Merge tag 'drm-next-2022-10-05' of git://anongit.freedesktop.org/drm/drm

Pull drm updates from Dave Airlie:
 "Lots of stuff all over, some new AMD IP support and gang submit
  support. i915 has further DG2 and Meteorlake pieces, and a bunch of
  i915 display refactoring. msm has a shrinker rework. There are also a
  bunch of conversions to use kunit.

  This has two external pieces, some MEI changes needed for future Intel
  discrete GPUs. These should be acked by Greg. There is also a cross
  maintainer shared tree with some backlight rework from Hans in here.

  Core:
   - convert selftests to kunit
   - managed init for more objects
   - move to idr_init_base
   - rename fb and gem cma helpers to dma
   - hide unregistered connectors from getconnector ioctl
   - DSC passthrough aux support
   - backlight handling improvements
   - add dma_resv_assert_held to vmap/vunmap

  edid:
   - move luminance calculation to core

  fbdev:
   - fix aperture helper usage

  fourcc:
   - add more format helpers
   - add DRM_FORMAT_Cxx, DRM_FORMAT_Rxx, DRM_FORMAT_Dxx
   - add packed AYUV8888, XYUV8888
   - add some kunit tests

  ttm:
   - allow bos without backing store
   - rewrite placement to use intersect/compatible functions

  dma-buf:
   - docs update
   - improve signalling when debugging

  udmabuf:
   - fix failure path GPF

  dp:
   - drop dp/mst legacy code
   - atomic mst state support
   - audio infoframe packing

  panel:
   - Samsung LTL101AL01
   - B120XAN01.0
   - R140NWF5 RH
   - DMT028VGHMCMI-1A T
   - AUO B133UAN02.1
   - IVO M133NW4J-R3
   - Innolux N120ACA-EA1

  amdgpu:
   - Gang submit support
   - Mode2 reset for RDNA2
   - New IP support:
        DCN 3.1.4, 3.2
        SMU 13.x
        NBIO 7.7
        GC 11.x
        PSP 13.x
        SDMA 6.x
        GMC 11.x
   - DSC passthrough support
   - PSP fixes for TA support
   - vangogh GFXOFF stats
   - clang fixes
   - gang submit CS cleanup prep work
   - fix VRAM eviction issues

  amdkfd:
   - GC 10.3 IP ISA fixes
   - fix CRIU regression
   - CPU fault on COW mapping fixes

  i915:
   - align fw versioning with kernel practices
   - add display substruct to i915 private
   - add initial runtime info to driver info
   - split out HDCP and backlight registers
   - MEI XeHP SDV GSC support
   - add per-gt sysfs defaults
   - TLB invalidation improvements
   - Disable PCI BAR resize on 32-bit
   - GuC firmware updates and compat changes
   - GuC log timestamp translation
   - DG2 preemption workaround changes
   - DG2 improved HDMI pixel clocks support
   - PCI BAR sanity checks
   - Enable DC5 on DG2
   - DG2 DMC fw bumped
   - ADL-S PCI ID added
   - Meteorlake enablement
   - Rename ggtt_view to gtt_view
   - host RPS fixes
   - release mmaps on rpm suspend on discrete
   - clocking and dpll refactoring
   - VBT definitions and parsing updates
   - SKL watermark code extracted to separate file
   - allow seamless M/N changes on eDP panels
   - BUG_ON removal and cleanups

  msm:
   - DPU:
       simplified VBIF configuration
       cleanup CTL interfaces
   - DSI:
       removed unused msm_display_dsc_config struct
       switch regulator calls to new API
       switched to PANEL_BRIDGE for direct attached panels
   - DSI_PHY: convert drivers to parent_hws
   - DP: cleanup pixel_rate handling
   - HDMI: turned hdmi-phy-8996 into OF clk provider
   - misc dt-bindings fixes
   - choose eDP as primary display if it's available
   - support getting interconnects from either the mdss or the mdp5/dpu
     device nodes
   - gem: Shrinker + LRU re-work:
   - adds a shared GEM LRU+shrinker helper and moves msm over to that
   - reduce lock contention between retire and submit by avoiding the
     need to acquire obj lock in retire path (and instead using resv
     seeing obj's busyness in the shrinker
   - fix reclaim vs submit issues
   - GEM fault injection for triggering userspace error paths
   - Map/unmap optimization
   - Improved robustness for a6xx GPU recovery

  virtio:
   - improve error and edge conditions handling
   - convert to use managed helpers
   - stop exposing LINEAR modifier

  mgag200:
   - split modeset handling per model

  udl:
   - suspend/disconnect handling improvements

  vc4:
   - rework HDMI power up
   - depend on PM
   - better unplugging support

  ast:
   - resolution handling improvements

  ingenic:
   - add JZ4760(B) support
   - avoid a modeset when sharpness property is unchanged
   - use the new PM ops

  it6505:
   - power seq and clock updates

  ssd130x:
   - regmap bulk write
   - use atomic helpers instead of simple helpers

  via:
   - rename via_drv to via_dri1, consolidate all code.

