Skip to contentSkip to conversation

Host1x/TegraDRM UAPI

PATCH Superseded Mikko Perttunen 2 participants 25 messages Nov 20, 2020 – Nov 25, 2020

Subsystems (7)

v4 of 7 Superseded Latest is v7 →

lore
Share & tools
lore

Mikko Perttunen

author maintainer
#0 lore raw

v4 cover Host1x/TegraDRM UAPI

Hi all,

here's the fourth revision of the Host1x/TegraDRM UAPI proposal.

The changes at a high level in this revision are:
* Small bugfixes for issues reported by CI bots
* Removal of not strictly required features like sync_file FDs, reservations, partial mappings etc. from the submit UAPI.
* All new UAPI placed under CONFIG_DRM_TEGRA_STAGING.

The test suite[1] has been updated for the changes in this revision,

The series can be also found in
https://github.com/cyndis/linux/commits/work/host1x-uapi-v4.

Older versions:
v1: https://www.spinics.net/lists/linux-tegra/msg51000.html
v2: https://www.spinics.net/lists/linux-tegra/msg53061.html
v3: https://www.spinics.net/lists/linux-tegra/msg54370.html

Thank you,
Mikko

[1] https://github.com/cyndis/uapi-test

Mikko Perttunen (21):
gpu: host1x: Use different lock classes for each client
gpu: host1x: Allow syncpoints without associated client
gpu: host1x: Show number of pending waiters in debugfs
gpu: host1x: Remove cancelled waiters immediately
gpu: host1x: Use HW-equivalent syncpoint expiration check
gpu: host1x: Cleanup and refcounting for syncpoints
gpu: host1x: Introduce UAPI header
gpu: host1x: Implement /dev/host1x device node
gpu: host1x: DMA fences and userspace fence creation
gpu: host1x: Add no-recovery mode
gpu: host1x: Add job release callback
gpu: host1x: Add support for syncpoint waits in CDMA pushbuffer
gpu: host1x: Reset max value when freeing a syncpoint
gpu: host1x: Reserve VBLANK syncpoints at initialization drm/tegra: Add new UAPI to header
drm/tegra: Boot VIC during runtime PM resume drm/tegra: Set resv fields when importing/exporting GEMs drm/tegra: Allocate per-engine channel in core code drm/tegra: Implement new UAPI
drm/tegra: Implement job submission part of new UAPI drm/tegra: Add job firewall

 drivers/gpu/drm/tegra/Makefile         |   4 +
 drivers/gpu/drm/tegra/dc.c             |  10 +-
 drivers/gpu/drm/tegra/drm.c            |  72 +++--
 drivers/gpu/drm/tegra/drm.h            |   9 +
 drivers/gpu/drm/tegra/gem.c            |   2 +
 drivers/gpu/drm/tegra/gr2d.c           |   4 +-
 drivers/gpu/drm/tegra/gr3d.c           |   4 +-
 drivers/gpu/drm/tegra/uapi.h           |  63 ++++
 drivers/gpu/drm/tegra/uapi/firewall.c  | 197 ++++++++++++
 drivers/gpu/drm/tegra/uapi/gather_bo.c |  86 +++++
 drivers/gpu/drm/tegra/uapi/gather_bo.h |  22 ++
 drivers/gpu/drm/tegra/uapi/submit.c    | 427 +++++++++++++++++++++++++
 drivers/gpu/drm/tegra/uapi/submit.h    |  20 ++
 drivers/gpu/drm/tegra/uapi/uapi.c      | 306 ++++++++++++++++++
 drivers/gpu/drm/tegra/vic.c            | 118 ++++---
 drivers/gpu/host1x/Makefile            |   2 +
 drivers/gpu/host1x/bus.c               |   7 +-
 drivers/gpu/host1x/cdma.c              |  69 +++-
 drivers/gpu/host1x/debug.c             |  14 +-
 drivers/gpu/host1x/dev.c               |  15 +
 drivers/gpu/host1x/dev.h               |  16 +-
 drivers/gpu/host1x/fence.c             | 208 ++++++++++++
 drivers/gpu/host1x/fence.h             |  13 +
 drivers/gpu/host1x/hw/cdma_hw.c        |   2 +-
 drivers/gpu/host1x/hw/channel_hw.c     |  63 ++--
 drivers/gpu/host1x/hw/debug_hw.c       |  11 +-
 drivers/gpu/host1x/intr.c              |  23 +-
 drivers/gpu/host1x/intr.h              |   2 +
 drivers/gpu/host1x/job.c               |  79 +++--
 drivers/gpu/host1x/job.h               |  14 +
 drivers/gpu/host1x/syncpt.c            | 185 ++++++-----
 drivers/gpu/host1x/syncpt.h            |  16 +-
 drivers/gpu/host1x/uapi.c              | 385 ++++++++++++++++++++++
 drivers/gpu/host1x/uapi.h              |  22 ++
 drivers/staging/media/tegra-video/vi.c |   8 +-
 include/linux/host1x.h                 |  47 ++-
 include/uapi/drm/tegra_drm.h           | 338 +++++++++++++++++--
 include/uapi/linux/host1x.h            | 134 ++++++++
 38 files changed, 2729 insertions(+), 288 deletions(-)
 create mode 100644 drivers/gpu/drm/tegra/uapi.h
 create mode 100644 drivers/gpu/drm/tegra/uapi/firewall.c
 create mode 100644 drivers/gpu/drm/tegra/uapi/gather_bo.c
 create mode 100644 drivers/gpu/drm/tegra/uapi/gather_bo.h
 create mode 100644 drivers/gpu/drm/tegra/uapi/submit.c
 create mode 100644 drivers/gpu/drm/tegra/uapi/submit.h
 create mode 100644 drivers/gpu/drm/tegra/uapi/uapi.c
 create mode 100644 drivers/gpu/host1x/fence.c
 create mode 100644 drivers/gpu/host1x/fence.h
 create mode 100644 drivers/gpu/host1x/uapi.c
 create mode 100644 drivers/gpu/host1x/uapi.h
 create mode 100644 include/uapi/linux/host1x.h
Signature · 1 line
2.29.2
author maintainer
#1 lore raw

PATCH v4 1/21 gpu: host1x: Use different lock classes for each client

To avoid false lockdep warnings, give each client lock a different lock class, passed from the initialization site by macro.

---
 drivers/gpu/host1x/bus.c | 7 ++++---
 include/linux/host1x.h   | 9 ++++++++-
 2 files changed, 12 insertions(+), 4 deletions(-)
drivers/gpu/host1x/bus.c +4 -3
index e201f62d62c0..4101f64bd545 100644
@@ -714,13 +714,14 @@ EXPORT_SYMBOL(host1x_driver_unregister);
  * device and call host1x_device_init(), which will in turn call each client's
  * &host1x_client_ops.init implementation.
  */
-int host1x_client_register(struct host1x_client *client)
+int __host1x_client_register(struct host1x_client *client,
+			   struct lock_class_key *key)
 {
 	struct host1x *host1x;
 	int err;
 
 	INIT_LIST_HEAD(&client->list);
-	mutex_init(&client->lock);
+	__mutex_init(&client->lock, "host1x client lock", key);
 	client->usecount = 0;
 
 	mutex_lock(&devices_lock);
@@ -741,7 +742,7 @@ int host1x_client_register(struct host1x_client *client)
 
 	return 0;
 }
-EXPORT_SYMBOL(host1x_client_register);
+EXPORT_SYMBOL(__host1x_client_register);
 
 /**
  * host1x_client_unregister() - unregister a host1x client
include/linux/host1x.h +8 -1
index 20c885d0bddc..f711fc0154f4 100644
@@ -320,7 +320,14 @@ static inline struct host1x_device *to_host1x_device(struct device *dev)
 int host1x_device_init(struct host1x_device *device);
 int host1x_device_exit(struct host1x_device *device);
 
-int host1x_client_register(struct host1x_client *client);
+int __host1x_client_register(struct host1x_client *client,
+			     struct lock_class_key *key);
+#define host1x_client_register(class) \
+	({ \
+		static struct lock_class_key __key; \
+		__host1x_client_register(class, &__key); \
+	})
+
 int host1x_client_unregister(struct host1x_client *client);
 
 int host1x_client_suspend(struct host1x_client *client);
Signature · 1 line
2.29.2
author maintainer
#2 lore raw

PATCH v4 2/21 gpu: host1x: Allow syncpoints without associated client

In reply to Mikko Perttunen

Syncpoints don't need to be associated with any client, so remove the property, and expose host1x_syncpt_alloc. This will allow allocating syncpoints without prior knowledge of the engine that it will be used with.

---
v3:
* Clean up host1x_syncpt_alloc signature to allow specifying a name for the syncpoint.
* Export the function.

---
 drivers/gpu/host1x/syncpt.c | 22 ++++++++++------------
 drivers/gpu/host1x/syncpt.h |  1 -
 include/linux/host1x.h      |  3 +++
 3 files changed, 13 insertions(+), 13 deletions(-)
drivers/gpu/host1x/syncpt.c +10 -12
index fce7892d5137..5982fdf64e1c 100644
@@ -42,13 +42,13 @@ static void host1x_syncpt_base_free(struct host1x_syncpt_base *base)
 		base->requested = false;
 }
 
-static struct host1x_syncpt *host1x_syncpt_alloc(struct host1x *host,
-						 struct host1x_client *client,
-						 unsigned long flags)
+struct host1x_syncpt *host1x_syncpt_alloc(struct host1x *host,
+					  unsigned long flags,
+					  const char *name)
 {
 	struct host1x_syncpt *sp = host->syncpt;
+	char *full_name;
 	unsigned int i;
-	char *name;
 
 	mutex_lock(&host->syncpt_mutex);
 
@@ -64,13 +64,11 @@ static struct host1x_syncpt *host1x_syncpt_alloc(struct host1x *host,
 			goto unlock;
 	}
 
-	name = kasprintf(GFP_KERNEL, "%02u-%s", sp->id,
-			 client ? dev_name(client->dev) : NULL);
-	if (!name)
+	full_name = kasprintf(GFP_KERNEL, "%u-%s", sp->id, name);
+	if (!full_name)
 		goto free_base;
 
-	sp->client = client;
-	sp->name = name;
+	sp->name = full_name;
 
 	if (flags & HOST1X_SYNCPT_CLIENT_MANAGED)
 		sp->client_managed = true;
@@ -87,6 +85,7 @@ static struct host1x_syncpt *host1x_syncpt_alloc(struct host1x *host,
 	mutex_unlock(&host->syncpt_mutex);
 	return NULL;
 }
+EXPORT_SYMBOL(host1x_syncpt_alloc);
 
 /**
  * host1x_syncpt_id() - retrieve syncpoint ID
@@ -401,7 +400,7 @@ int host1x_syncpt_init(struct host1x *host)
 	host1x_hw_syncpt_enable_protection(host);
 
 	/* Allocate sync point to use for clearing waits for expired fences */
-	host->nop_sp = host1x_syncpt_alloc(host, NULL, 0);
+	host->nop_sp = host1x_syncpt_alloc(host, 0, "reserved-nop");
 	if (!host->nop_sp)
 		return -ENOMEM;
 
@@ -423,7 +422,7 @@ struct host1x_syncpt *host1x_syncpt_request(struct host1x_client *client,
 {
 	struct host1x *host = dev_get_drvdata(client->host->parent);
 
-	return host1x_syncpt_alloc(host, client, flags);
+	return host1x_syncpt_alloc(host, flags, dev_name(client->dev));
 }
 EXPORT_SYMBOL(host1x_syncpt_request);
 
@@ -447,7 +446,6 @@ void host1x_syncpt_free(struct host1x_syncpt *sp)
 	host1x_syncpt_base_free(sp->base);
 	kfree(sp->name);
 	sp->base = NULL;
-	sp->client = NULL;
 	sp->name = NULL;
 	sp->client_managed = false;
 
drivers/gpu/host1x/syncpt.h +0 -1
index 8e1d04dacaa0..3aa6b25b1b9c 100644
@@ -33,7 +33,6 @@ struct host1x_syncpt {
 	const char *name;
 	bool client_managed;
 	struct host1x *host;
-	struct host1x_client *client;
 	struct host1x_syncpt_base *base;
 
 	/* interrupt data */
include/linux/host1x.h +3 -0
index f711fc0154f4..099eff8a06d2 100644
@@ -154,6 +154,9 @@ int host1x_syncpt_wait(struct host1x_syncpt *sp, u32 thresh, long timeout,
 struct host1x_syncpt *host1x_syncpt_request(struct host1x_client *client,
 					    unsigned long flags);
 void host1x_syncpt_free(struct host1x_syncpt *sp);
+struct host1x_syncpt *host1x_syncpt_alloc(struct host1x *host,
+					  unsigned long flags,
+					  const char *name);
 
 struct host1x_syncpt_base *host1x_syncpt_get_base(struct host1x_syncpt *sp);
 u32 host1x_syncpt_base_id(struct host1x_syncpt_base *base);
Signature · 1 line
2.29.2
author maintainer
#3 lore raw

PATCH v4 3/21 gpu: host1x: Show number of pending waiters in debugfs

In reply to Mikko Perttunen

Show the number of pending waiters in the debugfs status file. This is useful for testing to verify that waiters do not leak or accumulate incorrectly.

