* [PATCH v6 01/14] gpu: host1x: Add DMA fence implementation
2021-05-30 12:21 [PATCH v6 00/14] TegraDRM UAPI Mikko Perttunen
@ 2021-05-30 12:21 ` Mikko Perttunen
2021-05-30 13:45 ` kernel test robot
2021-05-30 12:21 ` [PATCH v6 02/14] gpu: host1x: Add no-recovery mode Mikko Perttunen
` (12 subsequent siblings)
13 siblings, 1 reply; 16+ messages in thread
From: Mikko Perttunen @ 2021-05-30 12:21 UTC (permalink / raw)
To: thierry.reding, jonathanh, digetx, airlied, daniel
Cc: linux-tegra, dri-devel, 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. The timeout can potentially be removed in the future
after job tracking code has been refactored.
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
---
v6:
* Removed userspace interface.
* Add host1x_ prefixes in various places.
v5:
* Update for change in put_ref prototype.
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 | 211 ++++++++++++++++++++++++++++++++++++
drivers/gpu/host1x/fence.h | 13 +++
drivers/gpu/host1x/intr.c | 9 ++
drivers/gpu/host1x/intr.h | 2 +
include/linux/host1x.h | 3 +
6 files changed, 239 insertions(+)
create mode 100644 drivers/gpu/host1x/fence.c
create mode 100644 drivers/gpu/host1x/fence.h
diff --git a/drivers/gpu/host1x/Makefile b/drivers/gpu/host1x/Makefile
index 096017b8789d..d2b6f7de0498 100644
--- a/drivers/gpu/host1x/Makefile
+++ b/drivers/gpu/host1x/Makefile
@@ -9,6 +9,7 @@ host1x-y = \
job.o \
debug.o \
mipi.o \
+ fence.o \
hw/host1x01.o \
hw/host1x02.o \
hw/host1x04.o \
diff --git a/drivers/gpu/host1x/fence.c b/drivers/gpu/host1x/fence.c
new file mode 100644
index 000000000000..fde59cd8487b
--- /dev/null
+++ b/drivers/gpu/host1x/fence.c
@@ -0,0 +1,211 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Syncpoint dma_fence implementation
+ *
+ * Copyright (c) 2020, NVIDIA Corporation.
+ */
+
+#include <linux/dma-fence.h>
+#include <linux/file.h>
+#include <linux/fs.h>
+#include <linux/slab.h>
+#include <linux/sync_file.h>
+
+#include "fence.h"
+#include "intr.h"
+#include "syncpt.h"
+
+DEFINE_SPINLOCK(lock);
+
+struct host1x_syncpt_fence {
+ struct dma_fence base;
+
+ atomic_t signaling;
+
+ struct host1x_syncpt *sp;
+ u32 threshold;
+
+ struct host1x_waitlist *waiter;
+ void *waiter_ref;
+
+ struct delayed_work timeout_work;
+};
+
+static const char *host1x_syncpt_fence_get_driver_name(struct dma_fence *f)
+{
+ return "host1x";
+}
+
+static const char *host1x_syncpt_fence_get_timeline_name(struct dma_fence *f)
+{
+ return "syncpoint";
+}
+
+static struct host1x_syncpt_fence *to_host1x_fence(struct dma_fence *f)
+{
+ return container_of(f, struct host1x_syncpt_fence, base);
+}
+
+static bool host1x_syncpt_fence_enable_signaling(struct dma_fence *f)
+{
+ struct host1x_syncpt_fence *sf = to_host1x_fence(f);
+ int err;
+
+ if (host1x_syncpt_is_expired(sf->sp, sf->threshold))
+ return false;
+
+ dma_fence_get(f);
+
+ /*
+ * The dma_fence framework requires the fence driver to keep a
+ * reference to any fences for which 'enable_signaling' has been
+ * called (and that have not been signalled).
+ *
+ * We provide a userspace API to create arbitrary syncpoint fences,
+ * so we cannot normally guarantee that all fences get signalled.
+ * As such, setup a timeout, so that long-lasting fences will get
+ * reaped eventually.
+ */
+ schedule_delayed_work(&sf->timeout_work, msecs_to_jiffies(30000));
+
+ err = host1x_intr_add_action(sf->sp->host, sf->sp, sf->threshold,
+ HOST1X_INTR_ACTION_SIGNAL_FENCE, f,
+ sf->waiter, &sf->waiter_ref);
+ if (err) {
+ cancel_delayed_work_sync(&sf->timeout_work);
+ dma_fence_put(f);
+ return false;
+ }
+
+ /* intr framework takes ownership of waiter */
+ sf->waiter = NULL;
+
+ /*
+ * The fence may get signalled at any time after the above call,
+ * so we need to initialize all state used by signalling
+ * before it.
+ */
+
+ return true;
+}
+
+static void host1x_syncpt_fence_release(struct dma_fence *f)
+{
+ struct host1x_syncpt_fence *sf = to_host1x_fence(f);
+
+ if (sf->waiter)
+ kfree(sf->waiter);
+
+ dma_fence_free(f);
+}
+
+const struct dma_fence_ops host1x_syncpt_fence_ops = {
+ .get_driver_name = host1x_syncpt_fence_get_driver_name,
+ .get_timeline_name = host1x_syncpt_fence_get_timeline_name,
+ .enable_signaling = host1x_syncpt_fence_enable_signaling,
+ .release = host1x_syncpt_fence_release,
+};
+
+void host1x_fence_signal(struct host1x_syncpt_fence *f)
+{
+ if (atomic_xchg(&f->signaling, 1))
+ return;
+
+ /*
+ * Cancel pending timeout work - if it races, it will
+ * not get 'f->signaling' and return.
+ */
+ cancel_delayed_work_sync(&f->timeout_work);
+
+ host1x_intr_put_ref(f->sp->host, f->sp->id, f->waiter_ref, false);
+
+ dma_fence_signal(&f->base);
+ dma_fence_put(&f->base);
+}
+
+static void do_fence_timeout(struct work_struct *work)
+{
+ struct delayed_work *dwork = (struct delayed_work *)work;
+ struct host1x_syncpt_fence *f =
+ container_of(dwork, struct host1x_syncpt_fence, timeout_work);
+
+ if (atomic_xchg(&f->signaling, 1))
+ return;
+
+ /*
+ * Cancel pending timeout work - if it races, it will
+ * not get 'f->signaling' and return.
+ */
+ host1x_intr_put_ref(f->sp->host, f->sp->id, f->waiter_ref, true);
+
+ dma_fence_set_error(&f->base, -ETIMEDOUT);
+ dma_fence_signal(&f->base);
+ dma_fence_put(&f->base);
+}
+
+struct dma_fence *host1x_fence_create(struct host1x_syncpt *sp, u32 threshold)
+{
+ struct host1x_syncpt_fence *fence;
+
+ fence = kzalloc(sizeof(*fence), GFP_KERNEL);
+ if (!fence)
+ return ERR_PTR(-ENOMEM);
+
+ fence->waiter = kzalloc(sizeof(*fence->waiter), GFP_KERNEL);
+ if (!fence->waiter)
+ return ERR_PTR(-ENOMEM);
+
+ fence->sp = sp;
+ fence->threshold = threshold;
+
+ dma_fence_init(&fence->base, &host1x_syncpt_fence_ops, &lock,
+ dma_fence_context_alloc(1), 0);
+
+ INIT_DELAYED_WORK(&fence->timeout_work, do_fence_timeout);
+
+ return &fence->base;
+}
+EXPORT_SYMBOL(host1x_fence_create);
+
+int host1x_fence_create_fd(struct host1x_syncpt *sp, u32 threshold)
+{
+ struct sync_file *file;
+ struct dma_fence *f;
+ int fd;
+
+ f = host1x_fence_create(sp, threshold);
+ if (IS_ERR(f))
+ return PTR_ERR(f);
+
+ fd = get_unused_fd_flags(O_CLOEXEC);
+ if (fd < 0) {
+ dma_fence_put(f);
+ return fd;
+ }
+
+ file = sync_file_create(f);
+ dma_fence_put(f);
+ if (!file)
+ return -ENOMEM;
+
+ fd_install(fd, file->file);
+
+ return fd;
+}
+EXPORT_SYMBOL(host1x_fence_create_fd);
+
+int host1x_fence_extract(struct dma_fence *fence, u32 *id, u32 *threshold)
+{
+ struct host1x_syncpt_fence *f;
+
+ if (fence->ops != &host1x_syncpt_fence_ops)
+ return -EINVAL;
+
+ f = container_of(fence, struct host1x_syncpt_fence, base);
+
+ *id = f->sp->id;
+ *threshold = f->threshold;
+
+ return 0;
+}
+EXPORT_SYMBOL(host1x_fence_extract);
diff --git a/drivers/gpu/host1x/fence.h b/drivers/gpu/host1x/fence.h
new file mode 100644
index 000000000000..70c91de82f14
--- /dev/null
+++ b/drivers/gpu/host1x/fence.h
@@ -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
diff --git a/drivers/gpu/host1x/intr.c b/drivers/gpu/host1x/intr.c
index 6d1f3c0fdbe7..45b6be927ec4 100644
--- a/drivers/gpu/host1x/intr.c
+++ b/drivers/gpu/host1x/intr.c
@@ -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])
diff --git a/drivers/gpu/host1x/intr.h b/drivers/gpu/host1x/intr.h
index 6ea55e615e3a..e4c346099273 100644
--- a/drivers/gpu/host1x/intr.h
+++ b/drivers/gpu/host1x/intr.h
@@ -33,6 +33,8 @@ enum host1x_intr_action {
*/
HOST1X_INTR_ACTION_WAKEUP_INTERRUPTIBLE,
+ HOST1X_INTR_ACTION_SIGNAL_FENCE,
+
HOST1X_INTR_ACTION_COUNT
};
diff --git a/include/linux/host1x.h b/include/linux/host1x.h
index e0a41c2b4c7a..f2ea24fe0b0d 100644
--- a/include/linux/host1x.h
+++ b/include/linux/host1x.h
@@ -166,6 +166,9 @@ u32 host1x_syncpt_base_id(struct host1x_syncpt_base *base);
void host1x_syncpt_release_vblank_reservation(struct host1x_client *client,
u32 syncpt_id);
+struct dma_fence *host1x_fence_create(struct host1x_syncpt *sp, u32 threshold);
+int host1x_fence_extract(struct dma_fence *fence, u32 *id, u32 *threshold);
+
/*
* host1x channel
*/
--
2.30.1
^ permalink raw reply related [flat|nested] 16+ messages in thread* Re: [PATCH v6 01/14] gpu: host1x: Add DMA fence implementation
2021-05-30 12:21 ` [PATCH v6 01/14] gpu: host1x: Add DMA fence implementation Mikko Perttunen
@ 2021-05-30 13:45 ` kernel test robot
0 siblings, 0 replies; 16+ messages in thread
From: kernel test robot @ 2021-05-30 13:45 UTC (permalink / raw)
To: Mikko Perttunen, thierry.reding, jonathanh, digetx, airlied,
daniel
Cc: linux-tegra, clang-built-linux, kbuild-all, dri-devel,
Mikko Perttunen
[-- Attachment #1: Type: text/plain, Size: 2905 bytes --]
Hi Mikko,
Thank you for the patch! Perhaps something to improve:
[auto build test WARNING on tegra-drm/drm/tegra/for-next]
[also build test WARNING on v5.13-rc3 next-20210528]
[cannot apply to tegra/for-next]
[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/TegraDRM-UAPI/20210530-205239
base: git://anongit.freedesktop.org/tegra/linux.git drm/tegra/for-next
config: arm-randconfig-r015-20210530 (attached as .config)
compiler: clang version 13.0.0 (https://github.com/llvm/llvm-project bc6799f2f79f0ae87e9f1ebf9d25ba799fbd25a9)
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/5de49ce9211cbf0fbbfeb200e72ce934f4574d4b
git remote add linux-review https://github.com/0day-ci/linux
git fetch --no-tags linux-review Mikko-Perttunen/TegraDRM-UAPI/20210530-205239
git checkout 5de49ce9211cbf0fbbfeb200e72ce934f4574d4b
# 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
Reported-by: kernel test robot <lkp@intel.com>
All warnings (new ones prefixed by >>):
>> drivers/gpu/host1x/fence.c:170:5: warning: no previous prototype for function 'host1x_fence_create_fd' [-Wmissing-prototypes]
int host1x_fence_create_fd(struct host1x_syncpt *sp, u32 threshold)
^
drivers/gpu/host1x/fence.c:170:1: note: declare 'static' if the function is not intended to be used outside of this translation unit
int host1x_fence_create_fd(struct host1x_syncpt *sp, u32 threshold)
^
static
1 warning generated.
vim +/host1x_fence_create_fd +170 drivers/gpu/host1x/fence.c
169
> 170 int host1x_fence_create_fd(struct host1x_syncpt *sp, u32 threshold)
171 {
172 struct sync_file *file;
173 struct dma_fence *f;
174 int fd;
175
176 f = host1x_fence_create(sp, threshold);
177 if (IS_ERR(f))
178 return PTR_ERR(f);
179
180 fd = get_unused_fd_flags(O_CLOEXEC);
181 if (fd < 0) {
182 dma_fence_put(f);
183 return fd;
184 }
185
186 file = sync_file_create(f);
187 dma_fence_put(f);
188 if (!file)
189 return -ENOMEM;
190
191 fd_install(fd, file->file);
192
193 return fd;
194 }
195 EXPORT_SYMBOL(host1x_fence_create_fd);
196
---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all@lists.01.org
[-- Attachment #2: .config.gz --]
[-- Type: application/gzip, Size: 35760 bytes --]
^ permalink raw reply [flat|nested] 16+ messages in thread
* [PATCH v6 02/14] gpu: host1x: Add no-recovery mode
2021-05-30 12:21 [PATCH v6 00/14] TegraDRM UAPI Mikko Perttunen
2021-05-30 12:21 ` [PATCH v6 01/14] gpu: host1x: Add DMA fence implementation Mikko Perttunen
@ 2021-05-30 12:21 ` Mikko Perttunen
2021-05-30 12:21 ` [PATCH v6 03/14] gpu: host1x: Add job release callback Mikko Perttunen
` (11 subsequent siblings)
13 siblings, 0 replies; 16+ messages in thread
From: Mikko Perttunen @ 2021-05-30 12:21 UTC (permalink / raw)
To: thierry.reding, jonathanh, digetx, airlied, daniel
Cc: linux-tegra, dri-devel, 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.
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
---
v5:
* Update for change in put_ref prototype.
* Fixed typo in comment.
