On 2022-02-07 at 20:52:33 +0530, Hellstrom, Thomas wrote:
On Mon, 2022-02-07 at 20:44 +0530, Ramalingam C wrote:
On 2022-02-07 at 20:25:42 +0530, Hellstrom, Thomas wrote:
Hi, Ram,
A couple of quick questions before starting a more detailed review:
- Does this also support migrating of compressed data LMEM->LMEM?
What-about inter-tile?
Honestly this series mainly facused on eviction of lmem into smem and restoration of same.
To cover migration, we need to handle this differently from eviction. Becasue when we migrate the compressed content we need to be able to use that from that new placement. can't keep the ccs data separately.
Migration of lmem->smem needs decompression incorportated. Migration of lmem_m->lmem_n needs to maintain the compressed/decompressed state as it is.
So we need to pass the information upto emit_copy to differentiate eviction and migration
If you dont have objection I would like to take the migration once we have the eviction of lmem in place.
Sure NP. I was thinking that in the final solution we might also need to think about the possibility that we might evict to another lmem region, although I figure that won't be enabled until we support multi- tile.
Yes we need it for multi tile enablement of XeHPSDV.
- Do we need to block faulting of compressed data in the fault
handler as a follow-up patch?
In case of evicted compressed data we dont need to treat it differently from the evicted normal data. So I dont think this needs a special treatment. Sorry if i dont understand your question.
My question wasn't directly related to eviction actually, but does user-space need to have mmap access to compressed data? If not, block it?
We shouldn't mmap the ccs data. As per my understanding we should be mmaping the obj size which doesn't count the ttm_tt inflated size.
I will verify this part and if needed will prepare a change to exclude increased pages from mmap range.
Ram.
Thanks, Thomas
Ram
/Thomas
On Mon, 2022-02-07 at 15:07 +0530, Ramalingam C wrote:
When we are swapping out the local memory obj on flat-ccs capable platform, we need to capture the ccs data too along with main meory and we need to restore it when we are swapping in the content.
Extracting and restoring the CCS data is done through a special cmd called XY_CTRL_SURF_COPY_BLT
Signed-off-by: Ramalingam C ramalingam.c@intel.com
drivers/gpu/drm/i915/gt/intel_migrate.c | 283 +++++++++++++-----
-- 1 file changed, 155 insertions(+), 128 deletions(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_migrate.c b/drivers/gpu/drm/i915/gt/intel_migrate.c index 5bdab0b3c735..e60ae6ff1847 100644 --- a/drivers/gpu/drm/i915/gt/intel_migrate.c +++ b/drivers/gpu/drm/i915/gt/intel_migrate.c @@ -449,14 +449,146 @@ static bool wa_1209644611_applies(int ver, u32 size) return height % 4 == 3 && height <= 8; }
+/**
- DOC: Flat-CCS - Memory compression for Local memory
- On Xe-HP and later devices, we use dedicated compression
control state (CCS)
- stored in local memory for each surface, to support the 3D
and media
- compression formats.
- The memory required for the CCS of the entire local memory is
1/256 of the
- local memory size. So before the kernel boot, the required
memory is reserved
- for the CCS data and a secure register will be programmed
with the CCS base
- address.
- Flat CCS data needs to be cleared when a lmem object is
allocated.
- And CCS data can be copied in and out of CCS region through
- XY_CTRL_SURF_COPY_BLT. CPU can't access the CCS data
directly.
- When we exaust the lmem, if the object's placements support
smem, then we can
- directly decompress the compressed lmem object into smem and
start using it
- from smem itself.
- But when we need to swapout the compressed lmem object into a
smem region
- though objects' placement doesn't support smem, then we copy
the lmem content
- as it is into smem region along with ccs data (using
XY_CTRL_SURF_COPY_BLT).
- When the object is referred, lmem content will be swaped in
along with
- restoration of the CCS data (using XY_CTRL_SURF_COPY_BLT) at
corresponding
- location.
- Flat-CCS Modifiers for different compression formats
- I915_FORMAT_MOD_F_TILED_DG2_RC_CCS - used to indicate the
buffers of Flat CCS
- render compression formats. Though the general layout is same
as
- I915_FORMAT_MOD_Y_TILED_GEN12_RC_CCS, new hashing/compression
algorithm is
- used. Render compression uses 128 byte compression blocks
- I915_FORMAT_MOD_F_TILED_DG2_MC_CCS -used to indicate the
buffers of Flat CCS
- media compression formats. Though the general layout is same
as
- I915_FORMAT_MOD_Y_TILED_GEN12_MC_CCS, new hashing/compression
algorithm is
- used. Media compression uses 256 byte compression blocks.
- I915_FORMAT_MOD_F_TILED_DG2_RC_CCS_CC - used to indicate the
buffers of Flat
- CCS clear color render compression formats. Unified
compression format for
- clear color render compression. The genral layout is a tiled
layout using
- 4Kb tiles i.e Tile4 layout.
