Merge tag 'drm-fixes-5.3-2019-08-07' of git://people.freedesktop.org/~agd5f/linux into drm-fixes

drm-fixes-5.3-2019-08-07:

amdgpu:
- Fixes VCN to handle the latest navi10 firmware
- Fixes for fan control on navi10
- Properly handle SMU metrics table on navi10
- Fix a resume regression on Stoney

amdkfd:
- Revert new GWS ioctl.  It's not ready.
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexdeucher@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190807184221.3323-1-alexander.deucher@amd.com

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

@@ -32,7 +32,6 @@ struct amdgpu_gds {
 	uint32_t gws_size;
 	uint32_t oa_size;
 	uint32_t gds_compute_max_wave_id;
-	uint32_t vgt_gs_max_wave_id;
 };
 
 struct amdgpu_gds_reg_offset {

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

@@ -30,6 +30,7 @@
 #define AMDGPU_VCN_FIRMWARE_OFFSET	256
 #define AMDGPU_VCN_MAX_ENC_RINGS	3
 
+#define VCN_DEC_KMD_CMD 		0x80000000
 #define VCN_DEC_CMD_FENCE		0x00000000
 #define VCN_DEC_CMD_TRAP		0x00000001
 #define VCN_DEC_CMD_WRITE_REG		0x00000004

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c

@@ -4206,15 +4206,6 @@ static void gfx_v10_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 	u32 header, control = 0;
 
-	/* Prevent a hw deadlock due to a wave ID mismatch between ME and GDS.
-	 * This resets the wave ID counters. (needed by transform feedback)
-	 * TODO: This might only be needed on a VMID switch when we change
-	 *       the GDS OA mapping, not sure.
-	 */
-	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
-	amdgpu_ring_write(ring, mmVGT_GS_MAX_WAVE_ID);
-	amdgpu_ring_write(ring, ring->adev->gds.vgt_gs_max_wave_id);
-
 	if (ib->flags & AMDGPU_IB_FLAG_CE)
 		header = PACKET3(PACKET3_INDIRECT_BUFFER_CNST, 2);
 	else
@@ -4961,7 +4952,7 @@ static const struct amdgpu_ring_funcs gfx_v10_0_ring_funcs_gfx = {
 		5 + /* HDP_INVL */
 		8 + 8 + /* FENCE x2 */
 		2, /* SWITCH_BUFFER */
-	.emit_ib_size =	7, /* gfx_v10_0_ring_emit_ib_gfx */
+	.emit_ib_size =	4, /* gfx_v10_0_ring_emit_ib_gfx */
 	.emit_ib = gfx_v10_0_ring_emit_ib_gfx,
 	.emit_fence = gfx_v10_0_ring_emit_fence,
 	.emit_pipeline_sync = gfx_v10_0_ring_emit_pipeline_sync,
@@ -5112,7 +5103,6 @@ static void gfx_v10_0_set_gds_init(struct amdgpu_device *adev)
 	default:
 		adev->gds.gds_size = 0x10000;
 		adev->gds.gds_compute_max_wave_id = 0x4ff;
-		adev->gds.vgt_gs_max_wave_id = 0x3ff;
 		break;
 	}
 

drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c

@@ -1321,6 +1321,39 @@ static int gfx_v8_0_rlc_init(struct amdgpu_device *adev)
 	return 0;
 }
 
+static int gfx_v8_0_csb_vram_pin(struct amdgpu_device *adev)
+{
+	int r;
+
+	r = amdgpu_bo_reserve(adev->gfx.rlc.clear_state_obj, false);
+	if (unlikely(r != 0))
+		return r;
+
+	r = amdgpu_bo_pin(adev->gfx.rlc.clear_state_obj,
+			AMDGPU_GEM_DOMAIN_VRAM);
+	if (!r)
+		adev->gfx.rlc.clear_state_gpu_addr =
+			amdgpu_bo_gpu_offset(adev->gfx.rlc.clear_state_obj);
+
+	amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj);
+
+	return r;
+}
+
+static void gfx_v8_0_csb_vram_unpin(struct amdgpu_device *adev)
+{
+	int r;
+
+	if (!adev->gfx.rlc.clear_state_obj)
+		return;
+
+	r = amdgpu_bo_reserve(adev->gfx.rlc.clear_state_obj, true);
+	if (likely(r == 0)) {
+		amdgpu_bo_unpin(adev->gfx.rlc.clear_state_obj);
+		amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj);
+	}
+}
+
 static void gfx_v8_0_mec_fini(struct amdgpu_device *adev)
 {
 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
@@ -4785,6 +4818,10 @@ static int gfx_v8_0_hw_init(void *handle)
 	gfx_v8_0_init_golden_registers(adev);
 	gfx_v8_0_constants_init(adev);
 
