Merge tag 'amd-drm-fixes-6.16-2025-06-05' of https://gitlab.freedesktop.org/agd5f/linux into drm-next

amd-drm-fixes-6.16-2025-06-05:

amdgpu:
- IP discovery fix
- Cleaner shader fix for GC 10.1.x
- OD fix
- UserQ fixes
- Non-OLED panel fix
- Misc display fixes
- Brightness fixes

amdkfd:
- Enable CONFIG_HSA_AMD on RISCV

Signed-off-by: Dave Airlie <[email protected]>

From: Alex Deucher <[email protected]>
Link: https://lore.kernel.org/r/[email protected]

Author: airlied

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

@@ -270,9 +270,10 @@ static int amdgpu_discovery_read_binary_from_sysmem(struct amdgpu_device *adev,
 static int amdgpu_discovery_read_binary_from_mem(struct amdgpu_device *adev,
 						 uint8_t *binary)
 {
+	bool sz_valid = true;
 	uint64_t vram_size;
-	u32 msg;
 	int i, ret = 0;
+	u32 msg;
 
 	if (!amdgpu_sriov_vf(adev)) {
 		/* It can take up to a second for IFWI init to complete on some dGPUs,
@@ -291,16 +292,25 @@ static int amdgpu_discovery_read_binary_from_mem(struct amdgpu_device *adev,
 		}
 	}
 
-	vram_size = (uint64_t)RREG32(mmRCC_CONFIG_MEMSIZE) << 20;
+	vram_size = RREG32(mmRCC_CONFIG_MEMSIZE);
+	if (!vram_size || vram_size == U32_MAX)
+		sz_valid = false;
+	else
+		vram_size <<= 20;
 
-	if (vram_size) {
+	if (sz_valid) {
 		uint64_t pos = vram_size - DISCOVERY_TMR_OFFSET;
 		amdgpu_device_vram_access(adev, pos, (uint32_t *)binary,
 					  adev->mman.discovery_tmr_size, false);
 	} else {
 		ret = amdgpu_discovery_read_binary_from_sysmem(adev, binary);
 	}
 
+	if (ret)
+		dev_err(adev->dev,
+			"failed to read discovery info from memory, vram size read: %llx",
+			vram_size);
+
 	return ret;
 }
 

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

@@ -58,7 +58,7 @@ amdgpu_gem_add_input_fence(struct drm_file *filp,
 		return 0;
 
 	syncobj_handles = memdup_user(u64_to_user_ptr(syncobj_handles_array),
-				      sizeof(uint32_t) * num_syncobj_handles);
+				      size_mul(sizeof(uint32_t), num_syncobj_handles));
 	if (IS_ERR(syncobj_handles))
 		return PTR_ERR(syncobj_handles);
 

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

@@ -2228,6 +2228,9 @@ void amdgpu_gfx_profile_ring_begin_use(struct amdgpu_ring *ring)
 	enum PP_SMC_POWER_PROFILE profile;
 	int r;
 
+	if (amdgpu_dpm_is_overdrive_enabled(adev))
+		return;
+
 	if (adev->gfx.num_gfx_rings)
 		profile = PP_SMC_POWER_PROFILE_FULLSCREEN3D;
 	else
@@ -2258,6 +2261,11 @@ void amdgpu_gfx_profile_ring_begin_use(struct amdgpu_ring *ring)
 
 void amdgpu_gfx_profile_ring_end_use(struct amdgpu_ring *ring)
 {
+	struct amdgpu_device *adev = ring->adev;
+
+	if (amdgpu_dpm_is_overdrive_enabled(adev))
+		return;
+
 	atomic_dec(&ring->adev->gfx.total_submission_cnt);
 
 	schedule_delayed_work(&ring->adev->gfx.idle_work, GFX_PROFILE_IDLE_TIMEOUT);

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

@@ -430,7 +430,7 @@ int amdgpu_userq_signal_ioctl(struct drm_device *dev, void *data,
 
 	num_syncobj_handles = args->num_syncobj_handles;
 	syncobj_handles = memdup_user(u64_to_user_ptr(args->syncobj_handles),
-				      sizeof(u32) * num_syncobj_handles);
+				      size_mul(sizeof(u32), num_syncobj_handles));
 	if (IS_ERR(syncobj_handles))
 		return PTR_ERR(syncobj_handles);
 
