Re: [Intel-gfx] [PATCH v2] drm/i915/mtl: enable local stolen memory
On Mon, 26 Sep 2022, Aravind Iddamsetty wrote:
> As an integrated GPU, MTL does not have local memory and
> HAS_LMEM() returns false. However the platform's stolen memory
> is presented via BAR2 (i.e., the BAR we traditionally consider
> to be the LMEM BAR) and should be managed by the driver the same
> way that local memory is on dgpu platforms (which includes
> setting the "lmem" bit on page table entries). We use the term
> "local stolen memory" to refer to this model.
>
> v2:
> 1. dropped is_dsm_invalid, updated valid_stolen_size check from Lucas
> (Jani, Lucas)
> 2. drop lmembar_is_igpu_stolen
> 3. revert to referring GFXMEM_BAR as GEN12_LMEM_BAR (Lucas)
>
> Cc: Matt Roper
> Cc: Lucas De Marchi
>
> Signed-off-by: CQ Tang
> Signed-off-by: Aravind Iddamsetty
> Original-author: CQ Tang
> ---
> drivers/gpu/drm/i915/gem/i915_gem_stolen.c | 88 ++
> drivers/gpu/drm/i915/gt/intel_ggtt.c | 2 +-
> drivers/gpu/drm/i915/i915_drv.h| 3 +
> 3 files changed, 76 insertions(+), 17 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
> b/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
> index c5a4035c99cd..582c4d7d2a9a 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
> @@ -77,9 +77,9 @@ void i915_gem_stolen_remove_node(struct drm_i915_private
> *i915,
> mutex_unlock(&i915->mm.stolen_lock);
> }
>
> -static bool valid_stolen_size(struct resource *dsm)
> +static bool valid_stolen_size(struct drm_i915_private *i915, struct resource
> *dsm)
> {
> - return dsm->start != 0 && dsm->end > dsm->start;
> + return (dsm->start != 0 || HAS_BAR2_SMEM_STOLEN(i915)) && dsm->end >
> dsm->start;
> }
>
> static int adjust_stolen(struct drm_i915_private *i915,
> @@ -88,7 +88,7 @@ static int adjust_stolen(struct drm_i915_private *i915,
> struct i915_ggtt *ggtt = to_gt(i915)->ggtt;
> struct intel_uncore *uncore = ggtt->vm.gt->uncore;
>
> - if (!valid_stolen_size(dsm))
> + if (!valid_stolen_size(i915, dsm))
> return -EINVAL;
>
> /*
> @@ -135,7 +135,7 @@ static int adjust_stolen(struct drm_i915_private *i915,
> }
> }
>
> - if (!valid_stolen_size(dsm))
> + if (!valid_stolen_size(i915, dsm))
> return -EINVAL;
>
> return 0;
> @@ -148,9 +148,10 @@ static int request_smem_stolen(struct drm_i915_private
> *i915,
>
> /*
>* With stolen lmem, we don't need to request system memory for the
> - * address range since it's local to the gpu.
> + * address range since it's local to the gpu and in some IGFX devices
> + * BAR2 is exposed as stolen
>*/
> - if (HAS_LMEM(i915))
> + if (HAS_LMEM(i915) || HAS_BAR2_SMEM_STOLEN(i915))
> return 0;
>
> /*
> @@ -385,8 +386,6 @@ static void icl_get_stolen_reserved(struct
> drm_i915_private *i915,
>
> drm_dbg(&i915->drm, "GEN6_STOLEN_RESERVED = 0x%016llx\n", reg_val);
>
> - *base = reg_val & GEN11_STOLEN_RESERVED_ADDR_MASK;
> -
> switch (reg_val & GEN8_STOLEN_RESERVED_SIZE_MASK) {
> case GEN8_STOLEN_RESERVED_1M:
> *size = 1024 * 1024;
> @@ -404,6 +403,12 @@ static void icl_get_stolen_reserved(struct
> drm_i915_private *i915,
> *size = 8 * 1024 * 1024;
> MISSING_CASE(reg_val & GEN8_STOLEN_RESERVED_SIZE_MASK);
> }
> +
> + if (HAS_BAR2_SMEM_STOLEN(i915))
> + /* the base is initialized to stolen top so subtract size to
> get base */
> + *base -= *size;
> + else
> + *base = reg_val & GEN11_STOLEN_RESERVED_ADDR_MASK;
> }
>
> /*
> @@ -833,6 +838,34 @@ static const struct intel_memory_region_ops
> i915_region_stolen_lmem_ops = {
> .init_object = _i915_gem_object_stolen_init,
> };
>
> +static int get_mtl_gms_size(struct intel_uncore *uncore)
Please always use platform TLA as prefix, not something in the middle of
the function name.
> +{
> + u16 ggc, gms;
> +
> + ggc = intel_uncore_read16(uncore, _MMIO(0x108040));
Please define the registers.
> +
> + /* check GGMS, should be fixed 0x3 (8MB) */
> + if ((ggc & 0xc0) != 0xc0)
> + return -EIO;
> +
> + /* return valid GMS value, -EIO if invalid */
> + gms = ggc >> 8;
> + switch (gms) {
> + case 0x0 ... 0x10:
> + return gms * 32;
> + case 0x20:
> + return 1024;
> + case 0x30:
> + return 1536;
> + case 0x40:
> + return 2048;
> + case 0xf0 ... 0xfe:
> + return (gms - 0xf0 + 1) * 4;
> + default:
> + return -EIO;
> + }
> +}
> +
> struct intel_memory_region *
> i915_gem_stolen_lmem_setup(struct drm_i915_private *i915, u16 type,
> u16 instance)
> @@ -843,6 +876,7 @@ i915_gem_stolen_lmem_setup(struct drm_i915_private *i915,
> u16 type,
> struct intel_memory_region *mem;
>
[Intel-gfx] [PATCH v2] drm/i915/mtl: enable local stolen memory
As an integrated GPU, MTL does not have local memory and
HAS_LMEM() returns false. However the platform's stolen memory
is presented via BAR2 (i.e., the BAR we traditionally consider
to be the LMEM BAR) and should be managed by the driver the same
way that local memory is on dgpu platforms (which includes
setting the "lmem" bit on page table entries). We use the term
"local stolen memory" to refer to this model.
