1 files changed, 248 insertions, 82 deletions

diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
index 4e6dc16e4a4c..5903cd891b49 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
@@ -224,14 +224,19 @@ unsigned long a6xx_gmu_get_freq(struct msm_gpu *gpu)
 
 static bool a6xx_gmu_check_idle_level(struct a6xx_gmu *gmu)
 {
-	u32 val;
+	struct a6xx_gpu *a6xx_gpu = container_of(gmu, struct a6xx_gpu, gmu);
+	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
 	int local = gmu->idle_level;
+	u32 val;
 
 	/* SPTP and IFPC both report as IFPC */
 	if (gmu->idle_level == GMU_IDLE_STATE_SPTP)
 		local = GMU_IDLE_STATE_IFPC;
 
-	val = gmu_read(gmu, REG_A6XX_GPU_GMU_CX_GMU_RPMH_POWER_STATE);
+	if (adreno_is_a8xx(adreno_gpu))
+		val = gmu_read(gmu, REG_A8XX_GPU_GMU_CX_GMU_RPMH_POWER_STATE);
+	else
+		val = gmu_read(gmu, REG_A6XX_GPU_GMU_CX_GMU_RPMH_POWER_STATE);
 
 	if (val == local) {
 		if (gmu->idle_level != GMU_IDLE_STATE_IFPC ||
@@ -269,7 +274,9 @@ static int a6xx_gmu_start(struct a6xx_gmu *gmu)
 	/* Set the log wptr index
 	 * note: downstream saves the value in poweroff and restores it here
 	 */
-	if (adreno_is_a7xx(adreno_gpu))
+	if (adreno_is_a8xx(adreno_gpu))
+		gmu_write(gmu, REG_A8XX_GMU_GENERAL_9, 0);
+	else if (adreno_is_a7xx(adreno_gpu))
 		gmu_write(gmu, REG_A7XX_GMU_GENERAL_9, 0);
 	else
 		gmu_write(gmu, REG_A6XX_GPU_GMU_CX_GMU_PWR_COL_CP_RESP, 0);
@@ -350,12 +357,18 @@ static const struct a6xx_gmu_oob_bits a6xx_gmu_oob_bits[] = {
 /* Trigger a OOB (out of band) request to the GMU */
 int a6xx_gmu_set_oob(struct a6xx_gmu *gmu, enum a6xx_gmu_oob_state state)
 {
+	struct a6xx_gpu *a6xx_gpu = container_of(gmu, struct a6xx_gpu, gmu);
+	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
 	int ret;
 	u32 val;
 	int request, ack;
 
 	WARN_ON_ONCE(!mutex_is_locked(&gmu->lock));
 
+	/* Skip OOB calls since RGMU is not enabled */
+	if (adreno_has_rgmu(adreno_gpu))
+		return 0;
+
 	if (state >= ARRAY_SIZE(a6xx_gmu_oob_bits))
 		return -EINVAL;
 
@@ -376,9 +389,23 @@ int a6xx_gmu_set_oob(struct a6xx_gmu *gmu, enum a6xx_gmu_oob_state state)
 	/* Trigger the equested OOB operation */
 	gmu_write(gmu, REG_A6XX_GMU_HOST2GMU_INTR_SET, 1 << request);
 
-	/* Wait for the acknowledge interrupt */
-	ret = gmu_poll_timeout(gmu, REG_A6XX_GMU_GMU2HOST_INTR_INFO, val,
-		val & (1 << ack), 100, 10000);
+	do {
+		/* Wait for the acknowledge interrupt */
+		ret = gmu_poll_timeout(gmu, REG_A6XX_GMU_GMU2HOST_INTR_INFO, val,
+			val & (1 << ack), 100, 10000);
+
+		if (!ret)
+			break;
+
+		if (completion_done(&a6xx_gpu->base.fault_coredump_done))
+			break;
+
+		/* We may timeout because the GMU is temporarily wedged from
+		 * pending faults from the GPU and we are taking a devcoredump.
+		 * Wait until the MMU is resumed and try again.
+		 */
+		wait_for_completion(&a6xx_gpu->base.fault_coredump_done);
+	} while (true);
 
 	if (ret)
 		DRM_DEV_ERROR(gmu->dev,
@@ -395,10 +422,16 @@ int a6xx_gmu_set_oob(struct a6xx_gmu *gmu, enum a6xx_gmu_oob_state state)
 /* Clear a pending OOB state in the GMU */
 void a6xx_gmu_clear_oob(struct a6xx_gmu *gmu, enum a6xx_gmu_oob_state state)
 {
+	struct a6xx_gpu *a6xx_gpu = container_of(gmu, struct a6xx_gpu, gmu);
+	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
 	int bit;
 
