_add_tail(&props->list, &dev->mem_props);
-
-			break;
-		}
-		i++;
-	}
-
-	return 0;
-}
-
-/*
- * kfd_parse_subtype_cache is called when the topology mutex
- * is already acquired
- */
-static int kfd_parse_subtype_cache(struct crat_subtype_cache *cache)
-{
-	struct kfd_cache_properties *props;
-	struct kfd_topology_device *dev;
-	uint32_t id;
-
-	id = cache->processor_id_low;
-
-	pr_info("Found cache entry in CRAT table with processor_id=%d\n", id);
-	list_for_each_entry(dev, &topology_device_list, list)
-		if (id == dev->node_props.cpu_core_id_base ||
-		    id == dev->node_props.simd_id_base) {
-			props = kfd_alloc_struct(props);
-			if (props == NULL)
-				return -ENOMEM;
-
-			props->processor_id_low = id;
-			props->cache_level = cache->cache_level;
-			props->cache_size = cache->cache_size;
-			props->cacheline_size = cache->cache_line_size;
-			props->cachelines_per_tag = cache->lines_per_tag;
-			props->cache_assoc = cache->associativity;
-			props->cache_latency = cache->cache_latency;
-
-			if (cache->flags & CRAT_CACHE_FLAGS_DATA_CACHE)
-				props->cache_type |= HSA_CACHE_TYPE_DATA;
-			if (cache->flags & CRAT_CACHE_FLAGS_INST_CACHE)
-				props->cache_type |= HSA_CACHE_TYPE_INSTRUCTION;
-			if (cache->flags & CRAT_CACHE_FLAGS_CPU_CACHE)
-				props->cache_type |= HSA_CACHE_TYPE_CPU;
-			if (cache->flags & CRAT_CACHE_FLAGS_SIMD_CACHE)
-				props->cache_type |= HSA_CACHE_TYPE_HSACU;
-
-			dev->cache_count++;
-			dev->node_props.caches_count++;
-			list_add_tail(&props->list, &dev->cache_props);
-
-			break;
-		}
-
-	return 0;
-}
-
-/*
- * kfd_parse_subtype_iolink is called when the topology mutex
- * is already acquired
- */
-static int kfd_parse_subtype_iolink(struct crat_subtype_iolink *iolink)
+struct kfd_dev *kfd_device_by_pci_dev(const struct pci_dev *pdev)
 {
-	struct kfd_iolink_properties *props;
-	struct kfd_topology_device *dev;
-	uint32_t i = 0;
-	uint32_t id_from;
-	uint32_t id_to;
-
-	id_from = iolink->proximity_domain_from;
-	id_to = iolink->proximity_domain_to;
+	struct kfd_topology_device *top_dev;
+	struct kfd_dev *device = NULL;
 
-	pr_info("Found IO link entry in CRAT table with id_from=%d\n", id_from);
-	list_for_each_entry(dev, &topology_device_list, list) {
-		if (id_from == i) {
-			props = kfd_alloc_struct(props);
-			if (props == NULL)
-				return -ENOMEM;
-
-			props->node_from = id_from;
-			props->node_to = id_to;
-			props->ver_maj = iolink->version_major;
-			props->ver_min = iolink->version_minor;
-
-			/*
-			 * weight factor (derived from CDIR), currently always 1
-			 */
-			props->weight = 1;
-
-			props->min_latency = iolink->minimum_latency;
-			props->max_latency = iolink->maximum_latency;
-			props->min_bandwidth = iolink->minimum_bandwidth_mbs;
-			props->max_bandwidth = iolink->maximum_bandwidth_mbs;
-			props->rec_transfer_size =
-					iolink->recommended_transfer_size;
-
-			dev->io_link_count++;
-			dev->node_props.io_links_count++;
-			list_add_tail(&props->list, &dev->io_link_props);
+	down_read(&topology_lock);
 
+	list_for_each_entry(top_dev, &topology_device_list, list)
+		if (top_dev->gpu->pdev == pdev) {
+			device = top_dev->gpu;
 			break;
 		}
-		i++;
-	}
 
-	return 0;
-}
-
-static int kfd_parse_subtype(struct crat_subtype_generic *sub_type_hdr)
-{
-	struct crat_subtype_computeunit *cu;
-	struct crat_subtype_memory *mem;
-	struct crat_subtype_cache *cache;
-	struct crat_subtype_iolink *iolink;
-	int ret = 0;
-
-	switch (sub_type_hdr->type) {
-	case CRAT_SUBTYPE_COMPUTEUNIT_AFFINITY:
-		cu = (struct crat_subtype_computeunit *)sub_type_hdr;
-		ret = kfd_parse_subtype_cu(cu);
-		break;
-	case CRAT_SUBTYPE_MEMORY_AFFINITY:
-		mem = (struct crat_subtype_memory *)sub_type_hdr;
-		ret = kfd_parse_subtype_mem(mem);
-		break;
-	case CRAT_SUBTYPE_CACHE_AFFINITY:
-		cache = (struct crat_subtype_cache *)sub_type_hdr;
-		ret = kfd_parse_subtype_cache(cache);
-		break;
-	case CRAT_SUBTYPE_TLB_AFFINITY:
-		/*
-		 * For now, nothing to do here
-		 */
-		pr_info("Found TLB entry in CRAT table (not processing)\n");
-		break;
-	case CRAT_SUBTYPE_CCOMPUTE_AFFINITY:
-		/*
-		 * For now, nothing to do here
-		 */
-		pr_info("Found CCOMPUTE entry in CRAT table (not processing)\n");
-		break;
-	case CRAT_SUBTYPE_IOLINK_AFFINITY:
-		iolink = (struct crat_subtype_iolink *)sub_type_hdr;
-		ret = kfd_parse_subtype_iolink(iolink);
-		break;
-	default:
-		pr_warn("Unknown subtype (%d) in CRAT\n",
-				sub_type_hdr->type);
-	}
+	up_read(&topology_lock);
 
