Hi Boris,
On 02.12.2024 12:20, Boris Brezillon wrote:
> On Thu, 28 Nov 2024 21:06:21 +0000
> Adrián Larumbe <adrian.larumbe@xxxxxxxxxxxxx> wrote:
>
> > Rather than remasking interrupts after a device reset in the main reset
> > path, allow selecting whether to do this with an additional bool parameter.
> >
> > To this end, split reenabling job interrupts into two functions, one that
> > clears the interrupts and another one which unmasks them conditionally.
> >
> > Signed-off-by: Adrián Larumbe <adrian.larumbe@xxxxxxxxxxxxx>
> > ---
> > drivers/gpu/drm/panfrost/panfrost_device.c | 11 +++++--
> > drivers/gpu/drm/panfrost/panfrost_device.h | 2 +-
> > drivers/gpu/drm/panfrost/panfrost_job.c | 34 ++++++++++++----------
> > drivers/gpu/drm/panfrost/panfrost_job.h | 3 +-
> > 4 files changed, 29 insertions(+), 21 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/panfrost/panfrost_device.c b/drivers/gpu/drm/panfrost/panfrost_device.c
> > index 4fe29286bbe3..7e5d39699982 100644
> > --- a/drivers/gpu/drm/panfrost/panfrost_device.c
> > +++ b/drivers/gpu/drm/panfrost/panfrost_device.c
> > @@ -395,20 +395,25 @@ bool panfrost_exception_needs_reset(const struct panfrost_device *pfdev,
> > return false;
> > }
> >
> > -void panfrost_device_reset(struct panfrost_device *pfdev)
> > +void panfrost_device_reset(struct panfrost_device *pfdev, bool enable_job_int)
> > {
> > + u32 irq_mask;
> > +
> > panfrost_gpu_soft_reset(pfdev);
> >
> > panfrost_gpu_power_on(pfdev);
> > panfrost_mmu_reset(pfdev);
> > - panfrost_job_enable_interrupts(pfdev);
> > +
> > + irq_mask = panfrost_job_reset_interrupts(pfdev);
> > + if (enable_job_int)
> > + panfrost_enable_interrupts(pfdev, irq_mask);
> > }
> >
> > static int panfrost_device_runtime_resume(struct device *dev)
> > {
> > struct panfrost_device *pfdev = dev_get_drvdata(dev);
> >
> > - panfrost_device_reset(pfdev);
> > + panfrost_device_reset(pfdev, true);
> > panfrost_devfreq_resume(pfdev);
> >
> > return 0;
> > diff --git a/drivers/gpu/drm/panfrost/panfrost_device.h b/drivers/gpu/drm/panfrost/panfrost_device.h
> > index d9aea2c2cbe5..fc83d5e104a3 100644
> > --- a/drivers/gpu/drm/panfrost/panfrost_device.h
> > +++ b/drivers/gpu/drm/panfrost/panfrost_device.h
> > @@ -205,7 +205,7 @@ int panfrost_unstable_ioctl_check(void);
> >
> > int panfrost_device_init(struct panfrost_device *pfdev);
> > void panfrost_device_fini(struct panfrost_device *pfdev);
> > -void panfrost_device_reset(struct panfrost_device *pfdev);
> > +void panfrost_device_reset(struct panfrost_device *pfdev, bool enable_job_int);
> >
> > extern const struct dev_pm_ops panfrost_pm_ops;
> >
> > diff --git a/drivers/gpu/drm/panfrost/panfrost_job.c b/drivers/gpu/drm/panfrost/panfrost_job.c
> > index fba1a376f593..79d2ee3625b2 100644
> > --- a/drivers/gpu/drm/panfrost/panfrost_job.c
> > +++ b/drivers/gpu/drm/panfrost/panfrost_job.c
> > @@ -27,6 +27,10 @@
> > #define job_write(dev, reg, data) writel(data, dev->iomem + (reg))
> > #define job_read(dev, reg) readl(dev->iomem + (reg))
> >
> > +#define JOB_INT_ENABLE_MASK \
> > + (GENMASK(16 + NUM_JOB_SLOTS - 1, 16) | \
> > + GENMASK(NUM_JOB_SLOTS - 1, 0))
> > +
> > struct panfrost_queue_state {
> > struct drm_gpu_scheduler sched;
> > u64 fence_context;
> > @@ -421,18 +425,23 @@ static struct dma_fence *panfrost_job_run(struct drm_sched_job *sched_job)
> > return fence;
> > }
> >
> > -void panfrost_job_enable_interrupts(struct panfrost_device *pfdev)
> > +u32 panfrost_job_reset_interrupts(struct panfrost_device *pfdev)
> > {
> > int j;
> > u32 irq_mask = 0;
> >
> > - clear_bit(PANFROST_COMP_BIT_JOB, pfdev->is_suspended);
> > -
> > for (j = 0; j < NUM_JOB_SLOTS; j++) {
> > irq_mask |= MK_JS_MASK(j);
> > }
> >
> > job_write(pfdev, JOB_INT_CLEAR, irq_mask);
> > +
> > + return irq_mask;
> > +}
>
> Let's define an ALL_JS_INT_MASK so we can get rid of the for loop and
> can avoid returning a value that's constant.