---
 drivers/gpu/host1x/debug.c | 14 +++++++++++---
 1 file changed, 11 insertions(+), 3 deletions(-)
drivers/gpu/host1x/debug.c +11 -3
index 3eee4318b158..2d06a7406b3b 100644
@@ -69,6 +69,7 @@ static int show_channel(struct host1x_channel *ch, void *data, bool show_fifo)
 
 static void show_syncpts(struct host1x *m, struct output *o)
 {
+	struct list_head *pos;
 	unsigned int i;
 
 	host1x_debug_output(o, "---- syncpts ----\n");
@@ -76,12 +77,19 @@ static void show_syncpts(struct host1x *m, struct output *o)
 	for (i = 0; i < host1x_syncpt_nb_pts(m); i++) {
 		u32 max = host1x_syncpt_read_max(m->syncpt + i);
 		u32 min = host1x_syncpt_load(m->syncpt + i);
+		unsigned int waiters = 0;
 
-		if (!min && !max)
+		spin_lock(&m->syncpt[i].intr.lock);
+		list_for_each(pos, &m->syncpt[i].intr.wait_head)
+			waiters++;
+		spin_unlock(&m->syncpt[i].intr.lock);
+
+		if (!min && !max && !waiters)
 			continue;
 
-		host1x_debug_output(o, "id %u (%s) min %d max %d\n",
-				    i, m->syncpt[i].name, min, max);
+		host1x_debug_output(o,
+				    "id %u (%s) min %d max %d (%d waiters)\n",
+				    i, m->syncpt[i].name, min, max, waiters);
 	}
 
 	for (i = 0; i < host1x_syncpt_nb_bases(m); i++) {
Signature · 1 line
2.29.2
author maintainer
#4 lore raw

PATCH v4 4/21 gpu: host1x: Remove cancelled waiters immediately

In reply to Mikko Perttunen

Before this patch, cancelled waiters would only be cleaned up once their threshold value was reached. Make host1x_intr_put_ref process the cancellation immediately to fix this.

---
 drivers/gpu/host1x/intr.c | 14 +++++++++-----
 1 file changed, 9 insertions(+), 5 deletions(-)
drivers/gpu/host1x/intr.c +9 -5
index 9245add23b5d..5d328d20ce6d 100644
@@ -247,13 +247,17 @@ void host1x_intr_put_ref(struct host1x *host, unsigned int id, void *ref)
 	struct host1x_waitlist *waiter = ref;
 	struct host1x_syncpt *syncpt;
 
-	while (atomic_cmpxchg(&waiter->state, WLS_PENDING, WLS_CANCELLED) ==
-	       WLS_REMOVED)
-		schedule();
+	atomic_cmpxchg(&waiter->state, WLS_PENDING, WLS_CANCELLED);
 
 	syncpt = host->syncpt + id;
-	(void)process_wait_list(host, syncpt,
-				host1x_syncpt_load(host->syncpt + id));
+
+	spin_lock(&syncpt->intr.lock);
+	if (atomic_cmpxchg(&waiter->state, WLS_CANCELLED, WLS_HANDLED) ==
+	    WLS_CANCELLED) {
+		list_del(&waiter->list);
+		kref_put(&waiter->refcount, waiter_release);
+	}
+	spin_unlock(&syncpt->intr.lock);
 
 	kref_put(&waiter->refcount, waiter_release);
 }
Signature · 1 line
2.29.2
author maintainer
#5 lore raw

PATCH v4 5/21 gpu: host1x: Use HW-equivalent syncpoint expiration check

In reply to Mikko Perttunen

Make syncpoint expiration checks always use the same logic used by the hardware. This ensures that there are no race conditions that could occur because of the hardware triggering a syncpoint interrupt and then the driver disagreeing.

One situation where this could occur is if a job incremented a syncpoint too many times -- then the hardware would trigger an interrupt, but the driver would assume that a syncpoint value greater than the syncpoint's max value is in the future, and not clean up the job.

---
 drivers/gpu/host1x/syncpt.c | 51 ++-----------------------------------
 1 file changed, 2 insertions(+), 49 deletions(-)
drivers/gpu/host1x/syncpt.c +2 -49
index 5982fdf64e1c..9ca0d852e32f 100644
@@ -306,59 +306,12 @@ EXPORT_SYMBOL(host1x_syncpt_wait);
 bool host1x_syncpt_is_expired(struct host1x_syncpt *sp, u32 thresh)
 {
 	u32 current_val;
-	u32 future_val;
 
 	smp_rmb();
 
 	current_val = (u32)atomic_read(&sp->min_val);
-	future_val = (u32)atomic_read(&sp->max_val);
-
-	/* Note the use of unsigned arithmetic here (mod 1<<32).
-	 *
-	 * c = current_val = min_val	= the current value of the syncpoint.
-	 * t = thresh			= the value we are checking
-	 * f = future_val  = max_val	= the value c will reach when all
-	 *				  outstanding increments have completed.
-	 *
-	 * Note that c always chases f until it reaches f.
-	 *
-	 * Dtf = (f - t)
-	 * Dtc = (c - t)
-	 *
-	 *  Consider all cases:
-	 *
-	 *	A) .....c..t..f.....	Dtf < Dtc	need to wait
-	 *	B) .....c.....f..t..	Dtf > Dtc	expired
-	 *	C) ..t..c.....f.....	Dtf > Dtc	expired	   (Dct very large)
-	 *
-	 *  Any case where f==c: always expired (for any t).	Dtf == Dcf
-	 *  Any case where t==c: always expired (for any f).	Dtf >= Dtc (because Dtc==0)
-	 *  Any case where t==f!=c: always wait.		Dtf <  Dtc (because Dtf==0,
-	 *							Dtc!=0)
-	 *
-	 *  Other cases:
-	 *
-	 *	A) .....t..f..c.....	Dtf < Dtc	need to wait
-	 *	A) .....f..c..t.....	Dtf < Dtc	need to wait
-	 *	A) .....f..t..c.....	Dtf > Dtc	expired
-	 *
-	 *   So:
-	 *	   Dtf >= Dtc implies EXPIRED	(return true)
-	 *	   Dtf <  Dtc implies WAIT	(return false)
-	 *
-	 * Note: If t is expired then we *cannot* wait on it. We would wait
-	 * forever (hang the system).
-	 *
-	 * Note: do NOT get clever and remove the -thresh from both sides. It
-	 * is NOT the same.
-	 *
-	 * If future valueis zero, we have a client managed sync point. In that
-	 * case we do a direct comparison.
-	 */
-	if (!host1x_syncpt_client_managed(sp))
-		return future_val - thresh >= current_val - thresh;
-	else
-		return (s32)(current_val - thresh) >= 0;
+
+	return ((current_val - thresh) & 0x80000000U) == 0U;
 }
 
 int host1x_syncpt_init(struct host1x *host)
Signature · 1 line
2.29.2
author maintainer
#6 lore raw

PATCH v4 6/21 gpu: host1x: Cleanup and refcounting for syncpoints

In reply to Mikko Perttunen

Add reference counting for allocated syncpoints to allow keeping them allocated while jobs are referencing them. Additionally, clean up various places using syncpoint IDs to use host1x_syncpt pointers instead.

---
v4: Update from _free to _put in VI driver as well

---
 drivers/gpu/drm/tegra/dc.c             |  4 +-
 drivers/gpu/drm/tegra/drm.c            | 17 +++---
 drivers/gpu/drm/tegra/gr2d.c           |  4 +-
 drivers/gpu/drm/tegra/gr3d.c           |  4 +-
 drivers/gpu/drm/tegra/vic.c            |  4 +-
 drivers/gpu/host1x/cdma.c              | 11 ++--
 drivers/gpu/host1x/dev.h               |  7 ++-
 drivers/gpu/host1x/hw/cdma_hw.c        |  2 +-
 drivers/gpu/host1x/hw/channel_hw.c     | 10 ++--
 drivers/gpu/host1x/hw/debug_hw.c       |  2 +-
 drivers/gpu/host1x/job.c               |  5 +-
 drivers/gpu/host1x/syncpt.c            | 75 +++++++++++++++++++-------
 drivers/gpu/host1x/syncpt.h            |  3 ++
 drivers/staging/media/tegra-video/vi.c |  8 +--
 include/linux/host1x.h                 |  8 +--
 15 files changed, 103 insertions(+), 61 deletions(-)
drivers/gpu/drm/tegra/dc.c +2 -2
index 9a0b3240bc58..efb41c10dad4 100644
@@ -2127,7 +2127,7 @@ static int tegra_dc_init(struct host1x_client *client)
 		drm_plane_cleanup(primary);
 
 	host1x_client_iommu_detach(client);
-	host1x_syncpt_free(dc->syncpt);
+	host1x_syncpt_put(dc->syncpt);
 
 	return err;
 }
@@ -2152,7 +2152,7 @@ static int tegra_dc_exit(struct host1x_client *client)
 	}
 
 	host1x_client_iommu_detach(client);
-	host1x_syncpt_free(dc->syncpt);
+	host1x_syncpt_put(dc->syncpt);
 
 	return 0;
 }
drivers/gpu/drm/tegra/drm.c +10 -7
index ba9d1c3e7cac..ceea9db341f0 100644
@@ -171,7 +171,7 @@ int tegra_drm_submit(struct tegra_drm_context *context,
 	struct drm_tegra_syncpt syncpt;
 	struct host1x *host1x = dev_get_drvdata(drm->dev->parent);
 	struct drm_gem_object **refs;
-	struct host1x_syncpt *sp;
+	struct host1x_syncpt *sp = NULL;
 	struct host1x_job *job;
 	unsigned int num_refs;
 	int err;
@@ -298,8 +298,8 @@ int tegra_drm_submit(struct tegra_drm_context *context,
 		goto fail;
 	}
 
-	/* check whether syncpoint ID is valid */
-	sp = host1x_syncpt_get(host1x, syncpt.id);
+	/* Syncpoint ref will be dropped on job release. */
+	sp = host1x_syncpt_get_by_id(host1x, syncpt.id);
 	if (!sp) {
 		err = -ENOENT;
 		goto fail;
@@ -308,7 +308,7 @@ int tegra_drm_submit(struct tegra_drm_context *context,
 	job->is_addr_reg = context->client->ops->is_addr_reg;
 	job->is_valid_class = context->client->ops->is_valid_class;
 	job->syncpt_incrs = syncpt.incrs;
-	job->syncpt_id = syncpt.id;
+	job->syncpt = sp;
 	job->timeout = 10000;
 
 	if (args->timeout && args->timeout < 10000)
@@ -327,6 +327,9 @@ int tegra_drm_submit(struct tegra_drm_context *context,
 	args->fence = job->syncpt_end;
 
 fail:
+	if (sp)
+		host1x_syncpt_put(sp);
+
 	while (num_refs--)
 		drm_gem_object_put(refs[num_refs]);
 
@@ -380,7 +383,7 @@ static int tegra_syncpt_read(struct drm_device *drm, void *data,
 	struct drm_tegra_syncpt_read *args = data;
 	struct host1x_syncpt *sp;
 
-	sp = host1x_syncpt_get(host, args->id);
+	sp = host1x_syncpt_get_by_id_noref(host, args->id);
 	if (!sp)
 		return -EINVAL;
 
@@ -395,7 +398,7 @@ static int tegra_syncpt_incr(struct drm_device *drm, void *data,
 	struct drm_tegra_syncpt_incr *args = data;
 	struct host1x_syncpt *sp;
 
-	sp = host1x_syncpt_get(host1x, args->id);
+	sp = host1x_syncpt_get_by_id_noref(host1x, args->id);
 	if (!sp)
 		return -EINVAL;
 
@@ -409,7 +412,7 @@ static int tegra_syncpt_wait(struct drm_device *drm, void *data,
 	struct drm_tegra_syncpt_wait *args = data;
 	struct host1x_syncpt *sp;
 
-	sp = host1x_syncpt_get(host1x, args->id);
+	sp = host1x_syncpt_get_by_id_noref(host1x, args->id);
 	if (!sp)
 		return -EINVAL;
 
drivers/gpu/drm/tegra/gr2d.c +2 -2
index 1a0d3ba6e525..d857a99b21a7 100644
@@ -67,7 +67,7 @@ static int gr2d_init(struct host1x_client *client)
 detach:
 	host1x_client_iommu_detach(client);
 free:
-	host1x_syncpt_free(client->syncpts[0]);
+	host1x_syncpt_put(client->syncpts[0]);
 put:
 	host1x_channel_put(gr2d->channel);
 	return err;
@@ -86,7 +86,7 @@ static int gr2d_exit(struct host1x_client *client)
 		return err;
 
 	host1x_client_iommu_detach(client);
-	host1x_syncpt_free(client->syncpts[0]);
+	host1x_syncpt_put(client->syncpts[0]);
 	host1x_channel_put(gr2d->channel);
 
 	return 0;
drivers/gpu/drm/tegra/gr3d.c +2 -2
index b0b8154e8104..24442ade0da3 100644
@@ -76,7 +76,7 @@ static int gr3d_init(struct host1x_client *client)
 detach:
 	host1x_client_iommu_detach(client);
 free:
-	host1x_syncpt_free(client->syncpts[0]);
+	host1x_syncpt_put(client->syncpts[0]);
 put:
 	host1x_channel_put(gr3d->channel);
 	return err;
@@ -94,7 +94,7 @@ static int gr3d_exit(struct host1x_client *client)
 		return err;
 
 	host1x_client_iommu_detach(client);
-	host1x_syncpt_free(client->syncpts[0]);
+	host1x_syncpt_put(client->syncpts[0]);
 	host1x_channel_put(gr3d->channel);
 
 	return 0;
drivers/gpu/drm/tegra/vic.c +2 -2
index ade56b860cf9..cb476da59adc 100644
@@ -197,7 +197,7 @@ static int vic_init(struct host1x_client *client)
 	return 0;
 
 free_syncpt:
-	host1x_syncpt_free(client->syncpts[0]);
+	host1x_syncpt_put(client->syncpts[0]);
 free_channel:
 	host1x_channel_put(vic->channel);
 detach:
@@ -221,7 +221,7 @@ static int vic_exit(struct host1x_client *client)
 	if (err < 0)
 		return err;
 
-	host1x_syncpt_free(client->syncpts[0]);
+	host1x_syncpt_put(client->syncpts[0]);
 	host1x_channel_put(vic->channel);
 	host1x_client_iommu_detach(client);
 
drivers/gpu/host1x/cdma.c +3 -8
index e8d3fda91d8a..6e6ca774f68d 100644
@@ -273,15 +273,13 @@ static int host1x_cdma_wait_pushbuffer_space(struct host1x *host1x,
 static void cdma_start_timer_locked(struct host1x_cdma *cdma,
 				    struct host1x_job *job)
 {
-	struct host1x *host = cdma_to_host1x(cdma);
-
 	if (cdma->timeout.client) {
 		/* timer already started */
 		return;
 	}
 
 	cdma->timeout.client = job->client;
-	cdma->timeout.syncpt = host1x_syncpt_get(host, job->syncpt_id);
+	cdma->timeout.syncpt = job->syncpt;
 	cdma->timeout.syncpt_val = job->syncpt_end;
 	cdma->timeout.start_ktime = ktime_get();
 