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(-)
diff --git a/drivers/gpu/drm/tegra/drm.c b/drivers/gpu/drm/tegra/drm.c
index ce5bdc58d315..992aa5cde510 100644
--- a/drivers/gpu/drm/tegra/drm.c
+++ b/drivers/gpu/drm/tegra/drm.c
@@ -200,6 +200,7 @@ 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
diff --git a/drivers/gpu/host1x/cdma.c b/drivers/gpu/host1x/cdma.c
index 6e6ca774f68d..765e5aa64eb6 100644
--- a/drivers/gpu/host1x/cdma.c
+++ b/drivers/gpu/host1x/cdma.c
@@ -312,10 +312,6 @@ static void update_cdma_locked(struct host1x_cdma *cdma)
bool signal = false;
struct host1x_job *job, *n;
- /* If CDMA is stopped, queue is cleared and we can return */
- if (!cdma->running)
- return;
-
/*
* Walk the sync queue, reading the sync point registers as necessary,
* to consume as many sync queue entries as possible without blocking
@@ -324,7 +320,8 @@ static void update_cdma_locked(struct host1x_cdma *cdma)
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)) {
+ if (!host1x_syncpt_is_expired(sp, job->syncpt_end) &&
+ !job->cancelled) {
/* Start timer on next pending syncpt */
if (job->timeout)
cdma_start_timer_locked(cdma, job);
@@ -413,8 +410,11 @@ void host1x_cdma_update_sync_queue(struct host1x_cdma *cdma,
else
restart_addr = cdma->last_pos;
+ if (!job)
+ goto resume;
+
/* do CPU increments for the remaining syncpts */
- if (job) {
+ if (job->syncpt_recovery) {
dev_dbg(dev, "%s: perform CPU incr on pending buffers\n",
__func__);
@@ -433,8 +433,44 @@ void host1x_cdma_update_sync_queue(struct host1x_cdma *cdma,
dev_dbg(dev, "%s: finished sync_queue modification\n",
__func__);
+ } else {
+ struct host1x_job *failed_job = job;
+
+ host1x_job_dump(dev, job);
+
+ host1x_syncpt_set_locked(job->syncpt);
+ failed_job->cancelled = true;
+
+ list_for_each_entry_continue(job, &cdma->sync_queue, list) {
+ unsigned int i;
+
+ if (job->syncpt != failed_job->syncpt)
+ continue;
+
+ for (i = 0; i < job->num_slots; i++) {
+ unsigned int slot = (job->first_get/8 + i) %
+ HOST1X_PUSHBUFFER_SLOTS;
+ u32 *mapped = cdma->push_buffer.mapped;
+
+ /*
+ * Overwrite opcodes with 0 word writes
+ * to offset 0xbad. This does nothing but
+ * has a easily detected signature in debug
+ * traces.
+ */
+ mapped[2*slot+0] = 0x1bad0000;
+ mapped[2*slot+1] = 0x1bad0000;
+ }
+
+ job->cancelled = true;
+ }
+
+ wmb();
+
+ update_cdma_locked(cdma);
}
+resume:
/* roll back DMAGET and start up channel again */
host1x_hw_cdma_resume(host1x, cdma, restart_addr);
}
@@ -490,6 +526,16 @@ int host1x_cdma_begin(struct host1x_cdma *cdma, struct host1x_job *job)
mutex_lock(&cdma->lock);
+ /*
+ * Check if syncpoint was locked due to previous job timeout.
+ * This needs to be done within the cdma lock to avoid a race
+ * with the timeout handler.
+ */
+ if (job->syncpt->locked) {
+ mutex_unlock(&cdma->lock);
+ return -EPERM;
+ }
+
if (job->timeout) {
/* init state on first submit with timeout value */
if (!cdma->timeout.initialized) {
diff --git a/drivers/gpu/host1x/hw/channel_hw.c b/drivers/gpu/host1x/hw/channel_hw.c
index d4c28faf27d1..bf21512e5078 100644
--- a/drivers/gpu/host1x/hw/channel_hw.c
+++ b/drivers/gpu/host1x/hw/channel_hw.c
@@ -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");
diff --git a/drivers/gpu/host1x/job.c b/drivers/gpu/host1x/job.c
index adbdc225de8d..8f59b34672c2 100644
--- a/drivers/gpu/host1x/job.c
+++ b/drivers/gpu/host1x/job.c
@@ -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, false);
+
if (job->syncpt)
host1x_syncpt_put(job->syncpt);
diff --git a/drivers/gpu/host1x/syncpt.c b/drivers/gpu/host1x/syncpt.c
index e648ebbb2027..d198a10848c6 100644
--- a/drivers/gpu/host1x/syncpt.c
+++ b/drivers/gpu/host1x/syncpt.c
@@ -407,6 +407,8 @@ static void syncpt_release(struct kref *ref)
atomic_set(&sp->max_val, host1x_syncpt_read(sp));
+ sp->locked = false;
+
mutex_lock(&sp->host->syncpt_mutex);
host1x_syncpt_base_free(sp->base);
diff --git a/drivers/gpu/host1x/syncpt.h b/drivers/gpu/host1x/syncpt.h
index a6766f8d55ee..93e894677d89 100644
--- a/drivers/gpu/host1x/syncpt.h
+++ b/drivers/gpu/host1x/syncpt.h
@@ -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
diff --git a/include/linux/host1x.h b/include/linux/host1x.h
index f2ea24fe0b0d..98dfb72e6c77 100644
--- a/include/linux/host1x.h
+++ b/include/linux/host1x.h
@@ -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,
--
2.30.1
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v6 03/14] gpu: host1x: Add job release callback
2021-05-30 12:21 [PATCH v6 00/14] TegraDRM UAPI Mikko Perttunen
2021-05-30 12:21 ` [PATCH v6 01/14] gpu: host1x: Add DMA fence implementation Mikko Perttunen
2021-05-30 12:21 ` [PATCH v6 02/14] gpu: host1x: Add no-recovery mode Mikko Perttunen
@ 2021-05-30 12:21 ` Mikko Perttunen
2021-05-30 12:21 ` [PATCH v6 04/14] gpu: host1x: Add support for syncpoint waits in CDMA pushbuffer Mikko Perttunen
` (10 subsequent siblings)
13 siblings, 0 replies; 16+ messages in thread
From: Mikko Perttunen @ 2021-05-30 12:21 UTC (permalink / raw)
To: thierry.reding, jonathanh, digetx, airlied, daniel
Cc: linux-tegra, dri-devel, 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.
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
---
drivers/gpu/host1x/job.c | 3 +++
include/linux/host1x.h | 4 ++++
2 files changed, 7 insertions(+)
diff --git a/drivers/gpu/host1x/job.c b/drivers/gpu/host1x/job.c
index 8f59b34672c2..09097e19c0d0 100644
--- a/drivers/gpu/host1x/job.c
+++ b/drivers/gpu/host1x/job.c
@@ -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, false);
diff --git a/include/linux/host1x.h b/include/linux/host1x.h
index 98dfb72e6c77..432125804208 100644
--- a/include/linux/host1x.h
+++ b/include/linux/host1x.h
@@ -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,
--
2.30.1
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v6 04/14] gpu: host1x: Add support for syncpoint waits in CDMA pushbuffer
2021-05-30 12:21 [PATCH v6 00/14] TegraDRM UAPI Mikko Perttunen
` (2 preceding siblings ...)
2021-05-30 12:21 ` [PATCH v6 03/14] gpu: host1x: Add job release callback Mikko Perttunen
@ 2021-05-30 12:21 ` Mikko Perttunen
2021-05-30 12:21 ` [PATCH v6 05/14] drm/tegra: Extract tegra_gem_lookup Mikko Perttunen
` (9 subsequent siblings)
13 siblings, 0 replies; 16+ messages in thread
From: Mikko Perttunen @ 2021-05-30 12:21 UTC (permalink / raw)
To: thierry.reding, jonathanh, digetx, airlied, daniel
Cc: linux-tegra, dri-devel, 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.
Also supported are waits relative to the start of the job,
which are useful for jobs doing multiple things with an engine
that doesn't natively support pipelining.
While at it, use 32-bit waits on chips that support them.
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
---
v6:
- Change into specified class after wait
- Support waits relative to start of job
- Support 32-bit waits
---
drivers/gpu/host1x/hw/channel_hw.c | 85 +++++++++++++++++-----
drivers/gpu/host1x/hw/debug_hw.c | 9 ++-
drivers/gpu/host1x/hw/hw_host1x02_uclass.h | 12 +++
drivers/gpu/host1x/hw/hw_host1x04_uclass.h | 12 +++
drivers/gpu/host1x/hw/hw_host1x05_uclass.h | 12 +++
drivers/gpu/host1x/hw/hw_host1x06_uclass.h | 12 +++
drivers/gpu/host1x/hw/hw_host1x07_uclass.h | 12 +++
drivers/gpu/host1x/job.c | 70 +++++++++++++-----
drivers/gpu/host1x/job.h | 16 ++++
include/linux/host1x.h | 6 +-
10 files changed, 203 insertions(+), 43 deletions(-)
diff --git a/drivers/gpu/host1x/hw/channel_hw.c b/drivers/gpu/host1x/hw/channel_hw.c
index bf21512e5078..1999780a7203 100644
--- a/drivers/gpu/host1x/hw/channel_hw.c
+++ b/drivers/gpu/host1x/hw/channel_hw.c
@@ -47,39 +47,84 @@ static void trace_write_gather(struct host1x_cdma *cdma, struct host1x_bo *bo,
}
}
-static void submit_gathers(struct host1x_job *job)
+static void submit_wait(struct host1x_cdma *cdma, u32 id, u32 threshold,
+ u32 next_class)
+{
+#if HOST1X_HW >= 2
+ host1x_cdma_push_wide(cdma,
+ host1x_opcode_setclass(
+ HOST1X_CLASS_HOST1X,
+ HOST1X_UCLASS_LOAD_SYNCPT_PAYLOAD_32,
+ /* WAIT_SYNCPT_32 is at SYNCPT_PAYLOAD_32+2 */
+ BIT(0) | BIT(2)
+ ),
+ threshold,
+ id,
+ host1x_opcode_setclass(next_class, 0, 0)
+ );
+#else
+ /* TODO add waitchk or use waitbases or other mitigation */
+ host1x_cdma_push(cdma,
+ host1x_opcode_setclass(
+ HOST1X_CLASS_HOST1X,
+ host1x_uclass_wait_syncpt_r(),
+ BIT(0)
+ ),
+ host1x_class_host_wait_syncpt(id, threshold)
+ );
+ host1x_cdma_push(cdma,
+ host1x_opcode_setclass(next_class, 0, 0),
+ HOST1X_OPCODE_NOP
+ );
+#endif
+}
+
+static void submit_gathers(struct host1x_job *job, u32 job_syncpt_base)
{
struct host1x_cdma *cdma = &job->channel->cdma;
#if HOST1X_HW < 6
struct device *dev = job->channel->dev;
#endif
unsigned int i;
+ u32 threshold;
- 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) {
+ if (cmd->wait.relative)
+ threshold = job_syncpt_base + cmd->wait.threshold;
+ else
+ threshold = cmd->wait.threshold;
- trace_write_gather(cdma, g->bo, g->offset, g->words);
+ submit_wait(cdma, cmd->wait.id, threshold, cmd->wait.next_class);
+ } else {
+ struct host1x_job_gather *g = &cmd->gather;
+
+ dma_addr_t addr = g->base + g->offset;
+ u32 op2, op3;
+
+ op2 = lower_32_bits(addr);
+ op3 = upper_32_bits(addr);
- if (op3 != 0) {
+ 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 +171,7 @@ 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 */
@@ -181,7 +226,7 @@ static int channel_submit(struct host1x_job *job)
host1x_opcode_setclass(job->class, 0, 0),
HOST1X_OPCODE_NOP);
- submit_gathers(job);
+ submit_gathers(job, syncval - user_syncpt_incrs);
/* end CDMA submit & stash pinned hMems into sync queue */
host1x_cdma_end(&ch->cdma, job);
diff --git a/drivers/gpu/host1x/hw/debug_hw.c b/drivers/gpu/host1x/hw/debug_hw.c
index ceb48229d14b..35952fd5597e 100644
--- a/drivers/gpu/host1x/hw/debug_hw.c
+++ b/drivers/gpu/host1x/hw/debug_hw.c
@@ -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
diff --git a/drivers/gpu/host1x/hw/hw_host1x02_uclass.h b/drivers/gpu/host1x/hw/hw_host1x02_uclass.h
index 4fc51f70496b..0a2ab8f1da6f 100644
--- a/drivers/gpu/host1x/hw/hw_host1x02_uclass.h
+++ b/drivers/gpu/host1x/hw/hw_host1x02_uclass.h
@@ -165,5 +165,17 @@ static inline u32 host1x_uclass_indoff_rwn_read_v(void)
}
#define HOST1X_UCLASS_INDOFF_INDROFFSET_F(v) \
host1x_uclass_indoff_indroffset_f(v)
+static inline u32 host1x_uclass_load_syncpt_payload_32_r(void)
+{
+ return 0x4e;
+}
+#define HOST1X_UCLASS_LOAD_SYNCPT_PAYLOAD_32 \
+ host1x_uclass_load_syncpt_payload_32_r()
+static inline u32 host1x_uclass_wait_syncpt_32_r(void)
+{
+ return 0x50;
+}
+#define HOST1X_UCLASS_WAIT_SYNCPT_32 \
+ host1x_uclass_wait_syncpt_32_r()
#endif
diff --git a/drivers/gpu/host1x/hw/hw_host1x04_uclass.h b/drivers/gpu/host1x/hw/hw_host1x04_uclass.h
index 9e84a4adca9f..60c692b92955 100644
--- a/drivers/gpu/host1x/hw/hw_host1x04_uclass.h
+++ b/drivers/gpu/host1x/hw/hw_host1x04_uclass.h
@@ -165,5 +165,17 @@ static inline u32 host1x_uclass_indoff_rwn_read_v(void)
}
#define HOST1X_UCLASS_INDOFF_INDROFFSET_F(v) \
host1x_uclass_indoff_indroffset_f(v)
+static inline u32 host1x_uclass_load_syncpt_payload_32_r(void)
+{
+ return 0x4e;
+}
+#define HOST1X_UCLASS_LOAD_SYNCPT_PAYLOAD_32 \
+ host1x_uclass_load_syncpt_payload_32_r()
+static inline u32 host1x_uclass_wait_syncpt_32_r(void)
+{
+ return 0x50;
+}
+#define HOST1X_UCLASS_WAIT_SYNCPT_32 \
+ host1x_uclass_wait_syncpt_32_r()
#endif
diff --git a/drivers/gpu/host1x/hw/hw_host1x05_uclass.h b/drivers/gpu/host1x/hw/hw_host1x05_uclass.h
index aee5a4e32877..2fcc9a2ad3ef 100644
--- a/drivers/gpu/host1x/hw/hw_host1x05_uclass.h
+++ b/drivers/gpu/host1x/hw/hw_host1x05_uclass.h
@@ -165,5 +165,17 @@ static inline u32 host1x_uclass_indoff_rwn_read_v(void)
}
#define HOST1X_UCLASS_INDOFF_INDROFFSET_F(v) \
host1x_uclass_indoff_indroffset_f(v)
+static inline u32 host1x_uclass_load_syncpt_payload_32_r(void)
+{
+ return 0x4e;
+}
+#define HOST1X_UCLASS_LOAD_SYNCPT_PAYLOAD_32 \
+ host1x_uclass_load_syncpt_payload_32_r()
+static inline u32 host1x_uclass_wait_syncpt_32_r(void)
+{
+ return 0x50;
+}
+#define HOST1X_UCLASS_WAIT_SYNCPT_32 \
+ host1x_uclass_wait_syncpt_32_r()
#endif
diff --git a/drivers/gpu/host1x/hw/hw_host1x06_uclass.h b/drivers/gpu/host1x/hw/hw_host1x06_uclass.h
index c4bacdb7155f..5f831438d19b 100644
--- a/drivers/gpu/host1x/hw/hw_host1x06_uclass.h
+++ b/drivers/gpu/host1x/hw/hw_host1x06_uclass.h
@@ -165,5 +165,17 @@ static inline u32 host1x_uclass_indoff_rwn_read_v(void)
}
#define HOST1X_UCLASS_INDOFF_INDROFFSET_F(v) \