Because ALL_JS_INT_MASK would be defined as an OR of MK_JS_MASK() applications,
and this macro is defined in panfrost_regs.h, I can't think of a nice location
for it that would be suitable for all the files where it would be called.
Maybe I could implement the same loop inside panfrost_job_init() where it would be
called only once, and store the result in a const struct panfrost_job_slot field?
> > +
> > +void panfrost_enable_interrupts(struct panfrost_device *pfdev, u32 irq_mask)
>
> Let's drop the irq_mask mask (use ALL_JS_INT_MASK), and call the
> function panfrost_job_enable_interrupts() instead.
I made a mistake here naming this function, it should've been panfrost_job_enable_interrupts().
Other than that, I decided to keep the irq_mask argument because I'm also calling it from
the very end of panfrost_reset() with JOB_INT_ENABLE_MASK, where I defined it to be
(GENMASK(16 + NUM_JOB_SLOTS - 1, 16) | GENMASK(NUM_JOB_SLOTS - 1, 0))
That raises the question, what is the functional difference between writing either
this or MK_JS_MASK(0) | MK_JS_MASK(1) | MK_JS_MASK(2) into the JOB_INT_MASK register?
It's also being done in panfrost_job_irq_handler_thread() so I'm not quite sure of this.
> > +{
> > + clear_bit(PANFROST_COMP_BIT_JOB, pfdev->is_suspended);
> > job_write(pfdev, JOB_INT_MASK, irq_mask);
> > }
> >
> > @@ -725,12 +734,7 @@ panfrost_reset(struct panfrost_device *pfdev,
> > spin_unlock(&pfdev->js->job_lock);
> >
> > /* Proceed with reset now. */
> > - panfrost_device_reset(pfdev);
> > -
> > - /* panfrost_device_reset() unmasks job interrupts, but we want to
> > - * keep them masked a bit longer.
> > - */
> > - job_write(pfdev, JOB_INT_MASK, 0);
> > + panfrost_device_reset(pfdev, false);
> >
> > /* GPU has been reset, we can clear the reset pending bit. */
> > atomic_set(&pfdev->reset.pending, 0);
> > @@ -752,9 +756,7 @@ panfrost_reset(struct panfrost_device *pfdev,
> > drm_sched_start(&pfdev->js->queue[i].sched, 0);
> >
> > /* Re-enable job interrupts now that everything has been restarted. */
> > - job_write(pfdev, JOB_INT_MASK,
> > - GENMASK(16 + NUM_JOB_SLOTS - 1, 16) |
> > - GENMASK(NUM_JOB_SLOTS - 1, 0));
> > + panfrost_enable_interrupts(pfdev, JOB_INT_ENABLE_MASK);
> >
> > dma_fence_end_signalling(cookie);
> > }
> > @@ -827,9 +829,7 @@ static irqreturn_t panfrost_job_irq_handler_thread(int irq, void *data)
> >
> > /* Enable interrupts only if we're not about to get suspended */
> > if (!test_bit(PANFROST_COMP_BIT_JOB, pfdev->is_suspended))
> > - job_write(pfdev, JOB_INT_MASK,
> > - GENMASK(16 + NUM_JOB_SLOTS - 1, 16) |
> > - GENMASK(NUM_JOB_SLOTS - 1, 0));
> > + job_write(pfdev, JOB_INT_MASK, JOB_INT_ENABLE_MASK);
> >
> > return IRQ_HANDLED;
> > }
> > @@ -854,6 +854,7 @@ int panfrost_job_init(struct panfrost_device *pfdev)
> > {
> > struct panfrost_job_slot *js;
> > unsigned int nentries = 2;
> > + u32 irq_mask;
> > int ret, j;
> >
> > /* All GPUs have two entries per queue, but without jobchain
> > @@ -905,7 +906,8 @@ int panfrost_job_init(struct panfrost_device *pfdev)
> > }
> > }
> >
> > - panfrost_job_enable_interrupts(pfdev);
> > + irq_mask = panfrost_job_reset_interrupts(pfdev);
> > + panfrost_enable_interrupts(pfdev, irq_mask);
> >
> > return 0;
> >
> > diff --git a/drivers/gpu/drm/panfrost/panfrost_job.h b/drivers/gpu/drm/panfrost/panfrost_job.h
> > index ec581b97852b..c09d42ee5039 100644
> > --- a/drivers/gpu/drm/panfrost/panfrost_job.h
> > +++ b/drivers/gpu/drm/panfrost/panfrost_job.h
> > @@ -46,7 +46,8 @@ void panfrost_job_close(struct panfrost_file_priv *panfrost_priv);
> > int panfrost_job_get_slot(struct panfrost_job *job);
> > int panfrost_job_push(struct panfrost_job *job);
> > void panfrost_job_put(struct panfrost_job *job);
> > -void panfrost_job_enable_interrupts(struct panfrost_device *pfdev);
> > +u32 panfrost_job_reset_interrupts(struct panfrost_device *pfdev);
> > +void panfrost_enable_interrupts(struct panfrost_device *pfdev, u32 irq_mask);
> > void panfrost_job_suspend_irq(struct panfrost_device *pfdev);
> > int panfrost_job_is_idle(struct panfrost_device *pfdev);
> >