-	return ret;
+	return device;
 }
 
+/* Called with write topology_lock acquired */
 static void kfd_release_topology_device(struct kfd_topology_device *dev)
 {
 	struct kfd_mem_properties *mem;
 	struct kfd_cache_properties *cache;
 	struct kfd_iolink_properties *iolink;
+	struct kfd_perf_properties *perf;
 
 	list_del(&dev->list);
 
@@ -375,25 +129,35 @@ static void kfd_release_topology_device(struct kfd_topology_device *dev)
 		kfree(iolink);
 	}
 
-	kfree(dev);
+	while (dev->perf_props.next != &dev->perf_props) {
+		perf = container_of(dev->perf_props.next,
+				struct kfd_perf_properties, list);
+		list_del(&perf->list);
+		kfree(perf);
+	}
 
-	sys_props.num_devices--;
+	kfree(dev);
 }
 
-static void kfd_release_live_view(void)
+void kfd_release_topology_device_list(struct list_head *device_list)
 {
 	struct kfd_topology_device *dev;
 
-	while (topology_device_list.next != &topology_device_list) {
-		dev = container_of(topology_device_list.next,
-				 struct kfd_topology_device, list);
+	while (!list_empty(device_list)) {
+		dev = list_first_entry(device_list,
+				       struct kfd_topology_device, list);
 		kfd_release_topology_device(dev);
+	}
 }
 
+static void kfd_release_live_view(void)
+{
+	kfd_release_topology_device_list(&topology_device_list);
 	memset(&sys_props, 0, sizeof(sys_props));
 }
 
-static struct kfd_topology_device *kfd_create_topology_device(void)
+struct kfd_topology_device *kfd_create_topology_device(
+				struct list_head *device_list)
 {
 	struct kfd_topology_device *dev;
 
@@ -406,65 +170,13 @@ static struct kfd_topology_device *kfd_create_topology_device(void)
 	INIT_LIST_HEAD(&dev->mem_props);
 	INIT_LIST_HEAD(&dev->cache_props);
 	INIT_LIST_HEAD(&dev->io_link_props);
+	INIT_LIST_HEAD(&dev->perf_props);
 
-	list_add_tail(&dev->list, &topology_device_list);
-	sys_props.num_devices++;
+	list_add_tail(&dev->list, device_list);
 
 	return dev;
 }
 
-static int kfd_parse_crat_table(void *crat_image)
-{
-	struct kfd_topology_device *top_dev;
-	struct crat_subtype_generic *sub_type_hdr;
-	uint16_t node_id;
-	int ret;
-	struct crat_header *crat_table = (struct crat_header *)crat_image;
-	uint16_t num_nodes;
-	uint32_t image_len;
-
-	if (!crat_image)
-		return -EINVAL;
-
-	num_nodes = crat_table->num_domains;
-	image_len = crat_table->length;
-
-	pr_info("Parsing CRAT table with %d nodes\n", num_nodes);
-
-	for (node_id = 0; node_id < num_nodes; node_id++) {
-		top_dev = kfd_create_topology_device();
-		if (!top_dev) {
-			kfd_release_live_view();
-			return -ENOMEM;
-		}
-	}
-
-	sys_props.platform_id =
-		(*((uint64_t *)crat_table->oem_id)) & CRAT_OEMID_64BIT_MASK;
-	sys_props.platform_oem = *((uint64_t *)crat_table->oem_table_id);
-	sys_props.platform_rev = crat_table->revision;
-
-	sub_type_hdr = (struct crat_subtype_generic *)(crat_table+1);
-	while ((char *)sub_type_hdr + sizeof(struct crat_subtype_generic) <
-			((char *)crat_image) + image_len) {
-		if (sub_type_hdr->flags & CRAT_SUBTYPE_FLAGS_ENABLED) {
-			ret = kfd_parse_subtype(sub_type_hdr);
-			if (ret != 0) {
-				kfd_release_live_view();
-				return ret;
-			}
-		}
-
-		sub_type_hdr = (typeof(sub_type_hdr))((char *)sub_type_hdr +
-				sub_type_hdr->length);
-	}
-
-	sys_props.generation_count++;
-	topology_crat_parsed = 1;
-
-	return 0;
-}
-
 
 #define sysfs_show_gen_prop(buffer, fmt, ...) \
 		snprintf(buffer, PAGE_SIZE, "%s"fmt, buffer, __VA_ARGS__)
@@ -501,11 +213,17 @@ static ssize_t sysprops_show(struct kobject *kobj, struct attribute *attr,
 	return ret;
 }
 
+static void kfd_topology_kobj_release(struct kobject *kobj)
+{
+	kfree(kobj);
+}
+
 static const struct sysfs_ops sysprops_ops = {
 	.show = sysprops_show,
 };
 
 static struct kobj_type sysprops_type = {
+	.release = kfd_topology_kobj_release,
 	.sysfs_ops = &sysprops_ops,
 };
 
@@ -541,6 +259,7 @@ static const struct sysfs_ops iolink_ops = {
 };
 
 static struct kobj_type iolink_type = {
+	.release = kfd_topology_kobj_release,
 	.sysfs_ops = &iolink_ops,
 };
 
@@ -568,6 +287,7 @@ static const struct sysfs_ops mem_ops = {
 };
 
 static struct kobj_type mem_type = {
+	.release = kfd_topology_kobj_release,
 	.sysfs_ops = &mem_ops,
 };
 
@@ -575,7 +295,7 @@ static ssize_t kfd_cache_show(struct kobject *kobj, struct attribute *attr,
 		char *buffer)
 {
 	ssize_t ret;
-	uint32_t i;
+	uint32_t i, j;
 	struct kfd_cache_properties *cache;
 