 	WARN_ON_ONCE(!mutex_is_locked(&gmu->lock));
 
+	/* Skip OOB calls since RGMU is not enabled */
+	if (adreno_has_rgmu(adreno_gpu))
+		return;
+
 	if (state >= ARRAY_SIZE(a6xx_gmu_oob_bits))
 		return;
 
@@ -485,17 +518,25 @@ static void a6xx_gemnoc_workaround(struct a6xx_gmu *gmu)
 	 * in the power down sequence not being fully executed. That in turn can
 	 * prevent CX_GDSC from collapsing. Assert Qactive to avoid this.
 	 */
-	if (adreno_is_a621(adreno_gpu) || adreno_is_7c3(adreno_gpu))
-		gmu_write(gmu, REG_A6XX_GMU_AO_AHB_FENCE_CTRL, BIT(0));
+	if (adreno_is_a8xx(adreno_gpu))
+		gmu_write(gmu, REG_A8XX_GPU_GMU_CX_GMU_CX_FALNEXT_INTF, BIT(0));
+	else if (adreno_is_a7xx(adreno_gpu) || (adreno_is_a621(adreno_gpu) ||
+				adreno_is_7c3(adreno_gpu)))
+		gmu_write(gmu, REG_A6XX_GPU_GMU_CX_GMU_CX_FALNEXT_INTF, BIT(0));
 }
 
 /* Let the GMU know that we are about to go into slumber */
 static int a6xx_gmu_notify_slumber(struct a6xx_gmu *gmu)
 {
+	struct a6xx_gpu *a6xx_gpu = container_of(gmu, struct a6xx_gpu, gmu);
+	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
 	int ret;
 
 	/* Disable the power counter so the GMU isn't busy */
-	gmu_write(gmu, REG_A6XX_GMU_CX_GMU_POWER_COUNTER_ENABLE, 0);
+	if (adreno_is_a8xx(adreno_gpu))
+		gmu_write(gmu, REG_A8XX_GMU_CX_GMU_POWER_COUNTER_ENABLE, 0);
+	else
+		gmu_write(gmu, REG_A6XX_GMU_CX_GMU_POWER_COUNTER_ENABLE, 0);
 
 	/* Disable SPTP_PC if the CPU is responsible for it */
 	if (gmu->idle_level < GMU_IDLE_STATE_SPTP)
@@ -522,10 +563,9 @@ static int a6xx_gmu_notify_slumber(struct a6xx_gmu *gmu)
 	}
 
 out:
-	a6xx_gemnoc_workaround(gmu);
-
 	/* Put fence into allow mode */
 	gmu_write(gmu, REG_A6XX_GMU_AO_AHB_FENCE_CTRL, 0);
+	a6xx_gemnoc_workaround(gmu);
 	return ret;
 }
 
@@ -561,16 +601,22 @@ static int a6xx_rpmh_start(struct a6xx_gmu *gmu)
 
 static void a6xx_rpmh_stop(struct a6xx_gmu *gmu)
 {
+	struct a6xx_gpu *a6xx_gpu = container_of(gmu, struct a6xx_gpu, gmu);
+	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
+	u32 bitmask = BIT(16);
 	int ret;
 	u32 val;
 
 	if (test_and_clear_bit(GMU_STATUS_FW_START, &gmu->status))
 		return;
 
+	if (adreno_is_a840(adreno_gpu))
+		bitmask = BIT(30);
+
 	gmu_write(gmu, REG_A6XX_GMU_RSCC_CONTROL_REQ, 1);
 
 	ret = gmu_poll_timeout_rscc(gmu, REG_A6XX_GPU_RSCC_RSC_STATUS0_DRV0,
-		val, val & (1 << 16), 100, 10000);
+		val, val & bitmask, 100, 10000);
 	if (ret)
 		DRM_DEV_ERROR(gmu->dev, "Unable to power off the GPU RSC\n");
 
@@ -584,22 +630,24 @@ static inline void pdc_write(void __iomem *ptr, u32 offset, u32 value)
 	writel(value, ptr + (offset << 2));
 }
 
