From 6eb3ecc33a6aaedda5ceb0824cafe34c47af2f55 Mon Sep 17 00:00:00 2001
From: Lucas Stach <l.stach@pengutronix.de>
Date: Thu, 28 Sep 2017 15:41:21 +0200
Subject: drm/etnaviv: rework clock initialization

The reset path wants to initialize the clock control register regardless
of the DYNAMIC_FREQUENCY_SCALING feature, so don't call clock update, but
explicitly load the register.

Also disabling of the debug registers is moved into the reset function,
so we always get to the same state after a GPU reset. This means the
clock update function should not touch the bits already set in the clock
control register, but instead only update the scaling bits.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
---
 drivers/gpu/drm/etnaviv/etnaviv_gpu.c | 15 ++++++++++-----
 1 file changed, 10 insertions(+), 5 deletions(-)

(limited to 'drivers/gpu/drm/etnaviv/etnaviv_gpu.c')

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
index fc9a6a83dfc7..de34e221c2fe 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
@@ -420,9 +420,10 @@ static void etnaviv_gpu_update_clock(struct etnaviv_gpu *gpu)
 			     gpu->base_rate_shader >> gpu->freq_scale);
 	} else {
 		unsigned int fscale = 1 << (6 - gpu->freq_scale);
-		u32 clock = VIVS_HI_CLOCK_CONTROL_DISABLE_DEBUG_REGISTERS |
-			    VIVS_HI_CLOCK_CONTROL_FSCALE_VAL(fscale);
+		u32 clock = gpu_read(gpu, VIVS_HI_CLOCK_CONTROL);
 
+		clock &= ~VIVS_HI_CLOCK_CONTROL_FSCALE_VAL__MASK;
+		clock |= VIVS_HI_CLOCK_CONTROL_FSCALE_VAL(fscale);
 		etnaviv_gpu_load_clock(gpu, clock);
 	}
 }
@@ -445,9 +446,9 @@ static int etnaviv_hw_reset(struct etnaviv_gpu *gpu)
 
 	while (time_is_after_jiffies(timeout)) {
 		/* enable clock */
-		etnaviv_gpu_update_clock(gpu);
-
-		control = gpu_read(gpu, VIVS_HI_CLOCK_CONTROL);
+		unsigned int fscale = 1 << (6 - gpu->freq_scale);
+		control = VIVS_HI_CLOCK_CONTROL_FSCALE_VAL(fscale);
+		etnaviv_gpu_load_clock(gpu, control);
 
 		/* Wait for stable clock.  Vivante's code waited for 1ms */
 		usleep_range(1000, 10000);
@@ -490,6 +491,10 @@ static int etnaviv_hw_reset(struct etnaviv_gpu *gpu)
 			continue;
 		}
 
+		/* disable debug registers, as they are not normally needed */
+		control |= VIVS_HI_CLOCK_CONTROL_DISABLE_DEBUG_REGISTERS;
+		gpu_write(gpu, VIVS_HI_CLOCK_CONTROL, control);
+
 		failed = false;
 		break;
 	}
-- 
cgit v1.2.3


From 355502e03ad26e3c872a0f5c408a4accca57ba7e Mon Sep 17 00:00:00 2001
From: Christian Gmeiner <christian.gmeiner@gmail.com>
Date: Sun, 24 Sep 2017 15:15:19 +0200
Subject: drm/etnaviv: use bitmap to keep track of events

This is prep work to be able to allocate multiple events in one go.

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
 drivers/gpu/drm/etnaviv/etnaviv_gpu.c | 31 +++++++++++++------------------
 drivers/gpu/drm/etnaviv/etnaviv_gpu.h |  6 ++++--
 2 files changed, 17 insertions(+), 20 deletions(-)

(limited to 'drivers/gpu/drm/etnaviv/etnaviv_gpu.c')

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
index de34e221c2fe..3b02814b9c52 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
@@ -744,10 +744,9 @@ int etnaviv_gpu_init(struct etnaviv_gpu *gpu)
 	/* Setup event management */
 	spin_lock_init(&gpu->event_spinlock);
 	init_completion(&gpu->event_free);
-	for (i = 0; i < ARRAY_SIZE(gpu->event); i++) {
-		gpu->event[i].used = false;
+	bitmap_zero(gpu->event_bitmap, ETNA_NR_EVENTS);
+	for (i = 0; i < ARRAY_SIZE(gpu->event); i++)
 		complete(&gpu->event_free);
-	}
 
 	/* Now program the hardware */
 	mutex_lock(&gpu->lock);
@@ -931,7 +930,7 @@ static void recover_worker(struct work_struct *work)
 	struct etnaviv_gpu *gpu = container_of(work, struct etnaviv_gpu,
 					       recover_work);
 	unsigned long flags;
-	unsigned int i;
+	unsigned int i = 0;
 
 	dev_err(gpu->dev, "hangcheck recover!\n");
 
@@ -950,14 +949,12 @@ static void recover_worker(struct work_struct *work)
 