 	/* Making sure that the buffer is an empty string */
@@ -593,12 +313,18 @@ static ssize_t kfd_cache_show(struct kobject *kobj, struct attribute *attr,
 	sysfs_show_32bit_prop(buffer, "latency", cache->cache_latency);
 	sysfs_show_32bit_prop(buffer, "type", cache->cache_type);
 	snprintf(buffer, PAGE_SIZE, "%ssibling_map ", buffer);
-	for (i = 0; i < KFD_TOPOLOGY_CPU_SIBLINGS; i++)
-		ret = snprintf(buffer, PAGE_SIZE, "%s%d%s",
-				buffer, cache->sibling_map[i],
-				(i == KFD_TOPOLOGY_CPU_SIBLINGS-1) ?
-						"\n" : ",");
-
+	for (i = 0; i < CRAT_SIBLINGMAP_SIZE; i++)
+		for (j = 0; j < sizeof(cache->sibling_map[0])*8; j++) {
+			/* Check each bit */
+			if (cache->sibling_map[i] & (1 << j))
+				ret = snprintf(buffer, PAGE_SIZE,
+					 "%s%d%s", buffer, 1, ",");
+			else
+				ret = snprintf(buffer, PAGE_SIZE,
+					 "%s%d%s", buffer, 0, ",");
+		}
+	/* Replace the last "," with end of line */
+	*(buffer + strlen(buffer) - 1) = 0xA;
 	return ret;
 }
 
@@ -607,9 +333,43 @@ static const struct sysfs_ops cache_ops = {
 };
 
 static struct kobj_type cache_type = {
+	.release = kfd_topology_kobj_release,
 	.sysfs_ops = &cache_ops,
 };
 
+/****** Sysfs of Performance Counters ******/
+
+struct kfd_perf_attr {
+	struct kobj_attribute attr;
+	uint32_t data;
+};
+
+static ssize_t perf_show(struct kobject *kobj, struct kobj_attribute *attrs,
+			char *buf)
+{
+	struct kfd_perf_attr *attr;
+
+	buf[0] = 0;
+	attr = container_of(attrs, struct kfd_perf_attr, attr);
+	if (!attr->data) /* invalid data for PMC */
+		return 0;
+	else
+		return sysfs_show_32bit_val(buf, attr->data);
+}
+
+#define KFD_PERF_DESC(_name, _data)			\
+{							\
+	.attr  = __ATTR(_name, 0444, perf_show, NULL),	\
+	.data = _data,					\
+}
+
+static struct kfd_perf_attr perf_attr_iommu[] = {
+	KFD_PERF_DESC(max_concurrent, 0),
+	KFD_PERF_DESC(num_counters, 0),
+	KFD_PERF_DESC(counter_ids, 0),
+};
+/****************************************/
+
 static ssize_t node_show(struct kobject *kobj, struct attribute *attr,
 		char *buffer)
 {
@@ -646,18 +406,8 @@ static ssize_t node_show(struct kobject *kobj, struct attribute *attr,
 			dev->node_props.cpu_cores_count);
 	sysfs_show_32bit_prop(buffer, "simd_count",
 			dev->node_props.simd_count);
-
-	if (dev->mem_bank_count < dev->node_props.mem_banks_count) {
-		pr_info_once("mem_banks_count truncated from %d to %d\n",
-				dev->node_props.mem_banks_count,
-				dev->mem_bank_count);
-		sysfs_show_32bit_prop(buffer, "mem_banks_count",
-				dev->mem_bank_count);
-	} else {
-		sysfs_show_32bit_prop(buffer, "mem_banks_count",
-				dev->node_props.mem_banks_count);
-	}
-
+	sysfs_show_32bit_prop(buffer, "mem_banks_count",
+			dev->node_props.mem_banks_count);
 	sysfs_show_32bit_prop(buffer, "caches_count",
 			dev->node_props.caches_count);
 	sysfs_show_32bit_prop(buffer, "io_links_count",
@@ -705,9 +455,12 @@ static ssize_t node_show(struct kobject *kobj, struct attribute *attr,
 				HSA_CAP_WATCH_POINTS_TOTALBITS_MASK);
 		}
 
+		if (dev->gpu->device_info->asic_family == CHIP_TONGA)
+			dev->node_props.capability |=
+					HSA_CAP_AQL_QUEUE_DOUBLE_MAP;
+
 		sysfs_show_32bit_prop(buffer, "max_engine_clk_fcompute",
-			dev->gpu->kfd2kgd->get_max_engine_clock_in_mhz(
-					dev->gpu->kgd));
+			dev->node_props.max_engine_clk_fcompute);
 
 		sysfs_show_64bit_prop(buffer, "local_mem_size",
 				(unsigned long long int) 0);
@@ -729,6 +482,7 @@ static const struct sysfs_ops node_ops = {
 };
 
 static struct kobj_type node_type = {
+	.release = kfd_topology_kobj_release,
 	.sysfs_ops = &node_ops,
 };
 
@@ -744,6 +498,7 @@ static void kfd_remove_sysfs_node_entry(struct kfd_topology_device *dev)
 	struct kfd_iolink_properties *iolink;
 	struct kfd_cache_properties *cache;
 	struct kfd_mem_properties *mem;
+	struct kfd_perf_properties *perf;
 
 	if (dev->kobj_iolink) {
 		list_for_each_entry(iolink, &dev->io_link_props, list)
@@ -780,6 +535,16 @@ static void kfd_remove_sysfs_node_entry(struct kfd_topology_device *dev)
 		dev->kobj_mem = NULL;
 	}
 
