[v2,6/6] drm/msm/a6xx: Add support for per-instance pagetables
diff mbox series

Message ID 20200626200414.14382-7-jcrouse@codeaurora.org
State New
Headers show
Series
  • iommu-arm-smmu: Add auxiliary domains and per-instance pagetables
Related show

Commit Message

Jordan Crouse June 26, 2020, 8:04 p.m. UTC
Add support for using per-instance pagetables if all the dependencies are
available.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
---

 drivers/gpu/drm/msm/adreno/a6xx_gpu.c | 43 +++++++++++++++++++++++++++
 drivers/gpu/drm/msm/msm_ringbuffer.h  |  1 +
 2 files changed, 44 insertions(+)

Comments

Rob Clark June 27, 2020, 7:56 p.m. UTC | #1
On Fri, Jun 26, 2020 at 1:04 PM Jordan Crouse <jcrouse@codeaurora.org> wrote:
>
> Add support for using per-instance pagetables if all the dependencies are
> available.
>
> Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
> ---
>
>  drivers/gpu/drm/msm/adreno/a6xx_gpu.c | 43 +++++++++++++++++++++++++++
>  drivers/gpu/drm/msm/msm_ringbuffer.h  |  1 +
>  2 files changed, 44 insertions(+)
>
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> index aa53f47b7e8b..95ed2ceac121 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> @@ -79,6 +79,34 @@ static void get_stats_counter(struct msm_ringbuffer *ring, u32 counter,
>         OUT_RING(ring, upper_32_bits(iova));
>  }
>
> +static void a6xx_set_pagetable(struct msm_gpu *gpu, struct msm_ringbuffer *ring,
> +               struct msm_file_private *ctx)
> +{
> +       phys_addr_t ttbr;
> +       u32 asid;
> +
> +       if (msm_iommu_pagetable_params(ctx->aspace->mmu, &ttbr, &asid))
> +               return;
> +
> +       /* Execute the table update */
> +       OUT_PKT7(ring, CP_SMMU_TABLE_UPDATE, 4);
> +       OUT_RING(ring, lower_32_bits(ttbr));
> +       OUT_RING(ring, (((u64) asid) << 48) | upper_32_bits(ttbr));
> +       /* CONTEXTIDR is currently unused */
> +       OUT_RING(ring, 0);
> +       /* CONTEXTBANK is currently unused */
> +       OUT_RING(ring, 0);
> +
> +       /*
> +        * Write the new TTBR0 to the memstore. This is good for debugging.
> +        */
> +       OUT_PKT7(ring, CP_MEM_WRITE, 4);
> +       OUT_RING(ring, lower_32_bits(rbmemptr(ring, ttbr0)));
> +       OUT_RING(ring, upper_32_bits(rbmemptr(ring, ttbr0)));
> +       OUT_RING(ring, lower_32_bits(ttbr));
> +       OUT_RING(ring, (((u64) asid) << 48) | upper_32_bits(ttbr));
> +}
> +
>  static void a6xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
>         struct msm_file_private *ctx)
>  {
> @@ -89,6 +117,8 @@ static void a6xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
>         struct msm_ringbuffer *ring = submit->ring;
>         unsigned int i;
>
> +       a6xx_set_pagetable(gpu, ring, ctx);
> +
>         get_stats_counter(ring, REG_A6XX_RBBM_PERFCTR_CP_0_LO,
>                 rbmemptr_stats(ring, index, cpcycles_start));
>
> @@ -872,6 +902,18 @@ static unsigned long a6xx_gpu_busy(struct msm_gpu *gpu)
>         return (unsigned long)busy_time;
>  }
>
> +struct msm_gem_address_space *a6xx_address_space_instance(struct msm_gpu *gpu)
> +{
> +       struct msm_mmu *mmu;
> +
> +       mmu = msm_iommu_pagetable_create(gpu->aspace->mmu);
> +       if (IS_ERR(mmu))
> +               return msm_gem_address_space_get(gpu->aspace);
> +
> +       return msm_gem_address_space_create(mmu,
> +               "gpu", 0x100000000ULL, 0x1ffffffffULL);
> +}
> +
>  static const struct adreno_gpu_funcs funcs = {
>         .base = {
>                 .get_param = adreno_get_param,
> @@ -895,6 +937,7 @@ static const struct adreno_gpu_funcs funcs = {
>                 .gpu_state_put = a6xx_gpu_state_put,
>  #endif
>                 .create_address_space = adreno_iommu_create_address_space,
> +               .address_space_instance = a6xx_address_space_instance,

Hmm, maybe instead of .address_space_instance, something like
.create_context_address_space?