 	/* complete all events, the GPU won't do it after the reset */
 	spin_lock_irqsave(&gpu->event_spinlock, flags);
-	for (i = 0; i < ARRAY_SIZE(gpu->event); i++) {
-		if (!gpu->event[i].used)
-			continue;
+	for_each_set_bit_from(i, gpu->event_bitmap, ETNA_NR_EVENTS) {
 		dma_fence_signal(gpu->event[i].fence);
 		gpu->event[i].fence = NULL;
-		gpu->event[i].used = false;
 		complete(&gpu->event_free);
 	}
+	bitmap_zero(gpu->event_bitmap, ETNA_NR_EVENTS);
 	spin_unlock_irqrestore(&gpu->event_spinlock, flags);
 	gpu->completed_fence = gpu->active_fence;
 
@@ -1148,7 +1145,7 @@ int etnaviv_gpu_fence_sync_obj(struct etnaviv_gem_object *etnaviv_obj,
 static unsigned int event_alloc(struct etnaviv_gpu *gpu)
 {
 	unsigned long ret, flags;
-	unsigned int i, event = ~0U;
+	unsigned int event;
 
 	ret = wait_for_completion_timeout(&gpu->event_free,
 					  msecs_to_jiffies(10 * 10000));
@@ -1158,13 +1155,11 @@ static unsigned int event_alloc(struct etnaviv_gpu *gpu)
 	spin_lock_irqsave(&gpu->event_spinlock, flags);
 
 	/* find first free event */
-	for (i = 0; i < ARRAY_SIZE(gpu->event); i++) {
-		if (gpu->event[i].used == false) {
-			gpu->event[i].used = true;
-			event = i;
-			break;
-		}
-	}
+	event = find_first_zero_bit(gpu->event_bitmap, ETNA_NR_EVENTS);
+	if (event < ETNA_NR_EVENTS)
+		set_bit(event, gpu->event_bitmap);
+	else
+		event = ~0U;
 
 	spin_unlock_irqrestore(&gpu->event_spinlock, flags);
 
@@ -1177,12 +1172,12 @@ static void event_free(struct etnaviv_gpu *gpu, unsigned int event)
 
 	spin_lock_irqsave(&gpu->event_spinlock, flags);
 
-	if (gpu->event[event].used == false) {
+	if (!test_bit(event, gpu->event_bitmap)) {
 		dev_warn(gpu->dev, "event %u is already marked as free",
 			 event);
 		spin_unlock_irqrestore(&gpu->event_spinlock, flags);
 	} else {
-		gpu->event[event].used = false;
+		clear_bit(event, gpu->event_bitmap);
 		spin_unlock_irqrestore(&gpu->event_spinlock, flags);
 
 		complete(&gpu->event_free);
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.h b/drivers/gpu/drm/etnaviv/etnaviv_gpu.h
index 689cb8f3680c..70e6590aacdf 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_gpu.h
+++ b/drivers/gpu/drm/etnaviv/etnaviv_gpu.h
@@ -88,13 +88,14 @@ struct etnaviv_chip_identity {
 };
 
 struct etnaviv_event {
-	bool used;
 	struct dma_fence *fence;
 };
 
 struct etnaviv_cmdbuf_suballoc;
 struct etnaviv_cmdbuf;
 
+#define ETNA_NR_EVENTS 30
+
 struct etnaviv_gpu {
 	struct drm_device *drm;
 	struct thermal_cooling_device *cooling;
@@ -112,7 +113,8 @@ struct etnaviv_gpu {
 	u32 memory_base;
 
 	/* event management: */
-	struct etnaviv_event event[30];
+	DECLARE_BITMAP(event_bitmap, ETNA_NR_EVENTS);
+	struct etnaviv_event event[ETNA_NR_EVENTS];
 	struct completion event_free;
 	spinlock_t event_spinlock;
 
-- 
cgit v1.2.3


From 95a428c17212ddef40909981e118acd6f518dc93 Mon Sep 17 00:00:00 2001
From: Christian Gmeiner <christian.gmeiner@gmail.com>
Date: Sun, 24 Sep 2017 15:15:20 +0200
Subject: drm/etnaviv: make it possible to allocate multiple events

This makes it possible to allocate multiple events under the event
spinlock. This change is needed to support 'sync'-points.

Changes v2 -> v3:
- wait for the completion of all events
- use 10sec timeout regardless of the number of events
- removed validation if there are enough free events
- fixed return value evaluation of event_alloc(..) in etnaviv_gpu_submit(..)

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
 drivers/gpu/drm/etnaviv/etnaviv_gpu.c | 47 +++++++++++++++++++++++------------
 1 file changed, 31 insertions(+), 16 deletions(-)

(limited to 'drivers/gpu/drm/etnaviv/etnaviv_gpu.c')

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
index 3b02814b9c52..8d961bc238d4 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
@@ -1142,28 +1142,44 @@ int etnaviv_gpu_fence_sync_obj(struct etnaviv_gem_object *etnaviv_obj,
  * event management:
  */
 
-static unsigned int event_alloc(struct etnaviv_gpu *gpu)
+static int event_alloc(struct etnaviv_gpu *gpu, unsigned nr_events,
+	unsigned int *events)
 {
-	unsigned long ret, flags;
-	unsigned int event;
+	unsigned long flags, timeout = msecs_to_jiffies(10 * 10000);
+	unsigned i, acquired = 0;
 