host1x_uclass_indoff_indroffset_f(v)
+static inline u32 host1x_uclass_load_syncpt_payload_32_r(void)
+{
+ return 0x4e;
+}
+#define HOST1X_UCLASS_LOAD_SYNCPT_PAYLOAD_32 \
+ host1x_uclass_load_syncpt_payload_32_r()
+static inline u32 host1x_uclass_wait_syncpt_32_r(void)
+{
+ return 0x50;
+}
+#define HOST1X_UCLASS_WAIT_SYNCPT_32 \
+ host1x_uclass_wait_syncpt_32_r()
#endif
diff --git a/drivers/gpu/host1x/hw/hw_host1x07_uclass.h b/drivers/gpu/host1x/hw/hw_host1x07_uclass.h
index c74070f3f203..8cd2ef087d5d 100644
--- a/drivers/gpu/host1x/hw/hw_host1x07_uclass.h
+++ b/drivers/gpu/host1x/hw/hw_host1x07_uclass.h
@@ -165,5 +165,17 @@ static inline u32 host1x_uclass_indoff_rwn_read_v(void)
}
#define HOST1X_UCLASS_INDOFF_INDROFFSET_F(v) \
host1x_uclass_indoff_indroffset_f(v)
+static inline u32 host1x_uclass_load_syncpt_payload_32_r(void)
+{
+ return 0x4e;
+}
+#define HOST1X_UCLASS_LOAD_SYNCPT_PAYLOAD_32 \
+ host1x_uclass_load_syncpt_payload_32_r()
+static inline u32 host1x_uclass_wait_syncpt_32_r(void)
+{
+ return 0x50;
+}
+#define HOST1X_UCLASS_WAIT_SYNCPT_32 \
+ host1x_uclass_wait_syncpt_32_r()
#endif
diff --git a/drivers/gpu/host1x/job.c b/drivers/gpu/host1x/job.c
index 09097e19c0d0..32619b73a2fc 100644
--- a/drivers/gpu/host1x/job.c
+++ b/drivers/gpu/host1x/job.c
@@ -38,7 +38,7 @@ struct host1x_job *host1x_job_alloc(struct host1x_channel *ch,
total = sizeof(struct host1x_job) +
(u64)num_relocs * sizeof(struct host1x_reloc) +
(u64)num_unpins * sizeof(struct host1x_job_unpin_data) +
- (u64)num_cmdbufs * sizeof(struct host1x_job_gather) +
+ (u64)num_cmdbufs * sizeof(struct host1x_job_cmd) +
(u64)num_unpins * sizeof(dma_addr_t) +
(u64)num_unpins * sizeof(u32 *);
if (total > ULONG_MAX)
@@ -57,8 +57,8 @@ struct host1x_job *host1x_job_alloc(struct host1x_channel *ch,
mem += num_relocs * sizeof(struct host1x_reloc);
job->unpins = num_unpins ? mem : NULL;
mem += num_unpins * sizeof(struct host1x_job_unpin_data);
- job->gathers = num_cmdbufs ? mem : NULL;
- mem += num_cmdbufs * sizeof(struct host1x_job_gather);
+ job->cmds = num_cmdbufs ? mem : NULL;
+ mem += num_cmdbufs * sizeof(struct host1x_job_cmd);
job->addr_phys = num_unpins ? mem : NULL;
job->reloc_addr_phys = job->addr_phys;
@@ -101,22 +101,38 @@ EXPORT_SYMBOL(host1x_job_put);
void host1x_job_add_gather(struct host1x_job *job, struct host1x_bo *bo,
unsigned int words, unsigned int offset)
{
- struct host1x_job_gather *gather = &job->gathers[job->num_gathers];
+ struct host1x_job_gather *gather = &job->cmds[job->num_cmds].gather;
gather->words = words;
gather->bo = bo;
gather->offset = offset;
- job->num_gathers++;
+ job->num_cmds++;
}
EXPORT_SYMBOL(host1x_job_add_gather);
+void host1x_job_add_wait(struct host1x_job *job, u32 id, u32 thresh,
+ bool relative, u32 next_class)
+{
+ struct host1x_job_cmd *cmd = &job->cmds[job->num_cmds];
+
+ cmd->is_wait = true;
+ cmd->wait.id = id;
+ cmd->wait.threshold = thresh;
+ cmd->wait.next_class = next_class;
+ cmd->wait.relative = relative;
+
+ job->num_cmds++;
+}
+EXPORT_SYMBOL(host1x_job_add_wait);
+
static unsigned int pin_job(struct host1x *host, struct host1x_job *job)
{
struct host1x_client *client = job->client;
struct device *dev = client->dev;
struct host1x_job_gather *g;
struct iommu_domain *domain;
+ struct sg_table *sgt;
unsigned int i;
int err;
@@ -126,7 +142,6 @@ static unsigned int pin_job(struct host1x *host, struct host1x_job *job)
for (i = 0; i < job->num_relocs; i++) {
struct host1x_reloc *reloc = &job->relocs[i];
dma_addr_t phys_addr, *phys;
- struct sg_table *sgt;
reloc->target.bo = host1x_bo_get(reloc->target.bo);
if (!reloc->target.bo) {
@@ -202,17 +217,20 @@ static unsigned int pin_job(struct host1x *host, struct host1x_job *job)
if (IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL))
return 0;
- for (i = 0; i < job->num_gathers; i++) {
+ for (i = 0; i < job->num_cmds; i++) {
size_t gather_size = 0;
struct scatterlist *sg;
- struct sg_table *sgt;
dma_addr_t phys_addr;
unsigned long shift;
struct iova *alloc;
dma_addr_t *phys;
unsigned int j;
- g = &job->gathers[i];
+ if (job->cmds[i].is_wait)
+ continue;
+
+ g = &job->cmds[i].gather;
+
g->bo = host1x_bo_get(g->bo);
if (!g->bo) {
err = -EINVAL;
@@ -545,8 +563,13 @@ static inline int copy_gathers(struct device *host, struct host1x_job *job,
fw.num_relocs = job->num_relocs;
fw.class = job->class;
- 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;
+
+ if (job->cmds[i].is_wait)
+ continue;
+
+ g = &job->cmds[i].gather;
size += g->words * sizeof(u32);
}
@@ -568,10 +591,14 @@ static inline int copy_gathers(struct device *host, struct host1x_job *job,
job->gather_copy_size = size;
- 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;
void *gather;
+ if (job->cmds[i].is_wait)
+ continue;
+ g = &job->cmds[i].gather;
+
/* Copy the gather */
gather = host1x_bo_mmap(g->bo);
memcpy(job->gather_copy_mapped + offset, gather + g->offset,
@@ -614,8 +641,12 @@ int host1x_job_pin(struct host1x_job *job, struct device *dev)
}
/* patch gathers */
- 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;
+
+ if (job->cmds[i].is_wait)
+ continue;
+ g = &job->cmds[i].gather;
/* process each gather mem only once */
if (g->handled)
@@ -625,10 +656,11 @@ int host1x_job_pin(struct host1x_job *job, struct device *dev)
if (!IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL))
g->base = job->gather_addr_phys[i];
- for (j = i + 1; j < job->num_gathers; j++) {
- if (job->gathers[j].bo == g->bo) {
- job->gathers[j].handled = true;
- job->gathers[j].base = g->base;
+ for (j = i + 1; j < job->num_cmds; j++) {
+ if (!job->cmds[j].is_wait &&
+ job->cmds[j].gather.bo == g->bo) {
+ job->cmds[j].gather.handled = true;
+ job->cmds[j].gather.base = g->base;
}
}
diff --git a/drivers/gpu/host1x/job.h b/drivers/gpu/host1x/job.h
index 94bc2e4ae241..b4428c5495c9 100644
--- a/drivers/gpu/host1x/job.h
+++ b/drivers/gpu/host1x/job.h
@@ -18,6 +18,22 @@ struct host1x_job_gather {
bool handled;
};
+struct host1x_job_wait {
+ u32 id;
+ u32 threshold;
+ u32 next_class;
+ bool relative;
+};
+
+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;
diff --git a/include/linux/host1x.h b/include/linux/host1x.h
index 432125804208..9f4c01d45544 100644
--- a/include/linux/host1x.h
+++ b/include/linux/host1x.h
@@ -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 +275,8 @@ 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,
+ bool relative, u32 next_class);
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);
--
2.30.1
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v6 05/14] drm/tegra: Extract tegra_gem_lookup
2021-05-30 12:21 [PATCH v6 00/14] TegraDRM UAPI Mikko Perttunen
` (3 preceding siblings ...)
2021-05-30 12:21 ` [PATCH v6 04/14] gpu: host1x: Add support for syncpoint waits in CDMA pushbuffer Mikko Perttunen
@ 2021-05-30 12:21 ` Mikko Perttunen
2021-05-30 12:21 ` [PATCH v6 06/14] drm/tegra: Add new UAPI to header Mikko Perttunen
` (8 subsequent siblings)
13 siblings, 0 replies; 16+ messages in thread
From: Mikko Perttunen @ 2021-05-30 12:21 UTC (permalink / raw)
To: thierry.reding, jonathanh, digetx, airlied, daniel
Cc: linux-tegra, dri-devel, Mikko Perttunen
The static function host1x_bo_lookup in drm.c is also useful
elsewhere. Extract it as tegra_gem_lookup in gem.c.
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
---
v6:
- New patch
---
drivers/gpu/drm/tegra/drm.c | 20 +++-----------------
drivers/gpu/drm/tegra/gem.c | 13 +++++++++++++
drivers/gpu/drm/tegra/gem.h | 2 ++
3 files changed, 18 insertions(+), 17 deletions(-)
diff --git a/drivers/gpu/drm/tegra/drm.c b/drivers/gpu/drm/tegra/drm.c
index 992aa5cde510..c9a35c1dc970 100644
--- a/drivers/gpu/drm/tegra/drm.c
+++ b/drivers/gpu/drm/tegra/drm.c
@@ -106,20 +106,6 @@ static void tegra_drm_context_free(struct tegra_drm_context *context)
kfree(context);
}
-static struct host1x_bo *
-host1x_bo_lookup(struct drm_file *file, u32 handle)
-{
- struct drm_gem_object *gem;
- struct tegra_bo *bo;
-
- gem = drm_gem_object_lookup(file, handle);
- if (!gem)
- return NULL;
-
- bo = to_tegra_bo(gem);
- return &bo->base;
-}
-
static int host1x_reloc_copy_from_user(struct host1x_reloc *dest,
struct drm_tegra_reloc __user *src,
struct drm_device *drm,
@@ -150,11 +136,11 @@ static int host1x_reloc_copy_from_user(struct host1x_reloc *dest,
dest->flags = HOST1X_RELOC_READ | HOST1X_RELOC_WRITE;
- dest->cmdbuf.bo = host1x_bo_lookup(file, cmdbuf);
+ dest->cmdbuf.bo = tegra_gem_lookup(file, cmdbuf);
if (!dest->cmdbuf.bo)
return -ENOENT;
- dest->target.bo = host1x_bo_lookup(file, target);
+ dest->target.bo = tegra_gem_lookup(file, target);
if (!dest->target.bo)
return -ENOENT;
@@ -237,7 +223,7 @@ int tegra_drm_submit(struct tegra_drm_context *context,
goto fail;
}
- bo = host1x_bo_lookup(file, cmdbuf.handle);
+ bo = tegra_gem_lookup(file, cmdbuf.handle);
if (!bo) {
err = -ENOENT;
goto fail;
diff --git a/drivers/gpu/drm/tegra/gem.c b/drivers/gpu/drm/tegra/gem.c
index 26af8daa9a16..6ec598f5d5b3 100644
--- a/drivers/gpu/drm/tegra/gem.c
+++ b/drivers/gpu/drm/tegra/gem.c
@@ -707,3 +707,16 @@ struct drm_gem_object *tegra_gem_prime_import(struct drm_device *drm,
return &bo->gem;
}
+
+struct host1x_bo *tegra_gem_lookup(struct drm_file *file, u32 handle)
+{
+ struct drm_gem_object *gem;
+ struct tegra_bo *bo;
+
+ gem = drm_gem_object_lookup(file, handle);
+ if (!gem)
+ return NULL;
+
+ bo = to_tegra_bo(gem);
+ return &bo->base;
+}
diff --git a/drivers/gpu/drm/tegra/gem.h b/drivers/gpu/drm/tegra/gem.h
index fafb5724499b..2752c66e129c 100644
--- a/drivers/gpu/drm/tegra/gem.h
+++ b/drivers/gpu/drm/tegra/gem.h
@@ -74,4 +74,6 @@ struct dma_buf *tegra_gem_prime_export(struct drm_gem_object *gem,
struct drm_gem_object *tegra_gem_prime_import(struct drm_device *drm,
struct dma_buf *buf);
+struct host1x_bo *tegra_gem_lookup(struct drm_file *file, u32 handle);
+
#endif
--
2.30.1
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v6 06/14] drm/tegra: Add new UAPI to header
2021-05-30 12:21 [PATCH v6 00/14] TegraDRM UAPI Mikko Perttunen
` (4 preceding siblings ...)
2021-05-30 12:21 ` [PATCH v6 05/14] drm/tegra: Extract tegra_gem_lookup Mikko Perttunen
@ 2021-05-30 12:21 ` Mikko Perttunen
2021-05-30 12:21 ` [PATCH v6 07/14] drm/tegra: Boot VIC during runtime PM resume Mikko Perttunen
` (7 subsequent siblings)
13 siblings, 0 replies; 16+ messages in thread
From: Mikko Perttunen @ 2021-05-30 12:21 UTC (permalink / raw)
To: thierry.reding, jonathanh, digetx, airlied, daniel
Cc: linux-tegra, dri-devel, 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.
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
---
v6:
* Fix comment
* Add syncpoint allocation/freeing IOCTLs, use syncpoint ID
instead of syncpoint FD
* Add syncobj fields to submit IOCTL
* Add relative syncpoint wait
* Add hardware flags field
* Add syncpoint wait ioctl
* Remove new GEM_CREATE/GEM_MMAP as unneeded
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 | 414 +++++++++++++++++++++++++++++++++--
1 file changed, 391 insertions(+), 23 deletions(-)
diff --git a/include/uapi/drm/tegra_drm.h b/include/uapi/drm/tegra_drm.h
index c4df3c3668b3..4820ae9e65c5 100644
--- a/include/uapi/drm/tegra_drm.h
+++ b/include/uapi/drm/tegra_drm.h
@@ -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)
@@ -649,8 +632,8 @@ struct drm_tegra_gem_get_flags {
#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
@@ -674,6 +657,391 @@ 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 */
+
+/**
+ * Reported by the driver in the `hardware_flags` field.
+ *
+ * If set, the engine is cache coherent with regard to the system
+ * memory.
+ */
+#define DRM_TEGRA_CHANNEL_OPEN_HW_CACHE_COHERENT (1<<0)
+
+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;
+
+ /**
+ * @hardware_flags: [out]
+ *
+ * Flags describing the hardware capabilities.
+ */
+ __u32 hardware_flags;
+};
+
+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.
+ */
+ 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 {
+ /**
+ * @id: [in]
+ *
+ * ID of the syncpoint that the job will increment.
+ */
+ __u32 id;
+
+ /**
+ * @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
+/**
+ * Wait for a syncpoint to reach a value before continuing with further
+ * commands. The threshold is calculated in relation to the start of the
+ * job.
+ */
+#define DRM_TEGRA_SUBMIT_CMD_WAIT_SYNCPT_RELATIVE 2
+
+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;
+
+ /**
+ * @syncobj_in: [in]
+ *
+ * Handle for DRM syncobj that will be waited before submission.
