Merge tag 'amd-drm-next-5.8-2020-05-12' of git://people.freedesktop.org/~agd5f/linux into drm-next

amd-drm-next-5.8-2020-05-12:

amdgpu:
- Misc cleanups
- RAS fixes
- Expose FP16 for modesetting
- DP 1.4 compliance test fixes
- Clockgating fixes
- MAINTAINERS update
- Soft recovery for gfx10
- Runtime PM cleanups
- PSP code cleanups

amdkfd:
- Track GPU memory utilization per process
- Report PCI domain in topology

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexdeucher@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200512213703.4039-1-alexander.deucher@amd.com

1 files changed, 113 insertions, 24 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index b038ddbb2ece..cc41e8f5ad14 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -68,6 +68,7 @@
 
 #include <linux/suspend.h>
 #include <drm/task_barrier.h>
+#include <linux/pm_runtime.h>
 
 MODULE_FIRMWARE("amdgpu/vega10_gpu_info.bin");
 MODULE_FIRMWARE("amdgpu/vega12_gpu_info.bin");
@@ -254,6 +255,32 @@ void amdgpu_device_vram_access(struct amdgpu_device *adev, loff_t pos,
 	uint32_t hi = ~0;
 	uint64_t last;
 
+
+#ifdef CONFIG_64BIT
+	last = min(pos + size, adev->gmc.visible_vram_size);
+	if (last > pos) {
+		void __iomem *addr = adev->mman.aper_base_kaddr + pos;
+		size_t count = last - pos;
+
+		if (write) {
+			memcpy_toio(addr, buf, count);
+			mb();
+			amdgpu_asic_flush_hdp(adev, NULL);
+		} else {
+			amdgpu_asic_invalidate_hdp(adev, NULL);
+			mb();
+			memcpy_fromio(buf, addr, count);
+		}
+
+		if (count == size)
+			return;
+
+		pos += count;
+		buf += count / 4;
+		size -= count;
+	}
+#endif
+
 	spin_lock_irqsave(&adev->mmio_idx_lock, flags);
 	for (last = pos + size; pos < last; pos += 4) {
 		uint32_t tmp = pos >> 31;
@@ -2891,6 +2918,14 @@ static int amdgpu_device_get_job_timeout_settings(struct amdgpu_device *adev)
 	return ret;
 }
 
+static const struct attribute *amdgpu_dev_attributes[] = {
+	&dev_attr_product_name.attr,
+	&dev_attr_product_number.attr,
+	&dev_attr_serial_number.attr,
+	&dev_attr_pcie_replay_count.attr,
+	NULL
+};
+
 /**
  * amdgpu_device_init - initialize the driver
  *
@@ -3240,27 +3275,9 @@ fence_driver_init:
 	queue_delayed_work(system_wq, &adev->delayed_init_work,
 			   msecs_to_jiffies(AMDGPU_RESUME_MS));
 
-	r = device_create_file(adev->dev, &dev_attr_pcie_replay_count);
+	r = sysfs_create_files(&adev->dev->kobj, amdgpu_dev_attributes);
 	if (r) {
-		dev_err(adev->dev, "Could not create pcie_replay_count");
-		return r;
-	}
-
-	r = device_create_file(adev->dev, &dev_attr_product_name);
-	if (r) {
-		dev_err(adev->dev, "Could not create product_name");
-		return r;
-	}
-
-	r = device_create_file(adev->dev, &dev_attr_product_number);
-	if (r) {
-		dev_err(adev->dev, "Could not create product_number");
-		return r;
-	}
-
-	r = device_create_file(adev->dev, &dev_attr_serial_number);
-	if (r) {
-		dev_err(adev->dev, "Could not create serial_number");
+		dev_err(adev->dev, "Could not create amdgpu device attr\n");
 		return r;
 	}
 
@@ -3343,12 +3360,10 @@ void amdgpu_device_fini(struct amdgpu_device *adev)
 	adev->rmmio = NULL;
 	amdgpu_device_doorbell_fini(adev);
 