-	ret = wait_for_completion_timeout(&gpu->event_free,
-					  msecs_to_jiffies(10 * 10000));
-	if (!ret)
-		dev_err(gpu->dev, "wait_for_completion_timeout failed");
+	for (i = 0; i < nr_events; i++) {
+		unsigned long ret;
+
+		ret = wait_for_completion_timeout(&gpu->event_free, timeout);
+
+		if (!ret) {
+			dev_err(gpu->dev, "wait_for_completion_timeout failed");
+			goto out;
+		}
+
+		acquired++;
+		timeout = ret;
+	}
 
 	spin_lock_irqsave(&gpu->event_spinlock, flags);
 
-	/* find first free event */
-	event = find_first_zero_bit(gpu->event_bitmap, ETNA_NR_EVENTS);
-	if (event < ETNA_NR_EVENTS)
+	for (i = 0; i < nr_events; i++) {
+		int event = find_first_zero_bit(gpu->event_bitmap, ETNA_NR_EVENTS);
+
+		events[i] = event;
 		set_bit(event, gpu->event_bitmap);
-	else
-		event = ~0U;
+	}
 
 	spin_unlock_irqrestore(&gpu->event_spinlock, flags);
 
-	return event;
+	return 0;
+
+out:
+	for (i = 0; i < acquired; i++)
+		complete(&gpu->event_free);
+
+	return -EBUSY;
 }
 
 static void event_free(struct etnaviv_gpu *gpu, unsigned int event)
@@ -1332,10 +1348,9 @@ int etnaviv_gpu_submit(struct etnaviv_gpu *gpu,
 	 *
 	 */
 
-	event = event_alloc(gpu);
-	if (unlikely(event == ~0U)) {
+	ret = event_alloc(gpu, 1, &event);
+	if (ret) {
 		DRM_ERROR("no free event\n");
-		ret = -EBUSY;
 		goto out_pm_put;
 	}
 
-- 
cgit v1.2.3


From 4fc3e66adb291cc3ec4a360c75221e9fbf416938 Mon Sep 17 00:00:00 2001
From: Christian Gmeiner <christian.gmeiner@gmail.com>
Date: Sun, 24 Sep 2017 15:15:24 +0200
Subject: drm/etnaviv: extend etnaviv_gpu_cmdbuf_new(..) with nr_pmrs

This commits extends etnaviv_gpu_cmdbuf_new(..) to define the number
of struct etnaviv_perfmon elements gets used.

Changes from v1 -> v2:
- make use of goto as requested by Lucas

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
 drivers/gpu/drm/etnaviv/etnaviv_cmdbuf.c     | 15 ++++++++++++++-
 drivers/gpu/drm/etnaviv/etnaviv_cmdbuf.h     |  2 +-
 drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c |  2 +-
 drivers/gpu/drm/etnaviv/etnaviv_gpu.c        |  2 +-
 4 files changed, 17 insertions(+), 4 deletions(-)

(limited to 'drivers/gpu/drm/etnaviv/etnaviv_gpu.c')

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_cmdbuf.c b/drivers/gpu/drm/etnaviv/etnaviv_cmdbuf.c
index 633e0f07cbac..66ac79558bbd 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_cmdbuf.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_cmdbuf.c
@@ -19,6 +19,7 @@
 #include "etnaviv_cmdbuf.h"
 #include "etnaviv_gpu.h"
 #include "etnaviv_mmu.h"
+#include "etnaviv_perfmon.h"
 
 #define SUBALLOC_SIZE		SZ_256K
 #define SUBALLOC_GRANULE	SZ_4K
@@ -87,9 +88,10 @@ void etnaviv_cmdbuf_suballoc_destroy(struct etnaviv_cmdbuf_suballoc *suballoc)
 
 struct etnaviv_cmdbuf *
 etnaviv_cmdbuf_new(struct etnaviv_cmdbuf_suballoc *suballoc, u32 size,
-		   size_t nr_bos)
+		   size_t nr_bos, size_t nr_pmrs)
 {
 	struct etnaviv_cmdbuf *cmdbuf;
+	struct etnaviv_perfmon_request *pmrs;
 	size_t sz = size_vstruct(nr_bos, sizeof(cmdbuf->bo_map[0]),
 				 sizeof(*cmdbuf));
 	int granule_offs, order, ret;
@@ -98,6 +100,12 @@ etnaviv_cmdbuf_new(struct etnaviv_cmdbuf_suballoc *suballoc, u32 size,
 	if (!cmdbuf)
 		return NULL;
 
+	sz = sizeof(*pmrs) * nr_pmrs;
+	pmrs = kzalloc(sz, GFP_KERNEL);
+	if (!pmrs)
+		goto out_free_cmdbuf;
+
+	cmdbuf->pmrs = pmrs;
 	cmdbuf->suballoc = suballoc;
 	cmdbuf->size = size;
 
@@ -124,6 +132,10 @@ retry:
 	cmdbuf->vaddr = suballoc->vaddr + cmdbuf->suballoc_offset;
 
 	return cmdbuf;
+
+out_free_cmdbuf:
+	kfree(cmdbuf);
+	return NULL;
 }
 
 void etnaviv_cmdbuf_free(struct etnaviv_cmdbuf *cmdbuf)
@@ -139,6 +151,7 @@ void etnaviv_cmdbuf_free(struct etnaviv_cmdbuf *cmdbuf)
 	suballoc->free_space = 1;
 	mutex_unlock(&suballoc->lock);
 	wake_up_all(&suballoc->free_event);
+	kfree(cmdbuf->pmrs);
 	kfree(cmdbuf);
 }
 