+	if (dev->kobj_perf) {
+		list_for_each_entry(perf, &dev->perf_props, list) {
+			kfree(perf->attr_group);
+			perf->attr_group = NULL;
+		}
+		kobject_del(dev->kobj_perf);
+		kobject_put(dev->kobj_perf);
+		dev->kobj_perf = NULL;
+	}
+
 	if (dev->kobj_node) {
 		sysfs_remove_file(dev->kobj_node, &dev->attr_gpuid);
 		sysfs_remove_file(dev->kobj_node, &dev->attr_name);
@@ -796,8 +561,10 @@ static int kfd_build_sysfs_node_entry(struct kfd_topology_device *dev,
 	struct kfd_iolink_properties *iolink;
 	struct kfd_cache_properties *cache;
 	struct kfd_mem_properties *mem;
+	struct kfd_perf_properties *perf;
 	int ret;
-	uint32_t i;
+	uint32_t i, num_attrs;
+	struct attribute **attrs;
 
 	if (WARN_ON(dev->kobj_node))
 		return -EEXIST;
@@ -826,6 +593,10 @@ static int kfd_build_sysfs_node_entry(struct kfd_topology_device *dev,
 	if (!dev->kobj_iolink)
 		return -ENOMEM;
 
+	dev->kobj_perf = kobject_create_and_add("perf", dev->kobj_node);
+	if (!dev->kobj_perf)
+		return -ENOMEM;
+
 	/*
 	 * Creating sysfs files for node properties
 	 */
@@ -903,11 +674,38 @@ static int kfd_build_sysfs_node_entry(struct kfd_topology_device *dev,
 		if (ret < 0)
 			return ret;
 		i++;
-}
+	}
+
+	/* All hardware blocks have the same number of attributes. */
+	num_attrs = sizeof(perf_attr_iommu)/sizeof(struct kfd_perf_attr);
+	list_for_each_entry(perf, &dev->perf_props, list) {
+		perf->attr_group = kzalloc(sizeof(struct kfd_perf_attr)
+			* num_attrs + sizeof(struct attribute_group),
+			GFP_KERNEL);
+		if (!perf->attr_group)
+			return -ENOMEM;
+
+		attrs = (struct attribute **)(perf->attr_group + 1);
+		if (!strcmp(perf->block_name, "iommu")) {
+		/* Information of IOMMU's num_counters and counter_ids is shown
+		 * under /sys/bus/event_source/devices/amd_iommu. We don't
+		 * duplicate here.
+		 */
+			perf_attr_iommu[0].data = perf->max_concurrent;
+			for (i = 0; i < num_attrs; i++)
+				attrs[i] = &perf_attr_iommu[i].attr.attr;
+		}
+		perf->attr_group->name = perf->block_name;
+		perf->attr_group->attrs = attrs;
+		ret = sysfs_create_group(dev->kobj_perf, perf->attr_group);
+		if (ret < 0)
+			return ret;
+	}
 
 	return 0;
 }
 
+/* Called with write topology lock acquired */
 static int kfd_build_sysfs_node_tree(void)
 {
 	struct kfd_topology_device *dev;
@@ -924,6 +722,7 @@ static int kfd_build_sysfs_node_tree(void)
 	return 0;
 }
 
+/* Called with write topology lock acquired */
 static void kfd_remove_sysfs_node_tree(void)
 {
 	struct kfd_topology_device *dev;
@@ -995,75 +794,246 @@ static void kfd_topology_release_sysfs(void)
 	}
 }
 