  radeon:
   - drop DP MST experimental support
   - delayed work flush fix
   - use time_after

  ti-sn65dsi86:
   - DP support

  mediatek:
   - MT8195 DP support
   - drop of_gpio header
   - remove unneeded result
   - small DP code improvements

  vkms:
   - RGB565, XRGB64 and ARGB64 support

  sun4i:
   - tv: convert to atomic

  rcar-du:
   - Synopsys DW HDMI bridge DT bindings update

  exynos:
   - use drm_display_info.is_hdmi
   - correct return of mixer_mode_valid and hdmi_mode_valid

  omap:
   - refcounting fix

  rockchip:
   - RK3568 support
   - RK3399 gamma support"

* tag 'drm-next-2022-10-05' of git://anongit.freedesktop.org/drm/drm: (1374 commits)
  drm/amdkfd: Fix UBSAN shift-out-of-bounds warning
  drm/amdkfd: Track unified memory when switching xnack mode
  drm/amdgpu: Enable sram on vcn_4_0_2
  drm/amdgpu: Enable VCN DPG for GC11_0_1
  drm/msm: Fix build break with recent mm tree
  drm/panel: simple: Use dev_err_probe() to simplify code
  drm/panel: panel-edp: Use dev_err_probe() to simplify code
  drm/panel: simple: Add Multi-Inno Technology MI0800FT-9
  dt-bindings: display: simple: Add Multi-Inno Technology MI0800FT-9 panel
  drm/amdgpu: correct the memcpy size for ip discovery firmware
  drm/amdgpu: Skip put_reset_domain if it doesn't exist
  drm/amdgpu: remove switch from amdgpu_gmc_noretry_set
  drm/amdgpu: Fix mc_umc_status used uninitialized warning
  drm/amd/display: Prevent OTG shutdown during PSR SU
  drm/amdgpu: add page retirement handling for CPU RAS
  drm/amdgpu: use RAS error address convert api in mca notifier
  drm/amdgpu: support to convert dedicated umc mca address
  drm/amdgpu: export umc error address convert interface
  drm/amdgpu: fix sdma v4 init microcode error
  drm/amd/display: fix array-bounds error in dc_stream_remove_writeback()
  ...

1 files changed, 100 insertions, 22 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 25e1f5ed7ead..ab8f970b2849 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -2459,19 +2459,21 @@ static int amdgpu_device_ip_init(struct amdgpu_device *adev)
 	 */
 	if (adev->gmc.xgmi.num_physical_nodes > 1) {
 		if (amdgpu_xgmi_add_device(adev) == 0) {
-			struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev);
+			if (!amdgpu_sriov_vf(adev)) {
+				struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev);
 
-			if (!hive->reset_domain ||
-			    !amdgpu_reset_get_reset_domain(hive->reset_domain)) {
-				r = -ENOENT;
+				if (!hive->reset_domain ||
+				    !amdgpu_reset_get_reset_domain(hive->reset_domain)) {
+					r = -ENOENT;
+					amdgpu_put_xgmi_hive(hive);
+					goto init_failed;
+				}
+
+				/* Drop the early temporary reset domain we created for device */
+				amdgpu_reset_put_reset_domain(adev->reset_domain);
+				adev->reset_domain = hive->reset_domain;
 				amdgpu_put_xgmi_hive(hive);
-				goto init_failed;
 			}
-
-			/* Drop the early temporary reset domain we created for device */
-			amdgpu_reset_put_reset_domain(adev->reset_domain);
-			adev->reset_domain = hive->reset_domain;
-			amdgpu_put_xgmi_hive(hive);
 		}
 	}
 
@@ -3510,6 +3512,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	adev->gmc.gart_size = 512 * 1024 * 1024;
 	adev->accel_working = false;
 	adev->num_rings = 0;
+	RCU_INIT_POINTER(adev->gang_submit, dma_fence_get_stub());
 	adev->mman.buffer_funcs = NULL;
 	adev->mman.buffer_funcs_ring = NULL;
 	adev->vm_manager.vm_pte_funcs = NULL;
@@ -3588,6 +3591,8 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	INIT_WORK(&adev->xgmi_reset_work, amdgpu_device_xgmi_reset_func);
 
 	adev->gfx.gfx_off_req_count = 1;
+	adev->gfx.gfx_off_residency = 0;
+	adev->gfx.gfx_off_entrycount = 0;
 	adev->pm.ac_power = power_supply_is_system_supplied() > 0;
 
 	atomic_set(&adev->throttling_logging_enabled, 1);
@@ -3976,8 +3981,7 @@ void amdgpu_device_fini_hw(struct amdgpu_device *adev)
 
 	amdgpu_gart_dummy_page_fini(adev);
 