@@ -312,7 +310,6 @@ static void stop_cdma_timer_locked(struct host1x_cdma *cdma)
 static void update_cdma_locked(struct host1x_cdma *cdma)
 {
 	bool signal = false;
-	struct host1x *host1x = cdma_to_host1x(cdma);
 	struct host1x_job *job, *n;
 
 	/* If CDMA is stopped, queue is cleared and we can return */
@@ -324,8 +321,7 @@ static void update_cdma_locked(struct host1x_cdma *cdma)
 	 * to consume as many sync queue entries as possible without blocking
 	 */
 	list_for_each_entry_safe(job, n, &cdma->sync_queue, list) {
-		struct host1x_syncpt *sp =
-			host1x_syncpt_get(host1x, job->syncpt_id);
+		struct host1x_syncpt *sp = job->syncpt;
 
 		/* Check whether this syncpt has completed, and bail if not */
 		if (!host1x_syncpt_is_expired(sp, job->syncpt_end)) {
@@ -499,8 +495,7 @@ int host1x_cdma_begin(struct host1x_cdma *cdma, struct host1x_job *job)
 		if (!cdma->timeout.initialized) {
 			int err;
 
-			err = host1x_hw_cdma_timeout_init(host1x, cdma,
-							  job->syncpt_id);
+			err = host1x_hw_cdma_timeout_init(host1x, cdma);
 			if (err) {
 				mutex_unlock(&cdma->lock);
 				return err;
drivers/gpu/host1x/dev.h +3 -4
index f781a9b0f39d..63010ae37a97 100644
@@ -37,7 +37,7 @@ struct host1x_cdma_ops {
 	void (*start)(struct host1x_cdma *cdma);
 	void (*stop)(struct host1x_cdma *cdma);
 	void (*flush)(struct  host1x_cdma *cdma);
-	int (*timeout_init)(struct host1x_cdma *cdma, unsigned int syncpt);
+	int (*timeout_init)(struct host1x_cdma *cdma);
 	void (*timeout_destroy)(struct host1x_cdma *cdma);
 	void (*freeze)(struct host1x_cdma *cdma);
 	void (*resume)(struct host1x_cdma *cdma, u32 getptr);
@@ -261,10 +261,9 @@ static inline void host1x_hw_cdma_flush(struct host1x *host,
 }
 
 static inline int host1x_hw_cdma_timeout_init(struct host1x *host,
-					      struct host1x_cdma *cdma,
-					      unsigned int syncpt)
+					      struct host1x_cdma *cdma)
 {
-	return host->cdma_op->timeout_init(cdma, syncpt);
+	return host->cdma_op->timeout_init(cdma);
 }
 
 static inline void host1x_hw_cdma_timeout_destroy(struct host1x *host,
drivers/gpu/host1x/hw/cdma_hw.c +1 -1
index 2f3bf94cf365..e49cd5b8f735 100644
@@ -295,7 +295,7 @@ static void cdma_timeout_handler(struct work_struct *work)
 /*
  * Init timeout resources
  */
-static int cdma_timeout_init(struct host1x_cdma *cdma, unsigned int syncpt)
+static int cdma_timeout_init(struct host1x_cdma *cdma)
 {
 	INIT_DELAYED_WORK(&cdma->timeout.wq, cdma_timeout_handler);
 	cdma->timeout.initialized = true;
drivers/gpu/host1x/hw/channel_hw.c +4 -6
index 5eaa29d171c9..d4c28faf27d1 100644
@@ -86,8 +86,7 @@ static void submit_gathers(struct host1x_job *job)
 
 static inline void synchronize_syncpt_base(struct host1x_job *job)
 {
-	struct host1x *host = dev_get_drvdata(job->channel->dev->parent);
-	struct host1x_syncpt *sp = host->syncpt + job->syncpt_id;
+	struct host1x_syncpt *sp = job->syncpt;
 	unsigned int id;
 	u32 value;
 
@@ -118,7 +117,7 @@ static void host1x_channel_set_streamid(struct host1x_channel *channel)
 static int channel_submit(struct host1x_job *job)
 {
 	struct host1x_channel *ch = job->channel;
-	struct host1x_syncpt *sp;
+	struct host1x_syncpt *sp = job->syncpt;
 	u32 user_syncpt_incrs = job->syncpt_incrs;
 	u32 prev_max = 0;
 	u32 syncval;
@@ -126,10 +125,9 @@ static int channel_submit(struct host1x_job *job)
 	struct host1x_waitlist *completed_waiter = NULL;
 	struct host1x *host = dev_get_drvdata(ch->dev->parent);
 
-	sp = host->syncpt + job->syncpt_id;
 	trace_host1x_channel_submit(dev_name(ch->dev),
 				    job->num_gathers, job->num_relocs,
-				    job->syncpt_id, job->syncpt_incrs);
+				    job->syncpt->id, job->syncpt_incrs);
 
 	/* before error checks, return current max */
 	prev_max = job->syncpt_end = host1x_syncpt_read_max(sp);
@@ -163,7 +161,7 @@ static int channel_submit(struct host1x_job *job)
 		host1x_cdma_push(&ch->cdma,
 				 host1x_opcode_setclass(HOST1X_CLASS_HOST1X,
 					host1x_uclass_wait_syncpt_r(), 1),
-				 host1x_class_host_wait_syncpt(job->syncpt_id,
+				 host1x_class_host_wait_syncpt(job->syncpt->id,
 					host1x_syncpt_read_max(sp)));
 	}
 
drivers/gpu/host1x/hw/debug_hw.c +1 -1
index f31bcfa1b837..ceb48229d14b 100644
@@ -204,7 +204,7 @@ static void show_channel_gathers(struct output *o, struct host1x_cdma *cdma)
 		unsigned int i;
 
 		host1x_debug_output(o, "\n%p: JOB, syncpt_id=%d, syncpt_val=%d, first_get=%08x, timeout=%d num_slots=%d, num_handles=%d\n",
-				    job, job->syncpt_id, job->syncpt_end,
+				    job, job->syncpt->id, job->syncpt_end,
 				    job->first_get, job->timeout,
 				    job->num_slots, job->num_unpins);
 
drivers/gpu/host1x/job.c +4 -1
index 89b6c14b7392..d8345d3bf0b3 100644
@@ -79,6 +79,9 @@ static void job_free(struct kref *ref)
 {
 	struct host1x_job *job = container_of(ref, struct host1x_job, ref);
 
+	if (job->syncpt)
+		host1x_syncpt_put(job->syncpt);
+
 	kfree(job);
 }
 
@@ -680,7 +683,7 @@ EXPORT_SYMBOL(host1x_job_unpin);
  */
 void host1x_job_dump(struct device *dev, struct host1x_job *job)
 {
-	dev_dbg(dev, "    SYNCPT_ID   %d\n", job->syncpt_id);
+	dev_dbg(dev, "    SYNCPT_ID   %d\n", job->syncpt->id);
 	dev_dbg(dev, "    SYNCPT_VAL  %d\n", job->syncpt_end);
 	dev_dbg(dev, "    FIRST_GET   0x%x\n", job->first_get);
 	dev_dbg(dev, "    TIMEOUT     %d\n", job->timeout);
drivers/gpu/host1x/syncpt.c +57 -18

113 lines not on this page · raw message

drivers/gpu/host1x/syncpt.h +3 -0
index 3aa6b25b1b9c..a6766f8d55ee 100644
@@ -11,6 +11,7 @@
 #include <linux/atomic.h>
 #include <linux/host1x.h>
 #include <linux/kernel.h>
+#include <linux/kref.h>
 #include <linux/sched.h>
 
 #include "intr.h"
@@ -26,6 +27,8 @@ struct host1x_syncpt_base {
 };
 
 struct host1x_syncpt {
+	struct kref ref;
+
 	unsigned int id;
 	atomic_t min_val;
 	atomic_t max_val;
drivers/staging/media/tegra-video/vi.c +4 -4
index 1b5e660155f5..f52377e4959e 100644
@@ -618,8 +618,8 @@ static void tegra_channel_cleanup(struct tegra_vi_channel *chan)
 {
 	v4l2_ctrl_handler_free(&chan->ctrl_handler);
 	media_entity_cleanup(&chan->video.entity);
-	host1x_syncpt_free(chan->mw_ack_sp);
-	host1x_syncpt_free(chan->frame_start_sp);
+	host1x_syncpt_put(chan->mw_ack_sp);
+	host1x_syncpt_put(chan->frame_start_sp);
 	mutex_destroy(&chan->video_lock);
 }
 
@@ -733,9 +733,9 @@ static int tegra_channel_init(struct tegra_vi_channel *chan)
 cleanup_media:
 	media_entity_cleanup(&chan->video.entity);
 free_mw_ack_syncpt:
-	host1x_syncpt_free(chan->mw_ack_sp);
+	host1x_syncpt_put(chan->mw_ack_sp);
 free_fs_syncpt:
-	host1x_syncpt_free(chan->frame_start_sp);
+	host1x_syncpt_put(chan->frame_start_sp);
 	return ret;
 }
 
include/linux/host1x.h +5 -3
index 099eff8a06d2..c222cf61b70b 100644
@@ -142,7 +142,9 @@ struct host1x_syncpt_base;
 struct host1x_syncpt;
 struct host1x;
 
-struct host1x_syncpt *host1x_syncpt_get(struct host1x *host, u32 id);
+struct host1x_syncpt *host1x_syncpt_get_by_id(struct host1x *host, u32 id);
+struct host1x_syncpt *host1x_syncpt_get_by_id_noref(struct host1x *host, u32 id);
+struct host1x_syncpt *host1x_syncpt_get(struct host1x_syncpt *sp);
 u32 host1x_syncpt_id(struct host1x_syncpt *sp);
 u32 host1x_syncpt_read_min(struct host1x_syncpt *sp);
 u32 host1x_syncpt_read_max(struct host1x_syncpt *sp);
@@ -153,7 +155,7 @@ int host1x_syncpt_wait(struct host1x_syncpt *sp, u32 thresh, long timeout,
 		       u32 *value);
 struct host1x_syncpt *host1x_syncpt_request(struct host1x_client *client,
 					    unsigned long flags);
-void host1x_syncpt_free(struct host1x_syncpt *sp);
+void host1x_syncpt_put(struct host1x_syncpt *sp);
 struct host1x_syncpt *host1x_syncpt_alloc(struct host1x *host,
 					  unsigned long flags,
 					  const char *name);
@@ -221,7 +223,7 @@ struct host1x_job {
 	dma_addr_t *reloc_addr_phys;
 
 	/* Sync point id, number of increments and end related to the submit */
-	u32 syncpt_id;
+	struct host1x_syncpt *syncpt;
 	u32 syncpt_incrs;
 	u32 syncpt_end;
 
Signature · 1 line
2.29.2
author maintainer
#7 lore raw

PATCH v4 7/21 gpu: host1x: Introduce UAPI header

In reply to Mikko Perttunen

Add the userspace interface header, specifying interfaces for allocating and accessing syncpoints from userspace, and for creating sync_file based fences based on syncpoint thresholds.

---
 include/uapi/linux/host1x.h | 134 ++++++++++++++++++++++++++++++++++++
 1 file changed, 134 insertions(+)
 create mode 100644 include/uapi/linux/host1x.h
include/uapi/linux/host1x.h new file +134 -0
new file mode 100644
index 000000000000..9c8fb9425cb2
@@ -0,0 +1,134 @@
+/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
+/* Copyright (c) 2020 NVIDIA Corporation */
+
+#ifndef _UAPI__LINUX_HOST1X_H
+#define _UAPI__LINUX_HOST1X_H
+
+#include <linux/ioctl.h>
+#include <linux/types.h>
+
+#if defined(__cplusplus)
+extern "C" {
+#endif
+
+struct host1x_allocate_syncpoint {
+	/**
+	 * @fd: [out]
+	 *
+	 * New file descriptor representing the allocated syncpoint.
+	 */
+	__s32 fd;
+
+	__u32 reserved[3];
+};
+
+struct host1x_syncpoint_info {
+	/**
+	 * @id: [out]
+	 *
+	 * System-global ID of the syncpoint.
+	 */
+	__u32 id;
+
+	__u32 reserved[3];
+};
+
+struct host1x_syncpoint_increment {
+	/**
+	 * @count: [in]
+	 *
+	 * Number of times to increment the syncpoint. The syncpoint can
+	 * be observed at in-between values, but each increment is atomic.
+	 */
+	__u32 count;
+};
+
+struct host1x_read_syncpoint {
+	/**
+	 * @id: [in]
+	 *
+	 * ID of the syncpoint to read.
+	 */
+	__u32 id;
+
+	/**
+	 * @value: [out]
+	 *
+	 * Current value of the syncpoint.
+	 */
+	__u32 value;
+};
+
+struct host1x_create_fence {
+	/**
+	 * @id: [in]
+	 *
+	 * ID of the syncpoint to create a fence for.
+	 */
+	__u32 id;
+
+	/**
+	 * @threshold: [in]
+	 *
+	 * When the syncpoint reaches this value, the fence will be signaled.
+	 * The syncpoint is considered to have reached the threshold when the
+	 * following condition is true:
+	 *
+	 * 	((value - threshold) & 0x80000000U) == 0U
+	 *
+	 */
+	__u32 threshold;
+
+	/**
+	 * @fence_fd: [out]
+	 *
+	 * New sync_file file descriptor containing the created fence.
+	 */
+	__s32 fence_fd;
+
+	__u32 reserved[1];
+};
+
+struct host1x_fence_extract_fence {
+	__u32 id;
+	__u32 threshold;
+};
+
+struct host1x_fence_extract {
+	/**
+	 * @fence_fd: [in]
+	 *
+	 * sync_file file descriptor
+	 */
+	__s32 fence_fd;
+
+	/**
+	 * @num_fences: [in,out]
+	 *
+	 * In: size of the `fences_ptr` array counted in elements.
+	 * Out: required size of the `fences_ptr` array counted in elements.
+	 */
+	__u32 num_fences;
+
+	/**
+	 * @fences_ptr: [in]
+	 *
+	 * Pointer to array of `struct host1x_fence_extract_fence`.
+	 */
+	__u64 fences_ptr;
+
+	__u32 reserved[2];
+};
+
+#define HOST1X_IOCTL_ALLOCATE_SYNCPOINT  _IOWR('X', 0x00, struct host1x_allocate_syncpoint)
+#define HOST1X_IOCTL_READ_SYNCPOINT      _IOR ('X', 0x01, struct host1x_read_syncpoint)
+#define HOST1X_IOCTL_CREATE_FENCE        _IOWR('X', 0x02, struct host1x_create_fence)
+#define HOST1X_IOCTL_SYNCPOINT_INFO      _IOWR('X', 0x03, struct host1x_syncpoint_info)
+#define HOST1X_IOCTL_SYNCPOINT_INCREMENT _IOWR('X', 0x04, struct host1x_syncpoint_increment)
+#define HOST1X_IOCTL_FENCE_EXTRACT       _IOWR('X', 0x05, struct host1x_fence_extract)
+
+#if defined(__cplusplus)
+}
+#endif
+
+#endif
Signature · 1 line
2.29.2
author maintainer
#8 lore raw

PATCH v4 8/21 gpu: host1x: Implement /dev/host1x device node

In reply to Mikko Perttunen

Add the /dev/host1x device node, implementing the following functionality:

- Reading syncpoint values
- Allocating syncpoints (providing syncpoint FDs)
- Incrementing syncpoints (based on syncpoint FD)

---
v4:
* Put UAPI under CONFIG_DRM_TEGRA_STAGING
v3:
* Pass process name as syncpoint name when allocating syncpoint.