Since like .create_address_space, it is creating an address space..
the difference is that it is a per context/process aspace..

BR,
-R

>         },
>         .get_timestamp = a6xx_get_timestamp,
>  };
> diff --git a/drivers/gpu/drm/msm/msm_ringbuffer.h b/drivers/gpu/drm/msm/msm_ringbuffer.h
> index 7764373d0ed2..0987d6bf848c 100644
> --- a/drivers/gpu/drm/msm/msm_ringbuffer.h
> +++ b/drivers/gpu/drm/msm/msm_ringbuffer.h
> @@ -31,6 +31,7 @@ struct msm_rbmemptrs {
>         volatile uint32_t fence;
>
>         volatile struct msm_gpu_submit_stats stats[MSM_GPU_SUBMIT_STATS_COUNT];
> +       volatile u64 ttbr0;
>  };
>
>  struct msm_ringbuffer {
> --
> 2.17.1
>
Rob Clark June 27, 2020, 8:11 p.m. UTC | #2
On Sat, Jun 27, 2020 at 12:56 PM Rob Clark <robdclark@gmail.com> wrote:
>
> On Fri, Jun 26, 2020 at 1:04 PM Jordan Crouse <jcrouse@codeaurora.org> wrote:
> >
> > Add support for using per-instance pagetables if all the dependencies are
> > available.
> >
> > Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
> > ---
> >
> >  drivers/gpu/drm/msm/adreno/a6xx_gpu.c | 43 +++++++++++++++++++++++++++
> >  drivers/gpu/drm/msm/msm_ringbuffer.h  |  1 +
> >  2 files changed, 44 insertions(+)
> >
> > diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> > index aa53f47b7e8b..95ed2ceac121 100644
> > --- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> > +++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> > @@ -79,6 +79,34 @@ static void get_stats_counter(struct msm_ringbuffer *ring, u32 counter,
> >         OUT_RING(ring, upper_32_bits(iova));
> >  }
> >
> > +static void a6xx_set_pagetable(struct msm_gpu *gpu, struct msm_ringbuffer *ring,
> > +               struct msm_file_private *ctx)
> > +{
> > +       phys_addr_t ttbr;
> > +       u32 asid;
> > +
> > +       if (msm_iommu_pagetable_params(ctx->aspace->mmu, &ttbr, &asid))
> > +               return;
> > +
> > +       /* Execute the table update */
> > +       OUT_PKT7(ring, CP_SMMU_TABLE_UPDATE, 4);
> > +       OUT_RING(ring, lower_32_bits(ttbr));
> > +       OUT_RING(ring, (((u64) asid) << 48) | upper_32_bits(ttbr));
> > +       /* CONTEXTIDR is currently unused */
> > +       OUT_RING(ring, 0);
> > +       /* CONTEXTBANK is currently unused */
> > +       OUT_RING(ring, 0);
> > +
> > +       /*
> > +        * Write the new TTBR0 to the memstore. This is good for debugging.
> > +        */
> > +       OUT_PKT7(ring, CP_MEM_WRITE, 4);
> > +       OUT_RING(ring, lower_32_bits(rbmemptr(ring, ttbr0)));
> > +       OUT_RING(ring, upper_32_bits(rbmemptr(ring, ttbr0)));
> > +       OUT_RING(ring, lower_32_bits(ttbr));
> > +       OUT_RING(ring, (((u64) asid) << 48) | upper_32_bits(ttbr));
> > +}
> > +
> >  static void a6xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
> >         struct msm_file_private *ctx)
> >  {
> > @@ -89,6 +117,8 @@ static void a6xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
> >         struct msm_ringbuffer *ring = submit->ring;
> >         unsigned int i;
> >
> > +       a6xx_set_pagetable(gpu, ring, ctx);
> > +
> >         get_stats_counter(ring, REG_A6XX_RBBM_PERFCTR_CP_0_LO,
> >                 rbmemptr_stats(ring, index, cpcycles_start));
> >
> > @@ -872,6 +902,18 @@ static unsigned long a6xx_gpu_busy(struct msm_gpu *gpu)
> >         return (unsigned long)busy_time;
> >  }
> >
> > +struct msm_gem_address_space *a6xx_address_space_instance(struct msm_gpu *gpu)
> > +{
> > +       struct msm_mmu *mmu;
> > +
> > +       mmu = msm_iommu_pagetable_create(gpu->aspace->mmu);
> > +       if (IS_ERR(mmu))
> > +               return msm_gem_address_space_get(gpu->aspace);
> > +
> > +       return msm_gem_address_space_create(mmu,
> > +               "gpu", 0x100000000ULL, 0x1ffffffffULL);
> > +}
> > +
> >  static const struct adreno_gpu_funcs funcs = {
> >         .base = {
> >                 .get_param = adreno_get_param,
> > @@ -895,6 +937,7 @@ static const struct adreno_gpu_funcs funcs = {
> >                 .gpu_state_put = a6xx_gpu_state_put,
> >  #endif
> >                 .create_address_space = adreno_iommu_create_address_space,
> > +               .address_space_instance = a6xx_address_space_instance,
>
> Hmm, maybe instead of .address_space_instance, something like
> .create_context_address_space?
>
> Since like .create_address_space, it is creating an address space..
> the difference is that it is a per context/process aspace..
>