+	r = gfx_v8_0_csb_vram_pin(adev);
+	if (r)
+		return r;
+
 	r = adev->gfx.rlc.funcs->resume(adev);
 	if (r)
 		return r;
@@ -4901,6 +4938,9 @@ static int gfx_v8_0_hw_fini(void *handle)
 	else
 		pr_err("rlc is busy, skip halt rlc\n");
 	amdgpu_gfx_rlc_exit_safe_mode(adev);
+
+	gfx_v8_0_csb_vram_unpin(adev);
+
 	return 0;
 }
 

drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c

@@ -1485,7 +1485,7 @@ static void vcn_v2_0_dec_ring_insert_start(struct amdgpu_ring *ring)
 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET, 0));
 	amdgpu_ring_write(ring, 0);
 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET, 0));
-	amdgpu_ring_write(ring, VCN_DEC_CMD_PACKET_START << 1);
+	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_PACKET_START << 1));
 }
 
 /**
@@ -1498,7 +1498,7 @@ static void vcn_v2_0_dec_ring_insert_start(struct amdgpu_ring *ring)
 static void vcn_v2_0_dec_ring_insert_end(struct amdgpu_ring *ring)
 {
 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET, 0));
-	amdgpu_ring_write(ring, VCN_DEC_CMD_PACKET_END << 1);
+	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_PACKET_END << 1));
 }
 
 /**
@@ -1543,7 +1543,7 @@ static void vcn_v2_0_dec_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64
 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
 
 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET, 0));
-	amdgpu_ring_write(ring, VCN_DEC_CMD_FENCE << 1);
+	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_FENCE << 1));
 
 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET, 0));
 	amdgpu_ring_write(ring, 0);
@@ -1553,7 +1553,7 @@ static void vcn_v2_0_dec_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64
 
 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET, 0));
 
-	amdgpu_ring_write(ring, VCN_DEC_CMD_TRAP << 1);
+	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_TRAP << 1));
 }
 
 /**
@@ -1597,7 +1597,7 @@ static void vcn_v2_0_dec_ring_emit_reg_wait(struct amdgpu_ring *ring,
 
 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET, 0));
 
-	amdgpu_ring_write(ring, VCN_DEC_CMD_REG_READ_COND_WAIT << 1);
+	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_REG_READ_COND_WAIT << 1));
 }
 
 static void vcn_v2_0_dec_ring_emit_vm_flush(struct amdgpu_ring *ring,
@@ -1626,7 +1626,7 @@ static void vcn_v2_0_dec_ring_emit_wreg(struct amdgpu_ring *ring,
 
 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET, 0));
 
-	amdgpu_ring_write(ring, VCN_DEC_CMD_WRITE_REG << 1);
+	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_WRITE_REG << 1));
 }
 
 /**
@@ -2079,6 +2079,36 @@ static int vcn_v2_0_process_interrupt(struct amdgpu_device *adev,
 	return 0;
 }
 