+/* Called with write topology_lock acquired */
+static void kfd_topology_update_device_list(struct list_head *temp_list,
+					struct list_head *master_list)
+{
+	while (!list_empty(temp_list)) {
+		list_move_tail(temp_list->next, master_list);
+		sys_props.num_devices++;
+	}
+}
+
+static void kfd_debug_print_topology(void)
+{
+	struct kfd_topology_device *dev;
+
+	down_read(&topology_lock);
+
+	dev = list_last_entry(&topology_device_list,
+			struct kfd_topology_device, list);
+	if (dev) {
+		if (dev->node_props.cpu_cores_count &&
+				dev->node_props.simd_count) {
+			pr_info("Topology: Add APU node [0x%0x:0x%0x]\n",
+				dev->node_props.device_id,
+				dev->node_props.vendor_id);
+		} else if (dev->node_props.cpu_cores_count)
+			pr_info("Topology: Add CPU node\n");
+		else if (dev->node_props.simd_count)
+			pr_info("Topology: Add dGPU node [0x%0x:0x%0x]\n",
+				dev->node_props.device_id,
+				dev->node_props.vendor_id);
+	}
+	up_read(&topology_lock);
+}
+
+/* Helper function for intializing platform_xx members of
+ * kfd_system_properties. Uses OEM info from the last CPU/APU node.
+ */
+static void kfd_update_system_properties(void)
+{
+	struct kfd_topology_device *dev;
+
+	down_read(&topology_lock);
+	dev = list_last_entry(&topology_device_list,
+			struct kfd_topology_device, list);
+	if (dev) {
+		sys_props.platform_id =
+			(*((uint64_t *)dev->oem_id)) & CRAT_OEMID_64BIT_MASK;
+		sys_props.platform_oem = *((uint64_t *)dev->oem_table_id);
+		sys_props.platform_rev = dev->oem_revision;
+	}
+	up_read(&topology_lock);
+}
+
+static void find_system_memory(const struct dmi_header *dm,
+	void *private)
+{
+	struct kfd_mem_properties *mem;
+	u16 mem_width, mem_clock;
+	struct kfd_topology_device *kdev =
+		(struct kfd_topology_device *)private;
+	const u8 *dmi_data = (const u8 *)(dm + 1);
+
+	if (dm->type == DMI_ENTRY_MEM_DEVICE && dm->length >= 0x15) {
+		mem_width = (u16)(*(const u16 *)(dmi_data + 0x6));
+		mem_clock = (u16)(*(const u16 *)(dmi_data + 0x11));
+		list_for_each_entry(mem, &kdev->mem_props, list) {
+			if (mem_width != 0xFFFF && mem_width != 0)
+				mem->width = mem_width;
+			if (mem_clock != 0)
+				mem->mem_clk_max = mem_clock;
+		}
+	}
+}
+
+/*
+ * Performance counters information is not part of CRAT but we would like to
+ * put them in the sysfs under topology directory for Thunk to get the data.
+ * This function is called before updating the sysfs.
+ */
+static int kfd_add_perf_to_topology(struct kfd_topology_device *kdev)
+{
+	struct kfd_perf_properties *props;
+
+	if (amd_iommu_pc_supported()) {
+		props = kfd_alloc_struct(props);
+		if (!props)
+			return -ENOMEM;
+		strcpy(props->block_name, "iommu");
+		props->max_concurrent = amd_iommu_pc_get_max_banks(0) *
+			amd_iommu_pc_get_max_counters(0); /* assume one iommu */
+		list_add_tail(&props->list, &kdev->perf_props);
+	}
+
+	return 0;
+}
+
+/* kfd_add_non_crat_information - Add information that is not currently
+ *	defined in CRAT but is necessary for KFD topology
+ * @dev - topology device to which addition info is added
+ */
+static void kfd_add_non_crat_information(struct kfd_topology_device *kdev)
+{
+	/* Check if CPU only node. */
+	if (!kdev->gpu) {
+		/* Add system memory information */
+		dmi_walk(find_system_memory, kdev);
+	}
+	/* TODO: For GPU node, rearrange code from kfd_topology_add_device */
+}
+
+/* kfd_is_acpi_crat_invalid - CRAT from ACPI is valid only for AMD APU devices.
+ *	Ignore CRAT for all other devices. AMD APU is identified if both CPU
+ *	and GPU cores are present.
+ * @device_list - topology device list created by parsing ACPI CRAT table.
+ * @return - TRUE if invalid, FALSE is valid.
+ */
+static bool kfd_is_acpi_crat_invalid(struct list_head *device_list)
+{
+	struct kfd_topology_device *dev;
+
+	list_for_each_entry(dev, device_list, list) {
+		if (dev->node_props.cpu_cores_count &&
+			dev->node_props.simd_count)
+			return false;
+	}
+	pr_info("Ignoring ACPI CRAT on non-APU system\n");
+	return true;
+}
+
 int kfd_topology_init(void)
 {
 	void *crat_image = NULL;
 	size_t image_size = 0;
 	int ret;
-
-	/*
-	 * Initialize the head for the topology device list
+	struct list_head temp_topology_device_list;
+	int cpu_only_node = 0;
+	struct kfd_topology_device *kdev;
+	int proximity_domain;
+
+	/* topology_device_list - Master list of all topology devices
+	 * temp_topology_device_list - temporary list created while parsing CRAT
+	 * or VCRAT. Once parsing is complete the contents of list is moved to
+	 * topology_device_list
 	 */
+
+	/* Initialize the head for the both the lists */
 	INIT_LIST_HEAD(&topology_device_list);
+	INIT_LIST_HEAD(&temp_topology_device_list);
 	init_rwsem(&topology_lock);
-	topology_crat_parsed = 0;
 
 	memset(&sys_props, 0, sizeof(sys_props));
 
+	/* Proximity domains in ACPI CRAT tables start counting at
+	 * 0. The same should be true for virtual CRAT tables created
+	 * at this stage. GPUs added later in kfd_topology_add_device
+	 * use a counter.
+	 */
+	proximity_domain = 0;
+
 	/*
-	 * Get the CRAT image from the ACPI
+	 * Get the CRAT image from the ACPI. If ACPI doesn't have one
+	 * or if ACPI CRAT is invalid create a virtual CRAT.
+	 * NOTE: The current implementation expects all AMD APUs to have
+	 *	CRAT. If no CRAT is available, it is assumed to be a CPU
 	 */
-	ret = kfd_topology_get_crat_acpi(crat_image, &image_size);
-	if (ret == 0 && image_size > 0) {
-		pr_info("Found CRAT image with size=%zd\n", image_size);
-		crat_image = kmalloc(image_size, GFP_KERNEL);
-		if (!crat_image) {
-			ret = -ENOMEM;
-			pr_err("No memory for allocating CRAT image\n");
-			goto err;
+	ret = kfd_create_crat_image_acpi(&crat_image, &image_size);
+	if (!ret) {
+		ret = kfd_parse_crat_table(crat_image,
+					   &temp_topology_device_list,
+					   proximity_domain);
+		if (ret ||
+		    kfd_is_acpi_crat_invalid(&temp_topology_device_list)) {
+			kfd_release_topology_device_list(
+				&temp_topology_device_list);
+			kfd_destroy_crat_image(crat_image);
+			crat_image = NULL;
 		}
-		ret = kfd_topology_get_crat_acpi(crat_image, &image_size);
-
-		if (ret == 0) {
-			down_write(&topology_lock);
-			ret = kfd_parse_crat_table(crat_image);
-			if (ret == 0)
-				ret = kfd_topology_update_sysfs();
-			up_write(&topology_lock);
-		} else {
-			pr_err("Couldn't get CRAT table size from ACPI\n");
+	}
+
+	if (!crat_image) {
+		ret = kfd_create_crat_image_virtual(&crat_image, &image_size,
+						    COMPUTE_UNIT_CPU, NULL,
+						    proximity_domain);
+		cpu_only_node = 1;
+		if (ret) {
+			pr_err("Error creating VCRAT table for CPU\n");
+			return ret;
 		}
-		kfree(crat_image);
-	} else if (ret == -ENODATA) {
-		ret = 0;
-	} else {
-		pr_err("Couldn't get CRAT table size from ACPI\n");
+
+		ret = kfd_parse_crat_table(crat_image,
+					   &temp_topology_device_list,
+					   proximity_domain);
+		if (ret) {
+			pr_err("Error parsing VCRAT table for CPU\n");
+			goto err;
+		}
+	}
+
+	kdev = list_first_entry(&temp_topology_device_list,
+				struct kfd_topology_device, list);
+	kfd_add_perf_to_topology(kdev);
+
+	down_write(&topology_lock);
+	kfd_topology_update_device_list(&temp_topology_device_list,
+					&topology_device_list);
+	atomic_set(&topology_crat_proximity_domain, sys_props.num_devices-1);
+	ret = kfd_topology_update_sysfs();
+	up_write(&topology_lock);
+
+	if (!ret) {
+		sys_props.generation_count++;
+		kfd_update_system_properties();
+		kfd_debug_print_topology();
+		pr_info("Finished initializing topology\n");
+	} else
+		pr_err("Failed to update topology in sysfs ret=%d\n", ret);
+
+	/* For nodes with GPU, this information gets added
+	 * when GPU is detected (kfd_topology_add_device).
+	 */
+	if (cpu_only_node) {
+		/* Add additional information to CPU only node created above */
+		down_write(&topology_lock);
+		kdev = list_first_entry(&topology_device_list,
+				struct kfd_topology_device, list);
+		up_write(&topology_lock);
+		kfd_add_non_crat_information(kdev);
 	}
 
