Merge tag 'drm-etnaviv-next-2024-06-28' of https://git.pengutronix.de…

…/git/lst/linux into drm-next - fix i.MX8MP NPU clock gating - workaround FE register cdc issues on some cores - fix DMA sync handling for cached buffers - fix job timeout handling - keep TS enabled on MMUv2 cores for improved performance Signed-off-by: Daniel Vetter <[email protected]> From: Lucas Stach <[email protected]> Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]
antonovi · Jul 5, 2024 · 6cab3e2 · 6cab3e2
2 parents 3ccf1b8 + 704d3d6
commit 6cab3e2
Show file tree

Hide file tree

Showing 10 changed files with 238 additions and 92 deletions.
diff --git a/drivers/gpu/drm/etnaviv/cmdstream.xml.h b/drivers/gpu/drm/etnaviv/cmdstream.xml.h
@@ -8,11 +8,11 @@ This file was generated by the rules-ng-ng headergen tool in this git repository
 git clone git://0x04.net/rules-ng-ng
 
 The rules-ng-ng source files this header was generated from are:
-- cmdstream.xml (  14094 bytes, from 2016-11-11 06:55:14)
-- copyright.xml (   1597 bytes, from 2016-10-29 07:29:22)
-- common.xml    (  23344 bytes, from 2016-11-10 15:14:07)
+- cmdstream.xml (  16933 bytes, from 2023-12-11 15:50:17)
+- copyright.xml (   1597 bytes, from 2016-11-10 13:58:32)
+- common.xml    (  35664 bytes, from 2023-12-06 10:55:32)
 
-Copyright (C) 2012-2016 by the following authors:
+Copyright (C) 2012-2023 by the following authors:
 - Wladimir J. van der Laan <[email protected]>
 - Christian Gmeiner <[email protected]>
 - Lucas Stach <[email protected]>
@@ -52,6 +52,9 @@ DEALINGS IN THE SOFTWARE.
 #define FE_OPCODE_RETURN					0x0000000b
 #define FE_OPCODE_DRAW_INSTANCED				0x0000000c
 #define FE_OPCODE_CHIP_SELECT					0x0000000d
+#define FE_OPCODE_WAIT_FENCE					0x0000000f
+#define FE_OPCODE_DRAW_INDIRECT					0x00000010
+#define FE_OPCODE_SNAP_PAGES					0x00000013
 #define PRIMITIVE_TYPE_POINTS					0x00000001
 #define PRIMITIVE_TYPE_LINES					0x00000002
 #define PRIMITIVE_TYPE_LINE_STRIP				0x00000003
@@ -192,6 +195,9 @@ DEALINGS IN THE SOFTWARE.
 #define VIV_FE_STALL_TOKEN_TO__MASK				0x00001f00
 #define VIV_FE_STALL_TOKEN_TO__SHIFT				8
 #define VIV_FE_STALL_TOKEN_TO(x)				(((x) << VIV_FE_STALL_TOKEN_TO__SHIFT) & VIV_FE_STALL_TOKEN_TO__MASK)
+#define VIV_FE_STALL_TOKEN_UNK28__MASK				0x30000000
+#define VIV_FE_STALL_TOKEN_UNK28__SHIFT				28
+#define VIV_FE_STALL_TOKEN_UNK28(x)				(((x) << VIV_FE_STALL_TOKEN_UNK28__SHIFT) & VIV_FE_STALL_TOKEN_UNK28__MASK)
 
 #define VIV_FE_CALL						0x00000000
 
@@ -266,5 +272,43 @@ DEALINGS IN THE SOFTWARE.
 #define VIV_FE_DRAW_INSTANCED_START_INDEX__SHIFT		0
 #define VIV_FE_DRAW_INSTANCED_START_INDEX(x)			(((x) << VIV_FE_DRAW_INSTANCED_START_INDEX__SHIFT) & VIV_FE_DRAW_INSTANCED_START_INDEX__MASK)
 