+static int vcn_v2_0_dec_ring_test_ring(struct amdgpu_ring *ring)
+{
+	struct amdgpu_device *adev = ring->adev;
+	uint32_t tmp = 0;
+	unsigned i;
+	int r;
+
+	WREG32(adev->vcn.external.scratch9, 0xCAFEDEAD);
+	r = amdgpu_ring_alloc(ring, 4);
+	if (r)
+		return r;
+	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET, 0));
+	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_PACKET_START << 1));
+	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0));
+	amdgpu_ring_write(ring, 0xDEADBEEF);
+	amdgpu_ring_commit(ring);
+	for (i = 0; i < adev->usec_timeout; i++) {
+		tmp = RREG32(adev->vcn.external.scratch9);
+		if (tmp == 0xDEADBEEF)
+			break;
+		DRM_UDELAY(1);
+	}
+
+	if (i >= adev->usec_timeout)
+		r = -ETIMEDOUT;
+
+	return r;
+}
+
+
 static int vcn_v2_0_set_powergating_state(void *handle,
 					  enum amd_powergating_state state)
 {
@@ -2142,7 +2172,7 @@ static const struct amdgpu_ring_funcs vcn_v2_0_dec_ring_vm_funcs = {
 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
-	.test_ring = amdgpu_vcn_dec_ring_test_ring,
+	.test_ring = vcn_v2_0_dec_ring_test_ring,
 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
 	.insert_start = vcn_v2_0_dec_ring_insert_start,

drivers/gpu/drm/amd/amdkfd/kfd_chardev.c

@@ -1567,32 +1567,6 @@ static int kfd_ioctl_unmap_memory_from_gpu(struct file *filep,
 	return err;
 }
 
-static int kfd_ioctl_alloc_queue_gws(struct file *filep,
-		struct kfd_process *p, void *data)
-{
-	int retval;
-	struct kfd_ioctl_alloc_queue_gws_args *args = data;
-	struct kfd_dev *dev;
-
-	if (!hws_gws_support)
-		return -ENODEV;
-
-	dev = kfd_device_by_id(args->gpu_id);
-	if (!dev) {
-		pr_debug("Could not find gpu id 0x%x\n", args->gpu_id);
-		return -ENODEV;
-	}
-	if (dev->dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS)
-		return -ENODEV;
-
-	mutex_lock(&p->mutex);
-	retval = pqm_set_gws(&p->pqm, args->queue_id, args->num_gws ? dev->gws : NULL);
-	mutex_unlock(&p->mutex);
-
-	args->first_gws = 0;
-	return retval;
-}
-
 static int kfd_ioctl_get_dmabuf_info(struct file *filep,
 		struct kfd_process *p, void *data)
 {
@@ -1795,8 +1769,6 @@ static const struct amdkfd_ioctl_desc amdkfd_ioctls[] = {
 	AMDKFD_IOCTL_DEF(AMDKFD_IOC_IMPORT_DMABUF,
 				kfd_ioctl_import_dmabuf, 0),
 
-	AMDKFD_IOCTL_DEF(AMDKFD_IOC_ALLOC_QUEUE_GWS,
-			kfd_ioctl_alloc_queue_gws, 0),
 };
 
 #define AMDKFD_CORE_IOCTL_COUNT	ARRAY_SIZE(amdkfd_ioctls)

drivers/gpu/drm/amd/powerplay/amdgpu_smu.c

@@ -315,6 +315,8 @@ int smu_get_power_num_states(struct smu_context *smu,
 int smu_common_read_sensor(struct smu_context *smu, enum amd_pp_sensors sensor,
 			   void *data, uint32_t *size)
 {
+	struct smu_power_context *smu_power = &smu->smu_power;
+	struct smu_power_gate *power_gate = &smu_power->power_gate;
 	int ret = 0;
 
 	switch (sensor) {
@@ -339,7 +341,7 @@ int smu_common_read_sensor(struct smu_context *smu, enum amd_pp_sensors sensor,
 		*size = 4;
 		break;
 	case AMDGPU_PP_SENSOR_VCN_POWER_STATE:
-		*(uint32_t *)data = smu_feature_is_enabled(smu, SMU_FEATURE_VCN_PG_BIT) ? 1 : 0;
+		*(uint32_t *)data = power_gate->vcn_gated ? 0 : 1;
 		*size = 4;
 		break;
 	default:

drivers/gpu/drm/amd/powerplay/inc/amdgpu_smu.h

@@ -451,6 +451,7 @@ struct smu_dpm_context {
 struct smu_power_gate {
 	bool uvd_gated;
 	bool vce_gated;
+	bool vcn_gated;
 };
 