@@ -612,21 +612,21 @@ int amdgpu_userq_wait_ioctl(struct drm_device *dev, void *data,
 
 	num_read_bo_handles = wait_info->num_bo_read_handles;
 	bo_handles_read = memdup_user(u64_to_user_ptr(wait_info->bo_read_handles),
-				      sizeof(u32) * num_read_bo_handles);
+				      size_mul(sizeof(u32), num_read_bo_handles));
 	if (IS_ERR(bo_handles_read))
 		return PTR_ERR(bo_handles_read);
 
 	num_write_bo_handles = wait_info->num_bo_write_handles;
 	bo_handles_write = memdup_user(u64_to_user_ptr(wait_info->bo_write_handles),
-				       sizeof(u32) * num_write_bo_handles);
+				       size_mul(sizeof(u32), num_write_bo_handles));
 	if (IS_ERR(bo_handles_write)) {
 		r = PTR_ERR(bo_handles_write);
 		goto free_bo_handles_read;
 	}
 
 	num_syncobj = wait_info->num_syncobj_handles;
 	syncobj_handles = memdup_user(u64_to_user_ptr(wait_info->syncobj_handles),
-				      sizeof(u32) * num_syncobj);
+				      size_mul(sizeof(u32), num_syncobj));
 	if (IS_ERR(syncobj_handles)) {
 		r = PTR_ERR(syncobj_handles);
 		goto free_bo_handles_write;

drivers/gpu/drm/amd/amdgpu/gfx_v10_0_cleaner_shader.h

@@ -43,9 +43,9 @@ static const u32 gfx_10_1_10_cleaner_shader_hex[] = {
 	0xd70f6a01, 0x000202ff,
 	0x00000400, 0x80828102,
 	0xbf84fff7, 0xbefc03ff,
-	0x00000068, 0xbe803080,
-	0xbe813080, 0xbe823080,
-	0xbe833080, 0x80fc847c,
+	0x00000068, 0xbe803000,
+	0xbe813000, 0xbe823000,
+	0xbe833000, 0x80fc847c,
 	0xbf84fffa, 0xbeea0480,
 	0xbeec0480, 0xbeee0480,
 	0xbef00480, 0xbef20480,

drivers/gpu/drm/amd/amdgpu/gfx_v10_1_10_cleaner_shader.asm

@@ -40,7 +40,6 @@ shader main
   type(CS)
   wave_size(32)
 // Note: original source code from SQ team
-
 //
 // Create 32 waves in a threadgroup (CS waves)
 // Each allocates 64 VGPRs
@@ -71,8 +70,8 @@ label_0005:
   s_sub_u32     s2, s2, 8
   s_cbranch_scc0  label_0005
   //
-  s_mov_b32     s2, 0x80000000                     // Bit31 is first_wave
-  s_and_b32     s2, s2, s0                                  // sgpr0 has tg_size (first_wave) term as in ucode only COMPUTE_PGM_RSRC2.tg_size_en is set
+  s_mov_b32     s2, 0x80000000                       // Bit31 is first_wave
+  s_and_b32     s2, s2, s1                           // sgpr0 has tg_size (first_wave) term as in ucode only COMPUTE_PGM_RSRC2.tg_size_en is set
   s_cbranch_scc0  label_0023                         // Clean LDS if its first wave of ThreadGroup/WorkGroup
   // CLEAR LDS
   //
@@ -99,10 +98,10 @@ label_001F:
 label_0023:
   s_mov_b32     m0, 0x00000068  // Loop 108/4=27 times  (loop unrolled for performance)
 label_sgpr_loop:
-  s_movreld_b32     s0, 0
-  s_movreld_b32     s1, 0
-  s_movreld_b32     s2, 0
-  s_movreld_b32     s3, 0
+  s_movreld_b32     s0, s0
+  s_movreld_b32     s1, s0
+  s_movreld_b32     s2, s0
+  s_movreld_b32     s3, s0
   s_sub_u32         m0, m0, 4
   s_cbranch_scc0  label_sgpr_loop
 

drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c

@@ -36,7 +36,7 @@
 #include "gc/gc_12_0_0_offset.h"
 #include "gc/gc_12_0_0_sh_mask.h"
 #include "soc24_enum.h"
-#include "ivsrcid/gfx/irqsrcs_gfx_11_0_0.h"
+#include "ivsrcid/gfx/irqsrcs_gfx_12_0_0.h"
 
 #include "soc15.h"
 #include "clearstate_gfx12.h"
@@ -1453,28 +1453,28 @@ static int gfx_v12_0_sw_init(struct amdgpu_ip_block *ip_block)
 
