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>

4 files changed, 69 insertions, 0 deletions

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;