Merge tag 'amd-drm-next-6.2-2022-12-02' of https://gitlab.freedesktop…

….org/agd5f/linux into drm-next

amd-drm-next-6.2-2022-12-02:

amdgpu:
- Fix CPU stalls when allocating large amounts of system memory
- SR-IOV fixes
- BACO fixes
- Enable GC 11.0.4
- Enable PSP 13.0.11
- Enable SMU 13.0.11
- Enable NBIO 7.7.1
- Fix reported VCN capabilities for RDNA2
- Misc cleanups
- PCI ref count fixes
- DCN DPIA fixes
- DCN 3.2.x fixes
- Documentation updates
- GC 11.x fixes
- VCN RAS fixes
- APU fix for passthrough
- PSR fixes
- GFX preemption support for gfx9
- SDMA fix for S0ix

amdkfd:
- Enable KFD support for GC 11.0.4
- Misc cleanups
- Fix memory leak

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexander.deucher@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221202160659.5987-1-alexander.deucher@amd.com

Documentation/gpu/amdgpu/amdgpu-glossary.rst

@@ -30,12 +30,35 @@ we have a dedicated glossary for Display Core at
     EOP
       End Of Pipe/Pipeline
 
+    GART
+      Graphics Address Remapping Table.  This is the name we use for the GPUVM
+      page table used by the GPU kernel driver.  It remaps system resources
+      (memory or MMIO space) into the GPU's address space so the GPU can access
+      them.  The name GART harkens back to the days of AGP when the platform
+      provided an MMU that the GPU could use to get a contiguous view of
+      scattered pages for DMA.  The MMU has since moved on to the GPU, but the
+      name stuck.
+
     GC
       Graphics and Compute
 
     GMC
       Graphic Memory Controller
 
+    GPUVM
+      GPU Virtual Memory.  This is the GPU's MMU.  The GPU supports multiple
+      virtual address spaces that can be in flight at any given time.  These
+      allow the GPU to remap VRAM and system resources into GPU virtual address
+      spaces for use by the GPU kernel driver and applications using the GPU.
+      These provide memory protection for different applications using the GPU.
+
+    GTT
+      Graphics Translation Tables.  This is a memory pool managed through TTM
+      which provides access to system resources (memory or MMIO space) for
+      use by the GPU. These addresses can be mapped into the "GART" GPUVM page
+      table for use by the kernel driver or into per process GPUVM page tables
+      for application usage.
+
     IH
       Interrupt Handler
 

Documentation/gpu/amdgpu/driver-core.rst

@@ -148,10 +148,10 @@ PRIME Buffer Sharing
 MMU Notifier
 ============
 
-.. kernel-doc:: drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c
+.. kernel-doc:: drivers/gpu/drm/amd/amdgpu/amdgpu_hmm.c
    :doc: MMU Notifier
 
-.. kernel-doc:: drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c
+.. kernel-doc:: drivers/gpu/drm/amd/amdgpu/amdgpu_hmm.c
    :internal:
 
 AMDGPU Virtual Memory

Documentation/gpu/amdgpu/index.rst

@@ -3,7 +3,7 @@
 ==========================
 
 The drm/amdgpu driver supports all AMD Radeon GPUs based on the Graphics Core
-Next (GCN) architecture.
+Next (GCN), Radeon DNA (RDNA), and Compute DNA (CDNA) architectures.
 