+#define VIV_FE_WAIT_FENCE					0x00000000
+
+#define VIV_FE_WAIT_FENCE_HEADER				0x00000000
+#define VIV_FE_WAIT_FENCE_HEADER_OP__MASK			0xf8000000
+#define VIV_FE_WAIT_FENCE_HEADER_OP__SHIFT			27
+#define VIV_FE_WAIT_FENCE_HEADER_OP_WAIT_FENCE			0x78000000
+#define VIV_FE_WAIT_FENCE_HEADER_UNK16__MASK			0x00030000
+#define VIV_FE_WAIT_FENCE_HEADER_UNK16__SHIFT			16
+#define VIV_FE_WAIT_FENCE_HEADER_UNK16(x)			(((x) << VIV_FE_WAIT_FENCE_HEADER_UNK16__SHIFT) & VIV_FE_WAIT_FENCE_HEADER_UNK16__MASK)
+#define VIV_FE_WAIT_FENCE_HEADER_WAITCOUNT__MASK		0x0000ffff
+#define VIV_FE_WAIT_FENCE_HEADER_WAITCOUNT__SHIFT		0
+#define VIV_FE_WAIT_FENCE_HEADER_WAITCOUNT(x)			(((x) << VIV_FE_WAIT_FENCE_HEADER_WAITCOUNT__SHIFT) & VIV_FE_WAIT_FENCE_HEADER_WAITCOUNT__MASK)
+
+#define VIV_FE_WAIT_FENCE_ADDRESS				0x00000004
+
+#define VIV_FE_DRAW_INDIRECT					0x00000000
+
+#define VIV_FE_DRAW_INDIRECT_HEADER				0x00000000
+#define VIV_FE_DRAW_INDIRECT_HEADER_OP__MASK			0xf8000000
+#define VIV_FE_DRAW_INDIRECT_HEADER_OP__SHIFT			27
+#define VIV_FE_DRAW_INDIRECT_HEADER_OP_DRAW_INDIRECT		0x80000000
+#define VIV_FE_DRAW_INDIRECT_HEADER_INDEXED			0x00000100
+#define VIV_FE_DRAW_INDIRECT_HEADER_TYPE__MASK			0x0000000f
+#define VIV_FE_DRAW_INDIRECT_HEADER_TYPE__SHIFT			0
+#define VIV_FE_DRAW_INDIRECT_HEADER_TYPE(x)			(((x) << VIV_FE_DRAW_INDIRECT_HEADER_TYPE__SHIFT) & VIV_FE_DRAW_INDIRECT_HEADER_TYPE__MASK)
+
+#define VIV_FE_DRAW_INDIRECT_ADDRESS				0x00000004
+
+#define VIV_FE_SNAP_PAGES					0x00000000
+
+#define VIV_FE_SNAP_PAGES_HEADER				0x00000000
+#define VIV_FE_SNAP_PAGES_HEADER_OP__MASK			0xf8000000
+#define VIV_FE_SNAP_PAGES_HEADER_OP__SHIFT			27
+#define VIV_FE_SNAP_PAGES_HEADER_OP_SNAP_PAGES			0x98000000
+#define VIV_FE_SNAP_PAGES_HEADER_UNK0__MASK			0x0000001f
+#define VIV_FE_SNAP_PAGES_HEADER_UNK0__SHIFT			0
+#define VIV_FE_SNAP_PAGES_HEADER_UNK0(x)			(((x) << VIV_FE_SNAP_PAGES_HEADER_UNK0__SHIFT) & VIV_FE_SNAP_PAGES_HEADER_UNK0__MASK)
+
 
 #endif /* CMDSTREAM_XML */
diff --git a/drivers/gpu/drm/etnaviv/common.xml.h b/drivers/gpu/drm/etnaviv/common.xml.h
@@ -8,12 +8,12 @@ This file was generated by the rules-ng-ng headergen tool in this git repository
 git clone git://0x04.net/rules-ng-ng
 
 The rules-ng-ng source files this header was generated from are:
-- texdesc_3d.xml (   3183 bytes, from 2017-12-18 16:51:59)
-- copyright.xml  (   1597 bytes, from 2016-12-08 16:37:56)
-- common.xml     (  35468 bytes, from 2018-01-22 13:48:54)
-- common_3d.xml  (  14615 bytes, from 2017-12-18 16:51:59)
+- texdesc_3d.xml (   3183 bytes, from 2022-11-18 09:38:25)
+- copyright.xml  (   1597 bytes, from 2016-11-10 13:58:32)
+- common.xml     (  35664 bytes, from 2023-12-06 10:55:32)
+- common_3d.xml  (  15069 bytes, from 2023-11-22 10:05:24)
 
