From c3180ff11fcbdcd7da2809bad57bb8d9bfdfd44e Mon Sep 17 00:00:00 2001
From: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Date: Wed, 27 Feb 2019 14:58:05 +0100
Subject: [PATCH] drm/panfrost: Implement implicit out fences
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
---
drivers/gpu/drm/panfrost/panfrost_device.c | 2 +
drivers/gpu/drm/panfrost/panfrost_device.h | 2 +
drivers/gpu/drm/panfrost/panfrost_drv.c | 84 ++++++++++++-
drivers/gpu/drm/panfrost/panfrost_gem.c | 17 ++-
drivers/gpu/drm/panfrost/panfrost_gem.h | 4 +
drivers/gpu/drm/panfrost/panfrost_job.c | 133 ++++++++++++++++++++-
drivers/gpu/drm/panfrost/panfrost_job.h | 8 ++
include/uapi/drm/panfrost_drm.h | 2 +
8 files changed, 244 insertions(+), 8 deletions(-)
diff --git a/drivers/gpu/drm/panfrost/panfrost_device.c b/drivers/gpu/drm/panfrost/panfrost_device.c
index 3efedb23a01d7..34c3bccc621a5 100644
--- a/drivers/gpu/drm/panfrost/panfrost_device.c
+++ b/drivers/gpu/drm/panfrost/panfrost_device.c
@@ -68,6 +68,8 @@ int panfrost_device_init(struct panfrost_device *pfdev)
int err;
struct resource *res;
+ mutex_init(&pfdev->sched_lock);
+
err = panfrost_clk_init(pfdev);
if (err) {
dev_err(pfdev->dev, "clk init failed %d\n", err);
diff --git a/drivers/gpu/drm/panfrost/panfrost_device.h b/drivers/gpu/drm/panfrost/panfrost_device.h
index 82d2a0ac4d4e5..3aca60fa60a8c 100644
--- a/drivers/gpu/drm/panfrost/panfrost_device.h
+++ b/drivers/gpu/drm/panfrost/panfrost_device.h
@@ -66,6 +66,8 @@ struct panfrost_device {
struct panfrost_job_slot *js;
struct panfrost_job *jobs[3];
+
+ struct mutex sched_lock;
};
struct panfrost_file_priv {
diff --git a/drivers/gpu/drm/panfrost/panfrost_drv.c b/drivers/gpu/drm/panfrost/panfrost_drv.c
index 19821f15b98d6..3791c06a894bc 100644
--- a/drivers/gpu/drm/panfrost/panfrost_drv.c
+++ b/drivers/gpu/drm/panfrost/panfrost_drv.c
@@ -129,13 +129,86 @@ static int panfrost_ioctl_gem_info(struct drm_device *dev, void *data,
return drm_gem_dumb_map_offset(file, dev, args->handle, &args->offset);
}
+/**
+ * panfrost_lookup_bos() - Sets up job->bo[] with the GEM objects
+ * referenced by the job.
+ * @dev: DRM device
+ * @file_priv: DRM file for this fd
+ * @job: job being set up
+ *
+ * The command validator needs to reference BOs by their index within
+ * the submitted job's BO list. This does the validation of the job's
+ * BO list and reference counting for the lifetime of the job.
+ *
+ * Note that this function doesn't need to unreference the BOs on
+ * failure, because that will happen at panfrost_job_cleanup() time.