-	if (drm_dev_is_unplugged(adev_to_drm(adev)))
-		amdgpu_device_unmap_mmio(adev);
+	amdgpu_device_unmap_mmio(adev);
 
 }
 
@@ -3990,6 +3994,7 @@ void amdgpu_device_fini_sw(struct amdgpu_device *adev)
 	release_firmware(adev->firmware.gpu_info_fw);
 	adev->firmware.gpu_info_fw = NULL;
 	adev->accel_working = false;
+	dma_fence_put(rcu_dereference_protected(adev->gang_submit, true));
 
 	amdgpu_reset_fini(adev);
 
@@ -4542,14 +4547,15 @@ bool amdgpu_device_has_job_running(struct amdgpu_device *adev)
  */
 bool amdgpu_device_should_recover_gpu(struct amdgpu_device *adev)
 {
-	if (!amdgpu_device_ip_check_soft_reset(adev)) {
-		dev_info(adev->dev, "Timeout, but no hardware hang detected.\n");
-		return false;
-	}
 
 	if (amdgpu_gpu_recovery == 0)
 		goto disabled;
 
+	if (!amdgpu_device_ip_check_soft_reset(adev)) {
+		dev_info(adev->dev,"Timeout, but no hardware hang detected.\n");
+		return false;
+	}
+
 	if (amdgpu_sriov_vf(adev))
 		return true;
 
@@ -4674,7 +4680,7 @@ int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev,
 		if (!need_full_reset)
 			need_full_reset = amdgpu_device_ip_need_full_reset(adev);
 
-		if (!need_full_reset) {
+		if (!need_full_reset && amdgpu_gpu_recovery) {
 			amdgpu_device_ip_pre_soft_reset(adev);
 			r = amdgpu_device_ip_soft_reset(adev);
 			amdgpu_device_ip_post_soft_reset(adev);
@@ -4770,6 +4776,7 @@ int amdgpu_do_asic_reset(struct list_head *device_list_handle,
 	struct amdgpu_device *tmp_adev = NULL;
 	bool need_full_reset, skip_hw_reset, vram_lost = false;
 	int r = 0;
+	bool gpu_reset_for_dev_remove = 0;
 
 	/* Try reset handler method first */
 	tmp_adev = list_first_entry(device_list_handle, struct amdgpu_device,
@@ -4789,6 +4796,10 @@ int amdgpu_do_asic_reset(struct list_head *device_list_handle,
 		test_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags);
 	skip_hw_reset = test_bit(AMDGPU_SKIP_HW_RESET, &reset_context->flags);
 
+	gpu_reset_for_dev_remove =
+		test_bit(AMDGPU_RESET_FOR_DEVICE_REMOVE, &reset_context->flags) &&
+			test_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags);
+
 	/*
 	 * ASIC reset has to be done on all XGMI hive nodes ASAP
 	 * to allow proper links negotiation in FW (within 1 sec)
@@ -4833,6 +4844,18 @@ int amdgpu_do_asic_reset(struct list_head *device_list_handle,
 		amdgpu_ras_intr_cleared();
 	}
 
+	/* Since the mode1 reset affects base ip blocks, the
+	 * phase1 ip blocks need to be resumed. Otherwise there
+	 * will be a BIOS signature error and the psp bootloader
+	 * can't load kdb on the next amdgpu install.
+	 */
+	if (gpu_reset_for_dev_remove) {
+		list_for_each_entry(tmp_adev, device_list_handle, reset_list)
+			amdgpu_device_ip_resume_phase1(tmp_adev);
+
+		goto end;
+	}
+
 	list_for_each_entry(tmp_adev, device_list_handle, reset_list) {
 		if (need_full_reset) {
 			/* post card */
@@ -5072,6 +5095,7 @@ static void amdgpu_device_recheck_guilty_jobs(
 
 			/* set guilty */
 			drm_sched_increase_karma(s_job);
+			amdgpu_reset_prepare_hwcontext(adev, reset_context);
 retry:
 			/* do hw reset */
 			if (amdgpu_sriov_vf(adev)) {
@@ -5154,6 +5178,11 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
 	bool need_emergency_restart = false;
 	bool audio_suspended = false;
 	int tmp_vram_lost_counter;
+	bool gpu_reset_for_dev_remove = false;
+
+	gpu_reset_for_dev_remove =
+			test_bit(AMDGPU_RESET_FOR_DEVICE_REMOVE, &reset_context->flags) &&
+				test_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags);
 