 err:
-	pr_info("Finished initializing topology ret=%d\n", ret);
+	kfd_destroy_crat_image(crat_image);
 	return ret;
 }
 
 void kfd_topology_shutdown(void)
 {
+	down_write(&topology_lock);
 	kfd_topology_release_sysfs();
 	kfd_release_live_view();
-}
-
-static void kfd_debug_print_topology(void)
-{
-	struct kfd_topology_device *dev;
-	uint32_t i = 0;
-
-	pr_info("DEBUG PRINT OF TOPOLOGY:");
-	list_for_each_entry(dev, &topology_device_list, list) {
-		pr_info("Node: %d\n", i);
-		pr_info("\tGPU assigned: %s\n", (dev->gpu ? "yes" : "no"));
-		pr_info("\tCPU count: %d\n", dev->node_props.cpu_cores_count);
-		pr_info("\tSIMD count: %d", dev->node_props.simd_count);
-		i++;
-	}
+	up_write(&topology_lock);
 }
 
 static uint32_t kfd_generate_gpu_id(struct kfd_dev *gpu)
@@ -1072,11 +1042,15 @@ static uint32_t kfd_generate_gpu_id(struct kfd_dev *gpu)
 	uint32_t buf[7];
 	uint64_t local_mem_size;
 	int i;
+	struct kfd_local_mem_info local_mem_info;
 
 	if (!gpu)
 		return 0;
 
-	local_mem_size = gpu->kfd2kgd->get_vmem_size(gpu->kgd);
+	gpu->kfd2kgd->get_local_mem_info(gpu->kgd, &local_mem_info);
+
+	local_mem_size = local_mem_info.local_mem_size_private +
+			local_mem_info.local_mem_size_public;
 
 	buf[0] = gpu->pdev->devfn;
 	buf[1] = gpu->pdev->subsystem_vendor;
@@ -1091,19 +1065,26 @@ static uint32_t kfd_generate_gpu_id(struct kfd_dev *gpu)
 
 	return hashout;
 }
-
+/* kfd_assign_gpu - Attach @gpu to the correct kfd topology device. If
+ *		the GPU device is not already present in the topology device
+ *		list then return NULL. This means a new topology device has to
+ *		be created for this GPU.
+ * TODO: Rather than assiging @gpu to first topology device withtout
+ *		gpu attached, it will better to have more stringent check.
+ */
 static struct kfd_topology_device *kfd_assign_gpu(struct kfd_dev *gpu)
 {
 	struct kfd_topology_device *dev;
 	struct kfd_topology_device *out_dev = NULL;
 
+	down_write(&topology_lock);
 	list_for_each_entry(dev, &topology_device_list, list)
 		if (!dev->gpu && (dev->node_props.simd_count > 0)) {
 			dev->gpu = gpu;
 			out_dev = dev;
 			break;
 		}
-
+	up_write(&topology_lock);
 	return out_dev;
 }
 
@@ -1115,84 +1096,196 @@ static void kfd_notify_gpu_change(uint32_t gpu_id, int arrival)
 	 */
 }
 
+/* kfd_fill_mem_clk_max_info - Since CRAT doesn't have memory clock info,
+ *		patch this after CRAT parsing.
+ */
+static void kfd_fill_mem_clk_max_info(struct kfd_topology_device *dev)
+{
+	struct kfd_mem_properties *mem;
+	struct kfd_local_mem_info local_mem_info;
+
+	if (!dev)
+		return;
+
+	/* Currently, amdgpu driver (amdgpu_mc) deals only with GPUs with
+	 * single bank of VRAM local memory.
+	 * for dGPUs - VCRAT reports only one bank of Local Memory
+	 * for APUs - If CRAT from ACPI reports more than one bank, then
+	 *	all the banks will report the same mem_clk_max information
+	 */
+	dev->gpu->kfd2kgd->get_local_mem_info(dev->gpu->kgd,
+		&local_mem_info);
+
+	list_for_each_entry(mem, &dev->mem_props, list)
+		mem->mem_clk_max = local_mem_info.mem_clk_max;
+}
+
+static void kfd_fill_iolink_non_crat_info(struct kfd_topology_device *dev)
+{
+	struct kfd_iolink_properties *link;
+
+	if (!dev || !dev->gpu)
+		return;
+
+	/* GPU only creates direck links so apply flags setting to all */
+	if (dev->gpu->device_info->asic_family == CHIP_HAWAII)
+		list_for_each_entry(link, &dev->io_link_props, list)
+			link->flags = CRAT_IOLINK_FLAGS_ENABLED |
+				CRAT_IOLINK_FLAGS_NO_ATOMICS_32_BIT |
+				CRAT_IOLINK_FLAGS_NO_ATOMICS_64_BIT;
+}
+
 int kfd_topology_add_device(struct kfd_dev *gpu)
 {
 	uint32_t gpu_id;
 	struct kfd_topology_device *dev;
-	int res;
+	struct kfd_cu_info cu_info;
+	int res = 0;
+	struct list_head temp_topology_device_list;
+	void *crat_image = NULL;
+	size_t image_size = 0;
+	int proximity_domain;
+
+	INIT_LIST_HEAD(&temp_topology_device_list);
 
 	gpu_id = kfd_generate_gpu_id(gpu);
 
 	pr_debug("Adding new GPU (ID: 0x%x) to topology\n", gpu_id);
 