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_cmdbuf.h b/drivers/gpu/drm/etnaviv/etnaviv_cmdbuf.h
index 1b549f0d59a0..b6348b9f2a9d 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_cmdbuf.h
+++ b/drivers/gpu/drm/etnaviv/etnaviv_cmdbuf.h
@@ -53,7 +53,7 @@ void etnaviv_cmdbuf_suballoc_destroy(struct etnaviv_cmdbuf_suballoc *suballoc);
 
 struct etnaviv_cmdbuf *
 etnaviv_cmdbuf_new(struct etnaviv_cmdbuf_suballoc *suballoc, u32 size,
-		   size_t nr_bos);
+		   size_t nr_bos, size_t nr_pmrs);
 void etnaviv_cmdbuf_free(struct etnaviv_cmdbuf *cmdbuf);
 
 u32 etnaviv_cmdbuf_get_va(struct etnaviv_cmdbuf *buf);
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c b/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c
index 026ef4e02f85..41abdfc7cb00 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c
@@ -350,7 +350,7 @@ int etnaviv_ioctl_gem_submit(struct drm_device *dev, void *data,
 	stream = kvmalloc_array(1, args->stream_size, GFP_KERNEL);
 	cmdbuf = etnaviv_cmdbuf_new(gpu->cmdbuf_suballoc,
 				    ALIGN(args->stream_size, 8) + 8,
-				    args->nr_bos);
+				    args->nr_bos, 0);
 	if (!bos || !relocs || !stream || !cmdbuf) {
 		ret = -ENOMEM;
 		goto err_submit_cmds;
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
index 8d961bc238d4..7b61071af0de 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
@@ -726,7 +726,7 @@ int etnaviv_gpu_init(struct etnaviv_gpu *gpu)
 	}
 
 	/* Create buffer: */
-	gpu->buffer = etnaviv_cmdbuf_new(gpu->cmdbuf_suballoc, PAGE_SIZE, 0);
+	gpu->buffer = etnaviv_cmdbuf_new(gpu->cmdbuf_suballoc, PAGE_SIZE, 0, 0);
 	if (!gpu->buffer) {
 		ret = -ENOMEM;
 		dev_err(gpu->dev, "could not create command buffer\n");
-- 
cgit v1.2.3


From 357713ce9bc86c1ae7ba804731d8db542944463c Mon Sep 17 00:00:00 2001
From: Christian Gmeiner <christian.gmeiner@gmail.com>
Date: Sun, 24 Sep 2017 15:15:28 +0200
Subject: drm/etnaviv: add 'sync point' support

In order to support performance counters in a sane way we need to provide
a method to sync the GPU with the CPU. The GPU can process multpile command
buffers/events per irq. With the help of a 'sync point' we can trigger an event
and stop the GPU/FE immediately. When the CPU is done with is processing it
simply needs to restart the FE and the GPU will process the command stream.

Changes from v1 -> v2:
- process sync point with a work item to keep irq as fast as possible

Changes from v4 -> v5:
- renamed pmrs_* to sync_point_*
- call event_free(..) in sync_point_worker(..)

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
 drivers/gpu/drm/etnaviv/etnaviv_buffer.c | 36 ++++++++++++++++++++++++++++++++
 drivers/gpu/drm/etnaviv/etnaviv_drv.h    |  1 +
 drivers/gpu/drm/etnaviv/etnaviv_gpu.c    | 26 +++++++++++++++++++++++
 drivers/gpu/drm/etnaviv/etnaviv_gpu.h    |  6 ++++++
 4 files changed, 69 insertions(+)

(limited to 'drivers/gpu/drm/etnaviv/etnaviv_gpu.c')

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_buffer.c b/drivers/gpu/drm/etnaviv/etnaviv_buffer.c
index ed9588f36bc9..9e7098e3207f 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_buffer.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_buffer.c
@@ -250,6 +250,42 @@ void etnaviv_buffer_end(struct etnaviv_gpu *gpu)
 	}
 }
 