-static void __iomem *a6xx_gmu_get_mmio(struct platform_device *pdev,
-		const char *name);
-
 static void a6xx_gmu_rpmh_init(struct a6xx_gmu *gmu)
 {
 	struct a6xx_gpu *a6xx_gpu = container_of(gmu, struct a6xx_gpu, gmu);
 	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
 	struct platform_device *pdev = to_platform_device(gmu->dev);
-	void __iomem *pdcptr = a6xx_gmu_get_mmio(pdev, "gmu_pdc");
 	u32 seqmem0_drv0_reg = REG_A6XX_RSCC_SEQ_MEM_0_DRV0;
 	void __iomem *seqptr = NULL;
 	uint32_t pdc_address_offset;
+	void __iomem *pdcptr;
 	bool pdc_in_aop = false;
 
+	/* On A8x and above, RPMH/PDC configurations are entirely configured in AOP */
+	if (adreno_is_a8xx(adreno_gpu))
+		return;
+
+	pdcptr = devm_platform_ioremap_resource_byname(pdev, "gmu_pdc");
 	if (IS_ERR(pdcptr))
-		goto err;
+		return;
 
 	if (adreno_is_a650_family(adreno_gpu) ||
 	    adreno_is_a7xx(adreno_gpu))
@@ -612,9 +660,9 @@ static void a6xx_gmu_rpmh_init(struct a6xx_gmu *gmu)
 		pdc_address_offset = 0x30080;
 
 	if (!pdc_in_aop) {
-		seqptr = a6xx_gmu_get_mmio(pdev, "gmu_pdc_seq");
+		seqptr = devm_platform_ioremap_resource_byname(pdev, "gmu_pdc_seq");
 		if (IS_ERR(seqptr))
-			goto err;
+			return;
 	}
 
 	/* Disable SDE clock gating */
@@ -704,12 +752,6 @@ setup_pdc:
 
 	/* ensure no writes happen before the uCode is fully written */
 	wmb();
-
-err:
-	if (!IS_ERR_OR_NULL(pdcptr))
-		iounmap(pdcptr);
-	if (!IS_ERR_OR_NULL(seqptr))
-		iounmap(seqptr);
 }
 
 /*
@@ -732,7 +774,7 @@ static void a6xx_gmu_power_config(struct a6xx_gmu *gmu)
 	gmu_write(gmu, REG_A6XX_GMU_DCACHE_CONFIG, 0x1);
 
 	/* A7xx knows better by default! */
-	if (adreno_is_a7xx(adreno_gpu))
+	if (adreno_is_a7xx(adreno_gpu) || adreno_is_a8xx(adreno_gpu))
 		return;
 
 	gmu_write(gmu, REG_A6XX_GMU_PWR_COL_INTER_FRAME_CTRL, 0x9c40400);
@@ -795,7 +837,9 @@ static int a6xx_gmu_fw_load(struct a6xx_gmu *gmu)
 	u32 itcm_base = 0x00000000;
 	u32 dtcm_base = 0x00040000;
 
-	if (adreno_is_a650_family(adreno_gpu) || adreno_is_a7xx(adreno_gpu))
+	if (adreno_is_a650_family(adreno_gpu) ||
+	    adreno_is_a7xx(adreno_gpu) ||
+	    adreno_is_a8xx(adreno_gpu))
 		dtcm_base = 0x10004000;
 
 	if (gmu->legacy) {
@@ -850,7 +894,9 @@ static int a6xx_gmu_fw_start(struct a6xx_gmu *gmu, unsigned int state)
 {
 	struct a6xx_gpu *a6xx_gpu = container_of(gmu, struct a6xx_gpu, gmu);
 	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
+	struct msm_gpu *gpu = &adreno_gpu->base;
 	const struct a6xx_info *a6xx_info = adreno_gpu->info->a6xx;
+	const struct adreno_reglist *gbif_cx = a6xx_info->gbif_cx;
 	u32 fence_range_lower, fence_range_upper;
 	u32 chipid = 0;
 	int ret;
@@ -859,12 +905,15 @@ static int a6xx_gmu_fw_start(struct a6xx_gmu *gmu, unsigned int state)
 	if (adreno_is_a650_family(adreno_gpu) || adreno_is_a7xx(adreno_gpu)) {
 		gmu_write(gmu, REG_A6XX_GPU_GMU_CX_GMU_CX_FALNEXT_INTF, 1);
 		gmu_write(gmu, REG_A6XX_GPU_GMU_CX_GMU_CX_FAL_INTF, 1);
+	} else if (adreno_is_a8xx(adreno_gpu)) {
+		gmu_write(gmu, REG_A8XX_GPU_GMU_CX_GMU_CX_FALNEXT_INTF, 1);
+		gmu_write(gmu, REG_A8XX_GPU_GMU_CX_GMU_CX_FAL_INTF, 1);
 	}
 