+ * Ignored if zero.
+ */
+ __u32 syncobj_in;
+
+ /**
+ * @syncobj_out: [in]
+ *
+ * Handle for DRM syncobj that will have its fence replaced with
+ * the job's completion fence. Ignored if zero.
+ */
+ __u32 syncobj_out;
+
+ /**
+ * @syncpt_incr: [in,out]
+ *
+ * Information about the syncpoint the job will increment.
+ */
+ struct drm_tegra_submit_syncpt_incr syncpt_incr;
+};
+
+struct drm_tegra_syncpoint_allocate {
+ /**
+ * @id: [out]
+ *
+ * ID of allocated syncpoint.
+ */
+ __u32 id;
+};
+
+struct drm_tegra_syncpoint_free {
+ /**
+ * @id: [in]
+ *
+ * ID of syncpoint to free.
+ */
+ __u32 id;
+};
+
+struct drm_tegra_syncpoint_wait {
+ /**
+ * @timeout: [in]
+ *
+ * Absolute timestamp at which the wait will time out.
+ */
+ __s64 timeout_ns;
+
+ /**
+ * @id: [in]
+ *
+ * ID of syncpoint to wait on.
+ */
+ __u32 id;
+
+ /**
+ * @threshold: [in]
+ *
+ * Threshold to wait for.
+ */
+ __u32 threshold;
+
+ /**
+ * @value: [out]
+ *
+ * Value of the syncpoint upon wait completion.
+ */
+ __u32 value;
+
+ __u32 padding[1];
+};
+
+#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_SYNCPOINT_ALLOCATE DRM_IOWR(DRM_COMMAND_BASE + 0x20, struct drm_tegra_syncpoint_allocate)
+#define DRM_IOCTL_TEGRA_SYNCPOINT_FREE DRM_IOWR(DRM_COMMAND_BASE + 0x21, struct drm_tegra_syncpoint_free)
+#define DRM_IOCTL_TEGRA_SYNCPOINT_WAIT DRM_IOWR(DRM_COMMAND_BASE + 0x22, struct drm_tegra_syncpoint_wait)
+
#if defined(__cplusplus)
}
#endif
--
2.30.1
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v6 07/14] drm/tegra: Boot VIC during runtime PM resume
2021-05-30 12:21 [PATCH v6 00/14] TegraDRM UAPI Mikko Perttunen
` (5 preceding siblings ...)
2021-05-30 12:21 ` [PATCH v6 06/14] drm/tegra: Add new UAPI to header Mikko Perttunen
@ 2021-05-30 12:21 ` Mikko Perttunen
2021-05-30 12:21 ` [PATCH v6 08/14] drm/tegra: Allocate per-engine channel in core code Mikko Perttunen
` (6 subsequent siblings)
13 siblings, 0 replies; 16+ messages in thread
From: Mikko Perttunen @ 2021-05-30 12:21 UTC (permalink / raw)
To: thierry.reding, jonathanh, digetx, airlied, daniel
Cc: linux-tegra, dri-devel, 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.
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
---
v6:
* Use pm_runtime_resume_and_get
v3:
* runtime_get/put is now done directly from submit path, so no
callbacks are added
* Reworded.
---
drivers/gpu/drm/tegra/vic.c | 112 ++++++++++++++++--------------------
1 file changed, 51 insertions(+), 61 deletions(-)
diff --git a/drivers/gpu/drm/tegra/vic.c b/drivers/gpu/drm/tegra/vic.c
index 72aea1cc0cfa..35cd9974eea9 100644
--- a/drivers/gpu/drm/tegra/vic.c
+++ b/drivers/gpu/drm/tegra/vic.c
@@ -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;
@@ -170,8 +124,6 @@ static int vic_boot(struct vic *vic)
return err;
}
- vic->booted = true;
-
return 0;
}
@@ -325,35 +277,74 @@ 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_resume_and_get(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_resume_and_get(vic->dev);
+ if (err < 0)
+ 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)
@@ -361,7 +352,6 @@ 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);
}
--
2.30.1
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v6 08/14] drm/tegra: Allocate per-engine channel in core code
2021-05-30 12:21 [PATCH v6 00/14] TegraDRM UAPI Mikko Perttunen
` (6 preceding siblings ...)
2021-05-30 12:21 ` [PATCH v6 07/14] drm/tegra: Boot VIC during runtime PM resume Mikko Perttunen
@ 2021-05-30 12:21 ` Mikko Perttunen
2021-05-30 12:21 ` [PATCH v6 09/14] drm/tegra: Implement new UAPI Mikko Perttunen
` (5 subsequent siblings)
13 siblings, 0 replies; 16+ messages in thread
From: Mikko Perttunen @ 2021-05-30 12:21 UTC (permalink / raw)
To: thierry.reding, jonathanh, digetx, airlied, daniel
Cc: linux-tegra, dri-devel, Mikko Perttunen
To avoid code duplication, allocate the per-engine shared channel in
the core code instead. This is the usual channel that all jobs are
submitted to when MLOCKing is not in use. Once MLOCKs are implemented
on Host1x side, we can also update this to avoid allocating a shared
channel when MLOCKs are enabled.
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
---
v6:
- Reorder fields in drm.h to put core-controlled fields first
---
drivers/gpu/drm/tegra/drm.c | 11 +++++++++++
drivers/gpu/drm/tegra/drm.h | 2 ++
2 files changed, 13 insertions(+)
diff --git a/drivers/gpu/drm/tegra/drm.c b/drivers/gpu/drm/tegra/drm.c
index c9a35c1dc970..a27e5a89d676 100644
--- a/drivers/gpu/drm/tegra/drm.c
+++ b/drivers/gpu/drm/tegra/drm.c
@@ -869,6 +869,14 @@ static const 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;
@@ -885,6 +893,9 @@ 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;
}
diff --git a/drivers/gpu/drm/tegra/drm.h b/drivers/gpu/drm/tegra/drm.h
index f38de08e0c95..1ec07034727c 100644
--- a/drivers/gpu/drm/tegra/drm.h
+++ b/drivers/gpu/drm/tegra/drm.h
@@ -86,7 +86,9 @@ struct tegra_drm_client {
struct host1x_client base;
struct list_head list;
struct tegra_drm *drm;
+ struct host1x_channel *shared_channel;
+ /* Set by driver */
unsigned int version;
const struct tegra_drm_client_ops *ops;
};
--
2.30.1
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v6 09/14] drm/tegra: Implement new UAPI
2021-05-30 12:21 [PATCH v6 00/14] TegraDRM UAPI Mikko Perttunen
` (7 preceding siblings ...)
2021-05-30 12:21 ` [PATCH v6 08/14] drm/tegra: Allocate per-engine channel in core code Mikko Perttunen
@ 2021-05-30 12:21 ` Mikko Perttunen
2021-05-30 12:21 ` [PATCH v6 10/14] drm/tegra: Implement syncpoint management UAPI Mikko Perttunen
` (4 subsequent siblings)
13 siblings, 0 replies; 16+ messages in thread
From: Mikko Perttunen @ 2021-05-30 12:21 UTC (permalink / raw)
To: thierry.reding, jonathanh, digetx, airlied, daniel
Cc: linux-tegra, dri-devel, 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.
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
---
v6:
* Whitespace fixes
* Tell userspace if engine is cache coherent when opening
a channel
* Removed uapi subdirectory
* Reuse tegra_drm_context instead of creating new
tegra_drm_channel_ctx
* Remove tegra_drm_channel_ctx_lock
* Simplify id variable names
* Remove unused close_channel_ctx function
* Use tegra_gem_lookup
* Clean up channel_open by outlining client-finding
code
* Use DMA API path when there is no IOMMU
* Remove unnecessary comment
* Remove new GEM_CREATE/GEM_MMAP
* Move unlock in channel_map to avoid double unlock
on error path.
v5:
* Set iova_end in both mapping paths
v4:
* New patch, split out from combined UAPI + submit patch.
---
drivers/gpu/drm/tegra/Makefile | 1 +
drivers/gpu/drm/tegra/drm.c | 37 ++--
drivers/gpu/drm/tegra/drm.h | 10 ++
drivers/gpu/drm/tegra/uapi.c | 300 +++++++++++++++++++++++++++++++++
drivers/gpu/drm/tegra/uapi.h | 51 ++++++
5 files changed, 383 insertions(+), 16 deletions(-)
create mode 100644 drivers/gpu/drm/tegra/uapi.c
create mode 100644 drivers/gpu/drm/tegra/uapi.h
diff --git a/drivers/gpu/drm/tegra/Makefile b/drivers/gpu/drm/tegra/Makefile
index d6cf202414f0..783475ffd943 100644
--- a/drivers/gpu/drm/tegra/Makefile
+++ b/drivers/gpu/drm/tegra/Makefile
@@ -3,6 +3,7 @@ ccflags-$(CONFIG_DRM_TEGRA_DEBUG) += -DDEBUG
tegra-drm-y := \
drm.o \
+ uapi.o \
gem.o \
fb.o \
dp.o \
diff --git a/drivers/gpu/drm/tegra/drm.c b/drivers/gpu/drm/tegra/drm.c
index a27e5a89d676..a15954c1a047 100644
--- a/drivers/gpu/drm/tegra/drm.c
+++ b/drivers/gpu/drm/tegra/drm.c
@@ -20,6 +20,7 @@
#include <drm/drm_prime.h>
#include <drm/drm_vblank.h>
+#include "uapi.h"
#include "drm.h"
#include "gem.h"
@@ -33,11 +34,6 @@
#define CARVEOUT_SZ SZ_64M
#define CDMA_GATHER_FETCHES_MAX_NB 16383
-struct tegra_drm_file {
- struct idr contexts;
- struct mutex lock;
-};
-
static int tegra_atomic_check(struct drm_device *drm,
struct drm_atomic_state *state)
{
@@ -93,7 +89,8 @@ static int tegra_drm_open(struct drm_device *drm, struct drm_file *filp)
if (!fpriv)
return -ENOMEM;
- idr_init_base(&fpriv->contexts, 1);
+ idr_init_base(&fpriv->legacy_contexts, 1);
+ xa_init_flags(&fpriv->contexts, XA_FLAGS_ALLOC1);
mutex_init(&fpriv->lock);
filp->driver_priv = fpriv;
@@ -418,7 +415,7 @@ static int tegra_client_open(struct tegra_drm_file *fpriv,
if (err < 0)
return err;
- err = idr_alloc(&fpriv->contexts, context, 1, 0, GFP_KERNEL);
+ err = idr_alloc(&fpriv->legacy_contexts, context, 1, 0, GFP_KERNEL);
if (err < 0) {
client->ops->close_channel(context);
return err;
@@ -473,13 +470,13 @@ static int tegra_close_channel(struct drm_device *drm, void *data,
mutex_lock(&fpriv->lock);
- context = idr_find(&fpriv->contexts, args->context);
+ context = idr_find(&fpriv->legacy_contexts, args->context);
if (!context) {
err = -EINVAL;
goto unlock;
}
- idr_remove(&fpriv->contexts, context->id);
+ idr_remove(&fpriv->legacy_contexts, context->id);
tegra_drm_context_free(context);
unlock:
@@ -498,7 +495,7 @@ static int tegra_get_syncpt(struct drm_device *drm, void *data,
mutex_lock(&fpriv->lock);
- context = idr_find(&fpriv->contexts, args->context);
+ context = idr_find(&fpriv->legacy_contexts, args->context);
if (!context) {
err = -ENODEV;
goto unlock;
@@ -527,7 +524,7 @@ static int tegra_submit(struct drm_device *drm, void *data,
mutex_lock(&fpriv->lock);
- context = idr_find(&fpriv->contexts, args->context);
+ context = idr_find(&fpriv->legacy_contexts, args->context);
if (!context) {
err = -ENODEV;
goto unlock;
@@ -552,7 +549,7 @@ static int tegra_get_syncpt_base(struct drm_device *drm, void *data,
mutex_lock(&fpriv->lock);
- context = idr_find(&fpriv->contexts, args->context);
+ context = idr_find(&fpriv->legacy_contexts, args->context);
if (!context) {
err = -ENODEV;
goto unlock;
@@ -721,10 +718,17 @@ static int tegra_gem_get_flags(struct drm_device *drm, void *data,
static const struct drm_ioctl_desc tegra_drm_ioctls[] = {
#ifdef CONFIG_DRM_TEGRA_STAGING
- DRM_IOCTL_DEF_DRV(TEGRA_GEM_CREATE, tegra_gem_create,
+ DRM_IOCTL_DEF_DRV(TEGRA_CHANNEL_OPEN, tegra_drm_ioctl_channel_open,
+ DRM_RENDER_ALLOW),
+ DRM_IOCTL_DEF_DRV(TEGRA_CHANNEL_CLOSE, tegra_drm_ioctl_channel_close,
DRM_RENDER_ALLOW),
- DRM_IOCTL_DEF_DRV(TEGRA_GEM_MMAP, tegra_gem_mmap,
+ DRM_IOCTL_DEF_DRV(TEGRA_CHANNEL_MAP, tegra_drm_ioctl_channel_map,
DRM_RENDER_ALLOW),
+ DRM_IOCTL_DEF_DRV(TEGRA_CHANNEL_UNMAP, tegra_drm_ioctl_channel_unmap,
+ DRM_RENDER_ALLOW),
+
+ DRM_IOCTL_DEF_DRV(TEGRA_GEM_CREATE, tegra_gem_create, DRM_RENDER_ALLOW),
+ DRM_IOCTL_DEF_DRV(TEGRA_GEM_MMAP, tegra_gem_mmap, DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(TEGRA_SYNCPT_READ, tegra_syncpt_read,
DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(TEGRA_SYNCPT_INCR, tegra_syncpt_incr,
@@ -778,10 +782,11 @@ static void tegra_drm_postclose(struct drm_device *drm, struct drm_file *file)
struct tegra_drm_file *fpriv = file->driver_priv;
mutex_lock(&fpriv->lock);
- idr_for_each(&fpriv->contexts, tegra_drm_context_cleanup, NULL);
+ idr_for_each(&fpriv->legacy_contexts, tegra_drm_context_cleanup, NULL);
+ tegra_drm_uapi_close_file(fpriv);
mutex_unlock(&fpriv->lock);
- idr_destroy(&fpriv->contexts);
+ idr_destroy(&fpriv->legacy_contexts);
mutex_destroy(&fpriv->lock);
kfree(fpriv);
}
diff --git a/drivers/gpu/drm/tegra/drm.h b/drivers/gpu/drm/tegra/drm.h
index 1ec07034727c..99918a532809 100644
--- a/drivers/gpu/drm/tegra/drm.h
+++ b/drivers/gpu/drm/tegra/drm.h
@@ -59,12 +59,22 @@ 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 {
struct tegra_drm_client *client;
struct host1x_channel *channel;
+
+ /* Only used by legacy UAPI. */
unsigned int id;
+
+ /* Only used by new UAPI. */
+ struct xarray mappings;
};
struct tegra_drm_client_ops {
diff --git a/drivers/gpu/drm/tegra/uapi.c b/drivers/gpu/drm/tegra/uapi.c
new file mode 100644
index 000000000000..59b698817d5c
--- /dev/null
+++ b/drivers/gpu/drm/tegra/uapi.c
@@ -0,0 +1,300 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Copyright (c) 2020 NVIDIA Corporation */
+
+#include <linux/host1x.h>
+#include <linux/iommu.h>
+#include <linux/list.h>
+
+#include <drm/drm_drv.h>
+#include <drm/drm_file.h>
+
+#include "drm.h"
+#include "uapi.h"
+
+static void tegra_drm_mapping_release(struct kref *ref)
+{
+ struct tegra_drm_mapping *mapping =
+ container_of(ref, struct tegra_drm_mapping, ref);
+
+ if (mapping->sgt)
+ dma_unmap_sgtable(mapping->dev, mapping->sgt,
+ mapping->direction, DMA_ATTR_SKIP_CPU_SYNC);
+
+ host1x_bo_unpin(mapping->dev, mapping->bo, mapping->sgt);
+ host1x_bo_put(mapping->bo);
+
+ kfree(mapping);
+}
+
+void tegra_drm_mapping_put(struct tegra_drm_mapping *mapping)
+{