+/* Append a 'sync point' to the ring buffer. */
+void etnaviv_sync_point_queue(struct etnaviv_gpu *gpu, unsigned int event)
+{
+	struct etnaviv_cmdbuf *buffer = gpu->buffer;
+	unsigned int waitlink_offset = buffer->user_size - 16;
+	u32 dwords, target;
+
+	/*
+	 * We need at most 3 dwords in the return target:
+	 * 1 event + 1 end + 1 wait + 1 link.
+	 */
+	dwords = 4;
+	target = etnaviv_buffer_reserve(gpu, buffer, dwords);
+
+	/* Signal sync point event */
+	CMD_LOAD_STATE(buffer, VIVS_GL_EVENT, VIVS_GL_EVENT_EVENT_ID(event) |
+		       VIVS_GL_EVENT_FROM_PE);
+
+	/* Stop the FE to 'pause' the GPU */
+	CMD_END(buffer);
+
+	/* Append waitlink */
+	CMD_WAIT(buffer);
+	CMD_LINK(buffer, 2, etnaviv_cmdbuf_get_va(buffer) +
+			    buffer->user_size - 4);
+
+	/*
+	 * Kick off the 'sync point' command by replacing the previous
+	 * WAIT with a link to the address in the ring buffer.
+	 */
+	etnaviv_buffer_replace_wait(buffer, waitlink_offset,
+				    VIV_FE_LINK_HEADER_OP_LINK |
+				    VIV_FE_LINK_HEADER_PREFETCH(dwords),
+				    target);
+}
+
 /* Append a command buffer to the ring buffer. */
 void etnaviv_buffer_queue(struct etnaviv_gpu *gpu, unsigned int event,
 	struct etnaviv_cmdbuf *cmdbuf)
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_drv.h b/drivers/gpu/drm/etnaviv/etnaviv_drv.h
index d157d9379e68..203613ae24dc 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_drv.h
+++ b/drivers/gpu/drm/etnaviv/etnaviv_drv.h
@@ -100,6 +100,7 @@ int etnaviv_gem_new_userptr(struct drm_device *dev, struct drm_file *file,
 u16 etnaviv_buffer_init(struct etnaviv_gpu *gpu);
 u16 etnaviv_buffer_config_mmuv2(struct etnaviv_gpu *gpu, u32 mtlb_addr, u32 safe_addr);
 void etnaviv_buffer_end(struct etnaviv_gpu *gpu);
+void etnaviv_sync_point_queue(struct etnaviv_gpu *gpu, unsigned int event);
 void etnaviv_buffer_queue(struct etnaviv_gpu *gpu, unsigned int event,
 	struct etnaviv_cmdbuf *cmdbuf);
 void etnaviv_validate_init(void);
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
index 7b61071af0de..cd70e7c04305 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
@@ -25,6 +25,7 @@
 #include "etnaviv_gpu.h"
 #include "etnaviv_gem.h"
 #include "etnaviv_mmu.h"
+#include "etnaviv_perfmon.h"
 #include "common.xml.h"
 #include "state.xml.h"
 #include "state_hi.xml.h"
@@ -1364,6 +1365,7 @@ int etnaviv_gpu_submit(struct etnaviv_gpu *gpu,
 	}
 
 	gpu->event[event].fence = fence;
+	gpu->event[event].sync_point = NULL;
 	submit->fence = dma_fence_get(fence);
 	gpu->active_fence = submit->fence->seqno;
 
@@ -1409,6 +1411,24 @@ out_pm_put:
 	return ret;
 }
 
+static void etnaviv_process_sync_point(struct etnaviv_gpu *gpu,
+	struct etnaviv_event *event)
+{
+	u32 addr = gpu_read(gpu, VIVS_FE_DMA_ADDRESS);
+
+	event->sync_point(gpu, event);
+	etnaviv_gpu_start_fe(gpu, addr + 2, 2);
+}
+
+static void sync_point_worker(struct work_struct *work)
+{
+	struct etnaviv_gpu *gpu = container_of(work, struct etnaviv_gpu,
+					       sync_point_work);
+
+	etnaviv_process_sync_point(gpu, &gpu->event[gpu->sync_point_event]);
+	event_free(gpu, gpu->sync_point_event);
+}
+
 /*
  * Init/Cleanup:
  */
@@ -1455,6 +1475,11 @@ static irqreturn_t irq_handler(int irq, void *data)
 
 			dev_dbg(gpu->dev, "event %u\n", event);
 
+			if (gpu->event[event].sync_point) {
+				gpu->sync_point_event = event;
+				etnaviv_queue_work(gpu->drm, &gpu->sync_point_work);
+			}
+
 			fence = gpu->event[event].fence;
 			gpu->event[event].fence = NULL;
 			dma_fence_signal(fence);
@@ -1660,6 +1685,7 @@ static int etnaviv_gpu_bind(struct device *dev, struct device *master,
 
 	INIT_LIST_HEAD(&gpu->active_cmd_list);
 	INIT_WORK(&gpu->retire_work, retire_worker);
+	INIT_WORK(&gpu->sync_point_work, sync_point_worker);
 	INIT_WORK(&gpu->recover_work, recover_worker);
 	init_waitqueue_head(&gpu->fence_event);
 
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.h b/drivers/gpu/drm/etnaviv/etnaviv_gpu.h
index 70e6590aacdf..3be5cb53e89f 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_gpu.h
+++ b/drivers/gpu/drm/etnaviv/etnaviv_gpu.h
@@ -89,6 +89,8 @@ struct etnaviv_chip_identity {
 
 struct etnaviv_event {
 	struct dma_fence *fence;
+
+	void (*sync_point)(struct etnaviv_gpu *gpu, struct etnaviv_event *event);
 };
 
 struct etnaviv_cmdbuf_suballoc;
@@ -135,6 +137,10 @@ struct etnaviv_gpu {
 	/* worker for handling active-list retiring: */
 	struct work_struct retire_work;
 
+	/* worker for handling 'sync' points: */
+	struct work_struct sync_point_work;
+	int sync_point_event;
+
 	void __iomem *mmio;
 	int irq;
 
-- 
cgit v1.2.3


From 547d340d3f9bcd83a356c00044f260ae9d80d0e0 Mon Sep 17 00:00:00 2001
From: Christian Gmeiner <christian.gmeiner@gmail.com>
Date: Sun, 24 Sep 2017 15:15:29 +0200
Subject: drm/etnaviv: clear alloced event

Results in less code as the users do not set every struct member to 0/NULL.