 	/* Turn on TCM (Tightly Coupled Memory) retention */
 	if (adreno_is_a7xx(adreno_gpu))
 		a6xx_llc_write(a6xx_gpu, REG_A7XX_CX_MISC_TCM_RET_CNTL, 1);
-	else
+	else if (!adreno_is_a8xx(adreno_gpu))
 		gmu_write(gmu, REG_A6XX_GMU_GENERAL_7, 1);
 
 	ret = a6xx_rpmh_start(gmu);
@@ -889,7 +938,10 @@ static int a6xx_gmu_fw_start(struct a6xx_gmu *gmu, unsigned int state)
 	gmu_write(gmu, REG_A6XX_GMU_HFI_QTBL_ADDR, gmu->hfi.iova);
 	gmu_write(gmu, REG_A6XX_GMU_HFI_QTBL_INFO, 1);
 
-	if (adreno_is_a7xx(adreno_gpu)) {
+	if (adreno_is_a8xx(adreno_gpu)) {
+		fence_range_upper = 0x32;
+		fence_range_lower = 0x8c0;
+	} else if (adreno_is_a7xx(adreno_gpu)) {
 		fence_range_upper = 0x32;
 		fence_range_lower = 0x8a0;
 	} else {
@@ -923,7 +975,12 @@ static int a6xx_gmu_fw_start(struct a6xx_gmu *gmu, unsigned int state)
 		chipid |= (adreno_gpu->chip_id << 8) & 0x0f00; /* patchid */
 	}
 
-	if (adreno_is_a7xx(adreno_gpu)) {
+	if (adreno_is_a8xx(adreno_gpu)) {
+		gmu_write(gmu, REG_A8XX_GMU_GENERAL_10, chipid);
+		gmu_write(gmu, REG_A8XX_GMU_GENERAL_8,
+			  (gmu->log.iova & GENMASK(31, 12)) |
+			  ((gmu->log.size / SZ_4K - 1) & GENMASK(7, 0)));
+	} else if (adreno_is_a7xx(adreno_gpu)) {
 		gmu_write(gmu, REG_A7XX_GMU_GENERAL_10, chipid);
 		gmu_write(gmu, REG_A7XX_GMU_GENERAL_8,
 			  (gmu->log.iova & GENMASK(31, 12)) |
@@ -935,6 +992,15 @@ static int a6xx_gmu_fw_start(struct a6xx_gmu *gmu, unsigned int state)
 			  gmu->log.iova | (gmu->log.size / SZ_4K - 1));
 	}
 
+	/* For A7x and newer, do the CX GBIF configurations before GMU wake up */
+	for (int i = 0; (gbif_cx && gbif_cx[i].offset); i++)
+		gpu_write(gpu, gbif_cx[i].offset, gbif_cx[i].value);
+
+	if (adreno_is_a8xx(adreno_gpu)) {
+		gpu_write(gpu, REG_A8XX_GBIF_CX_CONFIG, 0x20023000);
+		gmu_write(gmu, REG_A6XX_GMU_MRC_GBIF_QOS_CTRL, 0x33);
+	}
+
 	/* Set up the lowest idle level on the GMU */
 	a6xx_gmu_power_config(gmu);
 
@@ -986,7 +1052,7 @@ static void a6xx_gmu_rpmh_off(struct a6xx_gmu *gmu)
 	u32 val, seqmem_off = 0;
 
 	/* The second spin of A7xx GPUs messed with some register offsets.. */
-	if (adreno_is_a740_family(adreno_gpu))
+	if (adreno_is_a740_family(adreno_gpu) || adreno_is_a8xx(adreno_gpu))
 		seqmem_off = 4;
 
 	/* Make sure there are no outstanding RPMh votes */
@@ -999,7 +1065,7 @@ static void a6xx_gmu_rpmh_off(struct a6xx_gmu *gmu)
 	gmu_poll_timeout_rscc(gmu, REG_A6XX_RSCC_TCS3_DRV0_STATUS + seqmem_off,
 		val, (val & 1), 100, 1000);
 