 	/* EOP Event */
 	r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GRBM_CP,
-			      GFX_11_0_0__SRCID__CP_EOP_INTERRUPT,
+			      GFX_12_0_0__SRCID__CP_EOP_INTERRUPT,
 			      &adev->gfx.eop_irq);
 	if (r)
 		return r;
 
 	/* Bad opcode Event */
 	r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GRBM_CP,
-			      GFX_11_0_0__SRCID__CP_BAD_OPCODE_ERROR,
+			      GFX_12_0_0__SRCID__CP_BAD_OPCODE_ERROR,
 			      &adev->gfx.bad_op_irq);
 	if (r)
 		return r;
 
 	/* Privileged reg */
 	r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GRBM_CP,
-			      GFX_11_0_0__SRCID__CP_PRIV_REG_FAULT,
+			      GFX_12_0_0__SRCID__CP_PRIV_REG_FAULT,
 			      &adev->gfx.priv_reg_irq);
 	if (r)
 		return r;
 
 	/* Privileged inst */
 	r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GRBM_CP,
-			      GFX_11_0_0__SRCID__CP_PRIV_INSTR_FAULT,
+			      GFX_12_0_0__SRCID__CP_PRIV_INSTR_FAULT,
 			      &adev->gfx.priv_inst_irq);
 	if (r)
 		return r;

drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c

@@ -33,7 +33,7 @@
 #include "gc/gc_12_0_0_offset.h"
 #include "gc/gc_12_0_0_sh_mask.h"
 #include "hdp/hdp_6_0_0_offset.h"
-#include "ivsrcid/gfx/irqsrcs_gfx_11_0_0.h"
+#include "ivsrcid/gfx/irqsrcs_gfx_12_0_0.h"
 
 #include "soc15_common.h"
 #include "soc15.h"
@@ -43,6 +43,7 @@
 #include "sdma_v7_0.h"
 #include "v12_structs.h"
 #include "mes_userqueue.h"
+#include "amdgpu_userq_fence.h"
 
 MODULE_FIRMWARE("amdgpu/sdma_7_0_0.bin");
 MODULE_FIRMWARE("amdgpu/sdma_7_0_1.bin");
@@ -910,6 +911,9 @@ static int sdma_v7_0_mqd_init(struct amdgpu_device *adev, void *mqd,
 	m->sdmax_rlcx_csa_addr_lo = lower_32_bits(prop->csa_addr);
 	m->sdmax_rlcx_csa_addr_hi = upper_32_bits(prop->csa_addr);
 
+	m->sdmax_rlcx_mcu_dbg0 = lower_32_bits(prop->fence_address);
+	m->sdmax_rlcx_mcu_dbg1 = upper_32_bits(prop->fence_address);
+
 	return 0;
 }
 
@@ -1296,11 +1300,18 @@ static int sdma_v7_0_sw_init(struct amdgpu_ip_block *ip_block)
 
 	/* SDMA trap event */
 	r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GFX,
-			      GFX_11_0_0__SRCID__SDMA_TRAP,
+			      GFX_12_0_0__SRCID__SDMA_TRAP,
 			      &adev->sdma.trap_irq);
 	if (r)
 		return r;
 
+	/* SDMA user fence event */
+	r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GFX,
+			      GFX_12_0_0__SRCID__SDMA_FENCE,
+			      &adev->sdma.fence_irq);
+	if (r)
+		return r;
+
 	for (i = 0; i < adev->sdma.num_instances; i++) {
 		ring = &adev->sdma.instance[i].ring;
 		ring->ring_obj = NULL;
@@ -1526,25 +1537,9 @@ static int sdma_v7_0_process_trap_irq(struct amdgpu_device *adev,
 				      struct amdgpu_iv_entry *entry)
 {
 	int instances, queue;
-	uint32_t mes_queue_id = entry->src_data[0];
 
 	DRM_DEBUG("IH: SDMA trap\n");
 
-	if (adev->enable_mes && (mes_queue_id & AMDGPU_FENCE_MES_QUEUE_FLAG)) {
-		struct amdgpu_mes_queue *queue;
-
-		mes_queue_id &= AMDGPU_FENCE_MES_QUEUE_ID_MASK;
-
-		spin_lock(&adev->mes.queue_id_lock);
-		queue = idr_find(&adev->mes.queue_id_idr, mes_queue_id);
-		if (queue) {
-			DRM_DEBUG("process smda queue id = %d\n", mes_queue_id);
-			amdgpu_fence_process(queue->ring);
-		}
-		spin_unlock(&adev->mes.queue_id_lock);
-		return 0;
-	}
-
 	queue = entry->ring_id & 0xf;
 	instances = (entry->ring_id & 0xf0) >> 4;
 	if (instances > 1) {
@@ -1566,6 +1561,29 @@ static int sdma_v7_0_process_trap_irq(struct amdgpu_device *adev,
 	return 0;
 }
 
