[RFC PATCH v1 14/15] drm/msm/a6xx: Support per-instance pagetables

From: Jordan Crouse
Date: Fri Mar 01 2019 - 14:39:55 EST


Add support for per-instance pagetables for a6xx targets. Add support
to handle split pagetables and create a new instance if the needed
IOMMU support exists and insert the necessary PM4 commands to trigger
a pagetable switch at the beginning of a user command.

Signed-off-by: Jordan Crouse <jcrouse@xxxxxxxxxxxxxx>
---

drivers/gpu/drm/msm/adreno/a6xx_gpu.c | 124 ++++++++++++++++++++++++++++++++--
drivers/gpu/drm/msm/adreno/a6xx_gpu.h | 1 +
2 files changed, 121 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
index f2e0800..57d1909 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
@@ -10,6 +10,63 @@

#include <linux/devfreq.h>

+static void a6xx_set_pagetable(struct msm_gpu *gpu, struct msm_ringbuffer *ring,
+ struct msm_file_private *ctx)
+{
+ u64 ttbr;
+ u32 asid;
+
+ if (!msm_iommu_get_ptinfo(ctx->aspace->mmu, &ttbr, &asid))
+ return;
+
+ ttbr = ttbr | ((u64) asid) << 48;
+
+ /* Turn off protected mode */
+ OUT_PKT7(ring, CP_SET_PROTECTED_MODE, 1);
+ OUT_RING(ring, 0);
+
+ /* Turn on APIV mode to access critical regions */
+ OUT_PKT4(ring, REG_A6XX_CP_MISC_CNTL, 1);
+ OUT_RING(ring, 1);
+
+ /* Make sure the ME is synchronized before staring the update */
+ OUT_PKT7(ring, CP_WAIT_FOR_ME, 0);
+
+ /* Execute the table update */
+ OUT_PKT7(ring, CP_SMMU_TABLE_UPDATE, 4);
+ OUT_RING(ring, lower_32_bits(ttbr));
+ OUT_RING(ring, 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 preemption records - this will be used to
+ * reload the pagetable if the current ring gets preempted out.
+ */
+ 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, upper_32_bits(ttbr));
+
+ /* Invalidate the draw state so we start off fresh */
+ OUT_PKT7(ring, CP_SET_DRAW_STATE, 3);
+ OUT_RING(ring, 0x40000);
+ OUT_RING(ring, 1);
+ OUT_RING(ring, 0);
+
+ /* Turn off APRIV */
+ OUT_PKT4(ring, REG_A6XX_CP_MISC_CNTL, 1);
+ OUT_RING(ring, 0);
+
+ /* Turn off protected mode */
+ OUT_PKT7(ring, CP_SET_PROTECTED_MODE, 1);
+ OUT_RING(ring, 1);
+}
+
+
static inline bool _a6xx_check_idle(struct msm_gpu *gpu)
{
struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
@@ -87,6 +144,8 @@ 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));

@@ -783,21 +842,77 @@ static unsigned long a6xx_gpu_busy(struct msm_gpu *gpu)
return (unsigned long)busy_time;
}

+static struct msm_gem_address_space *a6xx_new_address_space(struct msm_gpu *gpu)
+{
+ struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
+ struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
+ struct msm_gem_address_space *aspace;
+ int ret;
+
+ /* Return the default pagetable if per instance tables don't work */
+ if (!a6xx_gpu->per_instance_tables)
+ return gpu->aspace;
+
+ aspace = msm_gem_address_space_create_instance(&gpu->pdev->dev, "gpu",
+ 0x100000000ULL, 0x1ffffffffULL);
+ if (IS_ERR(aspace))
+ return aspace;
+
+ ret = aspace->mmu->funcs->attach(aspace->mmu, NULL, 0);
+ if (ret) {
+ /* -ENODEV means that aux domains aren't supported */
+ if (ret == -ENODEV)
+ return gpu->aspace;
+
+ return ERR_PTR(ret);
+ }
+
+ return aspace;
+}
+
static struct msm_gem_address_space *
a6xx_create_address_space(struct msm_gpu *gpu)
{
+ struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
+ struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
+ struct device *dev = &gpu->pdev->dev;
struct msm_gem_address_space *aspace;
struct iommu_domain *iommu;
- int ret;
+ int ret, val = 1;
+
+ a6xx_gpu->per_instance_tables = false;

iommu = iommu_domain_alloc(&platform_bus_type);
if (!iommu)
return NULL;

- iommu->geometry.aperture_start = 0x100000000ULL;
- iommu->geometry.aperture_end = 0x1ffffffffULL;
+ /* Try to enable split pagetables */
+ if (iommu_domain_set_attr(iommu, DOMAIN_ATTR_SPLIT_TABLES, &val)) {
+ /*
+ * If split pagetables aren't available we won't be able to do
+ * per-instance pagetables so set up the global va space at our
+ * susual location
+ */
+ iommu->geometry.aperture_start = 0x100000000ULL;
+ iommu->geometry.aperture_end = 0x1ffffffffULL;
+ } else {
+ /*
+ * If split pagetables are available then we might be able to do
+ * per-instance pagetables. Put the default va-space in TTBR1 to
+ * prepare
+ */
+ iommu->geometry.aperture_start = 0xfffffff100000000ULL;
+ iommu->geometry.aperture_end = 0xffffff1ffffffffULL;
+
+ /*
+ * If both split pagetables and aux domains are supported we can
+ * do per_instance pagetables
+ */
+ a6xx_gpu->per_instance_tables =
+ iommu_dev_has_feature(dev, IOMMU_DEV_FEAT_AUX);
+ }

- aspace = msm_gem_address_space_create(&gpu->pdev->dev, iommu, "gpu");
+ aspace = msm_gem_address_space_create(dev, iommu, "gpu");
if (IS_ERR(aspace)) {
iommu_domain_free(iommu);
DRM_DEV_ERROR(gpu->dev->dev, "failed to init mmu: %ld\n",
@@ -836,6 +951,7 @@ static const struct adreno_gpu_funcs funcs = {
.gpu_state_get = a6xx_gpu_state_get,
.gpu_state_put = a6xx_gpu_state_put,
.create_address_space = a6xx_create_address_space,
+ .new_address_space = a6xx_new_address_space,
},
.get_timestamp = a6xx_get_timestamp,
};
diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.h b/drivers/gpu/drm/msm/adreno/a6xx_gpu.h
index 528a4cf..4e9ce1c 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.h
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.h
@@ -21,6 +21,7 @@ struct a6xx_gpu {
struct msm_ringbuffer *cur_ring;

struct a6xx_gmu gmu;
+ bool per_instance_tables;
};

#define to_a6xx_gpu(x) container_of(x, struct a6xx_gpu, base)
--
2.7.4