---
 drivers/gpu/host1x/Makefile |   1 +
 drivers/gpu/host1x/dev.c    |   9 ++
 drivers/gpu/host1x/dev.h    |   3 +
 drivers/gpu/host1x/uapi.c   | 282 ++++++++++++++++++++++++++++++++++++
 drivers/gpu/host1x/uapi.h   |  22 +++
 include/linux/host1x.h      |   2 +
 6 files changed, 319 insertions(+)
 create mode 100644 drivers/gpu/host1x/uapi.c
 create mode 100644 drivers/gpu/host1x/uapi.h
drivers/gpu/host1x/Makefile +1 -0
index 096017b8789d..882f928d75e1 100644
@@ -9,6 +9,7 @@ host1x-y = \
 	job.o \
 	debug.o \
 	mipi.o \
+	uapi.o \
 	hw/host1x01.o \
 	hw/host1x02.o \
 	hw/host1x04.o \
drivers/gpu/host1x/dev.c +9 -0
index d0ebb70e2fdd..641317d23828 100644
@@ -461,6 +461,12 @@ static int host1x_probe(struct platform_device *pdev)
 		goto deinit_syncpt;
 	}
 
+	err = host1x_uapi_init(&host->uapi, host);
+	if (err) {
+		dev_err(&pdev->dev, "failed to initialize uapi\n");
+		goto deinit_intr;
+	}
+
 	host1x_debug_init(host);
 
 	if (host->info->has_hypervisor)
@@ -480,6 +486,8 @@ static int host1x_probe(struct platform_device *pdev)
 	host1x_unregister(host);
 deinit_debugfs:
 	host1x_debug_deinit(host);
+	host1x_uapi_deinit(&host->uapi);
+deinit_intr:
 	host1x_intr_deinit(host);
 deinit_syncpt:
 	host1x_syncpt_deinit(host);
@@ -501,6 +509,7 @@ static int host1x_remove(struct platform_device *pdev)
 
 	host1x_unregister(host);
 	host1x_debug_deinit(host);
+	host1x_uapi_deinit(&host->uapi);
 	host1x_intr_deinit(host);
 	host1x_syncpt_deinit(host);
 	reset_control_assert(host->rst);
drivers/gpu/host1x/dev.h +3 -0
index 63010ae37a97..7b8b7e20e32b 100644
@@ -17,6 +17,7 @@
 #include "intr.h"
 #include "job.h"
 #include "syncpt.h"
+#include "uapi.h"
 
 struct host1x_syncpt;
 struct host1x_syncpt_base;
@@ -143,6 +144,8 @@ struct host1x {
 	struct list_head list;
 
 	struct device_dma_parameters dma_parms;
+
+	struct host1x_uapi uapi;
 };
 
 void host1x_hypervisor_writel(struct host1x *host1x, u32 r, u32 v);
drivers/gpu/host1x/uapi.c new file +282 -0

282 lines not on this page · raw message

drivers/gpu/host1x/uapi.h new file +22 -0
new file mode 100644
index 000000000000..7beb5e44c1b1
@@ -0,0 +1,22 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2020, NVIDIA Corporation.
+ */
+
+#ifndef HOST1X_UAPI_H
+#define HOST1X_UAPI_H
+
+#include <linux/cdev.h>
+
+struct host1x_uapi {
+	struct class *class;
+
+	struct cdev cdev;
+	struct device *dev;
+	dev_t dev_num;
+};
+
+int host1x_uapi_init(struct host1x_uapi *uapi, struct host1x *host1x);
+void host1x_uapi_deinit(struct host1x_uapi *uapi);
+
+#endif
include/linux/host1x.h +2 -0
index c222cf61b70b..8f6d61e13e05 100644
@@ -163,6 +163,8 @@ struct host1x_syncpt *host1x_syncpt_alloc(struct host1x *host,
 struct host1x_syncpt_base *host1x_syncpt_get_base(struct host1x_syncpt *sp);
 u32 host1x_syncpt_base_id(struct host1x_syncpt_base *base);
 
+struct host1x_syncpt *host1x_syncpt_fd_get(int fd);
+
 /*
  * host1x channel
  */
Signature · 1 line
2.29.2
author maintainer
#9 lore raw

PATCH v4 9/21 gpu: host1x: DMA fences and userspace fence creation

In reply to Mikko Perttunen

Add an implementation of dma_fences based on syncpoints. Syncpoint interrupts are used to signal fences. Additionally, after software signaling has been enabled, a 30 second timeout is started. If the syncpoint threshold is not reached within this period, the fence is signalled with an -ETIMEDOUT error code. This is to allow fences that would never reach their syncpoint threshold to be cleaned up.

Additionally, add a new /dev/host1x IOCTL for creating sync_file file descriptors backed by syncpoint fences.

---
v4:
* Fix _signal prototype and include it to avoid warning
* Remove use of unused local in error path
v3:
* Move declaration of host1x_fence_extract to public header

---
 drivers/gpu/host1x/Makefile |   1 +
 drivers/gpu/host1x/fence.c  | 208 ++++++++++++++++++++++++++++++++++++
 drivers/gpu/host1x/fence.h  |  13 +++
 drivers/gpu/host1x/intr.c   |   9 ++
 drivers/gpu/host1x/intr.h   |   2 +
 drivers/gpu/host1x/uapi.c   | 103 ++++++++++++++++++
 include/linux/host1x.h      |   4 +
 7 files changed, 340 insertions(+)
 create mode 100644 drivers/gpu/host1x/fence.c
 create mode 100644 drivers/gpu/host1x/fence.h
drivers/gpu/host1x/Makefile +1 -0
index 882f928d75e1..a48af2cefae1 100644
@@ -10,6 +10,7 @@ host1x-y = \
 	debug.o \
 	mipi.o \
 	uapi.o \
+	fence.o \
 	hw/host1x01.o \
 	hw/host1x02.o \
 	hw/host1x04.o \
drivers/gpu/host1x/fence.c new file +208 -0

208 lines not on this page · raw message

drivers/gpu/host1x/fence.h new file +13 -0
new file mode 100644
index 000000000000..70c91de82f14
@@ -0,0 +1,13 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2020, NVIDIA Corporation.
+ */
+
+#ifndef HOST1X_FENCE_H
+#define HOST1X_FENCE_H
+
+struct host1x_syncpt_fence;
+
+void host1x_fence_signal(struct host1x_syncpt_fence *fence);
+
+#endif
drivers/gpu/host1x/intr.c +9 -0
index 5d328d20ce6d..19b59c5c94d0 100644
@@ -13,6 +13,7 @@
 #include <trace/events/host1x.h>
 #include "channel.h"
 #include "dev.h"
+#include "fence.h"
 #include "intr.h"
 
 /* Wait list management */
@@ -121,12 +122,20 @@ static void action_wakeup_interruptible(struct host1x_waitlist *waiter)
 	wake_up_interruptible(wq);
 }
 
+static void action_signal_fence(struct host1x_waitlist *waiter)
+{
+	struct host1x_syncpt_fence *f = waiter->data;
+
+	host1x_fence_signal(f);
+}
+
 typedef void (*action_handler)(struct host1x_waitlist *waiter);
 
 static const action_handler action_handlers[HOST1X_INTR_ACTION_COUNT] = {
 	action_submit_complete,
 	action_wakeup,
 	action_wakeup_interruptible,
+	action_signal_fence,
 };
 
 static void run_handlers(struct list_head completed[HOST1X_INTR_ACTION_COUNT])
drivers/gpu/host1x/intr.h +2 -0
index aac38194398f..dedbd0f700fb 100644
@@ -33,6 +33,8 @@ enum host1x_intr_action {
 	 */
 	HOST1X_INTR_ACTION_WAKEUP_INTERRUPTIBLE,
 
+	HOST1X_INTR_ACTION_SIGNAL_FENCE,
+
 	HOST1X_INTR_ACTION_COUNT
 };
 
drivers/gpu/host1x/uapi.c +103 -0

123 lines not on this page · raw message

include/linux/host1x.h +4 -0
index 8f6d61e13e05..bcfb1cc9a1c1 100644
@@ -165,6 +165,10 @@ u32 host1x_syncpt_base_id(struct host1x_syncpt_base *base);
 
 struct host1x_syncpt *host1x_syncpt_fd_get(int fd);
 
+struct dma_fence *host1x_fence_create(struct host1x_syncpt *sp, u32 threshold);
+int host1x_fence_create_fd(struct host1x_syncpt *sp, u32 threshold);
+int host1x_fence_extract(struct dma_fence *fence, u32 *id, u32 *threshold);
+
 /*
  * host1x channel
  */
Signature · 1 line
2.29.2
author maintainer
#10 lore raw

PATCH v4 10/21 gpu: host1x: Add no-recovery mode

In reply to Mikko Perttunen

Add a new property for jobs to enable or disable recovery i.e. CPU increments of syncpoints to max value on job timeout. This allows for a more solid model for hanged jobs, where userspace doesn't need to guess if a syncpoint increment happened because the job completed, or because job timeout was triggered.

On job timeout, we stop the channel, NOP all future jobs on the channel using the same syncpoint, mark the syncpoint as locked and resume the channel from the next job, if any.

The future jobs are NOPed, since because we don't do the CPU increments, the value of the syncpoint is no longer synchronized, and any waiters would become confused if a future job incremented the syncpoint. The syncpoint is marked locked to ensure that any future jobs cannot increment the syncpoint either, until the application has recognized the situation and reallocated the syncpoint.

---
v3:
* Move 'locked' check inside CDMA lock to prevent race
* Add clarifying comment to NOP-patching code

---
 drivers/gpu/drm/tegra/drm.c        |  1 +
 drivers/gpu/host1x/cdma.c          | 58 ++++++++++++++++++++++++++----
 drivers/gpu/host1x/hw/channel_hw.c |  2 +-
 drivers/gpu/host1x/job.c           |  4 +++
 drivers/gpu/host1x/syncpt.c        |  2 ++
 drivers/gpu/host1x/syncpt.h        | 12 +++++++
 include/linux/host1x.h             |  9 +++++
 7 files changed, 81 insertions(+), 7 deletions(-)
drivers/gpu/drm/tegra/drm.c +1 -0
index ceea9db341f0..7437c67924aa 100644
@@ -197,6 +197,7 @@ int tegra_drm_submit(struct tegra_drm_context *context,
 	job->client = client;
 	job->class = client->class;
 	job->serialize = true;
+	job->syncpt_recovery = true;
 
 	/*
 	 * Track referenced BOs so that they can be unreferenced after the
drivers/gpu/host1x/cdma.c +52 -6

91 lines not on this page · raw message

drivers/gpu/host1x/hw/channel_hw.c +1 -1
index d4c28faf27d1..bf21512e5078 100644
@@ -191,7 +191,7 @@ static int channel_submit(struct host1x_job *job)
 	/* schedule a submit complete interrupt */
 	err = host1x_intr_add_action(host, sp, syncval,
 				     HOST1X_INTR_ACTION_SUBMIT_COMPLETE, ch,
-				     completed_waiter, NULL);
+				     completed_waiter, &job->waiter);
 	completed_waiter = NULL;
 	WARN(err, "Failed to set submit complete interrupt");
 
drivers/gpu/host1x/job.c +4 -0
index d8345d3bf0b3..e4f16fc899b0 100644
@@ -79,6 +79,10 @@ static void job_free(struct kref *ref)
 {
 	struct host1x_job *job = container_of(ref, struct host1x_job, ref);
 
+	if (job->waiter)
+		host1x_intr_put_ref(job->syncpt->host, job->syncpt->id,
+				    job->waiter);
+
 	if (job->syncpt)
 		host1x_syncpt_put(job->syncpt);
 
drivers/gpu/host1x/syncpt.c +2 -0
index c7b910e413d8..8d658e5f7db2 100644
@@ -385,6 +385,8 @@ static void syncpt_release(struct kref *ref)
 {
 	struct host1x_syncpt *sp = container_of(ref, struct host1x_syncpt, ref);
 
+	sp->locked = false;
+
 	mutex_lock(&sp->host->syncpt_mutex);
 
 	host1x_syncpt_base_free(sp->base);
drivers/gpu/host1x/syncpt.h +12 -0
index a6766f8d55ee..93e894677d89 100644
@@ -40,6 +40,13 @@ struct host1x_syncpt {
 
 	/* interrupt data */
 	struct host1x_syncpt_intr intr;
+
+	/* 
+	 * If a submission incrementing this syncpoint fails, lock it so that
+	 * further submission cannot be made until application has handled the
+	 * failure.
+	 */
+	bool locked;
 };
 
 /* Initialize sync point array  */
@@ -115,4 +122,9 @@ static inline int host1x_syncpt_is_valid(struct host1x_syncpt *sp)
 	return sp->id < host1x_syncpt_nb_pts(sp->host);
 }
 
+static inline void host1x_syncpt_set_locked(struct host1x_syncpt *sp)
+{
+	sp->locked = true;
+}
+
 #endif
include/linux/host1x.h +9 -0
index bcfb1cc9a1c1..fb62cc8b77dd 100644
@@ -233,9 +233,15 @@ struct host1x_job {
 	u32 syncpt_incrs;
 	u32 syncpt_end;
 
+	/* Completion waiter ref */
+	void *waiter;
+
 	/* Maximum time to wait for this job */
 	unsigned int timeout;
 
+	/* Job has timed out and should be released */
+	bool cancelled;
+
 	/* Index and number of slots used in the push buffer */
 	unsigned int first_get;
 	unsigned int num_slots;
@@ -256,6 +262,9 @@ struct host1x_job {
 
 	/* Add a channel wait for previous ops to complete */
 	bool serialize;
+
+	/* Fast-forward syncpoint increments on job timeout */
+	bool syncpt_recovery;
 };
 
 struct host1x_job *host1x_job_alloc(struct host1x_channel *ch,
Signature · 1 line
2.29.2
author maintainer
#11 lore raw

PATCH v4 11/21 gpu: host1x: Add job release callback

In reply to Mikko Perttunen

Add a callback field to the job structure, to be called just before the job is to be freed. This allows the job's submitter to clean up any of its own state, like decrement runtime PM refcounts.