+ kref_put(&mapping->ref, tegra_drm_mapping_release);
+}
+
+static void tegra_drm_channel_ctx_close(struct tegra_drm_context *ctx)
+{
+ struct tegra_drm_mapping *mapping;
+ unsigned long id;
+
+ xa_for_each(&ctx->mappings, id, mapping)
+ tegra_drm_mapping_put(mapping);
+
+ xa_destroy(&ctx->mappings);
+
+ host1x_channel_put(ctx->channel);
+
+ kfree(ctx);
+}
+
+void tegra_drm_uapi_close_file(struct tegra_drm_file *file)
+{
+ unsigned long ctx_id;
+ struct tegra_drm_context *ctx;
+
+ xa_for_each(&file->contexts, ctx_id, ctx)
+ tegra_drm_channel_ctx_close(ctx);
+
+ xa_destroy(&file->contexts);
+}
+
+static struct tegra_drm_client *tegra_drm_find_client(struct tegra_drm *tegra,
+ u32 class)
+{
+ struct tegra_drm_client *client;
+
+ list_for_each_entry(client, &tegra->clients, list)
+ if (client->base.class == class)
+ return client;
+
+ return NULL;
+}
+
+int tegra_drm_ioctl_channel_open(struct drm_device *drm, void *data,
+ struct drm_file *file)
+{
+ struct tegra_drm_file *fpriv = file->driver_priv;
+ struct tegra_drm *tegra = drm->dev_private;
+ struct drm_tegra_channel_open *args = data;
+ struct tegra_drm_client *client = NULL;
+ struct tegra_drm_context *ctx;
+ int err;
+
+ if (args->flags)
+ return -EINVAL;
+
+ ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
+ if (!ctx)
+ return -ENOMEM;
+
+ client = tegra_drm_find_client(tegra, args->host1x_class);
+ if (!client) {
+ err = -ENODEV;
+ goto free_ctx;
+ }
+
+ if (client->shared_channel) {
+ ctx->channel = host1x_channel_get(client->shared_channel);
+ } else {
+ ctx->channel = host1x_channel_request(&client->base);
+ if (!ctx->channel) {
+ err = -EBUSY;
+ goto free_ctx;
+ }
+ }
+
+ err = xa_alloc(&fpriv->contexts, &args->channel_ctx, ctx,
+ XA_LIMIT(1, U32_MAX), GFP_KERNEL);
+ if (err < 0)
+ goto put_channel;
+
+ ctx->client = client;
+ xa_init_flags(&ctx->mappings, XA_FLAGS_ALLOC1);
+
+ args->hardware_version = client->version;
+
+ args->hardware_flags = 0;
+ if (device_get_dma_attr(client->base.dev) == DEV_DMA_COHERENT)
+ args->hardware_flags |= DRM_TEGRA_CHANNEL_OPEN_HW_CACHE_COHERENT;
+
+ return 0;
+
+put_channel:
+ host1x_channel_put(ctx->channel);
+free_ctx:
+ kfree(ctx);
+
+ return err;
+}
+
+int tegra_drm_ioctl_channel_close(struct drm_device *drm, void *data,
+ struct drm_file *file)
+{
+ struct tegra_drm_file *fpriv = file->driver_priv;
+ struct drm_tegra_channel_close *args = data;
+ struct tegra_drm_context *ctx;
+
+ mutex_lock(&fpriv->lock);
+ ctx = xa_load(&fpriv->contexts, args->channel_ctx);
+ if (!ctx) {
+ mutex_unlock(&fpriv->lock);
+ return -EINVAL;
+ }
+
+ xa_erase(&fpriv->contexts, args->channel_ctx);
+
+ mutex_unlock(&fpriv->lock);
+
+ tegra_drm_channel_ctx_close(ctx);
+
+ return 0;
+}
+
+int tegra_drm_ioctl_channel_map(struct drm_device *drm, void *data,
+ struct drm_file *file)
+{
+ struct tegra_drm_file *fpriv = file->driver_priv;
+ struct drm_tegra_channel_map *args = data;
+ struct tegra_drm_mapping *mapping;
+ struct tegra_drm_context *ctx;
+ int err = 0;
+
+ if (args->flags & ~DRM_TEGRA_CHANNEL_MAP_READWRITE)
+ return -EINVAL;
+
+ mutex_lock(&fpriv->lock);
+ ctx = xa_load(&fpriv->contexts, args->channel_ctx);
+ if (!ctx) {
+ mutex_unlock(&fpriv->lock);
+ return -EINVAL;
+ }
+
+ mapping = kzalloc(sizeof(*mapping), GFP_KERNEL);
+ if (!mapping) {
+ err = -ENOMEM;
+ goto unlock;
+ }
+
+ kref_init(&mapping->ref);
+
+ mapping->dev = ctx->client->base.dev;
+ mapping->bo = tegra_gem_lookup(file, args->handle);
+ if (!mapping->bo) {
+ err = -EINVAL;
+ goto unlock;
+ }
+
+ if (ctx->client->base.group) {
+ /* IOMMU domain managed directly using IOMMU API */
+ host1x_bo_pin(mapping->dev, mapping->bo,
+ &mapping->iova);
+ } else {
+ /* No IOMMU, or IOMMU managed through DMA API */
+ mapping->direction = DMA_TO_DEVICE;
+ if (args->flags & DRM_TEGRA_CHANNEL_MAP_READWRITE)
+ mapping->direction = DMA_BIDIRECTIONAL;
+
+ mapping->sgt =
+ host1x_bo_pin(mapping->dev, mapping->bo, NULL);
+ if (IS_ERR(mapping->sgt)) {
+ err = PTR_ERR(mapping->sgt);
+ goto put_gem;
+ }
+
+ err = dma_map_sgtable(mapping->dev, mapping->sgt,
+ mapping->direction,
+ DMA_ATTR_SKIP_CPU_SYNC);
+ if (err)
+ goto unpin;
+
+ mapping->iova = sg_dma_address(mapping->sgt->sgl);
+ }
+
+ mapping->iova_end = mapping->iova + host1x_to_tegra_bo(mapping->bo)->size;
+
+ err = xa_alloc(&ctx->mappings, &args->mapping_id, mapping,
+ XA_LIMIT(1, U32_MAX), GFP_KERNEL);
+ if (err < 0)
+ goto unmap;
+
+ mutex_unlock(&fpriv->lock);
+
+ return 0;
+
+unmap:
+ if (mapping->sgt) {
+ dma_unmap_sgtable(mapping->dev, mapping->sgt,
+ mapping->direction, DMA_ATTR_SKIP_CPU_SYNC);
+ }
+unpin:
+ host1x_bo_unpin(mapping->dev, mapping->bo, mapping->sgt);
+put_gem:
+ host1x_bo_put(mapping->bo);
+ kfree(mapping);
+unlock:
+ mutex_unlock(&fpriv->lock);
+ return err;
+}
+
+int tegra_drm_ioctl_channel_unmap(struct drm_device *drm, void *data,
+ struct drm_file *file)
+{
+ struct tegra_drm_file *fpriv = file->driver_priv;
+ struct drm_tegra_channel_unmap *args = data;
+ struct tegra_drm_mapping *mapping;
+ struct tegra_drm_context *ctx;
+
+ mutex_lock(&fpriv->lock);
+ ctx = xa_load(&fpriv->contexts, args->channel_ctx);
+ if (!ctx) {
+ mutex_unlock(&fpriv->lock);
+ return -EINVAL;
+ }
+
+ mapping = xa_erase(&ctx->mappings, args->mapping_id);
+
+ mutex_unlock(&fpriv->lock);
+
+ if (mapping) {
+ tegra_drm_mapping_put(mapping);
+ return 0;
+ } else {
+ return -EINVAL;
+ }
+}
+
+int tegra_drm_ioctl_gem_create(struct drm_device *drm, void *data,
+ struct drm_file *file)
+{
+ struct drm_tegra_gem_create *args = data;
+ struct tegra_bo *bo;
+
+ if (args->flags)
+ return -EINVAL;
+
+ bo = tegra_bo_create_with_handle(file, drm, args->size, args->flags,
+ &args->handle);
+ if (IS_ERR(bo))
+ return PTR_ERR(bo);
+
+ return 0;
+}
+
+int tegra_drm_ioctl_gem_mmap(struct drm_device *drm, void *data,
+ struct drm_file *file)
+{
+ struct drm_tegra_gem_mmap *args = data;
+ struct drm_gem_object *gem;
+ struct tegra_bo *bo;
+
+ gem = drm_gem_object_lookup(file, args->handle);
+ if (!gem)
+ return -EINVAL;
+
+ bo = to_tegra_bo(gem);
+
+ args->offset = drm_vma_node_offset_addr(&bo->gem.vma_node);
+
+ drm_gem_object_put(gem);
+
+ return 0;
+}
diff --git a/drivers/gpu/drm/tegra/uapi.h b/drivers/gpu/drm/tegra/uapi.h
new file mode 100644
index 000000000000..fbef39726c29
--- /dev/null
+++ b/drivers/gpu/drm/tegra/uapi.h
@@ -0,0 +1,51 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/* Copyright (c) 2020 NVIDIA Corporation */
+
+#ifndef _TEGRA_DRM_UAPI_H
+#define _TEGRA_DRM_UAPI_H
+
+#include <linux/dma-mapping.h>
+#include <linux/idr.h>
+#include <linux/kref.h>
+#include <linux/xarray.h>
+
+#include <drm/drm.h>
+
+struct drm_file;
+struct drm_device;
+
+struct tegra_drm_file {
+ /* Legacy UAPI state */
+ struct idr legacy_contexts;
+ struct mutex lock;
+
+ /* New UAPI state */
+ struct xarray contexts;
+};
+
+struct tegra_drm_mapping {
+ struct kref ref;
+
+ struct device *dev;
+ struct host1x_bo *bo;
+ struct sg_table *sgt;
+ enum dma_data_direction direction;
+ dma_addr_t iova;
+ dma_addr_t iova_end;
+};
+
+int tegra_drm_ioctl_channel_open(struct drm_device *drm, void *data,
+ struct drm_file *file);
+int tegra_drm_ioctl_channel_close(struct drm_device *drm, void *data,
+ struct drm_file *file);
+int tegra_drm_ioctl_channel_map(struct drm_device *drm, void *data,
+ struct drm_file *file);
+int tegra_drm_ioctl_channel_unmap(struct drm_device *drm, void *data,
+ struct drm_file *file);
+int tegra_drm_ioctl_channel_submit(struct drm_device *drm, void *data,
+ struct drm_file *file);
+
+void tegra_drm_uapi_close_file(struct tegra_drm_file *file);
+void tegra_drm_mapping_put(struct tegra_drm_mapping *mapping);
+
+#endif
--
2.30.1
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v6 10/14] drm/tegra: Implement syncpoint management UAPI
2021-05-30 12:21 [PATCH v6 00/14] TegraDRM UAPI Mikko Perttunen
` (8 preceding siblings ...)
2021-05-30 12:21 ` [PATCH v6 09/14] drm/tegra: Implement new UAPI Mikko Perttunen
@ 2021-05-30 12:21 ` Mikko Perttunen
2021-05-30 12:21 ` [PATCH v6 11/14] drm/tegra: Implement syncpoint wait UAPI Mikko Perttunen
` (3 subsequent siblings)
13 siblings, 0 replies; 16+ messages in thread
From: Mikko Perttunen @ 2021-05-30 12:21 UTC (permalink / raw)
To: thierry.reding, jonathanh, digetx, airlied, daniel
Cc: linux-tegra, dri-devel, Mikko Perttunen
Implement TegraDRM IOCTLs for allocating and freeing syncpoints.
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
---
v6:
- New patch
---
drivers/gpu/drm/tegra/drm.c | 5 ++++
drivers/gpu/drm/tegra/uapi.c | 56 ++++++++++++++++++++++++++++++++++--
drivers/gpu/drm/tegra/uapi.h | 5 ++++
3 files changed, 64 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/tegra/drm.c b/drivers/gpu/drm/tegra/drm.c
index a15954c1a047..7530c988f795 100644
--- a/drivers/gpu/drm/tegra/drm.c
+++ b/drivers/gpu/drm/tegra/drm.c
@@ -91,6 +91,7 @@ static int tegra_drm_open(struct drm_device *drm, struct drm_file *filp)
idr_init_base(&fpriv->legacy_contexts, 1);
xa_init_flags(&fpriv->contexts, XA_FLAGS_ALLOC1);
+ xa_init(&fpriv->syncpoints);
mutex_init(&fpriv->lock);
filp->driver_priv = fpriv;
@@ -726,6 +727,10 @@ 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_SYNCPOINT_ALLOCATE, tegra_drm_ioctl_syncpoint_allocate,
+ DRM_RENDER_ALLOW),
+ DRM_IOCTL_DEF_DRV(TEGRA_SYNCPOINT_FREE, tegra_drm_ioctl_syncpoint_free,
+ DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(TEGRA_GEM_CREATE, tegra_gem_create, DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(TEGRA_GEM_MMAP, tegra_gem_mmap, DRM_RENDER_ALLOW),
diff --git a/drivers/gpu/drm/tegra/uapi.c b/drivers/gpu/drm/tegra/uapi.c
index 59b698817d5c..2788c5b4ae4f 100644
--- a/drivers/gpu/drm/tegra/uapi.c
+++ b/drivers/gpu/drm/tegra/uapi.c
@@ -48,13 +48,18 @@ static void tegra_drm_channel_ctx_close(struct tegra_drm_context *ctx)
void tegra_drm_uapi_close_file(struct tegra_drm_file *file)
{
- unsigned long ctx_id;
struct tegra_drm_context *ctx;
+ struct host1x_syncpt *sp;
+ unsigned long id;
- xa_for_each(&file->contexts, ctx_id, ctx)
+ xa_for_each(&file->contexts, id, ctx)
tegra_drm_channel_ctx_close(ctx);
+ xa_for_each(&file->syncpoints, id, sp)
+ host1x_syncpt_put(sp);
+
xa_destroy(&file->contexts);
+ xa_destroy(&file->syncpoints);
}
static struct tegra_drm_client *tegra_drm_find_client(struct tegra_drm *tegra,
@@ -298,3 +303,50 @@ int tegra_drm_ioctl_gem_mmap(struct drm_device *drm, void *data,
return 0;
}
+
+int tegra_drm_ioctl_syncpoint_allocate(struct drm_device *drm, void *data,
+ struct drm_file *file)
+{
+ struct host1x *host1x = tegra_drm_to_host1x(drm->dev_private);
+ struct tegra_drm_file *fpriv = file->driver_priv;
+ struct drm_tegra_syncpoint_allocate *args = data;
+ struct host1x_syncpt *sp;
+ int err;
+
+ if (args->id)
+ return -EINVAL;
+
+ sp = host1x_syncpt_alloc(host1x, HOST1X_SYNCPT_CLIENT_MANAGED,
+ current->comm);
+ if (!sp)
+ return -EBUSY;
+
+ args->id = host1x_syncpt_id(sp);
+
+ err = xa_insert(&fpriv->syncpoints, args->id, sp, GFP_KERNEL);
+ if (err) {
+ host1x_syncpt_put(sp);
+ return err;
+ }
+
+ return 0;
+}
+
+int tegra_drm_ioctl_syncpoint_free(struct drm_device *drm, void *data,
+ struct drm_file *file)
+{
+ struct tegra_drm_file *fpriv = file->driver_priv;
+ struct drm_tegra_syncpoint_allocate *args = data;
+ struct host1x_syncpt *sp;
+
+ mutex_lock(&fpriv->lock);
+ sp = xa_erase(&fpriv->syncpoints, args->id);
+ mutex_unlock(&fpriv->lock);
+
+ if (!sp)
+ return -EINVAL;
+
+ host1x_syncpt_put(sp);
+
+ return 0;
+}
diff --git a/drivers/gpu/drm/tegra/uapi.h b/drivers/gpu/drm/tegra/uapi.h
index fbef39726c29..e4e498facf61 100644
--- a/drivers/gpu/drm/tegra/uapi.h
+++ b/drivers/gpu/drm/tegra/uapi.h
@@ -21,6 +21,7 @@ struct tegra_drm_file {
/* New UAPI state */
struct xarray contexts;
+ struct xarray syncpoints;
};
struct tegra_drm_mapping {
@@ -44,6 +45,10 @@ int tegra_drm_ioctl_channel_unmap(struct drm_device *drm, void *data,
struct drm_file *file);
int tegra_drm_ioctl_channel_submit(struct drm_device *drm, void *data,
struct drm_file *file);
+int tegra_drm_ioctl_syncpoint_allocate(struct drm_device *drm, void *data,
+ struct drm_file *file);
+int tegra_drm_ioctl_syncpoint_free(struct drm_device *drm, void *data,
+ struct drm_file *file);
void tegra_drm_uapi_close_file(struct tegra_drm_file *file);
void tegra_drm_mapping_put(struct tegra_drm_mapping *mapping);
--
2.30.1
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v6 11/14] drm/tegra: Implement syncpoint wait UAPI
2021-05-30 12:21 [PATCH v6 00/14] TegraDRM UAPI Mikko Perttunen
` (9 preceding siblings ...)