 struct smu_power_context {

drivers/gpu/drm/amd/powerplay/navi10_ppt.c

@@ -502,6 +502,8 @@ static int navi10_store_powerplay_table(struct smu_context *smu)
 
 static int navi10_tables_init(struct smu_context *smu, struct smu_table *tables)
 {
+	struct smu_table_context *smu_table = &smu->smu_table;
+
 	SMU_TABLE_INIT(tables, SMU_TABLE_PPTABLE, sizeof(PPTable_t),
 		       PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM);
 	SMU_TABLE_INIT(tables, SMU_TABLE_WATERMARKS, sizeof(Watermarks_t),
@@ -516,9 +518,35 @@ static int navi10_tables_init(struct smu_context *smu, struct smu_table *tables)
 		       sizeof(DpmActivityMonitorCoeffInt_t), PAGE_SIZE,
 		       AMDGPU_GEM_DOMAIN_VRAM);
 
+	smu_table->metrics_table = kzalloc(sizeof(SmuMetrics_t), GFP_KERNEL);
+	if (!smu_table->metrics_table)
+		return -ENOMEM;
+	smu_table->metrics_time = 0;
+
 	return 0;
 }
 
+static int navi10_get_metrics_table(struct smu_context *smu,
+				    SmuMetrics_t *metrics_table)
+{
+	struct smu_table_context *smu_table= &smu->smu_table;
+	int ret = 0;
+
+	if (!smu_table->metrics_time || time_after(jiffies, smu_table->metrics_time + HZ / 1000)) {
+		ret = smu_update_table(smu, SMU_TABLE_SMU_METRICS, 0,
+				(void *)smu_table->metrics_table, false);
+		if (ret) {
+			pr_info("Failed to export SMU metrics table!\n");
+			return ret;
+		}
+		smu_table->metrics_time = jiffies;
+	}
+
+	memcpy(metrics_table, smu_table->metrics_table, sizeof(SmuMetrics_t));
+
+	return ret;
+}
+
 static int navi10_allocate_dpm_context(struct smu_context *smu)
 {
 	struct smu_dpm_context *smu_dpm = &smu->smu_dpm;
@@ -577,19 +605,26 @@ static int navi10_set_default_dpm_table(struct smu_context *smu)
 
 static int navi10_dpm_set_uvd_enable(struct smu_context *smu, bool enable)
 {
+	struct smu_power_context *smu_power = &smu->smu_power;
+	struct smu_power_gate *power_gate = &smu_power->power_gate;
 	int ret = 0;
 
 	if (enable) {
+		/* vcn dpm on is a prerequisite for vcn power gate messages */
+		if (smu_feature_is_enabled(smu, SMU_FEATURE_VCN_PG_BIT)) {
 			ret = smu_send_smc_msg_with_param(smu, SMU_MSG_PowerUpVcn, 1);
 			if (ret)
 				return ret;
+		}
+		power_gate->vcn_gated = false;
 	} else {
+		if (smu_feature_is_enabled(smu, SMU_FEATURE_VCN_PG_BIT)) {
 			ret = smu_send_smc_msg(smu, SMU_MSG_PowerDownVcn);
 			if (ret)
 				return ret;
 		}
-
-	ret = smu_feature_set_enabled(smu, SMU_FEATURE_VCN_PG_BIT, enable);
+		power_gate->vcn_gated = true;
+	}
 
 	return ret;
 }
@@ -598,15 +633,10 @@ static int navi10_get_current_clk_freq_by_table(struct smu_context *smu,
 				       enum smu_clk_type clk_type,
 				       uint32_t *value)
 {
-	static SmuMetrics_t metrics;
 	int ret = 0, clk_id = 0;
+	SmuMetrics_t metrics;
 
-	if (!value)
-		return -EINVAL;
-
-	memset(&metrics, 0, sizeof(metrics));
-
-	ret = smu_update_table(smu, SMU_TABLE_SMU_METRICS, 0, (void *)&metrics, false);
+	ret = navi10_get_metrics_table(smu, &metrics);
 	if (ret)
 		return ret;
 