---
 drivers/gpu/host1x/job.c | 3 +++
 include/linux/host1x.h   | 4 ++++
 2 files changed, 7 insertions(+)
drivers/gpu/host1x/job.c +3 -0
index e4f16fc899b0..acf322beb56c 100644
@@ -79,6 +79,9 @@ static void job_free(struct kref *ref)
 {
 	struct host1x_job *job = container_of(ref, struct host1x_job, ref);
 
+	if (job->release)
+		job->release(job);
+
 	if (job->waiter)
 		host1x_intr_put_ref(job->syncpt->host, job->syncpt->id,
 				    job->waiter);
include/linux/host1x.h +4 -0
index fb62cc8b77dd..d7070fd65833 100644
@@ -265,6 +265,10 @@ struct host1x_job {
 
 	/* Fast-forward syncpoint increments on job timeout */
 	bool syncpt_recovery;
+
+	/* Callback called when job is freed */
+	void (*release)(struct host1x_job *job);
+	void *user_data;
 };
 
 struct host1x_job *host1x_job_alloc(struct host1x_channel *ch,
Signature · 1 line
2.29.2
author maintainer
#12 lore raw

PATCH v4 12/21 gpu: host1x: Add support for syncpoint waits in CDMA pushbuffer

In reply to Mikko Perttunen

Add support for inserting syncpoint waits in the CDMA pushbuffer. These waits need to be done in HOST1X class, while gather submitted by the application execute in engine class.

Support is added by converting the gather list of job into a command list that can include both gathers and waits. When the job is submitted, these commands are pushed as the appropriate opcodes on the CDMA pushbuffer.

---
 drivers/gpu/host1x/hw/channel_hw.c | 51 +++++++++++++++--------
 drivers/gpu/host1x/hw/debug_hw.c   |  9 +++-
 drivers/gpu/host1x/job.c           | 67 +++++++++++++++++++++---------
 drivers/gpu/host1x/job.h           | 14 +++++++
 include/linux/host1x.h             |  5 ++-
 5 files changed, 105 insertions(+), 41 deletions(-)
drivers/gpu/host1x/hw/channel_hw.c +33 -18
index bf21512e5078..d88a32f73f5e 100644
@@ -55,31 +55,46 @@ static void submit_gathers(struct host1x_job *job)
 #endif
 	unsigned int i;
 
-	for (i = 0; i < job->num_gathers; i++) {
-		struct host1x_job_gather *g = &job->gathers[i];
-		dma_addr_t addr = g->base + g->offset;
-		u32 op2, op3;
+	for (i = 0; i < job->num_cmds; i++) {
+		struct host1x_job_cmd *cmd = &job->cmds[i];
 
-		op2 = lower_32_bits(addr);
-		op3 = upper_32_bits(addr);
+		if (cmd->is_wait) {
+			/* TODO use modern wait */
+			host1x_cdma_push(cdma,
+				 host1x_opcode_setclass(HOST1X_CLASS_HOST1X,
+					host1x_uclass_wait_syncpt_r(), 1),
+				 host1x_class_host_wait_syncpt(cmd->wait.id,
+					cmd->wait.threshold));
+			host1x_cdma_push(
+				cdma, host1x_opcode_setclass(job->class, 0, 0),
+				HOST1X_OPCODE_NOP);
+		} else {
+			struct host1x_job_gather *g = &cmd->gather;
 
-		trace_write_gather(cdma, g->bo, g->offset, g->words);
+			dma_addr_t addr = g->base + g->offset;
+			u32 op2, op3;
 
-		if (op3 != 0) {
+			op2 = lower_32_bits(addr);
+			op3 = upper_32_bits(addr);
+
+			trace_write_gather(cdma, g->bo, g->offset, g->words);
+
+			if (op3 != 0) {
 #if HOST1X_HW >= 6
-			u32 op1 = host1x_opcode_gather_wide(g->words);
-			u32 op4 = HOST1X_OPCODE_NOP;
+				u32 op1 = host1x_opcode_gather_wide(g->words);
+				u32 op4 = HOST1X_OPCODE_NOP;
 
-			host1x_cdma_push_wide(cdma, op1, op2, op3, op4);
+				host1x_cdma_push_wide(cdma, op1, op2, op3, op4);
 #else
-			dev_err(dev, "invalid gather for push buffer %pad\n",
-				&addr);
-			continue;
+				dev_err(dev, "invalid gather for push buffer %pad\n",
+					&addr);
+				continue;
 #endif
-		} else {
-			u32 op1 = host1x_opcode_gather(g->words);
+			} else {
+				u32 op1 = host1x_opcode_gather(g->words);
 
-			host1x_cdma_push(cdma, op1, op2);
+				host1x_cdma_push(cdma, op1, op2);
+			}
 		}
 	}
 }
@@ -126,7 +141,7 @@ static int channel_submit(struct host1x_job *job)
 	struct host1x *host = dev_get_drvdata(ch->dev->parent);
 
 	trace_host1x_channel_submit(dev_name(ch->dev),
-				    job->num_gathers, job->num_relocs,
+				    job->num_cmds, job->num_relocs,
 				    job->syncpt->id, job->syncpt_incrs);
 
 	/* before error checks, return current max */
drivers/gpu/host1x/hw/debug_hw.c +7 -2
index ceb48229d14b..35952fd5597e 100644
@@ -208,10 +208,15 @@ static void show_channel_gathers(struct output *o, struct host1x_cdma *cdma)
 				    job->first_get, job->timeout,
 				    job->num_slots, job->num_unpins);
 
-		for (i = 0; i < job->num_gathers; i++) {
-			struct host1x_job_gather *g = &job->gathers[i];
+		for (i = 0; i < job->num_cmds; i++) {
+			struct host1x_job_gather *g;
 			u32 *mapped;
 
+			if (job->cmds[i].is_wait)
+				continue;
+
+			g = &job->cmds[i].gather;
+
 			if (job->gather_copy_mapped)
 				mapped = (u32 *)job->gather_copy_mapped;
 			else
drivers/gpu/host1x/job.c +48 -19

145 lines not on this page · raw message

drivers/gpu/host1x/job.h +14 -0
index 94bc2e4ae241..33adfaede842 100644
@@ -18,6 +18,20 @@ struct host1x_job_gather {
 	bool handled;
 };
 
+struct host1x_job_wait {
+	u32 id;
+	u32 threshold;
+};
+
+struct host1x_job_cmd {
+	bool is_wait;
+
+	union {
+		struct host1x_job_gather gather;
+		struct host1x_job_wait wait;
+	};
+};
+
 struct host1x_job_unpin_data {
 	struct host1x_bo *bo;
 	struct sg_table *sgt;
include/linux/host1x.h +3 -2
index d7070fd65833..59296d3346fe 100644
@@ -215,8 +215,8 @@ struct host1x_job {
 	struct host1x_client *client;
 
 	/* Gathers and their memory */
-	struct host1x_job_gather *gathers;
-	unsigned int num_gathers;
+	struct host1x_job_cmd *cmds;
+	unsigned int num_cmds;
 
 	/* Array of handles to be pinned & unpinned */
 	struct host1x_reloc *relocs;
@@ -275,6 +275,7 @@ struct host1x_job *host1x_job_alloc(struct host1x_channel *ch,
 				    u32 num_cmdbufs, u32 num_relocs);
 void host1x_job_add_gather(struct host1x_job *job, struct host1x_bo *bo,
 			   unsigned int words, unsigned int offset);
+void host1x_job_add_wait(struct host1x_job *job, u32 id, u32 thresh);
 struct host1x_job *host1x_job_get(struct host1x_job *job);
 void host1x_job_put(struct host1x_job *job);
 int host1x_job_pin(struct host1x_job *job, struct device *dev);
Signature · 1 line
2.29.2
author maintainer
#13 lore raw

PATCH v4 13/21 gpu: host1x: Reset max value when freeing a syncpoint

In reply to Mikko Perttunen

With job recovery becoming optional, syncpoints may have a mismatch between their value and max value when freed. As such, when freeing, set the max value to the current value of the syncpoint so that it is in a sane state for the next user.

---
v3:
* Use host1x_syncpt_read instead of read_min to ensure syncpoint value is current.

---
 drivers/gpu/host1x/syncpt.c | 1 +
 1 file changed, 1 insertion(+)
drivers/gpu/host1x/syncpt.c +1 -0
index 8d658e5f7db2..99d31932eb34 100644
@@ -385,6 +385,7 @@ static void syncpt_release(struct kref *ref)
 {
 	struct host1x_syncpt *sp = container_of(ref, struct host1x_syncpt, ref);
 
+	atomic_set(&sp->max_val, host1x_syncpt_read(sp));
 	sp->locked = false;
 
 	mutex_lock(&sp->host->syncpt_mutex);
Signature · 1 line
2.29.2
author maintainer
#14 lore raw

PATCH v4 14/21 gpu: host1x: Reserve VBLANK syncpoints at initialization

In reply to Mikko Perttunen

On T20-T148 chips, the bootloader can set up a boot splash screen with DC configured to increment syncpoint 26/27 at VBLANK. Because of this we shouldn't allow these syncpoints to be allocated until DC has been reset and will no longer increment them in the background.

As such, on these chips, reserve those two syncpoints at initialization, and only mark them free once the DC driver has indicated it's safe to do so.

---
v3:
* New patch

---
 drivers/gpu/drm/tegra/dc.c  |  6 ++++++
 drivers/gpu/host1x/dev.c    |  6 ++++++
 drivers/gpu/host1x/dev.h    |  6 ++++++
 drivers/gpu/host1x/syncpt.c | 34 +++++++++++++++++++++++++++++++++-
 include/linux/host1x.h      |  3 +++
 5 files changed, 54 insertions(+), 1 deletion(-)
drivers/gpu/drm/tegra/dc.c +6 -0
index efb41c10dad4..0b23e0922c25 100644
@@ -2031,6 +2031,12 @@ static int tegra_dc_init(struct host1x_client *client)
 	struct drm_plane *cursor = NULL;
 	int err;
 
+	/*
+	 * DC has been reset by now, so VBLANK syncpoint can be released
+	 * for general use.
+	 */
+	host1x_syncpt_release_vblank_reservation(client, 26 + dc->pipe);
+
 	/*
 	 * XXX do not register DCs with no window groups because we cannot
 	 * assign a primary plane to them, which in turn will cause KMS to
drivers/gpu/host1x/dev.c +6 -0
index 641317d23828..8b50fbb22846 100644
@@ -77,6 +77,7 @@ static const struct host1x_info host1x01_info = {
 	.has_hypervisor = false,
 	.num_sid_entries = 0,
 	.sid_table = NULL,
+	.reserve_vblank_syncpts = true,
 };
 
 static const struct host1x_info host1x02_info = {
@@ -91,6 +92,7 @@ static const struct host1x_info host1x02_info = {
 	.has_hypervisor = false,
 	.num_sid_entries = 0,
 	.sid_table = NULL,
+	.reserve_vblank_syncpts = true,
 };
 
 static const struct host1x_info host1x04_info = {
@@ -105,6 +107,7 @@ static const struct host1x_info host1x04_info = {
 	.has_hypervisor = false,
 	.num_sid_entries = 0,
 	.sid_table = NULL,
+	.reserve_vblank_syncpts = false,
 };
 
 static const struct host1x_info host1x05_info = {
@@ -119,6 +122,7 @@ static const struct host1x_info host1x05_info = {
 	.has_hypervisor = false,
 	.num_sid_entries = 0,
 	.sid_table = NULL,
+	.reserve_vblank_syncpts = false,
 };
 
 static const struct host1x_sid_entry tegra186_sid_table[] = {
@@ -142,6 +146,7 @@ static const struct host1x_info host1x06_info = {
 	.has_hypervisor = true,
 	.num_sid_entries = ARRAY_SIZE(tegra186_sid_table),
 	.sid_table = tegra186_sid_table,
+	.reserve_vblank_syncpts = false,
 };
 
 static const struct host1x_sid_entry tegra194_sid_table[] = {
@@ -165,6 +170,7 @@ static const struct host1x_info host1x07_info = {
 	.has_hypervisor = true,
 	.num_sid_entries = ARRAY_SIZE(tegra194_sid_table),
 	.sid_table = tegra194_sid_table,
+	.reserve_vblank_syncpts = false,
 };
 
 static const struct of_device_id host1x_of_match[] = {
drivers/gpu/host1x/dev.h +6 -0
index 7b8b7e20e32b..e360bc4a25f6 100644
@@ -102,6 +102,12 @@ struct host1x_info {
 	bool has_hypervisor; /* has hypervisor registers */
 	unsigned int num_sid_entries;
 	const struct host1x_sid_entry *sid_table;
+	/*
+	 * On T20-T148, the boot chain may setup DC to increment syncpoints
+	 * 26/27 on VBLANK. As such we cannot use these syncpoints until
+	 * the display driver disables VBLANK increments.
+	 */
+	bool reserve_vblank_syncpts;
 };
 
 struct host1x {
drivers/gpu/host1x/syncpt.c +33 -1
index 99d31932eb34..d0be7bdbc6c9 100644
@@ -52,7 +52,7 @@ struct host1x_syncpt *host1x_syncpt_alloc(struct host1x *host,
 
 	mutex_lock(&host->syncpt_mutex);
 