2021-05-30 12:21 ` [PATCH v6 10/14] drm/tegra: Implement syncpoint management UAPI Mikko Perttunen
@ 2021-05-30 12:21 ` Mikko Perttunen
2021-05-30 12:21 ` [PATCH v6 12/14] drm/tegra: Implement job submission part of new UAPI Mikko Perttunen
` (2 subsequent siblings)
13 siblings, 0 replies; 16+ messages in thread
From: Mikko Perttunen @ 2021-05-30 12:21 UTC (permalink / raw)
To: thierry.reding, jonathanh, digetx, airlied, daniel
Cc: linux-tegra, dri-devel, Mikko Perttunen
Implement new syncpoint wait UAPI. This is different from the
legacy one in taking an absolute timestamp in line with modern
DRM conventions.
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
---
v6:
- New patch
---
drivers/gpu/drm/tegra/drm.c | 2 ++
drivers/gpu/drm/tegra/uapi.c | 22 ++++++++++++++++++++++
drivers/gpu/drm/tegra/uapi.h | 2 ++
3 files changed, 26 insertions(+)
diff --git a/drivers/gpu/drm/tegra/drm.c b/drivers/gpu/drm/tegra/drm.c
index 7530c988f795..b76bfdd078df 100644
--- a/drivers/gpu/drm/tegra/drm.c
+++ b/drivers/gpu/drm/tegra/drm.c
@@ -731,6 +731,8 @@ static const struct drm_ioctl_desc tegra_drm_ioctls[] = {
DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(TEGRA_SYNCPOINT_FREE, tegra_drm_ioctl_syncpoint_free,
DRM_RENDER_ALLOW),
+ DRM_IOCTL_DEF_DRV(TEGRA_SYNCPOINT_WAIT, tegra_drm_ioctl_syncpoint_wait,
+ DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(TEGRA_GEM_CREATE, tegra_gem_create, DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(TEGRA_GEM_MMAP, tegra_gem_mmap, DRM_RENDER_ALLOW),
diff --git a/drivers/gpu/drm/tegra/uapi.c b/drivers/gpu/drm/tegra/uapi.c
index 2788c5b4ae4f..b411d050b7d8 100644
--- a/drivers/gpu/drm/tegra/uapi.c
+++ b/drivers/gpu/drm/tegra/uapi.c
@@ -7,6 +7,7 @@
#include <drm/drm_drv.h>
#include <drm/drm_file.h>
+#include <drm/drm_utils.h>
#include "drm.h"
#include "uapi.h"
@@ -350,3 +351,24 @@ int tegra_drm_ioctl_syncpoint_free(struct drm_device *drm, void *data,
return 0;
}
+
+int tegra_drm_ioctl_syncpoint_wait(struct drm_device *drm, void *data,
+ struct drm_file *file)
+{
+ struct host1x *host1x = tegra_drm_to_host1x(drm->dev_private);
+ struct drm_tegra_syncpoint_wait *args = data;
+ signed long timeout_jiffies;
+ struct host1x_syncpt *sp;
+
+ if (args->padding[0] != 0)
+ return -EINVAL;
+
+ sp = host1x_syncpt_get_by_id_noref(host1x, args->id);
+ if (!sp)
+ return -EINVAL;
+
+ timeout_jiffies = drm_timeout_abs_to_jiffies(args->timeout_ns);
+
+ return host1x_syncpt_wait(sp, args->threshold, timeout_jiffies,
+ &args->value);
+}
diff --git a/drivers/gpu/drm/tegra/uapi.h b/drivers/gpu/drm/tegra/uapi.h
index e4e498facf61..12adad770ad3 100644
--- a/drivers/gpu/drm/tegra/uapi.h
+++ b/drivers/gpu/drm/tegra/uapi.h
@@ -49,6 +49,8 @@ int tegra_drm_ioctl_syncpoint_allocate(struct drm_device *drm, void *data,
struct drm_file *file);
int tegra_drm_ioctl_syncpoint_free(struct drm_device *drm, void *data,
struct drm_file *file);
+int tegra_drm_ioctl_syncpoint_wait(struct drm_device *drm, void *data,
+ struct drm_file *file);
void tegra_drm_uapi_close_file(struct tegra_drm_file *file);
void tegra_drm_mapping_put(struct tegra_drm_mapping *mapping);
--
2.30.1
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v6 12/14] drm/tegra: Implement job submission part of new UAPI
2021-05-30 12:21 [PATCH v6 00/14] TegraDRM UAPI Mikko Perttunen
` (10 preceding siblings ...)
2021-05-30 12:21 ` [PATCH v6 11/14] drm/tegra: Implement syncpoint wait UAPI Mikko Perttunen
@ 2021-05-30 12:21 ` Mikko Perttunen
2021-05-30 12:21 ` [PATCH v6 13/14] drm/tegra: Add job firewall Mikko Perttunen
2021-05-30 12:21 ` [PATCH v6 14/14] drm/tegra: Bump driver version Mikko Perttunen
13 siblings, 0 replies; 16+ messages in thread
From: Mikko Perttunen @ 2021-05-30 12:21 UTC (permalink / raw)
To: thierry.reding, jonathanh, digetx, airlied, daniel
Cc: linux-tegra, dri-devel, Mikko Perttunen
Implement the job submission IOCTL with a minimum feature set.
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
---
v6:
* Remove sgt bypass path in gather_bo - this would cause
cache maintenance to be skipped and is unnecessary in
general.
* Changes related to moving to using syncpoint IDs
* Add syncobj related code
* Print warning on submit failure describing the issue
* Use ARCH_DMA_ADDR_T_64BIT to check if that is indeed
the case
* Add support for relative syncpoint wait
* Use pm_runtime_resume_and_get
* Only try to resume engines that support runtime PM
* Removed uapi subdirectory
* Don't use "copy_err" variables for copy_from_user
return value
* Fix setting of blocklinear flag
v5:
* Add 16K size limit to copies from userspace.
* Guard RELOC_BLOCKLINEAR flag handling to only exist in ARM64
to prevent oversized shift on 32-bit platforms.
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 | 4 +-
drivers/gpu/drm/tegra/gather_bo.c | 81 +++++
drivers/gpu/drm/tegra/gather_bo.h | 22 ++
drivers/gpu/drm/tegra/submit.c | 541 ++++++++++++++++++++++++++++++
drivers/gpu/drm/tegra/submit.h | 17 +
6 files changed, 666 insertions(+), 1 deletion(-)
create mode 100644 drivers/gpu/drm/tegra/gather_bo.c
create mode 100644 drivers/gpu/drm/tegra/gather_bo.h
create mode 100644 drivers/gpu/drm/tegra/submit.c
create mode 100644 drivers/gpu/drm/tegra/submit.h
diff --git a/drivers/gpu/drm/tegra/Makefile b/drivers/gpu/drm/tegra/Makefile
index 783475ffd943..ab4289d1c991 100644
--- a/drivers/gpu/drm/tegra/Makefile
+++ b/drivers/gpu/drm/tegra/Makefile
@@ -4,6 +4,8 @@ ccflags-$(CONFIG_DRM_TEGRA_DEBUG) += -DDEBUG
tegra-drm-y := \
drm.o \
uapi.o \
+ submit.o \
+ gather_bo.o \
gem.o \
fb.o \
dp.o \
diff --git a/drivers/gpu/drm/tegra/drm.c b/drivers/gpu/drm/tegra/drm.c
index b76bfdd078df..59f04431cae0 100644
--- a/drivers/gpu/drm/tegra/drm.c
+++ b/drivers/gpu/drm/tegra/drm.c
@@ -727,6 +727,8 @@ 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_SYNCPOINT_ALLOCATE, tegra_drm_ioctl_syncpoint_allocate,
DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(TEGRA_SYNCPOINT_FREE, tegra_drm_ioctl_syncpoint_free,
@@ -851,7 +853,7 @@ static void tegra_debugfs_init(struct drm_minor *minor)
static const struct drm_driver tegra_drm_driver = {
.driver_features = DRIVER_MODESET | DRIVER_GEM |
- DRIVER_ATOMIC | DRIVER_RENDER,
+ DRIVER_ATOMIC | DRIVER_RENDER | DRIVER_SYNCOBJ,
.open = tegra_drm_open,
.postclose = tegra_drm_postclose,
.lastclose = drm_fb_helper_lastclose,
diff --git a/drivers/gpu/drm/tegra/gather_bo.c b/drivers/gpu/drm/tegra/gather_bo.c
new file mode 100644
index 000000000000..755c2da62361
--- /dev/null
+++ b/drivers/gpu/drm/tegra/gather_bo.c
@@ -0,0 +1,81 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Copyright (c) 2020 NVIDIA Corporation */
+
+#include <linux/scatterlist.h>
+#include <linux/slab.h>
+
+#include "gather_bo.h"
+
+static struct host1x_bo *gather_bo_get(struct host1x_bo *host_bo)
+{
+ struct gather_bo *bo = container_of(host_bo, struct gather_bo, base);
+
+ kref_get(&bo->ref);
+
+ return host_bo;
+}
+
+static void gather_bo_release(struct kref *ref)
+{
+ struct gather_bo *bo = container_of(ref, struct gather_bo, ref);
+
+ kfree(bo->gather_data);
+ kfree(bo);
+}
+
+void gather_bo_put(struct host1x_bo *host_bo)
+{
+ struct gather_bo *bo = container_of(host_bo, struct gather_bo, base);
+
+ kref_put(&bo->ref, gather_bo_release);
+}
+
+static struct sg_table *
+gather_bo_pin(struct device *dev, struct host1x_bo *host_bo, dma_addr_t *phys)
+{
+ struct gather_bo *bo = container_of(host_bo, struct gather_bo, base);
+ struct sg_table *sgt;
+ int err;
+
+ sgt = kzalloc(sizeof(*sgt), GFP_KERNEL);
+ if (!sgt)
+ return ERR_PTR(-ENOMEM);
+
+ err = sg_alloc_table(sgt, 1, GFP_KERNEL);
+ if (err) {
+ kfree(sgt);
+ return ERR_PTR(err);
+ }
+
+ sg_init_one(sgt->sgl, bo->gather_data, bo->gather_data_words*4);
+
+ return sgt;
+}
+
+static void gather_bo_unpin(struct device *dev, struct sg_table *sgt)
+{
+ if (sgt) {
+ sg_free_table(sgt);
+ kfree(sgt);
+ }
+}
+
+static void *gather_bo_mmap(struct host1x_bo *host_bo)
+{
+ struct gather_bo *bo = container_of(host_bo, struct gather_bo, base);
+
+ return bo->gather_data;
+}
+
+static void gather_bo_munmap(struct host1x_bo *host_bo, void *addr)
+{
+}
+
+const struct host1x_bo_ops gather_bo_ops = {
+ .get = gather_bo_get,
+ .put = gather_bo_put,
+ .pin = gather_bo_pin,
+ .unpin = gather_bo_unpin,
+ .mmap = gather_bo_mmap,
+ .munmap = gather_bo_munmap,
+};
diff --git a/drivers/gpu/drm/tegra/gather_bo.h b/drivers/gpu/drm/tegra/gather_bo.h
new file mode 100644
index 000000000000..6b4c9d83ac91
--- /dev/null
+++ b/drivers/gpu/drm/tegra/gather_bo.h
@@ -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
diff --git a/drivers/gpu/drm/tegra/submit.c b/drivers/gpu/drm/tegra/submit.c
new file mode 100644
index 000000000000..ae1bf916b972
--- /dev/null
+++ b/drivers/gpu/drm/tegra/submit.c
@@ -0,0 +1,541 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Copyright (c) 2020 NVIDIA Corporation */
+
+#include <linux/dma-fence-array.h>
+#include <linux/file.h>
+#include <linux/host1x.h>
+#include <linux/iommu.h>
+#include <linux/kref.h>
+#include <linux/list.h>
+#include <linux/nospec.h>
+#include <linux/pm_runtime.h>
+#include <linux/sync_file.h>
+
+#include <drm/drm_drv.h>
+#include <drm/drm_file.h>
+#include <drm/drm_syncobj.h>
+
+#include "drm.h"
+#include "gather_bo.h"
+#include "gem.h"
+#include "submit.h"
+#include "uapi.h"
+
+#define SUBMIT_ERR(ctx, fmt, ...) \
+ dev_err_ratelimited(ctx->client->base.dev, \
+ "%s: job submission failed: " fmt "\n", \
+ current->comm __VA_OPT__(,) __VA_ARGS__)
+
+static struct tegra_drm_mapping *
+tegra_drm_mapping_get(struct tegra_drm_context *ctx, u32 id)
+{
+ struct tegra_drm_mapping *mapping;
+
+ xa_lock(&ctx->mappings);
+ mapping = xa_load(&ctx->mappings, id);
+ if (mapping)
+ kref_get(&mapping->ref);
+ xa_unlock(&ctx->mappings);
+
+ return mapping;
+}
+
+static void *alloc_copy_user_array(void __user *from, size_t count, size_t size)
+{
+ size_t copy_len;
+ void *data;
+
+ if (check_mul_overflow(count, size, ©_len))
+ return ERR_PTR(-EINVAL);
+
+ if (copy_len > 0x4000)
+ return ERR_PTR(-E2BIG);
+
+ data = kvmalloc(copy_len, GFP_KERNEL);
+ if (!data)
+ return ERR_PTR(-ENOMEM);
+
+ if (copy_from_user(data, from, copy_len)) {
+ kvfree(data);
+ return ERR_PTR(-EFAULT);
+ }
+
+ return data;
+}
+
+static int submit_copy_gather_data(struct gather_bo **pbo,
+ struct tegra_drm_context *ctx,
+ struct drm_tegra_channel_submit *args)
+{
+ struct gather_bo *bo;
+ size_t copy_len;
+
+ if (args->gather_data_words == 0) {
+ SUBMIT_ERR(ctx, "gather_data_words cannot be zero");