+ */
+static int
+panfrost_lookup_bos(struct drm_device *dev,
+ struct drm_file *file_priv,
+ struct drm_panfrost_gem_submit_atom *atom,
+ struct panfrost_job *job)
+{
+ u32 *handles;
+ int ret = 0;
+ int i;
+
+ job->bo_count = atom->bo_handle_count;
+
+ if (!job->bo_count)
+ return 0;
+
+ job->bos = kvmalloc_array(job->bo_count,
+ sizeof(struct drm_panfrost_gem_object *),
+ GFP_KERNEL | __GFP_ZERO);
+ if (!job->bos) {
+ DRM_DEBUG("Failed to allocate validated BO pointers\n");
+ return -ENOMEM;
+ }
+
+ handles = kvmalloc_array(job->bo_count, sizeof(u32), GFP_KERNEL);
+ if (!handles) {
+ ret = -ENOMEM;
+ DRM_DEBUG("Failed to allocate incoming GEM handles\n");
+ goto fail;
+ }
+
+ if (copy_from_user(handles,
+ (void __user *)(uintptr_t)atom->bo_handles,
+ job->bo_count * sizeof(u32))) {
+ ret = -EFAULT;
+ DRM_DEBUG("Failed to copy in GEM handles\n");
+ goto fail;
+ }
+
+ spin_lock(&file_priv->table_lock);
+ for (i = 0; i < job->bo_count; i++) {
+ struct drm_gem_object *bo = idr_find(&file_priv->object_idr,
+ handles[i]);
+ if (!bo) {
+ DRM_DEBUG("Failed to look up GEM BO %d: %d\n",
+ i, handles[i]);
+ ret = -ENOENT;
+ spin_unlock(&file_priv->table_lock);
+ goto fail;
+ }
+ drm_gem_object_get(bo);
+ job->bos[i] = to_panfrost_bo(bo);
+ }
+ spin_unlock(&file_priv->table_lock);
+
+fail:
+ kvfree(handles);
+ return ret;
+}
+
static int panfrost_ioctl_gem_submit(struct drm_device *dev, void *data,
struct drm_file *file)
{
- struct drm_panfrost_gem_submit *args = data;
struct panfrost_device *pfdev = dev->dev_private;
- struct panfrost_job *job;
+ struct drm_panfrost_gem_submit *args = data;
struct drm_panfrost_gem_submit_atom *atoms;
+ struct panfrost_job *job;
u32 latest_flush_id;
int i, ret = 0;
@@ -163,6 +236,8 @@ static int panfrost_ioctl_gem_submit(struct drm_device *dev, void *data,
if (!job)
return -ENOMEM;
+ kref_init(&job->refcount);
+
job->pfdev = pfdev;
job->file_priv = file->driver_priv;
job->jc = atoms[i].jc;
@@ -170,6 +245,10 @@ static int panfrost_ioctl_gem_submit(struct drm_device *dev, void *data,
job->requirements = atoms[i].requirements;
job->flush_id = latest_flush_id;
+ ret = panfrost_lookup_bos(dev, file, &atoms[i], job);
+ if (ret)
+ goto err_free_atoms;
+
ret = panfrost_job_push(job);
if (ret)
goto err_free_atoms;
@@ -261,6 +340,7 @@ static struct drm_driver panfrost_drm_driver = {
.prime_fd_to_handle = drm_gem_prime_fd_to_handle,
.gem_prime_import_sg_table = panfrost_gem_prime_import_sg_table,
.gem_prime_mmap = drm_gem_prime_mmap,
+ .gem_prime_res_obj = panfrost_prime_res_obj,
};
static int panfrost_pdev_probe(struct platform_device *pdev)
diff --git a/drivers/gpu/drm/panfrost/panfrost_gem.c b/drivers/gpu/drm/panfrost/panfrost_gem.c
index 6ca69a4966f89..9f74b4d063da7 100644
--- a/drivers/gpu/drm/panfrost/panfrost_gem.c
+++ b/drivers/gpu/drm/panfrost/panfrost_gem.c
@@ -6,6 +6,7 @@
#include <linux/err.h>
#include <linux/ktime.h>
#include <linux/slab.h>
+#include <linux/dma-buf.h>
#include <drm/panfrost_drm.h>
#include "panfrost_device.h"
@@ -79,6 +80,9 @@ struct drm_gem_object *panfrost_gem_create_object(struct drm_device *dev, size_t
obj->base.base.funcs = &panfrost_gem_funcs;
+ obj->resv = &obj->_resv;
+ reservation_object_init(obj->resv);
+
spin_lock(&pfdev->mm_lock);
ret = drm_mm_insert_node(&pfdev->mm, &obj->node,
roundup(size, PAGE_SIZE) >> PAGE_SHIFT);
@@ -99,14 +103,25 @@ panfrost_gem_prime_import_sg_table(struct drm_device *dev,
struct sg_table *sgt)
{
struct drm_gem_object *obj;
+ struct panfrost_gem_object *pobj;
obj = drm_gem_shmem_prime_import_sg_table(dev, attach, sgt);
+ pobj = to_panfrost_bo(obj);
+
+ pobj->resv = attach->dmabuf->resv;
- panfrost_mmu_map(to_panfrost_bo(obj));
+ panfrost_mmu_map(pobj);
return obj;
}