 .. toctree::
 

drivers/gpu/drm/amd/amdgpu/Makefile

@@ -58,7 +58,8 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
 	amdgpu_vm_sdma.o amdgpu_discovery.o amdgpu_ras_eeprom.o amdgpu_nbio.o \
 	amdgpu_umc.o smu_v11_0_i2c.o amdgpu_fru_eeprom.o amdgpu_rap.o \
 	amdgpu_fw_attestation.o amdgpu_securedisplay.o \
-	amdgpu_eeprom.o amdgpu_mca.o amdgpu_psp_ta.o amdgpu_lsdma.o
+	amdgpu_eeprom.o amdgpu_mca.o amdgpu_psp_ta.o amdgpu_lsdma.o \
+	amdgpu_ring_mux.o
 
 amdgpu-$(CONFIG_PROC_FS) += amdgpu_fdinfo.o
 

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c

@@ -2204,7 +2204,7 @@ int amdgpu_amdkfd_gpuvm_import_dmabuf(struct amdgpu_device *adev,
 
 	ret = drm_vma_node_allow(&obj->vma_node, drm_priv);
 	if (ret) {
-		kfree(mem);
+		kfree(*mem);
 		return ret;
 	}
 

drivers/gpu/drm/amd/amdgpu/amdgpu_atomfirmware.c

@@ -104,7 +104,7 @@ void amdgpu_atomfirmware_scratch_regs_init(struct amdgpu_device *adev)
 static int amdgpu_atomfirmware_allocate_fb_v2_1(struct amdgpu_device *adev,
 	struct vram_usagebyfirmware_v2_1 *fw_usage, int *usage_bytes)
 {
-	uint32_t start_addr, fw_size, drv_size;
+	u32 start_addr, fw_size, drv_size;
 
 	start_addr = le32_to_cpu(fw_usage->start_address_in_kb);
 	fw_size = le16_to_cpu(fw_usage->used_by_firmware_in_kb);
@@ -116,7 +116,7 @@ static int amdgpu_atomfirmware_allocate_fb_v2_1(struct amdgpu_device *adev,
 			  drv_size);
 
 	if ((start_addr & ATOM_VRAM_OPERATION_FLAGS_MASK) ==
-		(uint32_t)(ATOM_VRAM_BLOCK_SRIOV_MSG_SHARE_RESERVATION <<
+		(u32)(ATOM_VRAM_BLOCK_SRIOV_MSG_SHARE_RESERVATION <<
 		ATOM_VRAM_OPERATION_FLAGS_SHIFT)) {
 		/* Firmware request VRAM reservation for SR-IOV */
 		adev->mman.fw_vram_usage_start_offset = (start_addr &
@@ -133,7 +133,7 @@ static int amdgpu_atomfirmware_allocate_fb_v2_1(struct amdgpu_device *adev,
 static int amdgpu_atomfirmware_allocate_fb_v2_2(struct amdgpu_device *adev,
 		struct vram_usagebyfirmware_v2_2 *fw_usage, int *usage_bytes)
 {
-	uint32_t fw_start_addr, fw_size, drv_start_addr, drv_size;
+	u32 fw_start_addr, fw_size, drv_start_addr, drv_size;
 
 	fw_start_addr = le32_to_cpu(fw_usage->fw_region_start_address_in_kb);
 	fw_size = le16_to_cpu(fw_usage->used_by_firmware_in_kb);
@@ -147,14 +147,18 @@ static int amdgpu_atomfirmware_allocate_fb_v2_2(struct amdgpu_device *adev,
 			  drv_start_addr,
 			  drv_size);
 
-	if ((fw_start_addr & (ATOM_VRAM_BLOCK_NEEDS_NO_RESERVATION << 30)) == 0) {
+	if (amdgpu_sriov_vf(adev) &&
+	    ((fw_start_addr & (ATOM_VRAM_BLOCK_NEEDS_NO_RESERVATION <<
+		ATOM_VRAM_OPERATION_FLAGS_SHIFT)) == 0)) {
 		/* Firmware request VRAM reservation for SR-IOV */
 		adev->mman.fw_vram_usage_start_offset = (fw_start_addr &
 			(~ATOM_VRAM_OPERATION_FLAGS_MASK)) << 10;
 		adev->mman.fw_vram_usage_size = fw_size << 10;
 	}
 