or maybe just .create_pgtable and return the 'struct msm_mmu' (which
is itself starting to become less of a great name)..

The only other thing a6xx_address_space_instance() adds is knowing
where the split is between the kernel and user pgtables, and I suppose
that isn't a thing that would really be changing between gens?

BR,
-R

> BR,
> -R
>
> >         },
> >         .get_timestamp = a6xx_get_timestamp,
> >  };
> > diff --git a/drivers/gpu/drm/msm/msm_ringbuffer.h b/drivers/gpu/drm/msm/msm_ringbuffer.h
> > index 7764373d0ed2..0987d6bf848c 100644
> > --- a/drivers/gpu/drm/msm/msm_ringbuffer.h
> > +++ b/drivers/gpu/drm/msm/msm_ringbuffer.h
> > @@ -31,6 +31,7 @@ struct msm_rbmemptrs {
> >         volatile uint32_t fence;
> >
> >         volatile struct msm_gpu_submit_stats stats[MSM_GPU_SUBMIT_STATS_COUNT];
> > +       volatile u64 ttbr0;
> >  };
> >
> >  struct msm_ringbuffer {
> > --
> > 2.17.1
> >
Jordan Crouse June 29, 2020, 2:56 p.m. UTC | #3
On Sat, Jun 27, 2020 at 01:11:14PM -0700, Rob Clark wrote:
> On Sat, Jun 27, 2020 at 12:56 PM Rob Clark <robdclark@gmail.com> wrote:
> >
> > On Fri, Jun 26, 2020 at 1:04 PM Jordan Crouse <jcrouse@codeaurora.org> wrote:
> > >
> > > Add support for using per-instance pagetables if all the dependencies are
> > > available.
> > >
> > > Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
> > > ---
> > >
> > >  drivers/gpu/drm/msm/adreno/a6xx_gpu.c | 43 +++++++++++++++++++++++++++
> > >  drivers/gpu/drm/msm/msm_ringbuffer.h  |  1 +
> > >  2 files changed, 44 insertions(+)
> > >
> > > diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> > > index aa53f47b7e8b..95ed2ceac121 100644
> > > --- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> > > +++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> > > @@ -79,6 +79,34 @@ static void get_stats_counter(struct msm_ringbuffer *ring, u32 counter,
> > >         OUT_RING(ring, upper_32_bits(iova));
> > >  }
> > >
> > > +static void a6xx_set_pagetable(struct msm_gpu *gpu, struct msm_ringbuffer *ring,
> > > +               struct msm_file_private *ctx)
> > > +{
> > > +       phys_addr_t ttbr;
> > > +       u32 asid;
> > > +
> > > +       if (msm_iommu_pagetable_params(ctx->aspace->mmu, &ttbr, &asid))
> > > +               return;
> > > +
> > > +       /* Execute the table update */
> > > +       OUT_PKT7(ring, CP_SMMU_TABLE_UPDATE, 4);
> > > +       OUT_RING(ring, lower_32_bits(ttbr));
> > > +       OUT_RING(ring, (((u64) asid) << 48) | upper_32_bits(ttbr));
> > > +       /* CONTEXTIDR is currently unused */
> > > +       OUT_RING(ring, 0);
> > > +       /* CONTEXTBANK is currently unused */
> > > +       OUT_RING(ring, 0);
> > > +
> > > +       /*
> > > +        * Write the new TTBR0 to the memstore. This is good for debugging.
> > > +        */
> > > +       OUT_PKT7(ring, CP_MEM_WRITE, 4);
> > > +       OUT_RING(ring, lower_32_bits(rbmemptr(ring, ttbr0)));
> > > +       OUT_RING(ring, upper_32_bits(rbmemptr(ring, ttbr0)));
> > > +       OUT_RING(ring, lower_32_bits(ttbr));
> > > +       OUT_RING(ring, (((u64) asid) << 48) | upper_32_bits(ttbr));
> > > +}
> > > +