v2:
1. dropped is_dsm_invalid, updated valid_stolen_size check from Lucas
(Jani, Lucas)
2. drop lmembar_is_igpu_stolen
3. revert to referring GFXMEM_BAR as GEN12_LMEM_BAR (Lucas)
Cc: Matt Roper
Cc: Lucas De Marchi
Signed-off-by: CQ Tang
Signed-off-by: Aravind Iddamsetty
Original-author: CQ Tang
---
drivers/gpu/drm/i915/gem/i915_gem_stolen.c | 88 ++
drivers/gpu/drm/i915/gt/intel_ggtt.c | 2 +-
drivers/gpu/drm/i915/i915_drv.h| 3 +
3 files changed, 76 insertions(+), 17 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
b/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
index c5a4035c99cd..582c4d7d2a9a 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
@@ -77,9 +77,9 @@ void i915_gem_stolen_remove_node(struct drm_i915_private
*i915,
mutex_unlock(&i915->mm.stolen_lock);
}
-static bool valid_stolen_size(struct resource *dsm)
+static bool valid_stolen_size(struct drm_i915_private *i915, struct resource
*dsm)
{
- return dsm->start != 0 && dsm->end > dsm->start;
+ return (dsm->start != 0 || HAS_BAR2_SMEM_STOLEN(i915)) && dsm->end >
dsm->start;
}
static int adjust_stolen(struct drm_i915_private *i915,
@@ -88,7 +88,7 @@ static int adjust_stolen(struct drm_i915_private *i915,
struct i915_ggtt *ggtt = to_gt(i915)->ggtt;
struct intel_uncore *uncore = ggtt->vm.gt->uncore;
- if (!valid_stolen_size(dsm))
+ if (!valid_stolen_size(i915, dsm))
return -EINVAL;
/*
@@ -135,7 +135,7 @@ static int adjust_stolen(struct drm_i915_private *i915,
}
}
- if (!valid_stolen_size(dsm))
+ if (!valid_stolen_size(i915, dsm))
return -EINVAL;
return 0;
@@ -148,9 +148,10 @@ static int request_smem_stolen(struct drm_i915_private
*i915,
/*
* With stolen lmem, we don't need to request system memory for the
-* address range since it's local to the gpu.
+* address range since it's local to the gpu and in some IGFX devices
+* BAR2 is exposed as stolen
*/
- if (HAS_LMEM(i915))
+ if (HAS_LMEM(i915) || HAS_BAR2_SMEM_STOLEN(i915))
return 0;
/*
@@ -385,8 +386,6 @@ static void icl_get_stolen_reserved(struct drm_i915_private
*i915,
drm_dbg(&i915->drm, "GEN6_STOLEN_RESERVED = 0x%016llx\n", reg_val);
- *base = reg_val & GEN11_STOLEN_RESERVED_ADDR_MASK;
-
switch (reg_val & GEN8_STOLEN_RESERVED_SIZE_MASK) {
case GEN8_STOLEN_RESERVED_1M:
*size = 1024 * 1024;
@@ -404,6 +403,12 @@ static void icl_get_stolen_reserved(struct
drm_i915_private *i915,
*size = 8 * 1024 * 1024;
MISSING_CASE(reg_val & GEN8_STOLEN_RESERVED_SIZE_MASK);
}
+
+ if (HAS_BAR2_SMEM_STOLEN(i915))
+ /* the base is initialized to stolen top so subtract size to
get base */
+ *base -= *size;
+ else
+ *base = reg_val & GEN11_STOLEN_RESERVED_ADDR_MASK;
}
/*
@@ -833,6 +838,34 @@ static const struct intel_memory_region_ops
i915_region_stolen_lmem_ops = {
.init_object = _i915_gem_object_stolen_init,
};
+static int get_mtl_gms_size(struct intel_uncore *uncore)
+{
+ u16 ggc, gms;
+
+ ggc = intel_uncore_read16(uncore, _MMIO(0x108040));
+
+ /* check GGMS, should be fixed 0x3 (8MB) */
+ if ((ggc & 0xc0) != 0xc0)
+ return -EIO;
+
+ /* return valid GMS value, -EIO if invalid */
+ gms = ggc >> 8;
+ switch (gms) {
+ case 0x0 ... 0x10:
+ return gms * 32;
+ case 0x20:
+ return 1024;
+ case 0x30:
+ return 1536;
+ case 0x40:
+ return 2048;
+ case 0xf0 ... 0xfe:
+ return (gms - 0xf0 + 1) * 4;
+ default:
+ return -EIO;
+ }
+}
+
struct intel_memory_region *
i915_gem_stolen_lmem_setup(struct drm_i915_private *i915, u16 type,
u16 instance)
@@ -843,6 +876,7 @@ i915_gem_stolen_lmem_setup(struct drm_i915_private *i915,
u16 type,
struct intel_memory_region *mem;
resource_size_t io_start, io_size;
resource_size_t min_page_size;
+ int ret;
if (WARN_ON_ONCE(instance))
return ERR_PTR(-ENODEV);
@@ -850,12 +884,8 @@ i915_gem_stolen_lmem_setup(struct drm_i915_private *i915,
u16 type,
if (!i915_pci_resource_valid(pdev, GEN12_LMEM_BAR))