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Lucas Stach <l.stach@pengutronix.de>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
 drivers/gpu/drm/etnaviv/etnaviv_gpu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'drivers/gpu/drm/etnaviv/etnaviv_gpu.c')

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
index cd70e7c04305..8aabdac8d00b 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
@@ -1169,6 +1169,7 @@ static int event_alloc(struct etnaviv_gpu *gpu, unsigned nr_events,
 		int event = find_first_zero_bit(gpu->event_bitmap, ETNA_NR_EVENTS);
 
 		events[i] = event;
+		memset(&gpu->event[event], 0, sizeof(struct etnaviv_event));
 		set_bit(event, gpu->event_bitmap);
 	}
 
@@ -1365,7 +1366,6 @@ int etnaviv_gpu_submit(struct etnaviv_gpu *gpu,
 	}
 
 	gpu->event[event].fence = fence;
-	gpu->event[event].sync_point = NULL;
 	submit->fence = dma_fence_get(fence);
 	gpu->active_fence = submit->fence->seqno;
 
-- 
cgit v1.2.3


From 68dc0b295dcbadbaf2747627dc535629e17eb664 Mon Sep 17 00:00:00 2001
From: Christian Gmeiner <christian.gmeiner@gmail.com>
Date: Sun, 24 Sep 2017 15:15:30 +0200
Subject: drm/etnaviv: use 'sync points' for performance monitor requests

With 'sync points' we can sample the reqeustes perform signals
before and/or after the submited command buffer.

Changes v2 -> v3:
- fixed indentation and init nr_events to 1

Changes v4 -> v5:
- simplify logic around fence handling.

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
 drivers/gpu/drm/etnaviv/etnaviv_gpu.c | 75 ++++++++++++++++++++++++++++++++---
 drivers/gpu/drm/etnaviv/etnaviv_gpu.h |  1 +
 2 files changed, 70 insertions(+), 6 deletions(-)

(limited to 'drivers/gpu/drm/etnaviv/etnaviv_gpu.c')

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
index 8aabdac8d00b..56764b4b94aa 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
@@ -1329,12 +1329,48 @@ void etnaviv_gpu_pm_put(struct etnaviv_gpu *gpu)
 	pm_runtime_put_autosuspend(gpu->dev);
 }
 
+static void sync_point_perfmon_sample(struct etnaviv_gpu *gpu,
+	struct etnaviv_event *event, unsigned int flags)
+{
+	const struct etnaviv_cmdbuf *cmdbuf = event->cmdbuf;
+	unsigned int i;
+
+	for (i = 0; i < cmdbuf->nr_pmrs; i++) {
+		const struct etnaviv_perfmon_request *pmr = cmdbuf->pmrs + i;
+
+		if (pmr->flags == flags)
+			etnaviv_perfmon_process(gpu, pmr);
+	}
+}
+
+static void sync_point_perfmon_sample_pre(struct etnaviv_gpu *gpu,
+	struct etnaviv_event *event)
+{
+	sync_point_perfmon_sample(gpu, event, ETNA_PM_PROCESS_PRE);
+}
+
+static void sync_point_perfmon_sample_post(struct etnaviv_gpu *gpu,
+	struct etnaviv_event *event)
+{
+	const struct etnaviv_cmdbuf *cmdbuf = event->cmdbuf;
+	unsigned int i;
+
+	sync_point_perfmon_sample(gpu, event, ETNA_PM_PROCESS_POST);
+
+	for (i = 0; i < cmdbuf->nr_pmrs; i++) {
+		const struct etnaviv_perfmon_request *pmr = cmdbuf->pmrs + i;
+
+		*pmr->bo_vma = pmr->sequence;
+	}
+}
+
+
 /* add bo's to gpu's ring, and kick gpu: */
 int etnaviv_gpu_submit(struct etnaviv_gpu *gpu,
 	struct etnaviv_gem_submit *submit, struct etnaviv_cmdbuf *cmdbuf)
 {
 	struct dma_fence *fence;
-	unsigned int event, i;
+	unsigned int i, nr_events = 1, event[3];
 	int ret;
 
 	ret = etnaviv_gpu_pm_get_sync(gpu);
@@ -1350,9 +1386,19 @@ int etnaviv_gpu_submit(struct etnaviv_gpu *gpu,
 	 *
 	 */
 
-	ret = event_alloc(gpu, 1, &event);
+	/*
+	 * if there are performance monitor requests we need to have
+	 * - a sync point to re-configure gpu and process ETNA_PM_PROCESS_PRE
+	 *   requests.
+	 * - a sync point to re-configure gpu, process ETNA_PM_PROCESS_POST requests
+	 *   and update the sequence number for userspace.
+	 */
+	if (cmdbuf->nr_pmrs)
+		nr_events = 3;
+
+	ret = event_alloc(gpu, nr_events, event);
 	if (ret) {
-		DRM_ERROR("no free event\n");
+		DRM_ERROR("no free events\n");
 		goto out_pm_put;
 	}
 
@@ -1360,12 +1406,14 @@ int etnaviv_gpu_submit(struct etnaviv_gpu *gpu,
 
 	fence = etnaviv_gpu_fence_alloc(gpu);
 	if (!fence) {
-		event_free(gpu, event);
+		for (i = 0; i < nr_events; i++)
+			event_free(gpu, event[i]);
+
 		ret = -ENOMEM;
 		goto out_unlock;
 	}
 