-Copyright (C) 2012-2018 by the following authors:
+Copyright (C) 2012-2023 by the following authors:
 - Wladimir J. van der Laan <[email protected]>
 - Christian Gmeiner <[email protected]>
 - Lucas Stach <[email protected]>
@@ -65,6 +65,7 @@ DEALINGS IN THE SOFTWARE.
 #define chipModel_GC520						0x00000520
 #define chipModel_GC530						0x00000530
 #define chipModel_GC600						0x00000600
+#define chipModel_GC620						0x00000620
 #define chipModel_GC700						0x00000700
 #define chipModel_GC800						0x00000800
 #define chipModel_GC860						0x00000860
@@ -481,5 +482,6 @@ DEALINGS IN THE SOFTWARE.
 #define chipMinorFeatures11_NN_INTERLEVE8			0x00000008
 #define chipMinorFeatures11_TP_REORDER				0x00000010
 #define chipMinorFeatures11_PE_DEPTH_ONLY_OQFIX			0x00000020
+#define chipMinorFeatures12_G2D_DEC400EX			0x00000020
 
 #endif /* COMMON_XML */
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_dump.c b/drivers/gpu/drm/etnaviv/etnaviv_dump.c
@@ -159,8 +159,7 @@ void etnaviv_core_dump(struct etnaviv_gem_submit *submit)
 	file_size += sizeof(*iter.hdr) * n_obj;
 
 	/* Allocate the file in vmalloc memory, it's likely to be big */
-	iter.start = __vmalloc(file_size, GFP_KERNEL | __GFP_NOWARN |
-			__GFP_NORETRY);
+	iter.start = __vmalloc(file_size, GFP_NOWAIT);
 	if (!iter.start) {
 		mutex_unlock(&submit->mmu_context->lock);
 		dev_warn(gpu->dev, "failed to allocate devcoredump file\n");
@@ -230,5 +229,5 @@ void etnaviv_core_dump(struct etnaviv_gem_submit *submit)
 
 	etnaviv_core_dump_header(&iter, ETDUMP_BUF_END, iter.data);
 
-	dev_coredumpv(gpu->dev, iter.start, iter.data - iter.start, GFP_KERNEL);
+	dev_coredumpv(gpu->dev, iter.start, iter.data - iter.start, GFP_NOWAIT);
 }
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gem.c b/drivers/gpu/drm/etnaviv/etnaviv_gem.c
@@ -355,9 +355,11 @@ static void *etnaviv_gem_vmap_impl(struct etnaviv_gem_object *obj)
 
 static inline enum dma_data_direction etnaviv_op_to_dma_dir(u32 op)
 {
-	if (op & ETNA_PREP_READ)
+	op &= ETNA_PREP_READ | ETNA_PREP_WRITE;
+
+	if (op == ETNA_PREP_READ)
 		return DMA_FROM_DEVICE;
-	else if (op & ETNA_PREP_WRITE)
+	else if (op == ETNA_PREP_WRITE)
 		return DMA_TO_DEVICE;
 	else
 		return DMA_BIDIRECTIONAL;

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
@@ -172,10 +172,12 @@ int etnaviv_gpu_get_param(struct etnaviv_gpu *gpu, u32 param, u64 *value)
 	return 0;
 }
 
+static inline bool etnaviv_is_model_rev(struct etnaviv_gpu *gpu, u32 model, u32 revision)
+{
+	return gpu->identity.model == model &&
+	       gpu->identity.revision == revision;
+}
 
-#define etnaviv_is_model_rev(gpu, mod, rev) \
-	((gpu)->identity.model == chipModel_##mod && \
-	 (gpu)->identity.revision == rev)
 #define etnaviv_field(val, field) \
 	(((val) & field##__MASK) >> field##__SHIFT)
 
@@ -281,7 +283,7 @@ static void etnaviv_hw_specs(struct etnaviv_gpu *gpu)
 