+struct reservation_object *panfrost_prime_res_obj(struct drm_gem_object *obj)
+{
+ struct panfrost_gem_object *bo = to_panfrost_bo(obj);
+
+ return bo->resv;
+}
+
int panfrost_gem_cpu_prep(struct drm_gem_object *obj, uint32_t op, ktime_t *timeout)
{
bool write = !!(op & ETNA_PREP_WRITE);
diff --git a/drivers/gpu/drm/panfrost/panfrost_gem.h b/drivers/gpu/drm/panfrost/panfrost_gem.h
index e90a39a060b53..5c56617040798 100644
--- a/drivers/gpu/drm/panfrost/panfrost_gem.h
+++ b/drivers/gpu/drm/panfrost/panfrost_gem.h
@@ -32,6 +32,9 @@ struct panfrost_gem_object {
/* cache maintenance */
u32 last_cpu_prep_op;
+ /* normally (resv == &_resv) except for imported bo's */
+ struct reservation_object *resv;
+ struct reservation_object _resv;
};
static inline
@@ -84,6 +87,7 @@ struct drm_gem_object *
panfrost_gem_prime_import_sg_table(struct drm_device *dev,
struct dma_buf_attachment *attach,
struct sg_table *sgt);
+struct reservation_object *panfrost_prime_res_obj(struct drm_gem_object *obj);
void panfrost_submit_put(struct panfrost_gem_submit * submit);
diff --git a/drivers/gpu/drm/panfrost/panfrost_job.c b/drivers/gpu/drm/panfrost/panfrost_job.c
index 88c0353ee9bc3..fb098f1947a1f 100644
--- a/drivers/gpu/drm/panfrost/panfrost_job.c
+++ b/drivers/gpu/drm/panfrost/panfrost_job.c
@@ -1,6 +1,7 @@
// SPDX-License-Identifier: GPL-2.0
/* Copyright 2018 Panfrost Team */
+#include <linux/reservation.h>
#include <drm/gpu_scheduler.h>
#include <drm/panfrost_drm.h>
@@ -8,6 +9,7 @@
#include "panfrost_job.h"
#include "panfrost_regs.h"
#include "panfrost_features.h"
+#include "panfrost_gem.h"
#define JOB_BASE 0x1000
@@ -85,7 +87,7 @@ to_panfrost_job(struct drm_sched_job *sched_job)
struct panfrost_fence {
struct dma_fence base;
struct drm_device *dev;
- /* v3d seqno for signaled() test */
+ /* panfrost seqno for signaled() test */
u64 seqno;
int queue;
};
@@ -265,19 +267,140 @@ static void kbase_job_hw_submit(struct panfrost_job *job,
job_write(pfdev, JS_COMMAND_NEXT(js), JS_COMMAND_START);
}
+
+static void
+panfrost_unlock_bo_reservations(struct panfrost_gem_object **bos,
+ int bo_count,
+ struct ww_acquire_ctx *acquire_ctx)
+{
+ int i;
+
+ for (i = 0; i < bo_count; i++)
+ ww_mutex_unlock(&bos[i]->resv->lock);
+
+ ww_acquire_fini(acquire_ctx);
+}
+
+/* Takes the reservation lock on all the BOs being referenced, so that
+ * at queue submit time we can update the reservations.
+ *
+ * We don't lock the RCL the tile alloc/state BOs, or overflow memory
+ * (all of which are on exec->unref_list). They're entirely private
+ * to panfrost, so we don't attach dma-buf fences to them.
+ */
+static int
+panfrost_lock_bo_reservations(struct panfrost_gem_object **bos,
+ int bo_count,
+ struct ww_acquire_ctx *acquire_ctx)
+{
+ int contended_lock = -1;
+ int i, ret;
+
+ ww_acquire_init(acquire_ctx, &reservation_ww_class);
+
+retry:
+ if (contended_lock != -1) {
+ struct panfrost_gem_object *bo = bos[contended_lock];
+
+ ret = ww_mutex_lock_slow_interruptible(&bo->resv->lock,
+ acquire_ctx);
+ if (ret) {
+ ww_acquire_done(acquire_ctx);
+ return ret;
+ }
+ }
+
+ for (i = 0; i < bo_count; i++) {
+ if (i == contended_lock)
+ continue;
+
+ ret = ww_mutex_lock_interruptible(&bos[i]->resv->lock,
+ acquire_ctx);
+ if (ret) {
+ int j;
+
+ for (j = 0; j < i; j++)
+ ww_mutex_unlock(&bos[j]->resv->lock);
+
+ if (contended_lock != -1 && contended_lock >= i) {
+ struct panfrost_gem_object *bo = bos[contended_lock];
+
+ ww_mutex_unlock(&bo->resv->lock);
+ }
+
+ if (ret == -EDEADLK) {
+ contended_lock = i;
+ goto retry;
+ }
+
+ ww_acquire_done(acquire_ctx);
+ return ret;
+ }
+ }
+
+ ww_acquire_done(acquire_ctx);
+
+ /* Reserve space for our shared (read-only) fence references,
+ * before we commit the job to the hardware.