> > >  static void a6xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
> > >         struct msm_file_private *ctx)
> > >  {
> > > @@ -89,6 +117,8 @@ static void a6xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
> > >         struct msm_ringbuffer *ring = submit->ring;
> > >         unsigned int i;
> > >
> > > +       a6xx_set_pagetable(gpu, ring, ctx);
> > > +
> > >         get_stats_counter(ring, REG_A6XX_RBBM_PERFCTR_CP_0_LO,
> > >                 rbmemptr_stats(ring, index, cpcycles_start));
> > >
> > > @@ -872,6 +902,18 @@ static unsigned long a6xx_gpu_busy(struct msm_gpu *gpu)
> > >         return (unsigned long)busy_time;
> > >  }
> > >
> > > +struct msm_gem_address_space *a6xx_address_space_instance(struct msm_gpu *gpu)
> > > +{
> > > +       struct msm_mmu *mmu;
> > > +
> > > +       mmu = msm_iommu_pagetable_create(gpu->aspace->mmu);
> > > +       if (IS_ERR(mmu))
> > > +               return msm_gem_address_space_get(gpu->aspace);
> > > +
> > > +       return msm_gem_address_space_create(mmu,
> > > +               "gpu", 0x100000000ULL, 0x1ffffffffULL);
> > > +}
> > > +
> > >  static const struct adreno_gpu_funcs funcs = {
> > >         .base = {
> > >                 .get_param = adreno_get_param,
> > > @@ -895,6 +937,7 @@ static const struct adreno_gpu_funcs funcs = {
> > >                 .gpu_state_put = a6xx_gpu_state_put,
> > >  #endif
> > >                 .create_address_space = adreno_iommu_create_address_space,
> > > +               .address_space_instance = a6xx_address_space_instance,
> >
> > Hmm, maybe instead of .address_space_instance, something like
> > .create_context_address_space?
> >
> > Since like .create_address_space, it is creating an address space..
> > the difference is that it is a per context/process aspace..
> >

This is a good suggestion. I'm always open to changing function names.

> 
> 
> or maybe just .create_pgtable and return the 'struct msm_mmu' (which
> is itself starting to become less of a great name)..
> 
> The only other thing a6xx_address_space_instance() adds is knowing
> where the split is between the kernel and user pgtables, and I suppose
> that isn't a thing that would really be changing between gens?

In theory the split is determined by the hardware but its been the same for all
a5xx/a6xx targets.

Jordan

> BR,
> -R
> 
> > BR,
> > -R
> >
> > >         },
> > >         .get_timestamp = a6xx_get_timestamp,
> > >  };
> > > diff --git a/drivers/gpu/drm/msm/msm_ringbuffer.h b/drivers/gpu/drm/msm/msm_ringbuffer.h
> > > index 7764373d0ed2..0987d6bf848c 100644
> > > --- a/drivers/gpu/drm/msm/msm_ringbuffer.h
> > > +++ b/drivers/gpu/drm/msm/msm_ringbuffer.h
> > > @@ -31,6 +31,7 @@ struct msm_rbmemptrs {
> > >         volatile uint32_t fence;
> > >
> > >         volatile struct msm_gpu_submit_stats stats[MSM_GPU_SUBMIT_STATS_COUNT];
> > > +       volatile u64 ttbr0;
> > >  };
> > >
> > >  struct msm_ringbuffer {
> > > --
> > > 2.17.1
> > >
> _______________________________________________
> Freedreno mailing list
> Freedreno@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/freedreno