 	switch (gpu->identity.instruction_count) {
 	case 0:
-		if (etnaviv_is_model_rev(gpu, GC2000, 0x5108) ||
+		if (etnaviv_is_model_rev(gpu, 0x2000, 0x5108) ||
 		    gpu->identity.model == chipModel_GC880)
 			gpu->identity.instruction_count = 512;
 		else
@@ -315,17 +317,17 @@ static void etnaviv_hw_specs(struct etnaviv_gpu *gpu)
 	 * For some cores, two varyings are consumed for position, so the
 	 * maximum varying count needs to be reduced by one.
 	 */
-	if (etnaviv_is_model_rev(gpu, GC5000, 0x5434) ||
-	    etnaviv_is_model_rev(gpu, GC4000, 0x5222) ||
-	    etnaviv_is_model_rev(gpu, GC4000, 0x5245) ||
-	    etnaviv_is_model_rev(gpu, GC4000, 0x5208) ||
-	    etnaviv_is_model_rev(gpu, GC3000, 0x5435) ||
-	    etnaviv_is_model_rev(gpu, GC2200, 0x5244) ||
-	    etnaviv_is_model_rev(gpu, GC2100, 0x5108) ||
-	    etnaviv_is_model_rev(gpu, GC2000, 0x5108) ||
-	    etnaviv_is_model_rev(gpu, GC1500, 0x5246) ||
-	    etnaviv_is_model_rev(gpu, GC880, 0x5107) ||
-	    etnaviv_is_model_rev(gpu, GC880, 0x5106))
+	if (etnaviv_is_model_rev(gpu, 0x5000, 0x5434) ||
+	    etnaviv_is_model_rev(gpu, 0x4000, 0x5222) ||
+	    etnaviv_is_model_rev(gpu, 0x4000, 0x5245) ||
+	    etnaviv_is_model_rev(gpu, 0x4000, 0x5208) ||
+	    etnaviv_is_model_rev(gpu, 0x3000, 0x5435) ||
+	    etnaviv_is_model_rev(gpu, 0x2200, 0x5244) ||
+	    etnaviv_is_model_rev(gpu, 0x2100, 0x5108) ||
+	    etnaviv_is_model_rev(gpu, 0x2000, 0x5108) ||
+	    etnaviv_is_model_rev(gpu, 0x1500, 0x5246) ||
+	    etnaviv_is_model_rev(gpu, 0x880, 0x5107) ||
+	    etnaviv_is_model_rev(gpu, 0x880, 0x5106))
 		gpu->identity.varyings_count -= 1;
 }
 
@@ -351,7 +353,7 @@ static void etnaviv_hw_identify(struct etnaviv_gpu *gpu)
 		 * Reading these two registers on GC600 rev 0x19 result in a
 		 * unhandled fault: external abort on non-linefetch
 		 */
-		if (!etnaviv_is_model_rev(gpu, GC600, 0x19)) {
+		if (!etnaviv_is_model_rev(gpu, 0x600, 0x19)) {
 			gpu->identity.product_id = gpu_read(gpu, VIVS_HI_CHIP_PRODUCT_ID);
 			gpu->identity.eco_id = gpu_read(gpu, VIVS_HI_CHIP_ECO_ID);
 		}
@@ -368,7 +370,7 @@ static void etnaviv_hw_identify(struct etnaviv_gpu *gpu)
 		}
 
 		/* Another special case */
-		if (etnaviv_is_model_rev(gpu, GC300, 0x2201)) {
+		if (etnaviv_is_model_rev(gpu, 0x300, 0x2201)) {
 			u32 chipTime = gpu_read(gpu, VIVS_HI_CHIP_TIME);
 
 			if (chipDate == 0x20080814 && chipTime == 0x12051100) {
@@ -387,15 +389,15 @@ static void etnaviv_hw_identify(struct etnaviv_gpu *gpu)
 		 * Fix model/rev here, so all other places can refer to this
 		 * core by its real identity.
 		 */
-		if (etnaviv_is_model_rev(gpu, GC2000, 0xffff5450)) {
+		if (etnaviv_is_model_rev(gpu, 0x2000, 0xffff5450)) {
 			gpu->identity.model = chipModel_GC3000;
 			gpu->identity.revision &= 0xffff;
 		}
 
-		if (etnaviv_is_model_rev(gpu, GC1000, 0x5037) && (chipDate == 0x20120617))
+		if (etnaviv_is_model_rev(gpu, 0x1000, 0x5037) && (chipDate == 0x20120617))
 			gpu->identity.eco_id = 1;
 
-		if (etnaviv_is_model_rev(gpu, GC320, 0x5303) && (chipDate == 0x20140511))
+		if (etnaviv_is_model_rev(gpu, 0x320, 0x5303) && (chipDate == 0x20140511))
 			gpu->identity.eco_id = 1;
 	}
 
@@ -641,17 +643,23 @@ static void etnaviv_gpu_enable_mlcg(struct etnaviv_gpu *gpu)
 		pmc |= BIT(15); /* Unknown bit */
 
 	/* Disable TX clock gating on affected core revisions. */
-	if (etnaviv_is_model_rev(gpu, GC4000, 0x5222) ||
-	    etnaviv_is_model_rev(gpu, GC2000, 0x5108) ||
-	    etnaviv_is_model_rev(gpu, GC7000, 0x6202) ||
-	    etnaviv_is_model_rev(gpu, GC7000, 0x6203))
+	if (etnaviv_is_model_rev(gpu, 0x4000, 0x5222) ||
+	    etnaviv_is_model_rev(gpu, 0x2000, 0x5108) ||
+	    etnaviv_is_model_rev(gpu, 0x7000, 0x6202) ||
+	    etnaviv_is_model_rev(gpu, 0x7000, 0x6203))
 		pmc |= VIVS_PM_MODULE_CONTROLS_DISABLE_MODULE_CLOCK_GATING_TX;
 
