Merge tag 'amd-drm-next-5.6-2020-01-17' of git://people.freedesktop.org/~agd5f/linux into drm-next

amd-drm-next-5.6-2020-01-17:

amdgpu:
- Fix 32 bit harder
- Powerplay cleanups
- VCN fixes for Arcturus
- RAS fixes
- eDP/DP fixes
- SR-IOV fixes
- Re-enable S/G display for PCO/RV2
- Free stolen memory after init on gmc10
- DF hashing optimizations for Arcturus
- Properly handle runtime pm in sysfs and debugfs
- Unify more GC programming between amdgpu and amdkfd
- Golden settings updates for gfx10
- GDDR6 training fixes
- Freesync fixes
- DSC fixes
- TMDS fixes
- Renoir USB-C fixes
- DC dml updates from hw team
- Pollock support
- Mutex init regresson fix

amdkfd:
- Unify more GC programming between amdgpu and amdkfd
- Use KIQ to setup HIQ rather than using MMIO

scheduler:
- Documentation fixes
- Improve job distribution with load sharing

drm:
- DP MST fix

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexdeucher@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200117213625.4722-1-alexander.deucher@amd.com

1 files changed, 105 insertions, 9 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
index da9765ff45d6..bbede09983e1 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
@@ -30,6 +30,8 @@
 #include "hdp/hdp_5_0_0_sh_mask.h"
 #include "gc/gc_10_1_0_sh_mask.h"
 #include "mmhub/mmhub_2_0_0_sh_mask.h"
+#include "athub/athub_2_0_0_sh_mask.h"
+#include "athub/athub_2_0_0_offset.h"
 #include "dcn/dcn_2_0_0_offset.h"
 #include "dcn/dcn_2_0_0_sh_mask.h"
 #include "oss/osssys_5_0_0_offset.h"
@@ -37,6 +39,7 @@
 #include "navi10_enum.h"
 
 #include "soc15.h"
+#include "soc15d.h"
 #include "soc15_common.h"
 
 #include "nbio_v2_3.h"
@@ -234,6 +237,19 @@ static bool gmc_v10_0_use_invalidate_semaphore(struct amdgpu_device *adev,
 		(!amdgpu_sriov_vf(adev)));
 }
 
+static bool gmc_v10_0_get_atc_vmid_pasid_mapping_info(
+					struct amdgpu_device *adev,
+					uint8_t vmid, uint16_t *p_pasid)
+{
+	uint32_t value;
+
+	value = RREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING)
+		     + vmid);
+	*p_pasid = value & ATC_VMID0_PASID_MAPPING__PASID_MASK;
+
+	return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK);
+}
+
 /*
  * GART
  * VMID 0 is the physical GPU addresses as used by the kernel.
@@ -380,6 +396,63 @@ error_alloc:
 	DRM_ERROR("Error flushing GPU TLB using the SDMA (%d)!\n", r);
 }
 
+/**
+ * gmc_v10_0_flush_gpu_tlb_pasid - tlb flush via pasid
+ *
+ * @adev: amdgpu_device pointer
+ * @pasid: pasid to be flush
+ *
+ * Flush the TLB for the requested pasid.
+ */
+static int gmc_v10_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
+					uint16_t pasid, uint32_t flush_type,
+					bool all_hub)
+{
+	int vmid, i;
+	signed long r;
+	uint32_t seq;
+	uint16_t queried_pasid;
+	bool ret;
+	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
+
+	if (amdgpu_emu_mode == 0 && ring->sched.ready) {
+		spin_lock(&adev->gfx.kiq.ring_lock);
+		amdgpu_ring_alloc(ring, kiq->pmf->invalidate_tlbs_size);
+		kiq->pmf->kiq_invalidate_tlbs(ring,
+					pasid, flush_type, all_hub);
+		amdgpu_fence_emit_polling(ring, &seq);
+		amdgpu_ring_commit(ring);
+		spin_unlock(&adev->gfx.kiq.ring_lock);
+		r = amdgpu_fence_wait_polling(ring, seq, adev->usec_timeout);
+		if (r < 1) {
+			DRM_ERROR("wait for kiq fence error: %ld.\n", r);
+			return -ETIME;
+		}
+
+		return 0;
+	}
+
+	for (vmid = 1; vmid < 16; vmid++) {
+
+		ret = gmc_v10_0_get_atc_vmid_pasid_mapping_info(adev, vmid,
+				&queried_pasid);
+		if (ret	&& queried_pasid == pasid) {
+			if (all_hub) {
+				for (i = 0; i < adev->num_vmhubs; i++)
+					gmc_v10_0_flush_gpu_tlb(adev, vmid,
+							i, 0);
+			} else {
+				gmc_v10_0_flush_gpu_tlb(adev, vmid,
+						AMDGPU_GFXHUB_0, 0);
+			}
+			break;
+		}
+	}
+
+	return 0;
+}
+
 static uint64_t gmc_v10_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
 					     unsigned vmid, uint64_t pd_addr)
 {
@@ -531,6 +604,7 @@ static void gmc_v10_0_get_vm_pte(struct amdgpu_device *adev,
 