-	if (!adreno_is_a740_family(adreno_gpu))
+	if (!adreno_is_a740_family(adreno_gpu) && !adreno_is_a8xx(adreno_gpu))
 		return;
 
 	gmu_poll_timeout_rscc(gmu, REG_A7XX_RSCC_TCS4_DRV0_STATUS + seqmem_off,
@@ -1027,7 +1093,10 @@ static void a6xx_gmu_force_off(struct a6xx_gmu *gmu)
 	 * Turn off keep alive that might have been enabled by the hang
 	 * interrupt
 	 */
-	gmu_write(&a6xx_gpu->gmu, REG_A6XX_GMU_GMU_PWR_COL_KEEPALIVE, 0);
+	if (adreno_is_a8xx(adreno_gpu))
+		gmu_write(&a6xx_gpu->gmu, REG_A8XX_GMU_GMU_PWR_COL_KEEPALIVE, 0);
+	else
+		gmu_write(&a6xx_gpu->gmu, REG_A6XX_GMU_GMU_PWR_COL_KEEPALIVE, 0);
 
 	/* Flush all the queues */
 	a6xx_hfi_stop(gmu);
@@ -1053,7 +1122,7 @@ static void a6xx_gmu_force_off(struct a6xx_gmu *gmu)
 	/* Halt the gmu cm3 core */
 	gmu_write(gmu, REG_A6XX_GMU_CM3_SYSRESET, 1);
 
-	a6xx_bus_clear_pending_transactions(adreno_gpu, true);
+	adreno_gpu->funcs->bus_halt(adreno_gpu, true);
 
 	/* Reset GPU core blocks */
 	a6xx_gpu_sw_reset(gpu, true);
@@ -1122,6 +1191,9 @@ int a6xx_gmu_resume(struct a6xx_gpu *a6xx_gpu)
 		return ret;
 	}
 
+	/* Read the slice info on A8x GPUs */
+	a8xx_gpu_get_slice_info(gpu);
+
 	/* Set the bus quota to a reasonable value for boot */
 	a6xx_gmu_set_initial_bw(gpu, gmu);
 
@@ -1131,7 +1203,7 @@ int a6xx_gmu_resume(struct a6xx_gpu *a6xx_gpu)
 	enable_irq(gmu->gmu_irq);
 
 	/* Check to see if we are doing a cold or warm boot */
-	if (adreno_is_a7xx(adreno_gpu)) {
+	if (adreno_is_a7xx(adreno_gpu) || adreno_is_a8xx(adreno_gpu)) {
 		status = a6xx_llc_read(a6xx_gpu, REG_A7XX_CX_MISC_TCM_RET_CNTL) == 1 ?
 			GMU_WARM_BOOT : GMU_COLD_BOOT;
 	} else if (gmu->legacy) {
@@ -1225,7 +1297,7 @@ static void a6xx_gmu_shutdown(struct a6xx_gmu *gmu)
 	if (ret)
 		goto force_off;
 
-	a6xx_bus_clear_pending_transactions(adreno_gpu, a6xx_gpu->hung);
+	adreno_gpu->funcs->bus_halt(adreno_gpu, a6xx_gpu->hung);
 
 	/* tell the GMU we want to slumber */
 	ret = a6xx_gmu_notify_slumber(gmu);
@@ -1460,7 +1532,7 @@ static int a6xx_gmu_rpmh_bw_votes_init(struct adreno_gpu *adreno_gpu,
 			vote = clamp(peak, 1, BCM_TCS_CMD_VOTE_MASK);
 
 			/* GMUs on A7xx votes on both x & y */
-			if (adreno_is_a7xx(adreno_gpu))
+			if (adreno_is_a7xx(adreno_gpu) || adreno_is_a8xx(adreno_gpu))
 				data[bcm_index] = BCM_TCS_CMD(commit, true, vote, vote);
 			else
 				data[bcm_index] = BCM_TCS_CMD(commit, true, 0, vote);
@@ -1492,13 +1564,14 @@ static unsigned int a6xx_gmu_get_arc_level(struct device *dev,
 }
 
 static int a6xx_gmu_rpmh_arc_votes_init(struct device *dev, u32 *votes,
-		unsigned long *freqs, int freqs_count, const char *id)
+		unsigned long *freqs, int freqs_count,
+		const char *pri_id, const char *sec_id)
 {
 	int i, j;
 	const u16 *pri, *sec;
 	size_t pri_count, sec_count;
 