 	/* Disable SE and RA clock gating on affected core revisions. */
-	if (etnaviv_is_model_rev(gpu, GC7000, 0x6202))
+	if (etnaviv_is_model_rev(gpu, 0x7000, 0x6202))
 		pmc |= VIVS_PM_MODULE_CONTROLS_DISABLE_MODULE_CLOCK_GATING_SE |
 		       VIVS_PM_MODULE_CONTROLS_DISABLE_MODULE_CLOCK_GATING_RA;
 
+	/* Disable SH_EU clock gating on affected core revisions. */
+	if (etnaviv_is_model_rev(gpu, 0x8000, 0x7200) ||
+	    etnaviv_is_model_rev(gpu, 0x8000, 0x8002) ||
+	    etnaviv_is_model_rev(gpu, 0x9200, 0x6304))
+		pmc |= VIVS_PM_MODULE_CONTROLS_DISABLE_MODULE_CLOCK_GATING_SH_EU;
+
 	pmc |= VIVS_PM_MODULE_CONTROLS_DISABLE_MODULE_CLOCK_GATING_RA_HZ;
 	pmc |= VIVS_PM_MODULE_CONTROLS_DISABLE_MODULE_CLOCK_GATING_RA_EZ;
 
@@ -701,14 +709,14 @@ static void etnaviv_gpu_setup_pulse_eater(struct etnaviv_gpu *gpu)
 	 */
 	u32 pulse_eater = 0x01590880;
 
-	if (etnaviv_is_model_rev(gpu, GC4000, 0x5208) ||
-	    etnaviv_is_model_rev(gpu, GC4000, 0x5222)) {
+	if (etnaviv_is_model_rev(gpu, 0x4000, 0x5208) ||
+	    etnaviv_is_model_rev(gpu, 0x4000, 0x5222)) {
 		pulse_eater |= BIT(23);
 
 	}
 
-	if (etnaviv_is_model_rev(gpu, GC1000, 0x5039) ||
-	    etnaviv_is_model_rev(gpu, GC1000, 0x5040)) {
+	if (etnaviv_is_model_rev(gpu, 0x1000, 0x5039) ||
+	    etnaviv_is_model_rev(gpu, 0x1000, 0x5040)) {
 		pulse_eater &= ~BIT(16);
 		pulse_eater |= BIT(17);
 	}
@@ -729,8 +737,8 @@ static void etnaviv_gpu_hw_init(struct etnaviv_gpu *gpu)
 	WARN_ON(!(gpu->state == ETNA_GPU_STATE_IDENTIFIED ||
 		  gpu->state == ETNA_GPU_STATE_RESET));
 
-	if ((etnaviv_is_model_rev(gpu, GC320, 0x5007) ||
-	     etnaviv_is_model_rev(gpu, GC320, 0x5220)) &&
+	if ((etnaviv_is_model_rev(gpu, 0x320, 0x5007) ||
+	     etnaviv_is_model_rev(gpu, 0x320, 0x5220)) &&
 	    gpu_read(gpu, VIVS_HI_CHIP_TIME) != 0x2062400) {
 		u32 mc_memory_debug;
 
@@ -756,7 +764,7 @@ static void etnaviv_gpu_hw_init(struct etnaviv_gpu *gpu)
 		  VIVS_HI_AXI_CONFIG_ARCACHE(2));
 
 	/* GC2000 rev 5108 needs a special bus config */
-	if (etnaviv_is_model_rev(gpu, GC2000, 0x5108)) {
+	if (etnaviv_is_model_rev(gpu, 0x2000, 0x5108)) {
 		u32 bus_config = gpu_read(gpu, VIVS_MC_BUS_CONFIG);
 		bus_config &= ~(VIVS_MC_BUS_CONFIG_FE_BUS_CONFIG__MASK |
 				VIVS_MC_BUS_CONFIG_TX_BUS_CONFIG__MASK);
@@ -855,12 +863,15 @@ int etnaviv_gpu_init(struct etnaviv_gpu *gpu)
 	 *
 	 * On MC1.0 cores the linear window offset is ignored by the TS engine,
 	 * leading to inconsistent memory views. Avoid using the offset on those
-	 * cores if possible, otherwise disable the TS feature.
+	 * cores if possible, otherwise disable the TS feature. MMUv2 doesn't
+	 * expose this issue, as all TS accesses are MMU translated, so the
+	 * linear window offset won't be used.
 	 */
 	cmdbuf_paddr = ALIGN_DOWN(etnaviv_cmdbuf_get_pa(&gpu->buffer), SZ_128M);
 