+ return -EINVAL;
+ }
+
+ if (check_mul_overflow((size_t)args->gather_data_words, (size_t)4, ©_len)) {
+ SUBMIT_ERR(ctx, "gather_data_words is too large");
+ return -EINVAL;
+ }
+
+ bo = kzalloc(sizeof(*bo), GFP_KERNEL);
+ if (!bo) {
+ SUBMIT_ERR(ctx, "failed to allocate memory for bo info");
+ return -ENOMEM;
+ }
+
+ kref_init(&bo->ref);
+ host1x_bo_init(&bo->base, &gather_bo_ops);
+
+ bo->gather_data = kmalloc(copy_len, GFP_KERNEL | __GFP_NOWARN);
+ if (!bo->gather_data) {
+ SUBMIT_ERR(ctx, "failed to allocate memory for gather data");
+ kfree(bo);
+ return -ENOMEM;
+ }
+
+ if (copy_from_user(bo->gather_data,
+ u64_to_user_ptr(args->gather_data_ptr), copy_len)) {
+ SUBMIT_ERR(ctx, "failed to copy gather data from userspace");
+ kfree(bo->gather_data);
+ kfree(bo);
+ return -EFAULT;
+ }
+
+ bo->gather_data_words = args->gather_data_words;
+
+ *pbo = bo;
+
+ return 0;
+}
+
+static int submit_write_reloc(struct tegra_drm_context *ctx,
+ struct gather_bo *bo,
+ struct drm_tegra_submit_buf *buf,
+ struct tegra_drm_mapping *mapping)
+{
+ /* TODO check that target_offset is within bounds */
+ dma_addr_t iova = mapping->iova + buf->reloc.target_offset;
+ u32 written_ptr;
+
+#ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
+ if (buf->flags & DRM_TEGRA_SUBMIT_BUF_RELOC_BLOCKLINEAR)
+ iova |= BIT_ULL(39);
+#endif
+
+ written_ptr = iova >> buf->reloc.shift;
+
+ if (buf->reloc.gather_offset_words >= bo->gather_data_words) {
+ SUBMIT_ERR(ctx,
+ "relocation has too large gather offset (%u vs gather length %zu)",
+ buf->reloc.gather_offset_words, bo->gather_data_words);
+ return -EINVAL;
+ }
+
+ buf->reloc.gather_offset_words = array_index_nospec(
+ buf->reloc.gather_offset_words, bo->gather_data_words);
+
+ bo->gather_data[buf->reloc.gather_offset_words] = written_ptr;
+
+ return 0;
+}
+
+static int submit_process_bufs(struct tegra_drm_context *ctx,
+ struct gather_bo *bo,
+ struct drm_tegra_channel_submit *args,
+ struct tegra_drm_submit_data *job_data)
+{
+ struct tegra_drm_used_mapping *mappings;
+ struct drm_tegra_submit_buf *bufs;
+ int err;
+ u32 i;
+
+ bufs = alloc_copy_user_array(u64_to_user_ptr(args->bufs_ptr),
+ args->num_bufs, sizeof(*bufs));
+ if (IS_ERR(bufs)) {
+ SUBMIT_ERR(ctx, "failed to copy bufs array from userspace");
+ return PTR_ERR(bufs);
+ }
+
+ mappings = kcalloc(args->num_bufs, sizeof(*mappings), GFP_KERNEL);
+ if (!mappings) {
+ SUBMIT_ERR(ctx, "failed to allocate memory for mapping info");
+ err = -ENOMEM;
+ goto done;
+ }
+
+ for (i = 0; i < args->num_bufs; i++) {
+ struct drm_tegra_submit_buf *buf = &bufs[i];
+ struct tegra_drm_mapping *mapping;
+
+ if (buf->flags & ~DRM_TEGRA_SUBMIT_BUF_RELOC_BLOCKLINEAR) {
+ SUBMIT_ERR(ctx, "invalid flag specified for buf");
+ err = -EINVAL;
+ goto drop_refs;
+ }
+
+ mapping = tegra_drm_mapping_get(ctx, buf->mapping_id);
+ if (!mapping) {
+ SUBMIT_ERR(ctx, "invalid mapping_id for buf '%u'",
+ buf->mapping_id);
+ err = -EINVAL;
+ goto drop_refs;
+ }
+
+ err = submit_write_reloc(ctx, bo, buf, mapping);
+ if (err) {
+ tegra_drm_mapping_put(mapping);
+ goto drop_refs;
+ }
+
+ mappings[i].mapping = mapping;
+ mappings[i].flags = buf->flags;
+ }
+
+ job_data->used_mappings = mappings;
+ job_data->num_used_mappings = i;
+
+ err = 0;
+
+ goto done;
+
+drop_refs:
+ for (;;) {
+ if (i-- == 0)
+ break;
+
+ tegra_drm_mapping_put(mappings[i].mapping);
+ }
+
+ kfree(mappings);
+ job_data->used_mappings = NULL;
+
+done:
+ kvfree(bufs);
+
+ return err;
+}
+
+static int submit_get_syncpt(struct tegra_drm_context *ctx,
+ struct host1x_job *job, struct xarray *syncpoints,
+ struct drm_tegra_channel_submit *args)
+{
+ struct host1x_syncpt *sp;
+
+ if (args->syncpt_incr.flags) {
+ SUBMIT_ERR(ctx, "invalid flag specified for syncpt_incr");
+ return -EINVAL;
+ }
+
+ /* Syncpt ref will be dropped on job release */
+ sp = xa_load(syncpoints, args->syncpt_incr.id);
+ if (!sp) {
+ SUBMIT_ERR(ctx, "syncpoint specified in syncpt_incr was not allocated");
+ return -EINVAL;
+ }
+
+ job->syncpt = host1x_syncpt_get(sp);
+ job->syncpt_incrs = args->syncpt_incr.num_incrs;
+
+ return 0;
+}
+
+static int submit_job_add_gather(struct host1x_job *job,
+ struct tegra_drm_context *ctx,
+ struct drm_tegra_submit_cmd_gather_uptr *cmd,
+ struct gather_bo *bo, u32 *offset,
+ struct tegra_drm_submit_data *job_data)
+{
+ u32 next_offset;
+
+ if (cmd->reserved[0] || cmd->reserved[1] || cmd->reserved[2]) {
+ SUBMIT_ERR(ctx, "non-zero reserved field in GATHER_UPTR command");
+ return -EINVAL;
+ }
+
+ /* Check for maximum gather size */
+ if (cmd->words > 16383) {
+ SUBMIT_ERR(ctx, "too many words in GATHER_UPTR command");
+ return -EINVAL;
+ }
+
+ if (check_add_overflow(*offset, cmd->words, &next_offset)) {
+ SUBMIT_ERR(ctx, "too many total words in job");
+ return -EINVAL;
+ }
+
+ if (next_offset > bo->gather_data_words) {
+ SUBMIT_ERR(ctx, "GATHER_UPTR command overflows gather data");
+ return -EINVAL;
+ }
+
+ host1x_job_add_gather(job, &bo->base, cmd->words, *offset * 4);
+
+ *offset = next_offset;
+
+ return 0;
+}
+
+static int submit_create_job(struct host1x_job **pjob,
+ struct tegra_drm_context *ctx,
+ struct gather_bo *bo,
+ struct drm_tegra_channel_submit *args,
+ struct tegra_drm_submit_data *job_data,
+ struct xarray *syncpoints)
+{
+ struct drm_tegra_submit_cmd *cmds;
+ u32 i, gather_offset = 0, class;
+ struct host1x_job *job;
+ int err;
+
+ /* Set initial class for firewall. */
+ class = ctx->client->base.class;
+
+ cmds = alloc_copy_user_array(u64_to_user_ptr(args->cmds_ptr),
+ args->num_cmds, sizeof(*cmds));
+ if (IS_ERR(cmds)) {
+ SUBMIT_ERR(ctx, "failed to copy cmds array from userspace");
+ return PTR_ERR(cmds);
+ }
+
+ job = host1x_job_alloc(ctx->channel, args->num_cmds, 0);
+ if (!job) {
+ SUBMIT_ERR(ctx, "failed to allocate memory for job");
+ err = -ENOMEM;
+ goto done;
+ }
+
+ err = submit_get_syncpt(ctx, job, syncpoints, args);
+ if (err < 0)
+ goto free_job;
+
+ job->client = &ctx->client->base;
+ job->class = ctx->client->base.class;
+ job->serialize = true;
+
+ for (i = 0; i < args->num_cmds; i++) {
+ struct drm_tegra_submit_cmd *cmd = &cmds[i];
+
+ if (cmd->flags) {
+ SUBMIT_ERR(ctx, "unknown flags given for cmd");
+ err = -EINVAL;
+ goto free_job;
+ }
+
+ if (cmd->type == DRM_TEGRA_SUBMIT_CMD_GATHER_UPTR) {
+ err = submit_job_add_gather(job, ctx,
+ &cmd->gather_uptr, bo,
+ &gather_offset, job_data);
+ if (err)
+ goto free_job;
+ } else if (cmd->type == DRM_TEGRA_SUBMIT_CMD_WAIT_SYNCPT) {
+ if (cmd->wait_syncpt.reserved[0] ||
+ cmd->wait_syncpt.reserved[1]) {
+ SUBMIT_ERR(ctx, "non-zero reserved value");
+ err = -EINVAL;
+ goto free_job;
+ }
+
+ host1x_job_add_wait(job, cmd->wait_syncpt.id,
+ cmd->wait_syncpt.threshold,
+ false,
+ class);
+ } else if (cmd->type == DRM_TEGRA_SUBMIT_CMD_WAIT_SYNCPT_RELATIVE) {
+ if (cmd->wait_syncpt.reserved[0] ||
+ cmd->wait_syncpt.reserved[1]) {
+ SUBMIT_ERR(ctx, "non-zero reserved value");
+ err = -EINVAL;
+ goto free_job;
+ }
+
+ if (cmd->wait_syncpt.id != args->syncpt_incr.id) {
+ SUBMIT_ERR(ctx, "syncpoint ID in CMD_WAIT_SYNCPT_RELATIVE is not used by the job");
+ err = -EINVAL;
+ goto free_job;
+ }
+
+ host1x_job_add_wait(job, cmd->wait_syncpt.id,
+ cmd->wait_syncpt.threshold,
+ true,
+ class);
+ } else {
+ SUBMIT_ERR(ctx, "unknown cmd type");
+ err = -EINVAL;
+ goto free_job;
+ }
+ }
+
+ if (gather_offset == 0) {
+ SUBMIT_ERR(ctx, "job must have at least one gather");
+ err = -EINVAL;
+ goto free_job;
+ }
+
+ *pjob = job;
+
+ err = 0;
+ goto done;
+
+free_job:
+ host1x_job_put(job);
+
+done:
+ kvfree(cmds);
+
+ return err;
+}
+
+static void release_job(struct host1x_job *job)
+{
+ struct tegra_drm_client *client =
+ container_of(job->client, struct tegra_drm_client, base);
+ struct tegra_drm_submit_data *job_data = job->user_data;
+ u32 i;
+
+ for (i = 0; i < job_data->num_used_mappings; i++)
+ tegra_drm_mapping_put(job_data->used_mappings[i].mapping);
+
+ kfree(job_data->used_mappings);
+ kfree(job_data);
+
+ if (pm_runtime_enabled(client->base.dev))
+ pm_runtime_put_autosuspend(client->base.dev);
+}
+
+int tegra_drm_ioctl_channel_submit(struct drm_device *drm, void *data,
+ struct drm_file *file)
+{
+ struct tegra_drm_file *fpriv = file->driver_priv;
+ struct drm_tegra_channel_submit *args = data;
+ struct tegra_drm_submit_data *job_data;
+ struct drm_syncobj *syncobj = NULL;
+ struct tegra_drm_context *ctx;
+ struct host1x_job *job;
+ struct gather_bo *bo;
+ u32 i;
+ int err;
+
+ mutex_lock(&fpriv->lock);
+ ctx = xa_load(&fpriv->contexts, args->channel_ctx);
+ if (!ctx) {
+ mutex_unlock(&fpriv->lock);
+ pr_err_ratelimited("%s: %s: invalid channel_ctx '%d'", __func__,
+ current->comm, args->channel_ctx);
+ return -EINVAL;
+ }
+
+ if (args->syncobj_in) {
+ struct dma_fence *fence;
+
+ err = drm_syncobj_find_fence(file, args->syncobj_in, 0, 0, &fence);
+ if (err) {
+ SUBMIT_ERR(ctx, "invalid syncobj_in '%d'", args->syncobj_in);
+ goto unlock;
+ }
+
+ err = dma_fence_wait_timeout(fence, true, msecs_to_jiffies(10000));
+ dma_fence_put(fence);
+ if (err) {
+ SUBMIT_ERR(ctx, "wait for syncobj_in timed out");
+ goto unlock;
+ }
+ }
+
+ if (args->syncobj_out) {
+ syncobj = drm_syncobj_find(file, args->syncobj_out);
+ if (!syncobj) {
+ SUBMIT_ERR(ctx, "invalid syncobj_out '%d'", args->syncobj_out);
+ err = -ENOENT;
+ goto unlock;
+ }
+ }
+
+ /* Allocate gather BO and copy gather words in. */
+ err = submit_copy_gather_data(&bo, ctx, args);
+ if (err)
+ goto unlock;
+
+ job_data = kzalloc(sizeof(*job_data), GFP_KERNEL);
+ if (!job_data) {
+ SUBMIT_ERR(ctx, "failed to allocate memory for job data");
+ err = -ENOMEM;
+ goto put_bo;
+ }
+
+ /* Get data buffer mappings and do relocation patching. */
+ err = submit_process_bufs(ctx, bo, args, job_data);
+ if (err)
+ goto free_job_data;
+
+ /* Allocate host1x_job and add gathers and waits to it. */
+ err = submit_create_job(&job, ctx, bo, args, job_data,
+ &fpriv->syncpoints);
+ if (err)
+ goto free_job_data;
+
+ /* Map gather data for Host1x. */
+ err = host1x_job_pin(job, ctx->client->base.dev);
+ if (err)
+ goto put_job;
+
+ /* Boot engine. */
+ if (pm_runtime_enabled(ctx->client->base.dev)) {
+ err = pm_runtime_resume_and_get(ctx->client->base.dev);
+ if (err < 0) {
+ SUBMIT_ERR(ctx, "could not power up engine: %d", err);
+ goto unpin_job;
+ }
+ }
+
+ job->user_data = job_data;
+ job->release = release_job;
+ job->timeout = 10000;
+
+ /*
+ * job_data is now part of job reference counting, so don't release
+ * it from here.