+ */
+ for (i = 0; i < bo_count; i++) {
+ ret = reservation_object_reserve_shared(bos[i]->resv, 1);
+ if (ret) {
+ panfrost_unlock_bo_reservations(bos, bo_count,
+ acquire_ctx);
+ return ret;
+ }
+ }
+
+ return 0;
+}
+
+static void
+panfrost_attach_object_fences(struct panfrost_gem_object **bos, int bo_count,
+ struct dma_fence *fence)
+{
+ int i;
+
+ for (i = 0; i < bo_count; i++) {
+ /* XXX: Use shared fences for read-only objects. */
+ reservation_object_add_excl_fence(bos[i]->resv, fence);
+ }
+}
+
int panfrost_job_push(struct panfrost_job *job)
{
+ struct panfrost_device *pfdev = job->pfdev;
int slot = panfrost_job_get_slot(job);
struct drm_sched_entity *entity = &job->file_priv->sched_entity[slot];
- int ret;
+ struct ww_acquire_ctx acquire_ctx;
+ int ret = 0;
- ret = drm_sched_job_init(&job->base, entity, NULL);
+ mutex_lock(&pfdev->sched_lock);
+
+ ret = panfrost_lock_bo_reservations(job->bos, job->bo_count,
+ &acquire_ctx);
if (ret)
return ret;
+ ret = drm_sched_job_init(&job->base, entity, NULL);
+ if (ret)
+ goto unlock;
+
+ job->render_done_fence = dma_fence_get(&job->base.s_fence->finished);
+
+ kref_get(&job->refcount); /* put by scheduler job completion */
+
drm_sched_entity_push_job(&job->base, entity);
- return 0;
+ mutex_unlock(&pfdev->sched_lock);
+
+ panfrost_attach_object_fences(job->bos, job->bo_count,
+ job->render_done_fence);
+
+unlock:
+ panfrost_unlock_bo_reservations(job->bos, job->bo_count, &acquire_ctx);
+
+ return ret;
}
#endif
@@ -290,7 +413,7 @@ panfrost_job_free(struct drm_sched_job *sched_job)
/**
* Returns the fences that the bin or render job depends on, one by one.
- * v3d_job_run() won't be called until all of them have been signaled.
+ * panfrost_job_run() won't be called until all of them have been signaled.
*/
static struct dma_fence *
panfrost_job_dependency(struct drm_sched_job *sched_job,
diff --git a/drivers/gpu/drm/panfrost/panfrost_job.h b/drivers/gpu/drm/panfrost/panfrost_job.h
index b5c8835e187c0..4a80fb11c09a7 100644
--- a/drivers/gpu/drm/panfrost/panfrost_job.h
+++ b/drivers/gpu/drm/panfrost/panfrost_job.h
@@ -9,11 +9,14 @@
#define NUM_JOB_SLOTS 2 /* Don't need 3rd one until we have compute support */
struct panfrost_device;
+struct panfrost_gem_object;
struct panfrost_file_priv;
struct panfrost_job {
struct drm_sched_job base;
+ struct kref refcount;
+
struct panfrost_device *pfdev;
struct panfrost_file_priv *file_priv;
@@ -23,6 +26,11 @@ struct panfrost_job {
__u32 atom_nr;
__u32 requirements;
__u32 flush_id;
+
+ struct panfrost_gem_object **bos;
+ u32 bo_count;
+
+ struct dma_fence *render_done_fence;
};
int panfrost_job_init(struct panfrost_device *pfdev);
diff --git a/include/uapi/drm/panfrost_drm.h b/include/uapi/drm/panfrost_drm.h
index cd3bb30175f8d..bb4c83b1065b1 100644
--- a/include/uapi/drm/panfrost_drm.h
+++ b/include/uapi/drm/panfrost_drm.h
@@ -84,6 +84,8 @@ struct drm_panfrost_gem_submit_atom {
__u64 jc; /* in, address to GPU mapping of job descriptor */
__u32 atom_nr; /* in, job ID */
__u32 requirements; /* in, a combination of PANFROST_JD_REQ_* */
+ __u64 bo_handles;
+ __u32 bo_handle_count;
};
struct drm_panfrost_gem_submit {
--
GitLab