-	for (i = 0; i < host->info->nb_pts && sp->name; i++, sp++)
+	for (i = 0; i < host->info->nb_pts && kref_read(&sp->ref); i++, sp++)
 		;
 
 	if (i >= host->info->nb_pts)
@@ -359,6 +359,11 @@ int host1x_syncpt_init(struct host1x *host)
 	if (!host->nop_sp)
 		return -ENOMEM;
 
+	if (host->info->reserve_vblank_syncpts) {
+		kref_init(&host->syncpt[26].ref);
+		kref_init(&host->syncpt[27].ref);
+	}
+
 	return 0;
 }
 
@@ -545,3 +550,30 @@ u32 host1x_syncpt_base_id(struct host1x_syncpt_base *base)
 	return base->id;
 }
 EXPORT_SYMBOL(host1x_syncpt_base_id);
+
+static void do_nothing(struct kref *ref)
+{
+}
+
+/**
+ * host1x_syncpt_release_vblank_reservation() - Make VBLANK syncpoint
+ *   available for allocation
+ *
+ * @client: host1x bus client
+ *
+ * Makes VBLANK<i> syncpoint available for allocatation if it was
+ * reserved at initialization time. This should be called by the display
+ * driver after it has ensured that any VBLANK increment programming configured
+ * by the boot chain has been disabled.
+ */
+void host1x_syncpt_release_vblank_reservation(struct host1x_client *client,
+					      u32 syncpt_id)
+{
+	struct host1x *host = dev_get_drvdata(client->host->parent);
+
+	if (!host->info->reserve_vblank_syncpts)
+		return;
+
+	kref_put(&host->syncpt[syncpt_id].ref, do_nothing);
+}
+EXPORT_SYMBOL(host1x_syncpt_release_vblank_reservation);
include/linux/host1x.h +3 -0
index 59296d3346fe..ff2a23cd6bb0 100644
@@ -163,6 +163,9 @@ struct host1x_syncpt *host1x_syncpt_alloc(struct host1x *host,
 struct host1x_syncpt_base *host1x_syncpt_get_base(struct host1x_syncpt *sp);
 u32 host1x_syncpt_base_id(struct host1x_syncpt_base *base);
 
+void host1x_syncpt_release_vblank_reservation(struct host1x_client *client,
+					      u32 syncpt_id);
+
 struct host1x_syncpt *host1x_syncpt_fd_get(int fd);
 
 struct dma_fence *host1x_fence_create(struct host1x_syncpt *sp, u32 threshold);
Signature · 1 line
2.29.2
author maintainer
#15 lore raw

PATCH v4 15/21 drm/tegra: Add new UAPI to header

In reply to Mikko Perttunen

Update the tegra_drm.h UAPI header, adding the new proposed UAPI. The old staging UAPI is left in for now, with minor modification to avoid name collisions.

---
v4:
* Remove features that are not strictly necessary
* Remove padding/reserved fields in IOCTL structs where DRM's zero extension feature allows future expansion
v3:
* Remove timeout field
* Inline the syncpt_incrs array to the submit structure
* Remove WRITE_RELOC (it is now implicit)

---
 include/uapi/drm/tegra_drm.h | 338 ++++++++++++++++++++++++++++++++---
 1 file changed, 311 insertions(+), 27 deletions(-)
include/uapi/drm/tegra_drm.h +311 -27
index c4df3c3668b3..014bc393c298 100644
@@ -1,24 +1,5 @@
-/*
- * Copyright (c) 2012-2013, NVIDIA CORPORATION.  All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
- * OTHER DEALINGS IN THE SOFTWARE.
- */
+/* SPDX-License-Identifier: MIT */
+/* Copyright (c) 2012-2020 NVIDIA Corporation */
 
 #ifndef _UAPI_TEGRA_DRM_H_
 #define _UAPI_TEGRA_DRM_H_
@@ -29,6 +10,8 @@
 extern "C" {
 #endif
 
+/* TegraDRM legacy UAPI. Only enabled with STAGING */
+
 #define DRM_TEGRA_GEM_CREATE_TILED     (1 << 0)
 #define DRM_TEGRA_GEM_CREATE_BOTTOM_UP (1 << 1)
 
@@ -644,13 +627,13 @@ struct drm_tegra_gem_get_flags {
 	__u32 flags;
 };
 
-#define DRM_TEGRA_GEM_CREATE		0x00
-#define DRM_TEGRA_GEM_MMAP		0x01
+#define DRM_TEGRA_GEM_CREATE_LEGACY	0x00
+#define DRM_TEGRA_GEM_MMAP_LEGACY	0x01
 #define DRM_TEGRA_SYNCPT_READ		0x02
 #define DRM_TEGRA_SYNCPT_INCR		0x03
 #define DRM_TEGRA_SYNCPT_WAIT		0x04
-#define DRM_TEGRA_OPEN_CHANNEL		0x05
-#define DRM_TEGRA_CLOSE_CHANNEL		0x06
+#define DRM_TEGRA_OPEN_CHANNEL	        0x05
+#define DRM_TEGRA_CLOSE_CHANNEL	        0x06
 #define DRM_TEGRA_GET_SYNCPT		0x07
 #define DRM_TEGRA_SUBMIT		0x08
 #define DRM_TEGRA_GET_SYNCPT_BASE	0x09
@@ -659,8 +642,8 @@ struct drm_tegra_gem_get_flags {
 #define DRM_TEGRA_GEM_SET_FLAGS		0x0c
 #define DRM_TEGRA_GEM_GET_FLAGS		0x0d
 
-#define DRM_IOCTL_TEGRA_GEM_CREATE DRM_IOWR(DRM_COMMAND_BASE + DRM_TEGRA_GEM_CREATE, struct drm_tegra_gem_create)
-#define DRM_IOCTL_TEGRA_GEM_MMAP DRM_IOWR(DRM_COMMAND_BASE + DRM_TEGRA_GEM_MMAP, struct drm_tegra_gem_mmap)
+#define DRM_IOCTL_TEGRA_GEM_CREATE_LEGACY DRM_IOWR(DRM_COMMAND_BASE + DRM_TEGRA_GEM_CREATE_LEGACY, struct drm_tegra_gem_create)
+#define DRM_IOCTL_TEGRA_GEM_MMAP_LEGACY DRM_IOWR(DRM_COMMAND_BASE + DRM_TEGRA_GEM_MMAP_LEGACY, struct drm_tegra_gem_mmap)
 #define DRM_IOCTL_TEGRA_SYNCPT_READ DRM_IOWR(DRM_COMMAND_BASE + DRM_TEGRA_SYNCPT_READ, struct drm_tegra_syncpt_read)
 #define DRM_IOCTL_TEGRA_SYNCPT_INCR DRM_IOWR(DRM_COMMAND_BASE + DRM_TEGRA_SYNCPT_INCR, struct drm_tegra_syncpt_incr)
 #define DRM_IOCTL_TEGRA_SYNCPT_WAIT DRM_IOWR(DRM_COMMAND_BASE + DRM_TEGRA_SYNCPT_WAIT, struct drm_tegra_syncpt_wait)
@@ -674,6 +657,307 @@ struct drm_tegra_gem_get_flags {
 #define DRM_IOCTL_TEGRA_GEM_SET_FLAGS DRM_IOWR(DRM_COMMAND_BASE + DRM_TEGRA_GEM_SET_FLAGS, struct drm_tegra_gem_set_flags)
 #define DRM_IOCTL_TEGRA_GEM_GET_FLAGS DRM_IOWR(DRM_COMMAND_BASE + DRM_TEGRA_GEM_GET_FLAGS, struct drm_tegra_gem_get_flags)
 
+/* New TegraDRM UAPI */
+
+struct drm_tegra_channel_open {
+	/**
+	 * @host1x_class: [in]
+	 *
+	 * Host1x class of the engine that will be programmed using this
+	 * channel.
+	 */
+	__u32 host1x_class;
+
+	/**
+	 * @flags: [in]
+	 *
+	 * Flags.
+	 */
+	__u32 flags;
+
+	/**
+	 * @channel_ctx: [out]
+	 *
+	 * Opaque identifier corresponding to the opened channel.
+	 */
+	__u32 channel_ctx;
+
+	/**
+	 * @hardware_version: [out]
+	 *
+	 * Version of the engine hardware. This can be used by userspace
+	 * to determine how the engine needs to be programmed.
+	 */
+	__u32 hardware_version;
+};
+
+struct drm_tegra_channel_close {
+	/**
+	 * @channel_ctx: [in]
+	 *
+	 * Identifier of the channel to close.
+	 */
+	__u32 channel_ctx;
+};
+
+#define DRM_TEGRA_CHANNEL_MAP_READWRITE			(1<<0)
+
+struct drm_tegra_channel_map {
+	/**
+	 * @channel_ctx: [in]
+	 *
+	 * Identifier of the channel to which make memory available for.
+	 */
+	__u32 channel_ctx;
+
+	/**
+	 * @handle: [in]
+	 *
+	 * GEM handle of the memory to map.
+	 */
+	__u32 handle;
+
+	/**
+	 * @flags: [in]
+	 *
+	 * Flags.
+	 */
+	__u32 flags;
+
+	/**
+	 * @mapping_id: [out]
+	 *
+	 * Identifier corresponding to the mapping, to be used for
+	 * relocations or unmapping later.
+	 */
+	__u32 mapping_id;
+};
+
+struct drm_tegra_channel_unmap {
+	/**
+	 * @channel_ctx: [in]
+	 *
+	 * Channel identifier of the channel to unmap memory from.
+	 */
+	__u32 channel_ctx;
+
+	/**
+	 * @mapping_id: [in]
+	 *
+	 * Mapping identifier of the memory mapping to unmap.
+	 */
+	__u32 mapping_id;
+};
+
+/* Submission */
+
+/**
+ * Specify that bit 39 of the patched-in address should be set to
+ * trigger layout swizzling between Tegra and non-Tegra Blocklinear
+ * layout on systems that store surfaces in system memory in non-Tegra
+ * Blocklinear layout.
+ */
+#define DRM_TEGRA_SUBMIT_BUF_RELOC_BLOCKLINEAR		(1<<0)
+
+struct drm_tegra_submit_buf {
+	/**
+	 * @mapping_id: [in]
+	 *
+	 * Identifier of the mapping to use in the submission.
+	 */
+	__u32 mapping_id;
+
+	/**
+	 * @flags: [in]
+	 *
+	 * Flags.
+	 */
+	__u32 flags;
+
+	/**
+	 * Information for relocation patching. Relocation patching will
+	 * be done if the MAP IOCTL that created `mapping_id` did not
+	 * return an IOVA. If an IOVA was returned, the application is
+	 * responsible for patching the address into the gather.
+	 */
+	struct {
+		/**
+		 * @target_offset: [in]
+		 *
+		 * Offset from the start of the mapping of the data whose
+		 * address is to be patched into the gather.
+		 */
+		__u64 target_offset;
+
+		/**
+		 * @gather_offset_words: [in]
+		 *
+		 * Offset in words from the start of the gather data to
+		 * where the address should be patched into.
+		 */
+		__u32 gather_offset_words;
+
+		/**
+		 * @shift: [in]
+		 *
+		 * Number of bits the address should be shifted right before
+		 * patching in.
+		 */
+		__u32 shift;
+	} reloc;
+};
+
+struct drm_tegra_submit_syncpt_incr {
+	/**
+	 * @syncpt_fd: [in]
+	 *
+	 * Syncpoint file descriptor of the syncpoint that the job will
+	 * increment.
+	 */
+	__s32 syncpt_fd;
+
+	/**
+	 * @flags: [in]
+	 *
+	 * Flags.
+	 */
+	__u32 flags;
+
+	/**
+	 * @num_incrs: [in]
+	 *
+	 * Number of times the job will increment this syncpoint.
+	 */
+	__u32 num_incrs;
+
+	/**
+	 * @fence_value: [out]
+	 *
+	 * Value the syncpoint will have once the job has completed all
+	 * its specified syncpoint increments.
+	 *
+	 * Note that the kernel may increment the syncpoint before or after
+	 * the job. These increments are not reflected in this field.
+	 *
+	 * If the job hangs or times out, not all of the increments may
+	 * get executed.
+	 */
+	__u32 fence_value;
+};
+
+/**
+ * Execute `words` words of Host1x opcodes specified in the `gather_data_ptr`
+ * buffer. Each GATHER_UPTR command uses successive words from the buffer.
+ */
+#define DRM_TEGRA_SUBMIT_CMD_GATHER_UPTR		0
+/**
+ * Wait for a syncpoint to reach a value before continuing with further
+ * commands.
+ */
+#define DRM_TEGRA_SUBMIT_CMD_WAIT_SYNCPT		1
+
+struct drm_tegra_submit_cmd_gather_uptr {
+	__u32 words;
+	__u32 reserved[3];
+};
+
+struct drm_tegra_submit_cmd_wait_syncpt {
+	__u32 id;
+	__u32 threshold;
+	__u32 reserved[2];
+};
+
+struct drm_tegra_submit_cmd {
+	/**
+	 * @type: [in]
+	 *
+	 * Command type to execute. One of the DRM_TEGRA_SUBMIT_CMD*
+	 * defines.
+	 */
+	__u32 type;
+
+	/**
+	 * @flags: [in]
+	 *
+	 * Flags.
+	 */
+	__u32 flags;
+
+	union {
+		struct drm_tegra_submit_cmd_gather_uptr gather_uptr;
+		struct drm_tegra_submit_cmd_wait_syncpt wait_syncpt;
+		__u32 reserved[4];
+	};
+};
+
+struct drm_tegra_channel_submit {
+	/**
+	 * @channel_ctx: [in]
+	 *
+	 * Identifier of the channel to submit this job to.
+	 */
+	__u32 channel_ctx;
+
+	/**
+	 * @num_bufs: [in]
+	 *
+	 * Number of elements in the `bufs_ptr` array.
+	 */
+	__u32 num_bufs;
+
+	/**
+	 * @num_cmds: [in]
+	 *
+	 * Number of elements in the `cmds_ptr` array.
+	 */
+	__u32 num_cmds;
+
+	/**
+	 * @gather_data_words: [in]
+	 *
+	 * Number of 32-bit words in the `gather_data_ptr` array.
+	 */
+	__u32 gather_data_words;
+
+	/**
+	 * @bufs_ptr: [in]
+	 *
+	 * Pointer to an array of drm_tegra_submit_buf structures.
+	 */
+	__u64 bufs_ptr;
+
+	/**
+	 * @cmds_ptr: [in]
+	 *
+	 * Pointer to an array of drm_tegra_submit_cmd structures.
+	 */
+	__u64 cmds_ptr;
+
+	/**
+	 * @gather_data_ptr: [in]
+	 *
+	 * Pointer to an array of Host1x opcodes to be used by GATHER_UPTR
+	 * commands.
+	 */
+	__u64 gather_data_ptr;
+
+	/**
+	 * @syncpt_incr: [in,out]
+	 *
+	 * Information about the syncpoint the job will increment.
+	 */
+	struct drm_tegra_submit_syncpt_incr syncpt_incr;
+};
+
+#define DRM_IOCTL_TEGRA_CHANNEL_OPEN     DRM_IOWR(DRM_COMMAND_BASE + 0x10, struct drm_tegra_channel_open)
+#define DRM_IOCTL_TEGRA_CHANNEL_CLOSE    DRM_IOWR(DRM_COMMAND_BASE + 0x11, struct drm_tegra_channel_close)
+#define DRM_IOCTL_TEGRA_CHANNEL_MAP      DRM_IOWR(DRM_COMMAND_BASE + 0x12, struct drm_tegra_channel_map)
+#define DRM_IOCTL_TEGRA_CHANNEL_UNMAP    DRM_IOWR(DRM_COMMAND_BASE + 0x13, struct drm_tegra_channel_unmap)
+#define DRM_IOCTL_TEGRA_CHANNEL_SUBMIT   DRM_IOWR(DRM_COMMAND_BASE + 0x14, struct drm_tegra_channel_submit)
+
+#define DRM_IOCTL_TEGRA_GEM_CREATE       DRM_IOWR(DRM_COMMAND_BASE + 0x15, struct drm_tegra_gem_create)
+#define DRM_IOCTL_TEGRA_GEM_MMAP         DRM_IOWR(DRM_COMMAND_BASE + 0x16, struct drm_tegra_gem_mmap)
+
 #if defined(__cplusplus)
 }
 #endif
Signature · 1 line
2.29.2
author maintainer
#16 lore raw