+ */
+ job_data = NULL;
+
+ /* Submit job to hardware. */
+ err = host1x_job_submit(job);
+ if (err) {
+ SUBMIT_ERR(ctx, "host1x job submission failed: %d", err);
+ goto unpin_job;
+ }
+
+ /* Return postfences to userspace and add fences to DMA reservations. */
+ args->syncpt_incr.fence_value = job->syncpt_end;
+
+ if (syncobj) {
+ struct dma_fence *fence = host1x_fence_create(job->syncpt, job->syncpt_end);
+ if (IS_ERR(fence))
+ err = PTR_ERR(fence);
+
+ drm_syncobj_replace_fence(syncobj, fence);
+ }
+
+ goto put_job;
+
+unpin_job:
+ host1x_job_unpin(job);
+put_job:
+ host1x_job_put(job);
+free_job_data:
+ if (job_data && job_data->used_mappings) {
+ for (i = 0; i < job_data->num_used_mappings; i++)
+ tegra_drm_mapping_put(job_data->used_mappings[i].mapping);
+ kfree(job_data->used_mappings);
+ }
+ if (job_data)
+ kfree(job_data);
+put_bo:
+ gather_bo_put(&bo->base);
+unlock:
+ if (syncobj)
+ drm_syncobj_put(syncobj);
+ mutex_unlock(&fpriv->lock);
+ return err;
+}
diff --git a/drivers/gpu/drm/tegra/submit.h b/drivers/gpu/drm/tegra/submit.h
new file mode 100644
index 000000000000..0a165e9e4bda
--- /dev/null
+++ b/drivers/gpu/drm/tegra/submit.h
@@ -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
--
2.30.1
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v6 13/14] drm/tegra: Add job firewall
2021-05-30 12:21 [PATCH v6 00/14] TegraDRM UAPI Mikko Perttunen
` (11 preceding siblings ...)
2021-05-30 12:21 ` [PATCH v6 12/14] drm/tegra: Implement job submission part of new UAPI Mikko Perttunen
@ 2021-05-30 12:21 ` Mikko Perttunen
2021-05-30 12:21 ` [PATCH v6 14/14] drm/tegra: Bump driver version Mikko Perttunen
13 siblings, 0 replies; 16+ messages in thread
From: Mikko Perttunen @ 2021-05-30 12:21 UTC (permalink / raw)
To: thierry.reding, jonathanh, digetx, airlied, daniel
Cc: linux-tegra, dri-devel, 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.
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
---
v6:
* If .is_valid_class is not set, match against client's
class
* Fix indentation
* Return to previous class after a WAIT command
* Print warning detailing error when rejecting a command
buffer
* Removed uapi subdirectory
v5:
* Support SETCLASS opcode
v3:
* New patch
---
drivers/gpu/drm/tegra/Makefile | 1 +
drivers/gpu/drm/tegra/firewall.c | 254 +++++++++++++++++++++++++++++++
drivers/gpu/drm/tegra/submit.c | 12 +-
drivers/gpu/drm/tegra/submit.h | 4 +
4 files changed, 269 insertions(+), 2 deletions(-)
create mode 100644 drivers/gpu/drm/tegra/firewall.c
diff --git a/drivers/gpu/drm/tegra/Makefile b/drivers/gpu/drm/tegra/Makefile
index ab4289d1c991..5d2039f0c734 100644
--- a/drivers/gpu/drm/tegra/Makefile
+++ b/drivers/gpu/drm/tegra/Makefile
@@ -6,6 +6,7 @@ tegra-drm-y := \
uapi.o \
submit.o \
gather_bo.o \
+ firewall.o \
gem.o \
fb.o \
dp.o \
diff --git a/drivers/gpu/drm/tegra/firewall.c b/drivers/gpu/drm/tegra/firewall.c
new file mode 100644
index 000000000000..1824d2db0e2c
--- /dev/null
+++ b/drivers/gpu/drm/tegra/firewall.c
@@ -0,0 +1,254 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Copyright (c) 2010-2020 NVIDIA Corporation */
+
+#include "drm.h"
+#include "submit.h"
+#include "uapi.h"
+
+struct tegra_drm_firewall {
+ struct tegra_drm_submit_data *submit;
+ struct tegra_drm_client *client;
+ u32 *data;
+ u32 pos;
+ u32 end;
+ u32 class;
+};
+
+static int fw_next(struct tegra_drm_firewall *fw, u32 *word)
+{
+ if (fw->pos == fw->end)
+ return -EINVAL;
+
+ *word = fw->data[fw->pos++];
+
+ return 0;
+}
+
+static bool fw_check_addr_valid(struct tegra_drm_firewall *fw, u32 offset)
+{
+ u32 i;
+
+ for (i = 0; i < fw->submit->num_used_mappings; i++) {
+ struct tegra_drm_mapping *m = fw->submit->used_mappings[i].mapping;
+
+ if (offset >= m->iova && offset <= m->iova_end)
+ return true;
+ }
+
+ return false;
+}
+
+static int fw_check_reg(struct tegra_drm_firewall *fw, u32 offset)
+{
+ bool is_addr;
+ u32 word;
+ int err;
+
+ err = fw_next(fw, &word);
+ if (err)
+ return err;
+
+ if (!fw->client->ops->is_addr_reg)
+ return 0;
+
+ is_addr = fw->client->ops->is_addr_reg(fw->client->base.dev, fw->class,
+ offset);
+
+ if (!is_addr)
+ return 0;
+
+ if (!fw_check_addr_valid(fw, word))
+ return -EINVAL;
+
+ return 0;
+}
+
+static int fw_check_regs_seq(struct tegra_drm_firewall *fw, u32 offset,
+ u32 count, bool incr)
+{
+ u32 i;
+
+ for (i = 0; i < count; i++) {
+ if (fw_check_reg(fw, offset))
+ return -EINVAL;
+
+ if (incr)
+ offset++;
+ }
+
+ return 0;
+}
+
+static int fw_check_regs_mask(struct tegra_drm_firewall *fw, u32 offset,
+ u16 mask)
+{
+ unsigned long bmask = mask;
+ unsigned int bit;
+
+ for_each_set_bit(bit, &bmask, 16) {
+ if (fw_check_reg(fw, offset+bit))
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+static int fw_check_regs_imm(struct tegra_drm_firewall *fw, u32 offset)
+{
+ bool is_addr;
+
+ is_addr = fw->client->ops->is_addr_reg(fw->client->base.dev, fw->class,
+ offset);
+ if (is_addr)
+ return -EINVAL;
+
+ return 0;
+}
+
+static int fw_check_class(struct tegra_drm_firewall *fw, u32 class)
+{
+ if (!fw->client->ops->is_valid_class) {
+ if (class == fw->client->base.class)
+ return 0;
+ else
+ return -EINVAL;
+ }
+
+ if (!fw->client->ops->is_valid_class(class))
+ return -EINVAL;
+
+ return 0;
+}
+
+enum {
+ HOST1X_OPCODE_SETCLASS = 0x00,
+ HOST1X_OPCODE_INCR = 0x01,
+ HOST1X_OPCODE_NONINCR = 0x02,
+ HOST1X_OPCODE_MASK = 0x03,
+ HOST1X_OPCODE_IMM = 0x04,
+ HOST1X_OPCODE_RESTART = 0x05,
+ HOST1X_OPCODE_GATHER = 0x06,
+ HOST1X_OPCODE_SETSTRMID = 0x07,
+ HOST1X_OPCODE_SETAPPID = 0x08,
+ HOST1X_OPCODE_SETPYLD = 0x09,
+ HOST1X_OPCODE_INCR_W = 0x0a,
+ HOST1X_OPCODE_NONINCR_W = 0x0b,
+ HOST1X_OPCODE_GATHER_W = 0x0c,
+ HOST1X_OPCODE_RESTART_W = 0x0d,
+ HOST1X_OPCODE_EXTEND = 0x0e,
+};
+
+int tegra_drm_fw_validate(struct tegra_drm_client *client, u32 *data, u32 start,
+ u32 words, struct tegra_drm_submit_data *submit,
+ u32 *job_class)
+{
+ struct tegra_drm_firewall fw = {
+ .submit = submit,
+ .client = client,
+ .data = data,
+ .pos = start,
+ .end = start+words,
+ .class = *job_class,
+ };
+ bool payload_valid = false;
+ u32 payload;
+ int err;
+
+ while (fw.pos != fw.end) {
+ u32 word, opcode, offset, count, mask, class;
+
+ err = fw_next(&fw, &word);
+ if (err)
+ return err;
+
+ opcode = (word & 0xf0000000) >> 28;
+
+ switch (opcode) {
+ case HOST1X_OPCODE_SETCLASS:
+ offset = word >> 16 & 0xfff;
+ mask = word & 0x3f;
+ class = (word >> 6) & 0x3ff;
+ err = fw_check_class(&fw, class);
+ fw.class = class;
+ *job_class = class;
+ if (!err)
+ err = fw_check_regs_mask(&fw, offset, mask);
+ if (err)
+ dev_warn(client->base.dev,
+ "illegal SETCLASS(offset=0x%x, mask=0x%x, class=0x%x) at word %u",
+ offset, mask, class, fw.pos-1);
+ break;
+ case HOST1X_OPCODE_INCR:
+ offset = (word >> 16) & 0xfff;
+ count = word & 0xffff;
+ err = fw_check_regs_seq(&fw, offset, count, true);
+ if (err)
+ dev_warn(client->base.dev,
+ "illegal INCR(offset=0x%x, count=%u) in class 0x%x at word %u",
+ offset, count, fw.class, fw.pos-1);
+ break;
+ case HOST1X_OPCODE_NONINCR:
+ offset = (word >> 16) & 0xfff;
+ count = word & 0xffff;
+ err = fw_check_regs_seq(&fw, offset, count, false);
+ if (err)
+ dev_warn(client->base.dev,
+ "illegal NONINCR(offset=0x%x, count=%u) in class 0x%x at word %u",
+ offset, count, fw.class, fw.pos-1);
+ break;
+ case HOST1X_OPCODE_MASK:
+ offset = (word >> 16) & 0xfff;
+ mask = word & 0xffff;
+ err = fw_check_regs_mask(&fw, offset, mask);
+ if (err)
+ dev_warn(client->base.dev,
+ "illegal MASK(offset=0x%x, mask=0x%x) in class 0x%x at word %u",
+ offset, mask, fw.class, fw.pos-1);
+ break;
+ case HOST1X_OPCODE_IMM:
+ /* IMM cannot reasonably be used to write a pointer */
+ offset = (word >> 16) & 0xfff;
+ err = fw_check_regs_imm(&fw, offset);
+ if (err)
+ dev_warn(client->base.dev,
+ "illegal IMM(offset=0x%x) in class 0x%x at word %u",
+ offset, fw.class, fw.pos-1);
+ break;
+ case HOST1X_OPCODE_SETPYLD:
+ payload = word & 0xffff;
+ payload_valid = true;
+ break;
+ case HOST1X_OPCODE_INCR_W:
+ if (!payload_valid)
+ return -EINVAL;
+
+ offset = word & 0x3fffff;
+ err = fw_check_regs_seq(&fw, offset, payload, true);
+ if (err)
+ dev_warn(client->base.dev,
+ "illegal INCR_W(offset=0x%x) in class 0x%x at word %u",
+ offset, fw.class, fw.pos-1);
+ break;
+ case HOST1X_OPCODE_NONINCR_W:
+ if (!payload_valid)
+ return -EINVAL;
+
+ offset = word & 0x3fffff;
+ err = fw_check_regs_seq(&fw, offset, payload, false);
+ if (err)
+ dev_warn(client->base.dev,
+ "illegal NONINCR(offset=0x%x) in class 0x%x at word %u",
+ offset, fw.class, fw.pos-1);
+ break;
+ default:
+ dev_warn(client->base.dev, "illegal opcode at word %u",
+ fw.pos-1);
+ return -EINVAL;
+ }
+
+ if (err)
+ return err;
+ }
+
+ return 0;
+}
diff --git a/drivers/gpu/drm/tegra/submit.c b/drivers/gpu/drm/tegra/submit.c
index ae1bf916b972..c23536a46475 100644
--- a/drivers/gpu/drm/tegra/submit.c
+++ b/drivers/gpu/drm/tegra/submit.c
@@ -246,7 +246,8 @@ static int submit_job_add_gather(struct host1x_job *job,
struct tegra_drm_context *ctx,
struct drm_tegra_submit_cmd_gather_uptr *cmd,
struct gather_bo *bo, u32 *offset,
- struct tegra_drm_submit_data *job_data)
+ struct tegra_drm_submit_data *job_data,
+ u32 *class)
{
u32 next_offset;
@@ -271,6 +272,12 @@ static int submit_job_add_gather(struct host1x_job *job,
return -EINVAL;
}
+ if (tegra_drm_fw_validate(ctx->client, bo->gather_data, *offset,
+ cmd->words, job_data, class)) {
+ SUBMIT_ERR(ctx, "job was rejected by firewall");
+ return -EINVAL;
+ }
+
host1x_job_add_gather(job, &bo->base, cmd->words, *offset * 4);
*offset = next_offset;
@@ -327,7 +334,8 @@ static int submit_create_job(struct host1x_job **pjob,
if (cmd->type == DRM_TEGRA_SUBMIT_CMD_GATHER_UPTR) {
err = submit_job_add_gather(job, ctx,
&cmd->gather_uptr, bo,
- &gather_offset, job_data);
+ &gather_offset, job_data,
+ &class);
if (err)
goto free_job;
} else if (cmd->type == DRM_TEGRA_SUBMIT_CMD_WAIT_SYNCPT) {
diff --git a/drivers/gpu/drm/tegra/submit.h b/drivers/gpu/drm/tegra/submit.h
index 0a165e9e4bda..cf6a2f0a29fc 100644
--- a/drivers/gpu/drm/tegra/submit.h
+++ b/drivers/gpu/drm/tegra/submit.h
@@ -14,4 +14,8 @@ 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,
+ u32 *job_class);
+
#endif
--
2.30.1
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v6 14/14] drm/tegra: Bump driver version
2021-05-30 12:21 [PATCH v6 00/14] TegraDRM UAPI Mikko Perttunen
` (12 preceding siblings ...)
2021-05-30 12:21 ` [PATCH v6 13/14] drm/tegra: Add job firewall Mikko Perttunen
@ 2021-05-30 12:21 ` Mikko Perttunen
13 siblings, 0 replies; 16+ messages in thread
From: Mikko Perttunen @ 2021-05-30 12:21 UTC (permalink / raw)
To: thierry.reding, jonathanh, digetx, airlied, daniel
Cc: linux-tegra, dri-devel, Mikko Perttunen
Bump driver version to 1.0.0 to allow userspace to detect
availability of new interfaces.
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
---
drivers/gpu/drm/tegra/drm.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/tegra/drm.c b/drivers/gpu/drm/tegra/drm.c
index 59f04431cae0..f568bb4c3345 100644
--- a/drivers/gpu/drm/tegra/drm.c
+++ b/drivers/gpu/drm/tegra/drm.c
@@ -27,7 +27,7 @@
#define DRIVER_NAME "tegra"
#define DRIVER_DESC "NVIDIA Tegra graphics"
#define DRIVER_DATE "20120330"
-#define DRIVER_MAJOR 0
+#define DRIVER_MAJOR 1
#define DRIVER_MINOR 0
#define DRIVER_PATCHLEVEL 0
--
2.30.1
^ permalink raw reply related [flat|nested] 16+ messages in thread