 	/*
 	 * Special case: RAS triggered and full reset isn't supported
@@ -5181,6 +5210,7 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
 
 	reset_context->job = job;
 	reset_context->hive = hive;
+
 	/*
 	 * Build list of devices to reset.
 	 * In case we are in XGMI hive mode, resort the device list
@@ -5188,8 +5218,11 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
 	 */
 	INIT_LIST_HEAD(&device_list);
 	if (!amdgpu_sriov_vf(adev) && (adev->gmc.xgmi.num_physical_nodes > 1)) {
-		list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head)
+		list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) {
 			list_add_tail(&tmp_adev->reset_list, &device_list);
+			if (gpu_reset_for_dev_remove && adev->shutdown)
+				tmp_adev->shutdown = true;
+		}
 		if (!list_is_first(&adev->reset_list, &device_list))
 			list_rotate_to_front(&adev->reset_list, &device_list);
 		device_list_handle = &device_list;
@@ -5272,6 +5305,10 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
 
 retry:	/* Rest of adevs pre asic reset from XGMI hive. */
 	list_for_each_entry(tmp_adev, device_list_handle, reset_list) {
+		if (gpu_reset_for_dev_remove) {
+			/* Workaroud for ASICs need to disable SMC first */
+			amdgpu_device_smu_fini_early(tmp_adev);
+		}
 		r = amdgpu_device_pre_asic_reset(tmp_adev, reset_context);
 		/*TODO Should we stop ?*/
 		if (r) {
@@ -5300,8 +5337,14 @@ retry:	/* Rest of adevs pre asic reset from XGMI hive. */
 			amdgpu_ras_resume(adev);
 	} else {
 		r = amdgpu_do_asic_reset(device_list_handle, reset_context);
-		if (r && r == -EAGAIN)
+		if (r && r == -EAGAIN) {
+			set_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context->flags);
+			adev->asic_reset_res = 0;
 			goto retry;
+		}
+
+		if (!r && gpu_reset_for_dev_remove)
+			goto recover_end;
 	}
 
 skip_hw_reset:
@@ -5375,6 +5418,7 @@ skip_sched_resume:
 		amdgpu_device_unset_mp1_state(tmp_adev);
 	}
 
+recover_end:
 	tmp_adev = list_first_entry(device_list_handle, struct amdgpu_device,
 					    reset_list);
 	amdgpu_device_unlock_reset_domain(tmp_adev->reset_domain);
@@ -5557,9 +5601,9 @@ bool amdgpu_device_is_peer_accessible(struct amdgpu_device *adev,
 		~*peer_adev->dev->dma_mask : ~((1ULL << 32) - 1);
 	resource_size_t aper_limit =
 		adev->gmc.aper_base + adev->gmc.aper_size - 1;
-	bool p2p_access = !adev->gmc.xgmi.connected_to_cpu &&
-			  !(pci_p2pdma_distance_many(adev->pdev,
-					&peer_adev->dev, 1, true) < 0);
+	bool p2p_access =
+		!adev->gmc.xgmi.connected_to_cpu &&
+		!(pci_p2pdma_distance(adev->pdev, peer_adev->dev, false) < 0);
 
 	return pcie_p2p && p2p_access && (adev->gmc.visible_vram_size &&
 		adev->gmc.real_vram_size == adev->gmc.visible_vram_size &&
@@ -5733,6 +5777,7 @@ pci_ers_result_t amdgpu_pci_slot_reset(struct pci_dev *pdev)
 	reset_context.reset_req_dev = adev;
 	set_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags);
 	set_bit(AMDGPU_SKIP_HW_RESET, &reset_context.flags);
+	set_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context.flags);
 
 	adev->no_hw_access = true;
 	r = amdgpu_device_pre_asic_reset(adev, &reset_context);
@@ -5942,3 +5987,36 @@ void amdgpu_device_pcie_port_wreg(struct amdgpu_device *adev,
 	(void)RREG32(data);
 	spin_unlock_irqrestore(&adev->pcie_idx_lock, flags);
 }
+
+/**
+ * amdgpu_device_switch_gang - switch to a new gang
+ * @adev: amdgpu_device pointer
+ * @gang: the gang to switch to
+ *
+ * Try to switch to a new gang.
+ * Returns: NULL if we switched to the new gang or a reference to the current
+ * gang leader.
+ */
+struct dma_fence *amdgpu_device_switch_gang(struct amdgpu_device *adev,
+					    struct dma_fence *gang)
+{
+	struct dma_fence *old = NULL;
+
+	do {
+		dma_fence_put(old);
+		rcu_read_lock();
+		old = dma_fence_get_rcu_safe(&adev->gang_submit);
+		rcu_read_unlock();
+
+		if (old == gang)
+			break;
+
+		if (!dma_fence_is_signaled(old))
+			return old;
+
+	} while (cmpxchg((struct dma_fence __force **)&adev->gang_submit,
+			 old, gang) != old);
+
+	dma_fence_put(old);
+	return NULL;
+}