-	down_write(&topology_lock);
-	/*
-	 * Try to assign the GPU to existing topology device (generated from
-	 * CRAT table
+	proximity_domain = atomic_inc_return(&topology_crat_proximity_domain);
+
+	/* Check to see if this gpu device exists in the topology_device_list.
+	 * If so, assign the gpu to that device,
+	 * else create a Virtual CRAT for this gpu device and then parse that
+	 * CRAT to create a new topology device. Once created assign the gpu to
+	 * that topology device
 	 */
 	dev = kfd_assign_gpu(gpu);
 	if (!dev) {
-		pr_info("GPU was not found in the current topology. Extending.\n");
-		kfd_debug_print_topology();
-		dev = kfd_create_topology_device();
-		if (!dev) {
-			res = -ENOMEM;
+		res = kfd_create_crat_image_virtual(&crat_image, &image_size,
+						    COMPUTE_UNIT_GPU, gpu,
+						    proximity_domain);
+		if (res) {
+			pr_err("Error creating VCRAT for GPU (ID: 0x%x)\n",
+			       gpu_id);
+			return res;
+		}
+		res = kfd_parse_crat_table(crat_image,
+					   &temp_topology_device_list,
+					   proximity_domain);
+		if (res) {
+			pr_err("Error parsing VCRAT for GPU (ID: 0x%x)\n",
+			       gpu_id);
 			goto err;
 		}
-		dev->gpu = gpu;
 
-		/*
-		 * TODO: Make a call to retrieve topology information from the
-		 * GPU vBIOS
-		 */
+		down_write(&topology_lock);
+		kfd_topology_update_device_list(&temp_topology_device_list,
+			&topology_device_list);
 
 		/* Update the SYSFS tree, since we added another topology
 		 * device
 		 */
-		if (kfd_topology_update_sysfs() < 0)
-			kfd_topology_release_sysfs();
-
+		res = kfd_topology_update_sysfs();
+		up_write(&topology_lock);
+
+		if (!res)
+			sys_props.generation_count++;
+		else
+			pr_err("Failed to update GPU (ID: 0x%x) to sysfs topology. res=%d\n",
+						gpu_id, res);
+		dev = kfd_assign_gpu(gpu);
+		if (WARN_ON(!dev)) {
+			res = -ENODEV;
+			goto err;
+		}
 	}
 
 	dev->gpu_id = gpu_id;
 	gpu->id = gpu_id;
+
+	/* TODO: Move the following lines to function
+	 *	kfd_add_non_crat_information
+	 */
+
+	/* Fill-in additional information that is not available in CRAT but
+	 * needed for the topology
+	 */
+
+	dev->gpu->kfd2kgd->get_cu_info(dev->gpu->kgd, &cu_info);
+	dev->node_props.simd_arrays_per_engine =
+		cu_info.num_shader_arrays_per_engine;
+
 	dev->node_props.vendor_id = gpu->pdev->vendor;
 	dev->node_props.device_id = gpu->pdev->device;
-	dev->node_props.location_id = (gpu->pdev->bus->number << 24) +
-			(gpu->pdev->devfn & 0xffffff);
-	/*
-	 * TODO: Retrieve max engine clock values from KGD
-	 */
+	dev->node_props.location_id = PCI_DEVID(gpu->pdev->bus->number,
+		gpu->pdev->devfn);
+	dev->node_props.max_engine_clk_fcompute =
+		dev->gpu->kfd2kgd->get_max_engine_clock_in_mhz(dev->gpu->kgd);
+	dev->node_props.max_engine_clk_ccompute =
+		cpufreq_quick_get_max(0) / 1000;
+
+	kfd_fill_mem_clk_max_info(dev);
+	kfd_fill_iolink_non_crat_info(dev);
+
+	switch (dev->gpu->device_info->asic_family) {
+	case CHIP_KAVERI:
+	case CHIP_HAWAII:
+	case CHIP_TONGA:
+		dev->node_props.capability |= ((HSA_CAP_DOORBELL_TYPE_PRE_1_0 <<
+			HSA_CAP_DOORBELL_TYPE_TOTALBITS_SHIFT) &
+			HSA_CAP_DOORBELL_TYPE_TOTALBITS_MASK);
+		break;
+	case CHIP_CARRIZO:
+	case CHIP_FIJI:
+	case CHIP_POLARIS10:
+	case CHIP_POLARIS11:
+		pr_debug("Adding doorbell packet type capability\n");
+		dev->node_props.capability |= ((HSA_CAP_DOORBELL_TYPE_1_0 <<
+			HSA_CAP_DOORBELL_TYPE_TOTALBITS_SHIFT) &
+			HSA_CAP_DOORBELL_TYPE_TOTALBITS_MASK);
+		break;
+	default:
+		WARN(1, "Unexpected ASIC family %u",
+		     dev->gpu->device_info->asic_family);
+	}
 