-	pri = cmd_db_read_aux_data(id, &pri_count);
+	pri = cmd_db_read_aux_data(pri_id, &pri_count);
 	if (IS_ERR(pri))
 		return PTR_ERR(pri);
 	/*
@@ -1509,13 +1582,7 @@ static int a6xx_gmu_rpmh_arc_votes_init(struct device *dev, u32 *votes,
 	if (!pri_count)
 		return -EINVAL;
 
-	/*
-	 * Some targets have a separate gfx mxc rail. So try to read that first and then fall back
-	 * to regular mx rail if it is missing
-	 */
-	sec = cmd_db_read_aux_data("gmxc.lvl", &sec_count);
-	if (IS_ERR(sec) && sec != ERR_PTR(-EPROBE_DEFER))
-		sec = cmd_db_read_aux_data("mx.lvl", &sec_count);
+	sec = cmd_db_read_aux_data(sec_id, &sec_count);
 	if (IS_ERR(sec))
 		return PTR_ERR(sec);
 
@@ -1569,6 +1636,57 @@ static int a6xx_gmu_rpmh_arc_votes_init(struct device *dev, u32 *votes,
 	return 0;
 }
 
+static int a6xx_gmu_rpmh_dep_votes_init(struct device *dev, u32 *votes,
+		unsigned long *freqs, int freqs_count)
+{
+	const u16 *mx;
+	size_t count;
+
+	mx = cmd_db_read_aux_data("mx.lvl", &count);
+	if (IS_ERR(mx))
+		return PTR_ERR(mx);
+	/*
+	 * The data comes back as an array of unsigned shorts so adjust the
+	 * count accordingly
+	 */
+	count >>= 1;
+	if (!count)
+		return -EINVAL;
+
+	/* Fix the vote for zero frequency */
+	votes[0] = 0xffffffff;
+
+	/* Construct a vote for rest of the corners */
+	for (int i = 1; i < freqs_count; i++) {
+		unsigned int level = a6xx_gmu_get_arc_level(dev, freqs[i]);
+		u8 j, index = 0;
+
+		/* Get the primary index that matches the arc level */
+		for (j = 0; j < count; j++) {
+			if (mx[j] >= level) {
+				index = j;
+				break;
+			}
+		}
+
+		if (j == count) {
+			DRM_DEV_ERROR(dev,
+				      "Mx Level %u not found in the RPMh list\n",
+				      level);
+			DRM_DEV_ERROR(dev, "Available levels:\n");
+			for (j = 0; j < count; j++)
+				DRM_DEV_ERROR(dev, "  %u\n", mx[j]);
+
+			return -EINVAL;
+		}
+
+		/* Construct the vote */
+		votes[i] = (0x3fff << 14) | (index << 8) | (0xff);
+	}
+
+	return 0;
+}
+
 /*
  * The GMU votes with the RPMh for itself and on behalf of the GPU but we need
  * to construct the list of votes on the CPU and send it over. Query the RPMh
@@ -1583,15 +1701,27 @@ static int a6xx_gmu_rpmh_votes_init(struct a6xx_gmu *gmu)
 	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
 	const struct a6xx_info *info = adreno_gpu->info->a6xx;
 	struct msm_gpu *gpu = &adreno_gpu->base;
+	const char *sec_id;
+	const u16 *gmxc;
 	int ret;
 
+	gmxc = cmd_db_read_aux_data("gmxc.lvl", NULL);
+	if (gmxc == ERR_PTR(-EPROBE_DEFER))
+		return -EPROBE_DEFER;
+
+	/* If GMxC is present, prefer that as secondary rail for GX votes */
+	sec_id = IS_ERR_OR_NULL(gmxc) ? "mx.lvl" : "gmxc.lvl";
+
 	/* Build the GX votes */
 	ret = a6xx_gmu_rpmh_arc_votes_init(&gpu->pdev->dev, gmu->gx_arc_votes,
-		gmu->gpu_freqs, gmu->nr_gpu_freqs, "gfx.lvl");
+		gmu->gpu_freqs, gmu->nr_gpu_freqs, "gfx.lvl", sec_id);
 
 	/* Build the CX votes */
 	ret |= a6xx_gmu_rpmh_arc_votes_init(gmu->dev, gmu->cx_arc_votes,
-		gmu->gmu_freqs, gmu->nr_gmu_freqs, "cx.lvl");
+		gmu->gmu_freqs, gmu->nr_gmu_freqs, "cx.lvl", "mx.lvl");
+
+	ret |= a6xx_gmu_rpmh_dep_votes_init(gmu->dev, gmu->dep_arc_votes,
+		gmu->gpu_freqs, gmu->nr_gpu_freqs);
 