-	gpu->event[event].fence = fence;
+	gpu->event[event[0]].fence = fence;
 	submit->fence = dma_fence_get(fence);
 	gpu->active_fence = submit->fence->seqno;
 
@@ -1375,7 +1423,19 @@ int etnaviv_gpu_submit(struct etnaviv_gpu *gpu,
 		gpu->lastctx = cmdbuf->ctx;
 	}
 
-	etnaviv_buffer_queue(gpu, event, cmdbuf);
+	if (cmdbuf->nr_pmrs) {
+		gpu->event[event[1]].sync_point = &sync_point_perfmon_sample_pre;
+		gpu->event[event[1]].cmdbuf = cmdbuf;
+		etnaviv_sync_point_queue(gpu, event[1]);
+	}
+
+	etnaviv_buffer_queue(gpu, event[0], cmdbuf);
+
+	if (cmdbuf->nr_pmrs) {
+		gpu->event[event[2]].sync_point = &sync_point_perfmon_sample_post;
+		gpu->event[event[2]].cmdbuf = cmdbuf;
+		etnaviv_sync_point_queue(gpu, event[2]);
+	}
 
 	cmdbuf->fence = fence;
 	list_add_tail(&cmdbuf->node, &gpu->active_cmd_list);
@@ -1481,6 +1541,9 @@ static irqreturn_t irq_handler(int irq, void *data)
 			}
 
 			fence = gpu->event[event].fence;
+			if (!fence)
+				continue;
+
 			gpu->event[event].fence = NULL;
 			dma_fence_signal(fence);
 
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.h b/drivers/gpu/drm/etnaviv/etnaviv_gpu.h
index 3be5cb53e89f..4f10f147297a 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_gpu.h
+++ b/drivers/gpu/drm/etnaviv/etnaviv_gpu.h
@@ -89,6 +89,7 @@ struct etnaviv_chip_identity {
 
 struct etnaviv_event {
 	struct dma_fence *fence;
+	struct etnaviv_cmdbuf *cmdbuf;
 
 	void (*sync_point)(struct etnaviv_gpu *gpu, struct etnaviv_event *event);
 };
-- 
cgit v1.2.3


From 2c8b0c5a99054ba909036d7f9c7bb7391b48b6f5 Mon Sep 17 00:00:00 2001
From: Christian Gmeiner <christian.gmeiner@gmail.com>
Date: Sun, 24 Sep 2017 15:15:39 +0200
Subject: drm/etnaviv: need to disable clock gating when doing profiling

As done by Vivante kernel driver.

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Lucas Stach <l.stach@pengutronix.de>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
 drivers/gpu/drm/etnaviv/etnaviv_gpu.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

(limited to 'drivers/gpu/drm/etnaviv/etnaviv_gpu.c')

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
index 56764b4b94aa..c00086d75580 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
@@ -1346,6 +1346,13 @@ static void sync_point_perfmon_sample(struct etnaviv_gpu *gpu,
 static void sync_point_perfmon_sample_pre(struct etnaviv_gpu *gpu,
 	struct etnaviv_event *event)
 {
+	u32 val;
+
+	/* disable clock gating */
+	val = gpu_read(gpu, VIVS_PM_POWER_CONTROLS);
+	val &= ~VIVS_PM_POWER_CONTROLS_ENABLE_MODULE_CLOCK_GATING;
+	gpu_write(gpu, VIVS_PM_POWER_CONTROLS, val);
+
 	sync_point_perfmon_sample(gpu, event, ETNA_PM_PROCESS_PRE);
 }
 
@@ -1354,6 +1361,7 @@ static void sync_point_perfmon_sample_post(struct etnaviv_gpu *gpu,
 {
 	const struct etnaviv_cmdbuf *cmdbuf = event->cmdbuf;
 	unsigned int i;
+	u32 val;
 
 	sync_point_perfmon_sample(gpu, event, ETNA_PM_PROCESS_POST);
 
@@ -1362,6 +1370,11 @@ static void sync_point_perfmon_sample_post(struct etnaviv_gpu *gpu,
 
 		*pmr->bo_vma = pmr->sequence;
 	}
+
+	/* enable clock gating */
+	val = gpu_read(gpu, VIVS_PM_POWER_CONTROLS);
+	val |= VIVS_PM_POWER_CONTROLS_ENABLE_MODULE_CLOCK_GATING;
+	gpu_write(gpu, VIVS_PM_POWER_CONTROLS, val);
 }
 
 
-- 
cgit v1.2.3


From 04a7d18d7975f720bc0122957eafb32fe42e2dff Mon Sep 17 00:00:00 2001
From: Christian Gmeiner <christian.gmeiner@gmail.com>
Date: Sun, 24 Sep 2017 15:15:42 +0200
Subject: drm/etnaviv: enable debug registers on demand

Some performance register are debug register and they need to
be enabled in order to be functional.