-	if ((drv_start_addr & (ATOM_VRAM_BLOCK_NEEDS_NO_RESERVATION << 30)) == 0) {
+	if (amdgpu_sriov_vf(adev) &&
+	    ((drv_start_addr & (ATOM_VRAM_BLOCK_NEEDS_NO_RESERVATION <<
+		ATOM_VRAM_OPERATION_FLAGS_SHIFT)) == 0)) {
 		/* driver request VRAM reservation for SR-IOV */
 		adev->mman.drv_vram_usage_start_offset = (drv_start_addr &
 			(~ATOM_VRAM_OPERATION_FLAGS_MASK)) << 10;
@@ -172,8 +176,8 @@ int amdgpu_atomfirmware_allocate_fb_scratch(struct amdgpu_device *adev)
 						vram_usagebyfirmware);
 	struct vram_usagebyfirmware_v2_1 *fw_usage_v2_1;
 	struct vram_usagebyfirmware_v2_2 *fw_usage_v2_2;
-	uint16_t data_offset;
-	uint8_t frev, crev;
+	u16 data_offset;
+	u8 frev, crev;
 	int usage_bytes = 0;
 
 	if (amdgpu_atom_parse_data_header(ctx, index, NULL, &frev, &crev, &data_offset)) {

drivers/gpu/drm/amd/amdgpu/amdgpu_bios.c

@@ -317,6 +317,7 @@ static bool amdgpu_atrm_get_bios(struct amdgpu_device *adev)
 
 	if (!found)
 		return false;
+	pci_dev_put(pdev);
 
 	adev->bios = kmalloc(size, GFP_KERNEL);
 	if (!adev->bios) {

drivers/gpu/drm/amd/amdgpu/amdgpu_device.c

@@ -2473,6 +2473,11 @@ static int amdgpu_device_ip_init(struct amdgpu_device *adev)
 			if (!amdgpu_sriov_vf(adev)) {
 				struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev);
 
+				if (WARN_ON(!hive)) {
+					r = -ENOENT;
+					goto init_failed;
+				}
+
 				if (!hive->reset_domain ||
 				    !amdgpu_reset_get_reset_domain(hive->reset_domain)) {
 					r = -ENOENT;

drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c

@@ -1512,6 +1512,7 @@ static int amdgpu_discovery_set_common_ip_blocks(struct amdgpu_device *adev)
 	case IP_VERSION(11, 0, 1):
 	case IP_VERSION(11, 0, 2):
 	case IP_VERSION(11, 0, 3):
+	case IP_VERSION(11, 0, 4):
 		amdgpu_device_ip_block_add(adev, &soc21_common_ip_block);
 		break;
 	default:
@@ -1556,6 +1557,7 @@ static int amdgpu_discovery_set_gmc_ip_blocks(struct amdgpu_device *adev)
 	case IP_VERSION(11, 0, 1):
 	case IP_VERSION(11, 0, 2):
 	case IP_VERSION(11, 0, 3):
+	case IP_VERSION(11, 0, 4):
 		amdgpu_device_ip_block_add(adev, &gmc_v11_0_ip_block);
 		break;
 	default:
@@ -1641,6 +1643,7 @@ static int amdgpu_discovery_set_psp_ip_blocks(struct amdgpu_device *adev)
 	case IP_VERSION(13, 0, 7):
 	case IP_VERSION(13, 0, 8):
 	case IP_VERSION(13, 0, 10):
+	case IP_VERSION(13, 0, 11):
 		amdgpu_device_ip_block_add(adev, &psp_v13_0_ip_block);
 		break;
 	case IP_VERSION(13, 0, 4):
@@ -1691,6 +1694,7 @@ static int amdgpu_discovery_set_smu_ip_blocks(struct amdgpu_device *adev)
 	case IP_VERSION(13, 0, 7):
 	case IP_VERSION(13, 0, 8):
 	case IP_VERSION(13, 0, 10):
+	case IP_VERSION(13, 0, 11):
 		amdgpu_device_ip_block_add(adev, &smu_v13_0_ip_block);
 		break;
 	default:
@@ -1804,6 +1808,7 @@ static int amdgpu_discovery_set_gc_ip_blocks(struct amdgpu_device *adev)
 	case IP_VERSION(11, 0, 1):
 	case IP_VERSION(11, 0, 2):
 	case IP_VERSION(11, 0, 3):
+	case IP_VERSION(11, 0, 4):
 		amdgpu_device_ip_block_add(adev, &gfx_v11_0_ip_block);
 		break;
 	default:
@@ -1967,6 +1972,7 @@ static int amdgpu_discovery_set_mes_ip_blocks(struct amdgpu_device *adev)
 	case IP_VERSION(11, 0, 1):
 	case IP_VERSION(11, 0, 2):
 	case IP_VERSION(11, 0, 3):
+	case IP_VERSION(11, 0, 4):
 		amdgpu_device_ip_block_add(adev, &mes_v11_0_ip_block);
 		adev->enable_mes = true;
 		adev->enable_mes_kiq = true;
@@ -2197,6 +2203,7 @@ int amdgpu_discovery_set_ip_blocks(struct amdgpu_device *adev)
 		adev->family = AMDGPU_FAMILY_GC_11_0_0;
 		break;
 	case IP_VERSION(11, 0, 1):
+	case IP_VERSION(11, 0, 4):
 		adev->family = AMDGPU_FAMILY_GC_11_0_1;
 		break;
 	default:
@@ -2214,6 +2221,7 @@ int amdgpu_discovery_set_ip_blocks(struct amdgpu_device *adev)
 	case IP_VERSION(10, 3, 6):
 	case IP_VERSION(10, 3, 7):
 	case IP_VERSION(11, 0, 1):
+	case IP_VERSION(11, 0, 4):
 		adev->flags |= AMD_IS_APU;
 		break;
 	default:
@@ -2270,6 +2278,7 @@ int amdgpu_discovery_set_ip_blocks(struct amdgpu_device *adev)
 		adev->nbio.hdp_flush_reg = &nbio_v4_3_hdp_flush_reg;
 		break;
 	case IP_VERSION(7, 7, 0):
+	case IP_VERSION(7, 7, 1):
 		adev->nbio.funcs = &nbio_v7_7_funcs;
 		adev->nbio.hdp_flush_reg = &nbio_v7_7_hdp_flush_reg;
 		break;

drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c

@@ -231,16 +231,18 @@ module_param_named(vis_vramlimit, amdgpu_vis_vram_limit, int, 0444);
 
 /**
  * DOC: gartsize (uint)
- * Restrict the size of GART in Mib (32, 64, etc.) for testing. The default is -1 (The size depends on asic).
+ * Restrict the size of GART (for kernel use) in Mib (32, 64, etc.) for testing.
+ * The default is -1 (The size depends on asic).
  */
-MODULE_PARM_DESC(gartsize, "Size of GART to setup in megabytes (32, 64, etc., -1=auto)");
+MODULE_PARM_DESC(gartsize, "Size of kernel GART to setup in megabytes (32, 64, etc., -1=auto)");
 module_param_named(gartsize, amdgpu_gart_size, uint, 0600);
 
 /**
  * DOC: gttsize (int)
- * Restrict the size of GTT domain in MiB for testing. The default is -1 (Use 1/2 RAM, minimum value is 3GB).
+ * Restrict the size of GTT domain (for userspace use) in MiB for testing.
+ * The default is -1 (Use 1/2 RAM, minimum value is 3GB).
  */
-MODULE_PARM_DESC(gttsize, "Size of the GTT domain in megabytes (-1 = auto)");
+MODULE_PARM_DESC(gttsize, "Size of the GTT userspace domain in megabytes (-1 = auto)");
 module_param_named(gttsize, amdgpu_gtt_size, int, 0600);
 