 	/* Build the interconnect votes */
 	if (info->bcms && gmu->nr_gpu_bws > 1)
@@ -1795,27 +1925,6 @@ static int a6xx_gmu_clocks_probe(struct a6xx_gmu *gmu)
 	return 0;
 }
 
-static void __iomem *a6xx_gmu_get_mmio(struct platform_device *pdev,
-		const char *name)
-{
-	void __iomem *ret;
-	struct resource *res = platform_get_resource_byname(pdev,
-			IORESOURCE_MEM, name);
-
-	if (!res) {
-		DRM_DEV_ERROR(&pdev->dev, "Unable to find the %s registers\n", name);
-		return ERR_PTR(-EINVAL);
-	}
-
-	ret = ioremap(res->start, resource_size(res));
-	if (!ret) {
-		DRM_DEV_ERROR(&pdev->dev, "Unable to map the %s registers\n", name);
-		return ERR_PTR(-EINVAL);
-	}
-
-	return ret;
-}
-
 static int a6xx_gmu_get_irq(struct a6xx_gmu *gmu, struct platform_device *pdev,
 		const char *name, irq_handler_t handler)
 {
@@ -1866,7 +1975,6 @@ void a6xx_gmu_remove(struct a6xx_gpu *a6xx_gpu)
 {
 	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
-	struct platform_device *pdev = to_platform_device(gmu->dev);
 
 	mutex_lock(&gmu->lock);
 	if (!gmu->initialized) {
@@ -1895,12 +2003,11 @@ void a6xx_gmu_remove(struct a6xx_gpu *a6xx_gpu)
 		qmp_put(gmu->qmp);
 
 	iounmap(gmu->mmio);
-	if (platform_get_resource_byname(pdev, IORESOURCE_MEM, "rscc"))
-		iounmap(gmu->rscc);
 	gmu->mmio = NULL;
 	gmu->rscc = NULL;
 
-	if (!adreno_has_gmu_wrapper(adreno_gpu)) {
+	if (!adreno_has_gmu_wrapper(adreno_gpu) &&
+	    !adreno_has_rgmu(adreno_gpu)) {
 		a6xx_gmu_memory_free(gmu);
 
 		free_irq(gmu->gmu_irq, gmu);
@@ -1922,10 +2029,38 @@ static int cxpd_notifier_cb(struct notifier_block *nb,
 	return 0;
 }
 
+static void __iomem *a6xx_gmu_get_mmio(struct platform_device *pdev,
+		const char *name, resource_size_t *start)
+{
+	void __iomem *ret;
+	struct resource *res = platform_get_resource_byname(pdev,
+			IORESOURCE_MEM, name);
+
+	if (!res) {
+		DRM_DEV_ERROR(&pdev->dev, "Unable to find the %s registers\n", name);
+		return ERR_PTR(-EINVAL);
+	}
+
+	ret = ioremap(res->start, resource_size(res));
+	if (!ret) {
+		DRM_DEV_ERROR(&pdev->dev, "Unable to map the %s registers\n", name);
+		return ERR_PTR(-EINVAL);
+	}
+
+	if (start)
+		*start = res->start;
+
+	return ret;
+}
+
 int a6xx_gmu_wrapper_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
 {
 	struct platform_device *pdev = of_find_device_by_node(node);
+	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
+	struct msm_gpu *gpu = &adreno_gpu->base;
 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
+	resource_size_t start;
+	struct resource *res;
 	int ret;
 
 	if (!pdev)
@@ -1942,13 +2077,29 @@ int a6xx_gmu_wrapper_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
 	/* Mark legacy for manual SPTPRAC control */
 	gmu->legacy = true;
 
+	/* RGMU requires clocks */
+	ret = devm_clk_bulk_get_all(gmu->dev, &gmu->clocks);
+	if (ret < 0)
+		goto err_clk;
+
+	gmu->nr_clocks = ret;
+
 	/* Map the GMU registers */
-	gmu->mmio = a6xx_gmu_get_mmio(pdev, "gmu");
+	gmu->mmio = a6xx_gmu_get_mmio(pdev, "gmu", &start);
 	if (IS_ERR(gmu->mmio)) {
 		ret = PTR_ERR(gmu->mmio);
 		goto err_mmio;
 	}
 