Patch
diff mbox series

diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
index aa53f47b7e8b..95ed2ceac121 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
@@ -79,6 +79,34 @@  static void get_stats_counter(struct msm_ringbuffer *ring, u32 counter,
 	OUT_RING(ring, upper_32_bits(iova));
 }
 
+static void a6xx_set_pagetable(struct msm_gpu *gpu, struct msm_ringbuffer *ring,
+		struct msm_file_private *ctx)
+{
+	phys_addr_t ttbr;
+	u32 asid;
+
+	if (msm_iommu_pagetable_params(ctx->aspace->mmu, &ttbr, &asid))
+		return;
+
+	/* Execute the table update */
+	OUT_PKT7(ring, CP_SMMU_TABLE_UPDATE, 4);
+	OUT_RING(ring, lower_32_bits(ttbr));
+	OUT_RING(ring, (((u64) asid) << 48) | upper_32_bits(ttbr));
+	/* CONTEXTIDR is currently unused */
+	OUT_RING(ring, 0);
+	/* CONTEXTBANK is currently unused */
+	OUT_RING(ring, 0);
+
+	/*
+	 * Write the new TTBR0 to the memstore. This is good for debugging.
+	 */
+	OUT_PKT7(ring, CP_MEM_WRITE, 4);
+	OUT_RING(ring, lower_32_bits(rbmemptr(ring, ttbr0)));
+	OUT_RING(ring, upper_32_bits(rbmemptr(ring, ttbr0)));
+	OUT_RING(ring, lower_32_bits(ttbr));
+	OUT_RING(ring, (((u64) asid) << 48) | upper_32_bits(ttbr));
+}
+
 static void a6xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
 	struct msm_file_private *ctx)
 {
@@ -89,6 +117,8 @@  static void a6xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
 	struct msm_ringbuffer *ring = submit->ring;
 	unsigned int i;
 
+	a6xx_set_pagetable(gpu, ring, ctx);
+
 	get_stats_counter(ring, REG_A6XX_RBBM_PERFCTR_CP_0_LO,
 		rbmemptr_stats(ring, index, cpcycles_start));
 
@@ -872,6 +902,18 @@  static unsigned long a6xx_gpu_busy(struct msm_gpu *gpu)
 	return (unsigned long)busy_time;
 }
 
+struct msm_gem_address_space *a6xx_address_space_instance(struct msm_gpu *gpu)
+{
+	struct msm_mmu *mmu;
+
+	mmu = msm_iommu_pagetable_create(gpu->aspace->mmu);
+	if (IS_ERR(mmu))
+		return msm_gem_address_space_get(gpu->aspace);
+
+	return msm_gem_address_space_create(mmu,
+		"gpu", 0x100000000ULL, 0x1ffffffffULL);
+}
+
 static const struct adreno_gpu_funcs funcs = {
 	.base = {
 		.get_param = adreno_get_param,
@@ -895,6 +937,7 @@  static const struct adreno_gpu_funcs funcs = {
 		.gpu_state_put = a6xx_gpu_state_put,
 #endif
 		.create_address_space = adreno_iommu_create_address_space,
+		.address_space_instance = a6xx_address_space_instance,
 	},
 	.get_timestamp = a6xx_get_timestamp,
 };
diff --git a/drivers/gpu/drm/msm/msm_ringbuffer.h b/drivers/gpu/drm/msm/msm_ringbuffer.h
index 7764373d0ed2..0987d6bf848c 100644
--- a/drivers/gpu/drm/msm/msm_ringbuffer.h
+++ b/drivers/gpu/drm/msm/msm_ringbuffer.h
@@ -31,6 +31,7 @@  struct msm_rbmemptrs {
 	volatile uint32_t fence;
 
 	volatile struct msm_gpu_submit_stats stats[MSM_GPU_SUBMIT_STATS_COUNT];
+	volatile u64 ttbr0;
 };
 
 struct msm_ringbuffer {