 	if (!(gpu->identity.features & chipFeatures_PIPE_3D) ||
-	    (gpu->identity.minor_features0 & chipMinorFeatures0_MC20)) {
+	    (gpu->identity.minor_features0 & chipMinorFeatures0_MC20) ||
+	    (gpu->identity.minor_features1 & chipMinorFeatures1_MMU_VERSION)) {
 		if (cmdbuf_paddr >= SZ_2G)
 			priv->mmu_global->memory_base = SZ_2G;
 		else
@@ -1537,6 +1548,7 @@ static irqreturn_t irq_handler(int irq, void *data)
 	u32 intr = gpu_read(gpu, VIVS_HI_INTR_ACKNOWLEDGE);
 
 	if (intr != 0) {
+		ktime_t now = ktime_get();
 		int event;
 
 		pm_runtime_mark_last_busy(gpu->dev);
@@ -1586,7 +1598,7 @@ static irqreturn_t irq_handler(int irq, void *data)
 			 */
 			if (fence_after(fence->seqno, gpu->completed_fence))
 				gpu->completed_fence = fence->seqno;
-			dma_fence_signal(fence);
+			dma_fence_signal_timestamp(fence, now);
 
 			event_free(gpu, event);
 		}
@@ -1975,7 +1987,6 @@ static const struct dev_pm_ops etnaviv_gpu_pm_ops = {
 struct platform_driver etnaviv_gpu_driver = {
 	.driver = {
 		.name = "etnaviv-gpu",
-		.owner = THIS_MODULE,
 		.pm = pm_ptr(&etnaviv_gpu_pm_ops),
 		.of_match_table = etnaviv_gpu_match,
 	},

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.h b/drivers/gpu/drm/etnaviv/etnaviv_gpu.h
@@ -11,6 +11,7 @@
 #include "etnaviv_mmu.h"
 #include "etnaviv_drv.h"
 #include "common.xml.h"
+#include "state.xml.h"
 
 struct etnaviv_gem_submit;
 struct etnaviv_vram_mapping;
@@ -170,6 +171,13 @@ static inline void gpu_write(struct etnaviv_gpu *gpu, u32 reg, u32 data)
 
 static inline u32 gpu_read(struct etnaviv_gpu *gpu, u32 reg)
 {
+	/* On some variants, such as the GC7000r6009, some FE registers
+	 * need two reads to be consistent. Do that extra read here and
+	 * throw away the result.
+	 */
+	if (reg >= VIVS_FE_DMA_STATUS && reg <= VIVS_FE_AUTO_FLUSH)
+		readl(gpu->mmio + reg);
+
 	return readl(gpu->mmio + reg);
 }
 

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_sched.c b/drivers/gpu/drm/etnaviv/etnaviv_sched.c
@@ -38,9 +38,6 @@ static enum drm_gpu_sched_stat etnaviv_sched_timedout_job(struct drm_sched_job
 	u32 dma_addr;
 	int change;
 
-	/* block scheduler */
-	drm_sched_stop(&gpu->sched, sched_job);
-
 	/*
 	 * If the GPU managed to complete this jobs fence, the timout is
 	 * spurious. Bail out.
@@ -63,6 +60,9 @@ static enum drm_gpu_sched_stat etnaviv_sched_timedout_job(struct drm_sched_job
 		goto out_no_timeout;
 	}
 
+	/* block scheduler */
+	drm_sched_stop(&gpu->sched, sched_job);
+
 	if(sched_job)
 		drm_sched_increase_karma(sched_job);
 
@@ -76,8 +76,7 @@ static enum drm_gpu_sched_stat etnaviv_sched_timedout_job(struct drm_sched_job
 	return DRM_GPU_SCHED_STAT_NOMINAL;
 
 out_no_timeout:
-	/* restart scheduler after GPU is usable again */
-	drm_sched_start(&gpu->sched, true);
+	list_add(&sched_job->list, &sched_job->sched->pending_list);
 	return DRM_GPU_SCHED_STAT_NOMINAL;
 }