 /**
@@ -2569,6 +2571,8 @@ static int amdgpu_pmops_runtime_suspend(struct device *dev)
 		amdgpu_device_baco_enter(drm_dev);
 	}
 
+	dev_dbg(&pdev->dev, "asic/device is runtime suspended\n");
+
 	return 0;
 }
 

drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c

@@ -55,6 +55,7 @@ struct amdgpu_fence {
 
 	/* RB, DMA, etc. */
 	struct amdgpu_ring		*ring;
+	ktime_t				start_timestamp;
 };
 
 static struct kmem_cache *amdgpu_fence_slab;
@@ -199,6 +200,8 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f, struct amd
 		}
 	}
 
+	to_amdgpu_fence(fence)->start_timestamp = ktime_get();
+
 	/* This function can't be called concurrently anyway, otherwise
 	 * emitting the fence would mess up the hardware ring buffer.
 	 */
@@ -406,6 +409,57 @@ unsigned amdgpu_fence_count_emitted(struct amdgpu_ring *ring)
 	return lower_32_bits(emitted);
 }
 
+/**
+ * amdgpu_fence_last_unsignaled_time_us - the time fence emitted until now
+ * @ring: ring the fence is associated with
+ *
+ * Find the earliest fence unsignaled until now, calculate the time delta
+ * between the time fence emitted and now.
+ */
+u64 amdgpu_fence_last_unsignaled_time_us(struct amdgpu_ring *ring)
+{
+	struct amdgpu_fence_driver *drv = &ring->fence_drv;
+	struct dma_fence *fence;
+	uint32_t last_seq, sync_seq;
+
+	last_seq = atomic_read(&ring->fence_drv.last_seq);
+	sync_seq = READ_ONCE(ring->fence_drv.sync_seq);
+	if (last_seq == sync_seq)
+		return 0;
+
+	++last_seq;
+	last_seq &= drv->num_fences_mask;
+	fence = drv->fences[last_seq];
+	if (!fence)
+		return 0;
+
+	return ktime_us_delta(ktime_get(),
+		to_amdgpu_fence(fence)->start_timestamp);
+}
+
+/**
+ * amdgpu_fence_update_start_timestamp - update the timestamp of the fence
+ * @ring: ring the fence is associated with
+ * @seq: the fence seq number to update.
+ * @timestamp: the start timestamp to update.
+ *
+ * The function called at the time the fence and related ib is about to
+ * resubmit to gpu in MCBP scenario. Thus we do not consider race condition
+ * with amdgpu_fence_process to modify the same fence.
+ */
+void amdgpu_fence_update_start_timestamp(struct amdgpu_ring *ring, uint32_t seq, ktime_t timestamp)
+{
+	struct amdgpu_fence_driver *drv = &ring->fence_drv;
+	struct dma_fence *fence;
+
+	seq &= drv->num_fences_mask;
+	fence = drv->fences[seq];
+	if (!fence)
+		return;
+
+	to_amdgpu_fence(fence)->start_timestamp = timestamp;
+}
+
 /**
  * amdgpu_fence_driver_start_ring - make the fence driver
  * ready for use on the requested ring.

drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h

@@ -33,6 +33,7 @@
 #include "amdgpu_imu.h"
 #include "soc15.h"
 #include "amdgpu_ras.h"
+#include "amdgpu_ring_mux.h"
 
 /* GFX current status */
 #define AMDGPU_GFX_NORMAL_MODE			0x00000000L
@@ -352,6 +353,9 @@ struct amdgpu_gfx {
 	struct amdgpu_gfx_ras		*ras;
 
 	bool				is_poweron;
+
+	struct amdgpu_ring		sw_gfx_ring[AMDGPU_MAX_SW_GFX_RINGS];
+	struct amdgpu_ring_mux          muxer;
 };
 
 #define amdgpu_gfx_get_gpu_clock_counter(adev) (adev)->gfx.funcs->get_gpu_clock_counter((adev))