+static int sdma_v7_0_process_fence_irq(struct amdgpu_device *adev,
+				       struct amdgpu_irq_src *source,
+				       struct amdgpu_iv_entry *entry)
+{
+	u32 doorbell_offset = entry->src_data[0];
+
+	if (adev->enable_mes && doorbell_offset) {
+		struct amdgpu_userq_fence_driver *fence_drv = NULL;
+		struct xarray *xa = &adev->userq_xa;
+		unsigned long flags;
+
+		doorbell_offset >>= SDMA0_QUEUE0_DOORBELL_OFFSET__OFFSET__SHIFT;
+
+		xa_lock_irqsave(xa, flags);
+		fence_drv = xa_load(xa, doorbell_offset);
+		if (fence_drv)
+			amdgpu_userq_fence_driver_process(fence_drv);
+		xa_unlock_irqrestore(xa, flags);
+	}
+
+	return 0;
+}
+
 static int sdma_v7_0_process_illegal_inst_irq(struct amdgpu_device *adev,
 					      struct amdgpu_irq_src *source,
 					      struct amdgpu_iv_entry *entry)
@@ -1703,6 +1721,10 @@ static const struct amdgpu_irq_src_funcs sdma_v7_0_trap_irq_funcs = {
 	.process = sdma_v7_0_process_trap_irq,
 };
 
+static const struct amdgpu_irq_src_funcs sdma_v7_0_fence_irq_funcs = {
+	.process = sdma_v7_0_process_fence_irq,
+};
+
 static const struct amdgpu_irq_src_funcs sdma_v7_0_illegal_inst_irq_funcs = {
 	.process = sdma_v7_0_process_illegal_inst_irq,
 };
@@ -1712,6 +1734,7 @@ static void sdma_v7_0_set_irq_funcs(struct amdgpu_device *adev)
 	adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_INSTANCE0 +
 					adev->sdma.num_instances;
 	adev->sdma.trap_irq.funcs = &sdma_v7_0_trap_irq_funcs;
+	adev->sdma.fence_irq.funcs = &sdma_v7_0_fence_irq_funcs;
 	adev->sdma.illegal_inst_irq.funcs = &sdma_v7_0_illegal_inst_irq_funcs;
 }
 

drivers/gpu/drm/amd/amdkfd/Kconfig

@@ -5,7 +5,7 @@
 
 config HSA_AMD
 	bool "HSA kernel driver for AMD GPU devices"
-	depends on DRM_AMDGPU && (X86_64 || ARM64 || PPC64)
+	depends on DRM_AMDGPU && (X86_64 || ARM64 || PPC64 || (RISCV && 64BIT))
 	select HMM_MIRROR
 	select MMU_NOTIFIER
 	select DRM_AMDGPU_USERPTR

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c

@@ -4910,6 +4910,7 @@ amdgpu_dm_register_backlight_device(struct amdgpu_dm_connector *aconnector)
 	struct backlight_properties props = { 0 };
 	struct amdgpu_dm_backlight_caps caps = { 0 };
 	char bl_name[16];
+	int min, max;
 
 	if (aconnector->bl_idx == -1)
 		return;
@@ -4922,11 +4923,15 @@ amdgpu_dm_register_backlight_device(struct amdgpu_dm_connector *aconnector)
 	}
 
 	amdgpu_acpi_get_backlight_caps(&caps);
-	if (caps.caps_valid) {
+	if (caps.caps_valid && get_brightness_range(&caps, &min, &max)) {
 		if (power_supply_is_system_supplied() > 0)
-			props.brightness = caps.ac_level;
+			props.brightness = (max - min) * DIV_ROUND_CLOSEST(caps.ac_level, 100);
 		else
-			props.brightness = caps.dc_level;
+			props.brightness = (max - min) * DIV_ROUND_CLOSEST(caps.dc_level, 100);
+		/* min is zero, so max needs to be adjusted */
+		props.max_brightness = max - min;
+		drm_dbg(drm, "Backlight caps: min: %d, max: %d, ac %d, dc %d\n", min, max,
+			caps.ac_level, caps.dc_level);
 	} else
 		props.brightness = AMDGPU_MAX_BL_LEVEL;