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Lucas Stach <l.stach@pengutronix.de>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
 drivers/gpu/drm/etnaviv/etnaviv_gpu.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

(limited to 'drivers/gpu/drm/etnaviv/etnaviv_gpu.c')

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
index c00086d75580..51b3fbea1bf0 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
@@ -1353,6 +1353,11 @@ static void sync_point_perfmon_sample_pre(struct etnaviv_gpu *gpu,
 	val &= ~VIVS_PM_POWER_CONTROLS_ENABLE_MODULE_CLOCK_GATING;
 	gpu_write(gpu, VIVS_PM_POWER_CONTROLS, val);
 
+	/* enable debug register */
+	val = gpu_read(gpu, VIVS_HI_CLOCK_CONTROL);
+	val &= ~VIVS_HI_CLOCK_CONTROL_DISABLE_DEBUG_REGISTERS;
+	gpu_write(gpu, VIVS_HI_CLOCK_CONTROL, val);
+
 	sync_point_perfmon_sample(gpu, event, ETNA_PM_PROCESS_PRE);
 }
 
@@ -1371,6 +1376,11 @@ static void sync_point_perfmon_sample_post(struct etnaviv_gpu *gpu,
 		*pmr->bo_vma = pmr->sequence;
 	}
 
+	/* disable debug register */
+	val = gpu_read(gpu, VIVS_HI_CLOCK_CONTROL);
+	val |= VIVS_HI_CLOCK_CONTROL_DISABLE_DEBUG_REGISTERS;
+	gpu_write(gpu, VIVS_HI_CLOCK_CONTROL, val);
+
 	/* enable clock gating */
 	val = gpu_read(gpu, VIVS_PM_POWER_CONTROLS);
 	val |= VIVS_PM_POWER_CONTROLS_ENABLE_MODULE_CLOCK_GATING;
-- 
cgit v1.2.3


From 7f9d398c16040f32ba7bc6521a89c06c9cb09442 Mon Sep 17 00:00:00 2001
From: Lucas Stach <l.stach@pengutronix.de>
Date: Thu, 28 Sep 2017 15:45:14 +0200
Subject: drm/etnaviv: remove stale comment

This comment is outdated as the driver is taking care about clock
gating and the pulse eater for quite some time already.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
---
 drivers/gpu/drm/etnaviv/etnaviv_gpu.c | 7 -------
 1 file changed, 7 deletions(-)

(limited to 'drivers/gpu/drm/etnaviv/etnaviv_gpu.c')

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
index 51b3fbea1bf0..3e0fcc1fafb5 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
@@ -435,13 +435,6 @@ static int etnaviv_hw_reset(struct etnaviv_gpu *gpu)
 	unsigned long timeout;
 	bool failed = true;
 
-	/* TODO
-	 *
-	 * - clock gating
-	 * - puls eater
-	 * - what about VG?
-	 */
-
 	/* We hope that the GPU resets in under one second */
 	timeout = jiffies + msecs_to_jiffies(1000);
 
-- 
cgit v1.2.3


From 4046217954d4e79d57d34c56ce42dec4f87e2ce9 Mon Sep 17 00:00:00 2001
From: Philipp Zabel <p.zabel@pengutronix.de>
Date: Mon, 9 Oct 2017 12:03:30 +0200
Subject: drm/etnaviv: reduce reset delay

After reset assertion, we only have to wait for the reset signals to
propagate through the GPU before deasserting the reset again. A few
hundred clock cycles should be more than enough. Replace the msleep(1),
which can actually take about 30 ms on i.MX6Q in some configurations,
with an usleep_range of a few microseconds. If the delay was too short,
the FE would not be idle afterwards, and the reset would be retried.

Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
 drivers/gpu/drm/etnaviv/etnaviv_gpu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'drivers/gpu/drm/etnaviv/etnaviv_gpu.c')

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
index 3e0fcc1fafb5..95ee479c95cb 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
@@ -456,7 +456,7 @@ static int etnaviv_hw_reset(struct etnaviv_gpu *gpu)
 		gpu_write(gpu, VIVS_HI_CLOCK_CONTROL, control);
 
 		/* wait for reset. */
-		msleep(1);
+		usleep_range(10, 20);
 
 		/* reset soft reset bit. */
 		control &= ~VIVS_HI_CLOCK_CONTROL_SOFT_RESET;
-- 
cgit v1.2.3


From 8272170f7da34e2554bed10ab76582094ea6816d Mon Sep 17 00:00:00 2001
From: Philipp Zabel <p.zabel@pengutronix.de>
Date: Mon, 9 Oct 2017 12:03:31 +0200
Subject: drm/etnaviv: remove unnecessary clock stabilization delay

There is no reason to wait for clock stabilization here, as the clock
framework guarantees that PLL clock sources are stable before clk_enable
returns.

Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
 drivers/gpu/drm/etnaviv/etnaviv_gpu.c | 3 ---
 1 file changed, 3 deletions(-)

(limited to 'drivers/gpu/drm/etnaviv/etnaviv_gpu.c')

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
index 95ee479c95cb..8197e1d6ed11 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_gpu.c
@@ -444,9 +444,6 @@ static int etnaviv_hw_reset(struct etnaviv_gpu *gpu)
 		control = VIVS_HI_CLOCK_CONTROL_FSCALE_VAL(fscale);
 		etnaviv_gpu_load_clock(gpu, control);
 
-		/* Wait for stable clock.  Vivante's code waited for 1ms */
-		usleep_range(1000, 10000);
-
 		/* isolate the GPU. */
 		control |= VIVS_HI_CLOCK_CONTROL_ISOLATE_GPU;
 		gpu_write(gpu, VIVS_HI_CLOCK_CONTROL, control);
-- 
cgit v1.2.3