 static const struct amdgpu_gmc_funcs gmc_v10_0_gmc_funcs = {
 	.flush_gpu_tlb = gmc_v10_0_flush_gpu_tlb,
+	.flush_gpu_tlb_pasid = gmc_v10_0_flush_gpu_tlb_pasid,
 	.emit_flush_gpu_tlb = gmc_v10_0_emit_flush_gpu_tlb,
 	.emit_pasid_mapping = gmc_v10_0_emit_pasid_mapping,
 	.map_mtype = gmc_v10_0_map_mtype,
@@ -566,6 +640,13 @@ static int gmc_v10_0_late_init(void *handle)
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 	int r;
 
+	/*
+	 * Can't free the stolen VGA memory when it might be used for memory
+	 * training again.
+	 */
+	if (!adev->fw_vram_usage.mem_train_support)
+		amdgpu_bo_late_init(adev);
+
 	r = amdgpu_gmc_allocate_vm_inv_eng(adev);
 	if (r)
 		return r;
@@ -720,6 +801,10 @@ static int gmc_v10_0_sw_init(void *handle)
 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VMC,
 			      VMC_1_0__SRCID__VM_FAULT,
 			      &adev->gmc.vm_fault);
+
+	if (r)
+		return r;
+
 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_UTCL2,
 			      UTCL2_1_0__SRCID__FAULT,
 			      &adev->gmc.vm_fault);
@@ -732,15 +817,6 @@ static int gmc_v10_0_sw_init(void *handle)
 	 */
 	adev->gmc.mc_mask = 0xffffffffffffULL; /* 48 bit MC */
 
-	/*
-	 * Reserve 8M stolen memory for navi10 like vega10
-	 * TODO: will check if it's really needed on asic.
-	 */
-	if (amdgpu_emu_mode == 1)
-		adev->gmc.stolen_size = 0;
-	else
-		adev->gmc.stolen_size = 9 * 1024 *1024;
-
 	r = dma_set_mask_and_coherent(adev->dev, DMA_BIT_MASK(44));
 	if (r) {
 		printk(KERN_WARNING "amdgpu: No suitable DMA available.\n");
@@ -753,6 +829,19 @@ static int gmc_v10_0_sw_init(void *handle)
 
 	adev->gmc.stolen_size = gmc_v10_0_get_vbios_fb_size(adev);
 
+	/*
+	 * In dual GPUs scenario, stolen_size is assigned to zero on the
+	 * secondary GPU, since there is no pre-OS console using that memory.
+	 * Then the bottom region of VRAM was allocated as GTT, unfortunately a
+	 * small region of bottom VRAM was encroached by UMC firmware during
+	 * GDDR6 BIST training, this cause page fault.
+	 * The page fault can be fixed by forcing stolen_size to 3MB, then the
+	 * bottom region of VRAM was allocated as stolen memory, GTT corruption
+	 * avoid.
+	 */
+	adev->gmc.stolen_size = max(adev->gmc.stolen_size,
+				    AMDGPU_STOLEN_BIST_TRAINING_DEFAULT_SIZE);
+
 	/* Memory manager */
 	r = amdgpu_bo_init(adev);
 	if (r)
@@ -792,6 +881,13 @@ static void gmc_v10_0_gart_fini(struct amdgpu_device *adev)
 static int gmc_v10_0_sw_fini(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	void *stolen_vga_buf;
+
+	/*
+	 * Free the stolen memory if it wasn't already freed in late_init
+	 * because of memory training.
+	 */
+	amdgpu_bo_free_kernel(&adev->stolen_vga_memory, NULL, &stolen_vga_buf);
 
 	amdgpu_vm_manager_fini(adev);
 	gmc_v10_0_gart_fini(adev);