-	device_remove_file(adev->dev, &dev_attr_pcie_replay_count);
 	if (adev->ucode_sysfs_en)
 		amdgpu_ucode_sysfs_fini(adev);
-	device_remove_file(adev->dev, &dev_attr_product_name);
-	device_remove_file(adev->dev, &dev_attr_product_number);
-	device_remove_file(adev->dev, &dev_attr_serial_number);
+
+	sysfs_remove_files(&adev->dev->kobj, amdgpu_dev_attributes);
 	if (IS_ENABLED(CONFIG_PERF_EVENTS))
 		amdgpu_pmu_fini(adev);
 	if (amdgpu_discovery && adev->asic_type >= CHIP_NAVI10)
@@ -4116,6 +4131,64 @@ static void amdgpu_device_unlock_adev(struct amdgpu_device *adev)
 	mutex_unlock(&adev->lock_reset);
 }
 
+static void amdgpu_device_resume_display_audio(struct amdgpu_device *adev)
+{
+	struct pci_dev *p = NULL;
+
+	p = pci_get_domain_bus_and_slot(pci_domain_nr(adev->pdev->bus),
+			adev->pdev->bus->number, 1);
+	if (p) {
+		pm_runtime_enable(&(p->dev));
+		pm_runtime_resume(&(p->dev));
+	}
+}
+
+static int amdgpu_device_suspend_display_audio(struct amdgpu_device *adev)
+{
+	enum amd_reset_method reset_method;
+	struct pci_dev *p = NULL;
+	u64 expires;
+
+	/*
+	 * For now, only BACO and mode1 reset are confirmed
+	 * to suffer the audio issue without proper suspended.
+	 */
+	reset_method = amdgpu_asic_reset_method(adev);
+	if ((reset_method != AMD_RESET_METHOD_BACO) &&
+	     (reset_method != AMD_RESET_METHOD_MODE1))
+		return -EINVAL;
+
+	p = pci_get_domain_bus_and_slot(pci_domain_nr(adev->pdev->bus),
+			adev->pdev->bus->number, 1);
+	if (!p)
+		return -ENODEV;
+
+	expires = pm_runtime_autosuspend_expiration(&(p->dev));
+	if (!expires)
+		/*
+		 * If we cannot get the audio device autosuspend delay,
+		 * a fixed 4S interval will be used. Considering 3S is
+		 * the audio controller default autosuspend delay setting.
+		 * 4S used here is guaranteed to cover that.
+		 */
+		expires = ktime_get_mono_fast_ns() + NSEC_PER_SEC * 4ULL;
+
+	while (!pm_runtime_status_suspended(&(p->dev))) {
+		if (!pm_runtime_suspend(&(p->dev)))
+			break;
+
+		if (expires < ktime_get_mono_fast_ns()) {
+			dev_warn(adev->dev, "failed to suspend display audio\n");
+			/* TODO: abort the succeeding gpu reset? */
+			return -ETIMEDOUT;
+		}
+	}
+
+	pm_runtime_disable(&(p->dev));
+
+	return 0;
+}
+
 /**
  * amdgpu_device_gpu_recover - reset the asic and recover scheduler
  *
@@ -4140,6 +4213,7 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
 	bool use_baco =
 		(amdgpu_asic_reset_method(adev) == AMD_RESET_METHOD_BACO) ?
 		true : false;
+	bool audio_suspended = false;
 
 	/*
 	 * Flush RAM to disk so that after reboot
@@ -4197,6 +4271,19 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
 			return 0;
 		}
 
+		/*
+		 * Try to put the audio codec into suspend state
+		 * before gpu reset started.
+		 *
+		 * Due to the power domain of the graphics device
+		 * is shared with AZ power domain. Without this,
+		 * we may change the audio hardware from behind
+		 * the audio driver's back. That will trigger
+		 * some audio codec errors.
+		 */
+		if (!amdgpu_device_suspend_display_audio(tmp_adev))
+			audio_suspended = true;
+
 		amdgpu_ras_set_error_query_ready(tmp_adev, false);
 
 		cancel_delayed_work_sync(&tmp_adev->delayed_init_work);
@@ -4309,6 +4396,8 @@ skip_sched_resume:
 		/*unlock kfd: SRIOV would do it separately */
 		if (!(in_ras_intr && !use_baco) && !amdgpu_sriov_vf(tmp_adev))
 	                amdgpu_amdkfd_post_reset(tmp_adev);
+		if (audio_suspended)
+			amdgpu_device_resume_display_audio(tmp_adev);
 		amdgpu_device_unlock_adev(tmp_adev);
 	}