+	res = platform_get_resource_byname(gpu->pdev, IORESOURCE_MEM, "kgsl_3d0_reg_memory");
+	if (!res) {
+		ret = -EINVAL;
+		goto err_mmio;
+	}
+
+	/* Identify gmu base offset from gpu base address */
+	gmu->mmio_offset = (u32)(start - res->start);
+
 	gmu->cxpd = dev_pm_domain_attach_by_name(gmu->dev, "cx");
 	if (IS_ERR(gmu->cxpd)) {
 		ret = PTR_ERR(gmu->cxpd);
@@ -1981,6 +2132,7 @@ detach_cxpd:
 err_mmio:
 	iounmap(gmu->mmio);
 
+err_clk:
 	/* Drop reference taken in of_find_device_by_node */
 	put_device(gmu->dev);
 
@@ -1989,10 +2141,13 @@ err_mmio:
 
 int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
 {
+	struct platform_device *pdev = of_find_device_by_node(node);
 	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
+	struct msm_gpu *gpu = &adreno_gpu->base;
 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
-	struct platform_device *pdev = of_find_device_by_node(node);
 	struct device_link *link;
+	resource_size_t start;
+	struct resource *res;
 	int ret;
 
 	if (!pdev)
@@ -2028,13 +2183,14 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
 	 */
 	gmu->dummy.size = SZ_4K;
 	if (adreno_is_a660_family(adreno_gpu) ||
-	    adreno_is_a7xx(adreno_gpu)) {
+	    adreno_is_a7xx(adreno_gpu) ||
+	    adreno_is_a8xx(adreno_gpu)) {
 		ret = a6xx_gmu_memory_alloc(gmu, &gmu->debug, SZ_4K * 7,
 					    0x60400000, "debug");
 		if (ret)
 			goto err_memory;
 
-		gmu->dummy.size = SZ_8K;
+		gmu->dummy.size = SZ_16K;
 	}
 
 	/* Allocate memory for the GMU dummy page */
@@ -2045,7 +2201,8 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
 
 	/* Note that a650 family also includes a660 family: */
 	if (adreno_is_a650_family(adreno_gpu) ||
-	    adreno_is_a7xx(adreno_gpu)) {
+	    adreno_is_a7xx(adreno_gpu) ||
+	    adreno_is_a8xx(adreno_gpu)) {
 		ret = a6xx_gmu_memory_alloc(gmu, &gmu->icache,
 			SZ_16M - SZ_16K, 0x04000, "icache");
 		if (ret)
@@ -2087,19 +2244,30 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
 		goto err_memory;
 
 	/* Map the GMU registers */
-	gmu->mmio = a6xx_gmu_get_mmio(pdev, "gmu");
+	gmu->mmio = a6xx_gmu_get_mmio(pdev, "gmu", &start);
 	if (IS_ERR(gmu->mmio)) {
 		ret = PTR_ERR(gmu->mmio);
 		goto err_memory;
 	}
 
+	res = platform_get_resource_byname(gpu->pdev, IORESOURCE_MEM, "kgsl_3d0_reg_memory");
+	if (!res) {
+		ret = -EINVAL;
+		goto err_mmio;
+	}
+
+	/* Identify gmu base offset from gpu base address */
+	gmu->mmio_offset = (u32)(start - res->start);
+
 	if (adreno_is_a650_family(adreno_gpu) ||
 	    adreno_is_a7xx(adreno_gpu)) {
-		gmu->rscc = a6xx_gmu_get_mmio(pdev, "rscc");
+		gmu->rscc = devm_platform_ioremap_resource_byname(pdev, "rscc");
 		if (IS_ERR(gmu->rscc)) {
 			ret = -ENODEV;
 			goto err_mmio;
 		}
+	} else if (adreno_is_a8xx(adreno_gpu)) {
+		gmu->rscc = gmu->mmio + 0x19000;
 	} else {
 		gmu->rscc = gmu->mmio + 0x23000;
 	}
@@ -2173,8 +2341,6 @@ detach_cxpd:
 
 err_mmio:
 	iounmap(gmu->mmio);
-	if (platform_get_resource_byname(pdev, IORESOURCE_MEM, "rscc"))
-		iounmap(gmu->rscc);
 	free_irq(gmu->gmu_irq, gmu);
 	free_irq(gmu->hfi_irq, gmu);