@@ -894,8 +924,9 @@ static int navi10_get_gpu_power(struct smu_context *smu, uint32_t *value)
 	if (!value)
 		return -EINVAL;
 
-	ret = smu_update_table(smu, SMU_TABLE_SMU_METRICS, 0, (void *)&metrics,
-			       false);
+	ret = navi10_get_metrics_table(smu, &metrics);
+	if (ret)
+		return ret;
 	if (ret)
 		return ret;
 
@@ -914,10 +945,7 @@ static int navi10_get_current_activity_percent(struct smu_context *smu,
 	if (!value)
 		return -EINVAL;
 
-	msleep(1);
-
-	ret = smu_update_table(smu, SMU_TABLE_SMU_METRICS, 0,
-			       (void *)&metrics, false);
+	ret = navi10_get_metrics_table(smu, &metrics);
 	if (ret)
 		return ret;
 
@@ -956,10 +984,9 @@ static int navi10_get_fan_speed_rpm(struct smu_context *smu,
 	if (!speed)
 		return -EINVAL;
 
-	memset(&metrics, 0, sizeof(metrics));
-
-	ret = smu_update_table(smu, SMU_TABLE_SMU_METRICS, 0,
-			       (void *)&metrics, false);
+	ret = navi10_get_metrics_table(smu, &metrics);
+	if (ret)
+		return ret;
 	if (ret)
 		return ret;
 
@@ -1307,7 +1334,7 @@ static int navi10_thermal_get_temperature(struct smu_context *smu,
 	if (!value)
 		return -EINVAL;
 
-	ret = smu_update_table(smu, SMU_TABLE_SMU_METRICS, 0, (void *)&metrics, false);
+	ret = navi10_get_metrics_table(smu, &metrics);
 	if (ret)
 		return ret;
 

drivers/gpu/drm/amd/powerplay/smu_v11_0.c

@@ -1391,7 +1391,7 @@ smu_v11_0_smc_fan_control(struct smu_context *smu, bool start)
 {
 	int ret = 0;
 
-	if (smu_feature_is_supported(smu, SMU_FEATURE_FAN_CONTROL_BIT))
+	if (!smu_feature_is_supported(smu, SMU_FEATURE_FAN_CONTROL_BIT))
 		return 0;
 
 	ret = smu_feature_set_enabled(smu, SMU_FEATURE_FAN_CONTROL_BIT, start);

include/uapi/linux/kfd_ioctl.h

@@ -410,21 +410,6 @@ struct kfd_ioctl_unmap_memory_from_gpu_args {
 	__u32 n_success;		/* to/from KFD */
 };
 
-/* Allocate GWS for specific queue
- *
- * @gpu_id:      device identifier
- * @queue_id:    queue's id that GWS is allocated for
- * @num_gws:     how many GWS to allocate
- * @first_gws:   index of the first GWS allocated.
- *               only support contiguous GWS allocation
- */
-struct kfd_ioctl_alloc_queue_gws_args {
-	__u32 gpu_id;		/* to KFD */
-	__u32 queue_id;		/* to KFD */
-	__u32 num_gws;		/* to KFD */
-	__u32 first_gws;	/* from KFD */
-};
-
 struct kfd_ioctl_get_dmabuf_info_args {
 	__u64 size;		/* from KFD */
 	__u64 metadata_ptr;	/* to KFD */
@@ -544,10 +529,7 @@ enum kfd_mmio_remap {
 #define AMDKFD_IOC_IMPORT_DMABUF		\
 		AMDKFD_IOWR(0x1D, struct kfd_ioctl_import_dmabuf_args)
 
-#define AMDKFD_IOC_ALLOC_QUEUE_GWS		\
-		AMDKFD_IOWR(0x1E, struct kfd_ioctl_alloc_queue_gws_args)
-
 #define AMDKFD_COMMAND_START		0x01
-#define AMDKFD_COMMAND_END		0x1F
+#define AMDKFD_COMMAND_END		0x1E
 
 #endif