PATCH v4 16/21 drm/tegra: Boot VIC during runtime PM resume

In reply to Mikko Perttunen

With the new UAPI implementation, engines are powered on and off when there are active jobs, and the core code handles channel allocation. To accommodate that, boot the engine as part of runtime PM instead of using the open_channel callback, which is not used by the new submit path.

---
v3:
* runtime_get/put is now done directly from submit path, so no callbacks are added
* Reworded.

---
 drivers/gpu/drm/tegra/vic.c | 114 +++++++++++++++++-------------------
 1 file changed, 53 insertions(+), 61 deletions(-)
drivers/gpu/drm/tegra/vic.c +53 -61
index cb476da59adc..5d2ad125dca3 100644
@@ -29,7 +29,6 @@ struct vic_config {
 
 struct vic {
 	struct falcon falcon;
-	bool booted;
 
 	void __iomem *regs;
 	struct tegra_drm_client client;
@@ -52,48 +51,6 @@ static void vic_writel(struct vic *vic, u32 value, unsigned int offset)
 	writel(value, vic->regs + offset);
 }
 
-static int vic_runtime_resume(struct device *dev)
-{
-	struct vic *vic = dev_get_drvdata(dev);
-	int err;
-
-	err = clk_prepare_enable(vic->clk);
-	if (err < 0)
-		return err;
-
-	usleep_range(10, 20);
-
-	err = reset_control_deassert(vic->rst);
-	if (err < 0)
-		goto disable;
-
-	usleep_range(10, 20);
-
-	return 0;
-
-disable:
-	clk_disable_unprepare(vic->clk);
-	return err;
-}
-
-static int vic_runtime_suspend(struct device *dev)
-{
-	struct vic *vic = dev_get_drvdata(dev);
-	int err;
-
-	err = reset_control_assert(vic->rst);
-	if (err < 0)
-		return err;
-
-	usleep_range(2000, 4000);
-
-	clk_disable_unprepare(vic->clk);
-
-	vic->booted = false;
-
-	return 0;
-}
-
 static int vic_boot(struct vic *vic)
 {
 #ifdef CONFIG_IOMMU_API
@@ -103,9 +60,6 @@ static int vic_boot(struct vic *vic)
 	void *hdr;
 	int err = 0;
 
-	if (vic->booted)
-		return 0;
-
 #ifdef CONFIG_IOMMU_API
 	if (vic->config->supports_sid && spec) {
 		u32 value;
@@ -153,8 +107,6 @@ static int vic_boot(struct vic *vic)
 		return err;
 	}
 
-	vic->booted = true;
-
 	return 0;
 }
 
@@ -308,35 +260,76 @@ static int vic_load_firmware(struct vic *vic)
 	return err;
 }
 
-static int vic_open_channel(struct tegra_drm_client *client,
-			    struct tegra_drm_context *context)
+
+static int vic_runtime_resume(struct device *dev)
 {
-	struct vic *vic = to_vic(client);
+	struct vic *vic = dev_get_drvdata(dev);
 	int err;
 
-	err = pm_runtime_get_sync(vic->dev);
+	err = clk_prepare_enable(vic->clk);
 	if (err < 0)
 		return err;
 
+	usleep_range(10, 20);
+
+	err = reset_control_deassert(vic->rst);
+	if (err < 0)
+		goto disable;
+
+	usleep_range(10, 20);
+
 	err = vic_load_firmware(vic);
 	if (err < 0)
-		goto rpm_put;
+		goto assert;
 
 	err = vic_boot(vic);
 	if (err < 0)
-		goto rpm_put;
+		goto assert;
+
+	return 0;
+
+assert:
+	reset_control_assert(vic->rst);
+disable:
+	clk_disable_unprepare(vic->clk);
+	return err;
+}
+
+static int vic_runtime_suspend(struct device *dev)
+{
+	struct vic *vic = dev_get_drvdata(dev);
+	int err;
+
+	err = reset_control_assert(vic->rst);
+	if (err < 0)
+		return err;
+
+	usleep_range(2000, 4000);
+
+	clk_disable_unprepare(vic->clk);
+
+	return 0;
+}
+
+static int vic_open_channel(struct tegra_drm_client *client,
+			    struct tegra_drm_context *context)
+{
+	struct vic *vic = to_vic(client);
+	int err;
+
+	err = pm_runtime_get_sync(vic->dev);
+	if (err < 0) {
+		pm_runtime_put(vic->dev);
+		return err;
+	}
 
 	context->channel = host1x_channel_get(vic->channel);
 	if (!context->channel) {
-		err = -ENOMEM;
-		goto rpm_put;
+		pm_runtime_put(vic->dev);
+		return -ENOMEM;
 	}
 
 	return 0;
-
-rpm_put:
-	pm_runtime_put(vic->dev);
-	return err;
 }
 
 static void vic_close_channel(struct tegra_drm_context *context)
@@ -344,7 +337,6 @@ static void vic_close_channel(struct tegra_drm_context *context)
 	struct vic *vic = to_vic(context->client);
 
 	host1x_channel_put(context->channel);
-
 	pm_runtime_put(vic->dev);
 }
 
Signature · 1 line
2.29.2
author maintainer
#17 lore raw

PATCH v4 17/21 drm/tegra: Set resv fields when importing/exporting GEMs

In reply to Mikko Perttunen

To allow sharing of implicit fences when exporting/importing dma_buf objects, set the 'resv' fields when importing or exporting GEM objects.

---
 drivers/gpu/drm/tegra/gem.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/tegra/gem.c +2 -0
index 723df142a981..4a8acd4724bd 100644
@@ -423,6 +423,7 @@ static struct tegra_bo *tegra_bo_import(struct drm_device *drm,
 	}
 
 	bo->gem.import_attach = attach;
+	bo->gem.resv = buf->resv;
 
 	return bo;
 
@@ -675,6 +676,7 @@ struct dma_buf *tegra_gem_prime_export(struct drm_gem_object *gem,
 	exp_info.size = gem->size;
 	exp_info.flags = flags;
 	exp_info.priv = gem;
+	exp_info.resv = gem->resv;
 
 	return drm_gem_dmabuf_export(gem->dev, &exp_info);
 }
Signature · 1 line
2.29.2
author maintainer
#18 lore raw

PATCH v4 18/21 drm/tegra: Allocate per-engine channel in core code

In reply to Mikko Perttunen

To avoid duplication, allocate the per-engine shared channel in the core code instead. Once MLOCKs are implemented on Host1x side, we can also update this to avoid allocating a shared channel when MLOCKs are enabled.

---
 drivers/gpu/drm/tegra/drm.c | 11 +++++++++++
 drivers/gpu/drm/tegra/drm.h |  4 ++++
 2 files changed, 15 insertions(+)
drivers/gpu/drm/tegra/drm.c +11 -0
index 7437c67924aa..7124b0b0154b 100644
@@ -887,6 +887,14 @@ static struct drm_driver tegra_drm_driver = {
 int tegra_drm_register_client(struct tegra_drm *tegra,
 			      struct tegra_drm_client *client)
 {
+	/*
+	 * When MLOCKs are implemented, change to allocate a shared channel
+	 * only when MLOCKs are disabled.
+	 */
+	client->shared_channel = host1x_channel_request(&client->base);
+	if (!client->shared_channel)
+		return -EBUSY;
+
 	mutex_lock(&tegra->clients_lock);
 	list_add_tail(&client->list, &tegra->clients);
 	client->drm = tegra;
@@ -903,6 +911,9 @@ int tegra_drm_unregister_client(struct tegra_drm *tegra,
 	client->drm = NULL;
 	mutex_unlock(&tegra->clients_lock);
 
+	if (client->shared_channel)
+		host1x_channel_put(client->shared_channel);
+
 	return 0;
 }
 
drivers/gpu/drm/tegra/drm.h +4 -0
index b25443255be6..3fc42fd97911 100644
@@ -86,8 +86,12 @@ struct tegra_drm_client {
 	struct list_head list;
 	struct tegra_drm *drm;
 
+	/* Set by driver */
 	unsigned int version;
 	const struct tegra_drm_client_ops *ops;
+
+	/* Set by TegraDRM core */
+	struct host1x_channel *shared_channel;
 };
 
 static inline struct tegra_drm_client *
Signature · 1 line
2.29.2
author maintainer
#19 lore raw

PATCH v4 19/21 drm/tegra: Implement new UAPI

In reply to Mikko Perttunen

Implement the non-submission parts of the new UAPI, including channel management and memory mapping. The UAPI is under the CONFIG_DRM_TEGRA_STAGING config flag for now.

---
v4:
* New patch, split out from combined UAPI + submit patch.

---
 drivers/gpu/drm/tegra/Makefile    |   1 +
 drivers/gpu/drm/tegra/drm.c       |  41 ++--
 drivers/gpu/drm/tegra/drm.h       |   5 +
 drivers/gpu/drm/tegra/uapi.h      |  63 ++++++
 drivers/gpu/drm/tegra/uapi/uapi.c | 306 ++++++++++++++++++++++++++++++
 5 files changed, 400 insertions(+), 16 deletions(-)
 create mode 100644 drivers/gpu/drm/tegra/uapi.h
 create mode 100644 drivers/gpu/drm/tegra/uapi/uapi.c
drivers/gpu/drm/tegra/Makefile +1 -0
index d6cf202414f0..0abdb21b38b9 100644
@@ -3,6 +3,7 @@ ccflags-$(CONFIG_DRM_TEGRA_DEBUG) += -DDEBUG
 
 tegra-drm-y := \
 	drm.o \
+	uapi/uapi.o \
 	gem.o \
 	fb.o \
 	dp.o \
drivers/gpu/drm/tegra/drm.c +25 -16

110 lines not on this page · raw message

drivers/gpu/drm/tegra/drm.h +5 -0
index 3fc42fd97911..42f8a301555c 100644
@@ -58,6 +58,11 @@ struct tegra_drm {
 	struct tegra_display_hub *hub;
 };
 
+static inline struct host1x *tegra_drm_to_host1x(struct tegra_drm *tegra)
+{
+	return dev_get_drvdata(tegra->drm->dev->parent);
+}
+
 struct tegra_drm_client;
 
 struct tegra_drm_context {
drivers/gpu/drm/tegra/uapi.h new file +63 -0

63 lines not on this page · raw message

drivers/gpu/drm/tegra/uapi/uapi.c new file +306 -0

306 lines not on this page · raw message

Signature · 1 line
2.29.2
#20 lore raw

Re: [PATCH v4 19/21] drm/tegra: Implement new UAPI

Thank you for the patch! Perhaps something to improve:

Hi Mikko,

Thank you for the patch! Perhaps something to improve:

[auto build test WARNING on tegra/for-next]
[also build test WARNING on linus/master v5.10-rc4]
[cannot apply to tegra-drm/drm/tegra/for-next next-20201120] [If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]

url:    https://github.com/0day-ci/linux/commits/Mikko-Perttunen/Host1x-TegraDRM-UAPI/20201120-192927
base:   https://git.kernel.org/pub/scm/linux/kernel/git/tegra/linux.git for-next
config: arm-randconfig-r014-20201120 (attached as .config)
compiler: arm-linux-gnueabi-gcc (GCC) 9.3.0

reproduce (this is a W=1 build):
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# https://github.com/0day-ci/linux/commit/2b3b62240ac675fb308bbb5acbd68187f940d1e7
git remote add linux-review https://github.com/0day-ci/linux
git fetch --no-tags linux-review Mikko-Perttunen/Host1x-TegraDRM-UAPI/20201120-192927 git checkout 2b3b62240ac675fb308bbb5acbd68187f940d1e7
# save the attached .config to linux build tree
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-9.3.0 make.cross ARCH=arm

If you fix the issue, kindly add following tag as appropriate

All warnings (new ones prefixed by >>):

>> drivers/gpu/drm/tegra/uapi/uapi.c:62:5: warning: no previous prototype for 'close_channel_ctx' [-Wmissing-prototypes]

62 | int close_channel_ctx(int id, void *p, void *data)
| ^~~~~~~~~~~~~~~~~

vim +/close_channel_ctx +62 drivers/gpu/drm/tegra/uapi/uapi.c

    61	
  > 62	int close_channel_ctx(int id, void *p, void *data)
    63	{
    64		struct tegra_drm_channel_ctx *ctx = p;
    65	
    66		tegra_drm_channel_ctx_close(ctx);
    67	
    68		return 0;
    69	}
    70	

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org

#21 lore raw

Re: [PATCH v4 19/21] drm/tegra: Implement new UAPI

In reply to Mikko Perttunen
Thank you for the patch! Perhaps something to improve:

Hi Mikko,

Thank you for the patch! Perhaps something to improve:

[auto build test WARNING on tegra/for-next]
[also build test WARNING on linus/master v5.10-rc4]
[cannot apply to tegra-drm/drm/tegra/for-next next-20201120] [If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]

url:    https://github.com/0day-ci/linux/commits/Mikko-Perttunen/Host1x-TegraDRM-UAPI/20201120-192927
base:   https://git.kernel.org/pub/scm/linux/kernel/git/tegra/linux.git for-next
config: arm64-randconfig-r022-20201120 (attached as .config)
compiler: clang version 12.0.0 (https://github.com/llvm/llvm-project 3ded927cf80ac519f9f9c4664fef08787f7c537d)

reproduce (this is a W=1 build):
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# install arm64 cross compiling tool for clang build
# apt-get install binutils-aarch64-linux-gnu
# https://github.com/0day-ci/linux/commit/2b3b62240ac675fb308bbb5acbd68187f940d1e7
git remote add linux-review https://github.com/0day-ci/linux
git fetch --no-tags linux-review Mikko-Perttunen/Host1x-TegraDRM-UAPI/20201120-192927 git checkout 2b3b62240ac675fb308bbb5acbd68187f940d1e7
# save the attached .config to linux build tree
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross ARCH=arm64

If you fix the issue, kindly add following tag as appropriate

All warnings (new ones prefixed by >>):

>> drivers/gpu/drm/tegra/uapi/uapi.c:62:5: warning: no previous prototype for function 'close_channel_ctx' [-Wmissing-prototypes]
   int close_channel_ctx(int id, void *p, void *data)
       ^

drivers/gpu/drm/tegra/uapi/uapi.c:62:1: note: declare 'static' if the function is not intended to be used outside of this translation unit int close_channel_ctx(int id, void *p, void *data)
^
static
1 warning generated.

vim +/close_channel_ctx +62 drivers/gpu/drm/tegra/uapi/uapi.c

    61	
  > 62	int close_channel_ctx(int id, void *p, void *data)
    63	{
    64		struct tegra_drm_channel_ctx *ctx = p;
    65	
    66		tegra_drm_channel_ctx_close(ctx);
    67	
    68		return 0;
    69	}
    70	

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org

author maintainer
#22 lore raw

PATCH v4 20/21 drm/tegra: Implement job submission part of new UAPI

In reply to Mikko Perttunen

Implement the job submission IOCTL with a minimum feature set.

---
v4:
* Remove all features that are not strictly necessary.
* Split into two patches.
v3:
* Remove WRITE_RELOC. Relocations are now patched implicitly when patching is needed.
* Directly call PM runtime APIs on devices instead of using power_on/power_off callbacks.
* Remove incorrect mutex unlock in tegra_drm_ioctl_channel_open
* Use XA_FLAGS_ALLOC1 instead of XA_FLAGS_ALLOC
* Accommodate for removal of timeout field and inlining of syncpt_incrs array.
* Copy entire user arrays at a time instead of going through elements one-by-one.
* Implement waiting of DMA reservations.
* Split out gather_bo implementation into a separate file.
* Fix length parameter passed to sg_init_one in gather_bo
* Cosmetic cleanup.

---
 drivers/gpu/drm/tegra/Makefile         |   2 +
 drivers/gpu/drm/tegra/drm.c            |   2 +
 drivers/gpu/drm/tegra/uapi/gather_bo.c |  86 +++++
 drivers/gpu/drm/tegra/uapi/gather_bo.h |  22 ++
 drivers/gpu/drm/tegra/uapi/submit.c    | 423 +++++++++++++++++++++++++
 drivers/gpu/drm/tegra/uapi/submit.h    |  17 +
 6 files changed, 552 insertions(+)
 create mode 100644 drivers/gpu/drm/tegra/uapi/gather_bo.c
 create mode 100644 drivers/gpu/drm/tegra/uapi/gather_bo.h
 create mode 100644 drivers/gpu/drm/tegra/uapi/submit.c
 create mode 100644 drivers/gpu/drm/tegra/uapi/submit.h
drivers/gpu/drm/tegra/Makefile +2 -0
index 0abdb21b38b9..059322e88943 100644
@@ -4,6 +4,8 @@ ccflags-$(CONFIG_DRM_TEGRA_DEBUG) += -DDEBUG
 tegra-drm-y := \
 	drm.o \
 	uapi/uapi.o \
+	uapi/submit.o \
+	uapi/gather_bo.o \
 	gem.o \
 	fb.o \
 	dp.o \
drivers/gpu/drm/tegra/drm.c +2 -0
index 79416b6b6715..3077bd594560 100644
@@ -740,6 +740,8 @@ static const struct drm_ioctl_desc tegra_drm_ioctls[] = {
 			  DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(TEGRA_CHANNEL_UNMAP, tegra_drm_ioctl_channel_unmap,
 			  DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(TEGRA_CHANNEL_SUBMIT, tegra_drm_ioctl_channel_submit,
+			  DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(TEGRA_GEM_CREATE, tegra_drm_ioctl_gem_create,
 			  DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(TEGRA_GEM_MMAP, tegra_drm_ioctl_gem_mmap,
drivers/gpu/drm/tegra/uapi/gather_bo.c new file +86 -0

86 lines not on this page · raw message

drivers/gpu/drm/tegra/uapi/gather_bo.h new file +22 -0
new file mode 100644
index 000000000000..6b4c9d83ac91
@@ -0,0 +1,22 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/* Copyright (c) 2020 NVIDIA Corporation */
+
+#ifndef _TEGRA_DRM_SUBMIT_GATHER_BO_H
+#define _TEGRA_DRM_SUBMIT_GATHER_BO_H
+
+#include <linux/host1x.h>
+#include <linux/kref.h>
+
+struct gather_bo {
+	struct host1x_bo base;
+
+	struct kref ref;
+
+	u32 *gather_data;
+	size_t gather_data_words;
+};
+
+extern const struct host1x_bo_ops gather_bo_ops;
+void gather_bo_put(struct host1x_bo *host_bo);
+
+#endif
drivers/gpu/drm/tegra/uapi/submit.c new file +423 -0

423 lines not on this page · raw message

drivers/gpu/drm/tegra/uapi/submit.h new file +17 -0
new file mode 100644
index 000000000000..0a165e9e4bda
@@ -0,0 +1,17 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/* Copyright (c) 2020 NVIDIA Corporation */
+
+#ifndef _TEGRA_DRM_UAPI_SUBMIT_H
+#define _TEGRA_DRM_UAPI_SUBMIT_H
+
+struct tegra_drm_used_mapping {
+	struct tegra_drm_mapping *mapping;
+	u32 flags;
+};
+
+struct tegra_drm_submit_data {
+	struct tegra_drm_used_mapping *used_mappings;
+	u32 num_used_mappings;
+};
+
+#endif
Signature · 1 line
2.29.2
#23 lore raw

Re: [PATCH v4 20/21] drm/tegra: Implement job submission part of new UAPI

Thank you for the patch! Perhaps something to improve:

Hi Mikko,

Thank you for the patch! Perhaps something to improve:

[auto build test WARNING on tegra/for-next]
[also build test WARNING on linus/master v5.10-rc5]
[cannot apply to tegra-drm/drm/tegra/for-next next-20201124] [If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]

url:    https://github.com/0day-ci/linux/commits/Mikko-Perttunen/Host1x-TegraDRM-UAPI/20201120-192927
base:   https://git.kernel.org/pub/scm/linux/kernel/git/tegra/linux.git for-next
config: arm-randconfig-r036-20201124 (attached as .config)
compiler: clang version 12.0.0 (https://github.com/llvm/llvm-project df9ae5992889560a8f3c6760b54d5051b47c7bf5)

reproduce (this is a W=1 build):
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# install arm cross compiling tool for clang build
# apt-get install binutils-arm-linux-gnueabi
# https://github.com/0day-ci/linux/commit/d5a3a653d37f5c7e8dfda77be51cc68ed7192b70
git remote add linux-review https://github.com/0day-ci/linux
git fetch --no-tags linux-review Mikko-Perttunen/Host1x-TegraDRM-UAPI/20201120-192927 git checkout d5a3a653d37f5c7e8dfda77be51cc68ed7192b70
# save the attached .config to linux build tree
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross ARCH=arm

If you fix the issue, kindly add following tag as appropriate

All warnings (new ones prefixed by >>):

>> drivers/gpu/drm/tegra/uapi/submit.c:114:18: warning: shift count >= width of type [-Wshift-count-overflow]
                   written_ptr |= BIT(39);
                                  ^~~~~~~
   include/vdso/bits.h:7:26: note: expanded from macro 'BIT'
   #define BIT(nr)                 (UL(1) << (nr))
                                          ^  ~~~~

1 warning generated.

vim +114 drivers/gpu/drm/tegra/uapi/submit.c

104
105 static int submit_write_reloc(struct gather_bo *bo,
106 struct drm_tegra_submit_buf *buf,
107 struct tegra_drm_mapping *mapping)
108 {
109 /* TODO check that target_offset is within bounds */
110 dma_addr_t iova = mapping->iova + buf->reloc.target_offset;
111 u32 written_ptr = (u32)(iova >> buf->reloc.shift);
112
113 if (buf->flags & DRM_TEGRA_SUBMIT_BUF_RELOC_BLOCKLINEAR)

 > 114			written_ptr |= BIT(39);

115
116 if (buf->reloc.gather_offset_words >= bo->gather_data_words)
117 return -EINVAL;
118
119 buf->reloc.gather_offset_words = array_index_nospec(
120 buf->reloc.gather_offset_words, bo->gather_data_words);
121
122 bo->gather_data[buf->reloc.gather_offset_words] = written_ptr;
123
124 return 0;
125 }
126

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org

author maintainer
#24 lore raw

PATCH v4 21/21 drm/tegra: Add job firewall

In reply to Mikko Perttunen

Add a firewall that validates jobs before submission to ensure they don't do anything they aren't allowed to do, like accessing memory they should not access.

The firewall is functionality-wise a copy of the firewall already implemented in gpu/host1x. It is copied here as it makes more sense for it to live on the DRM side, as it is only needed for userspace job submissions, and generally the data it needs to do its job is easier to access here.

In the future, the other implementation will be removed.

---
v3:
* New patch

---
 drivers/gpu/drm/tegra/Makefile        |   1 +
 drivers/gpu/drm/tegra/uapi/firewall.c | 197 ++++++++++++++++++++++++++
 drivers/gpu/drm/tegra/uapi/submit.c   |   4 +
 drivers/gpu/drm/tegra/uapi/submit.h   |   3 +
 4 files changed, 205 insertions(+)
 create mode 100644 drivers/gpu/drm/tegra/uapi/firewall.c
drivers/gpu/drm/tegra/Makefile +1 -0
index 059322e88943..4e3295f436f1 100644
@@ -5,6 +5,7 @@ tegra-drm-y := \
 	drm.o \
 	uapi/uapi.o \
 	uapi/submit.o \
+	uapi/firewall.o \
 	uapi/gather_bo.o \
 	gem.o \
 	fb.o \
drivers/gpu/drm/tegra/uapi/firewall.c new file +197 -0

197 lines not on this page · raw message

drivers/gpu/drm/tegra/uapi/submit.c +4 -0
index d625776071e8..7755b31deab0 100644
@@ -236,6 +236,10 @@ static int submit_job_add_gather(struct host1x_job *job,
 	if (next_offset > bo->gather_data_words)
 		return -EINVAL;
 
+	if (tegra_drm_fw_validate(ctx->client, bo->gather_data, *offset,
+				  cmd->words, job_data))
+		return -EINVAL;
+
 	host1x_job_add_gather(job, &bo->base, cmd->words, *offset * 4);
 
 	*offset = next_offset;
drivers/gpu/drm/tegra/uapi/submit.h +3 -0
index 0a165e9e4bda..0e51627e73f8 100644
@@ -14,4 +14,7 @@ struct tegra_drm_submit_data {
 	u32 num_used_mappings;
 };
 
+int tegra_drm_fw_validate(struct tegra_drm_client *client, u32 *data, u32 start,
+			  u32 words, struct tegra_drm_submit_data *submit);
+
 #endif
Signature · 1 line
2.29.2