+	/* Fix errors in CZ CRAT.
+	 * simd_count: Carrizo CRAT reports wrong simd_count, probably
+	 *		because it doesn't consider masked out CUs
+	 * max_waves_per_simd: Carrizo reports wrong max_waves_per_simd
+	 * capability flag: Carrizo CRAT doesn't report IOMMU flags
+	 */
 	if (dev->gpu->device_info->asic_family == CHIP_CARRIZO) {
-		dev->node_props.capability |= HSA_CAP_DOORBELL_PACKET_TYPE;
-		pr_info("Adding doorbell packet type capability\n");
+		dev->node_props.simd_count =
+			cu_info.simd_per_cu * cu_info.cu_active_number;
+		dev->node_props.max_waves_per_simd = 10;
+		dev->node_props.capability |= HSA_CAP_ATS_PRESENT;
 	}
 
-	res = 0;
-
-err:
-	up_write(&topology_lock);
+	kfd_debug_print_topology();
 
-	if (res == 0)
+	if (!res)
 		kfd_notify_gpu_change(gpu_id, 1);
-
+err:
+	kfd_destroy_crat_image(crat_image);
 	return res;
 }
 
 int kfd_topology_remove_device(struct kfd_dev *gpu)
 {
-	struct kfd_topology_device *dev;
+	struct kfd_topology_device *dev, *tmp;
 	uint32_t gpu_id;
 	int res = -ENODEV;
 
 	down_write(&topology_lock);
 
-	list_for_each_entry(dev, &topology_device_list, list)
+	list_for_each_entry_safe(dev, tmp, &topology_device_list, list)
 		if (dev->gpu == gpu) {
 			gpu_id = dev->gpu_id;
 			kfd_remove_sysfs_node_entry(dev);
 			kfd_release_topology_device(dev);
+			sys_props.num_devices--;
 			res = 0;
 			if (kfd_topology_update_sysfs() < 0)
 				kfd_topology_release_sysfs();
@@ -1201,28 +1294,32 @@ int kfd_topology_remove_device(struct kfd_dev *gpu)
 
 	up_write(&topology_lock);
 
-	if (res == 0)
+	if (!res)
 		kfd_notify_gpu_change(gpu_id, 0);
 
 	return res;
 }
 
-/*
- * When idx is out of bounds, the function will return NULL
+/* kfd_topology_enum_kfd_devices - Enumerate through all devices in KFD
+ *	topology. If GPU device is found @idx, then valid kfd_dev pointer is
+ *	returned through @kdev
+ * Return -	0: On success (@kdev will be NULL for non GPU nodes)
+ *		-1: If end of list
  */
-struct kfd_dev *kfd_topology_enum_kfd_devices(uint8_t idx)
+int kfd_topology_enum_kfd_devices(uint8_t idx, struct kfd_dev **kdev)
 {
 
 	struct kfd_topology_device *top_dev;
-	struct kfd_dev *device = NULL;
 	uint8_t device_idx = 0;
 
+	*kdev = NULL;
 	down_read(&topology_lock);
 
 	list_for_each_entry(top_dev, &topology_device_list, list) {
 		if (device_idx == idx) {
-			device = top_dev->gpu;
-			break;
+			*kdev = top_dev->gpu;
+			up_read(&topology_lock);
+			return 0;
 		}
 
 		device_idx++;
@@ -1230,6 +1327,88 @@ struct kfd_dev *kfd_topology_enum_kfd_devices(uint8_t idx)
 
 	up_read(&topology_lock);
 
-	return device;
+	return -1;
+
+}
+
+static int kfd_cpumask_to_apic_id(const struct cpumask *cpumask)
+{
+	const struct cpuinfo_x86 *cpuinfo;
+	int first_cpu_of_numa_node;
+
+	if (!cpumask || cpumask == cpu_none_mask)
+		return -1;
+	first_cpu_of_numa_node = cpumask_first(cpumask);
+	if (first_cpu_of_numa_node >= nr_cpu_ids)
+		return -1;
+	cpuinfo = &cpu_data(first_cpu_of_numa_node);
 
+	return cpuinfo->apicid;
 }
+
+/* kfd_numa_node_to_apic_id - Returns the APIC ID of the first logical processor
+ *	of the given NUMA node (numa_node_id)
+ * Return -1 on failure
+ */
+int kfd_numa_node_to_apic_id(int numa_node_id)
+{
+	if (numa_node_id == -1) {
+		pr_warn("Invalid NUMA Node. Use online CPU mask\n");
+		return kfd_cpumask_to_apic_id(cpu_online_mask);
+	}
+	return kfd_cpumask_to_apic_id(cpumask_of_node(numa_node_id));
+}
+
+#if defined(CONFIG_DEBUG_FS)
+
+int kfd_debugfs_hqds_by_device(struct seq_file *m, void *data)
+{
+	struct kfd_topology_device *dev;
+	unsigned int i = 0;
+	int r = 0;
+
+	down_read(&topology_lock);
+
+	list_for_each_entry(dev, &topology_device_list, list) {
+		if (!dev->gpu) {
+			i++;
+			continue;
+		}
+
+		seq_printf(m, "Node %u, gpu_id %x:\n", i++, dev->gpu->id);
+		r = dqm_debugfs_hqds(m, dev->gpu->dqm);
+		if (r)
+			break;
+	}
+
+	up_read(&topology_lock);
+
+	return r;
+}
+
+int kfd_debugfs_rls_by_device(struct seq_file *m, void *data)
+{
+	struct kfd_topology_device *dev;
+	unsigned int i = 0;
+	int r = 0;
+
+	down_read(&topology_lock);
+
+	list_for_each_entry(dev, &topology_device_list, list) {
+		if (!dev->gpu) {
+			i++;
+			continue;
+		}
+
+		seq_printf(m, "Node %u, gpu_id %x:\n", i++, dev->gpu->id);
+		r = pm_debugfs_runlist(m, &dev->gpu->dqm->packets);
+		if (r)
+			break;
+	}
+
+	up_read(&topology_lock);
+
+	return r;
+}
+
+#endif