- */
+static inline u32 *i915_flush_dw(u32 *cmd, u64 dst, u32 flags) +{
/* Mask the 3 LSB to use the PPGTT address space */*cmd++ = MI_FLUSH_DW | flags;*cmd++ = lower_32_bits(dst);*cmd++ = upper_32_bits(dst);return cmd;+}
+static u32 calc_ctrl_surf_instr_size(struct drm_i915_private *i915, int size) +{
u32 num_cmds, num_blks, total_size;if (!GET_CCS_SIZE(i915, size))return 0;/** XY_CTRL_SURF_COPY_BLT transfers CCS in 256 byte* blocks. one XY_CTRL_SURF_COPY_BLT command can* trnasfer upto 1024 blocks.*/num_blks = GET_CCS_SIZE(i915, size);num_cmds = (num_blks + (NUM_CCS_BLKS_PER_XFER - 1)) >>10;
total_size = (XY_CTRL_SURF_INSTR_SIZE) * num_cmds;/** We need to add a flush before and after* XY_CTRL_SURF_COPY_BLT*/total_size += 2 * MI_FLUSH_DW_SIZE;return total_size;+}
+static u32 *_i915_ctrl_surf_copy_blt(u32 *cmd, u64 src_addr, u64 dst_addr,
u8 src_mem_access, u8dst_mem_access,
int src_mocs, int dst_mocs,u16 num_ccs_blocks)+{
int i = num_ccs_blocks;/** The XY_CTRL_SURF_COPY_BLT instruction is used to copythe CCS
* data in and out of the CCS region.** We can copy at most 1024 blocks of 256 bytes using one* XY_CTRL_SURF_COPY_BLT instruction.** In case we need to copy more than 1024 blocks, we needto add
* another instruction to the same batch buffer.** 1024 blocks of 256 bytes of CCS represent a total256KB of CCS.
** 256 KB of CCS represents 256 * 256 KB = 64 MB of LMEM.*/do {/** We use logical AND with 1023 since the sizefield
* takes values which is in the range of 0 - 1023*/*cmd++ = ((XY_CTRL_SURF_COPY_BLT) |(src_mem_access <<SRC_ACCESS_TYPE_SHIFT) |
(dst_mem_access <<DST_ACCESS_TYPE_SHIFT) |
(((i - 1) & 1023) << CCS_SIZE_SHIFT));*cmd++ = lower_32_bits(src_addr);*cmd++ = ((upper_32_bits(src_addr) & 0xFFFF) |(src_mocs << XY_CTRL_SURF_MOCS_SHIFT));*cmd++ = lower_32_bits(dst_addr);*cmd++ = ((upper_32_bits(dst_addr) & 0xFFFF) |(dst_mocs << XY_CTRL_SURF_MOCS_SHIFT));src_addr += SZ_64M;dst_addr += SZ_64M;i -= NUM_CCS_BLKS_PER_XFER;} while (i > 0);return cmd;+}
static int emit_copy(struct i915_request *rq,
u32 dst_offset, u32 src_offset, int size)
bool dst_is_lmem, u32 dst_offset,bool src_is_lmem, u32 src_offset, int size){
struct drm_i915_private *i915 = rq->engine->i915; const int ver = GRAPHICS_VER(rq->engine->i915); u32 instance = rq->engine->instance;u32 num_ccs_blks, ccs_ring_size;u8 src_access, dst_access; u32 *cs;
cs = intel_ring_begin(rq, ver >= 8 ? 10 : 6);
ccs_ring_size = ((src_is_lmem || dst_is_lmem) &&HAS_FLAT_CCS(i915)) ?
calc_ctrl_surf_instr_size(i915, size) :0;
cs = intel_ring_begin(rq, ver >= 8 ? 10 + ccs_ring_size :6); if (IS_ERR(cs)) return PTR_ERR(cs);
@@ -492,6 +624,25 @@ static int emit_copy(struct i915_request *rq, *cs++ = src_offset; }
if (ccs_ring_size) {/* TODO: Migration needs to be handled withresolve of compressed data */
num_ccs_blks = (GET_CCS_SIZE(i915, size) +NUM_CCS_BYTES_PER_BLOCK - 1) >>8;
src_access = !src_is_lmem && dst_is_lmem;dst_access = !src_access;if (src_access) /* Swapin of compressed data */src_offset += size;elsedst_offset += size;cs = _i915_ctrl_surf_copy_blt(cs, src_offset,dst_offset,
src_access,dst_access,
1, 1,num_ccs_blks);
cs = i915_flush_dw(cs, dst_offset, MI_FLUSH_LLC |MI_FLUSH_CCS);
}intel_ring_advance(rq, cs); return 0;} @@ -578,7 +729,8 @@ intel_context_migrate_copy(struct intel_context *ce, if (err) goto out_rq;
err = emit_copy(rq, dst_offset, src_offset, len);
err = emit_copy(rq, dst_is_lmem, dst_offset,src_is_lmem, src_offset, len); /* Arbitration is re-enabled between requests. */out_rq: @@ -596,131 +748,6 @@ intel_context_migrate_copy(struct intel_context *ce, return err; }
-/**
- DOC: Flat-CCS - Memory compression for Local memory
- On Xe-HP and later devices, we use dedicated compression
control state (CCS)
- stored in local memory for each surface, to support the 3D
and media
- compression formats.
- The memory required for the CCS of the entire local memory is
1/256 of the
- local memory size. So before the kernel boot, the required
memory is reserved
- for the CCS data and a secure register will be programmed
with the CCS base
- address.
- Flat CCS data needs to be cleared when a lmem object is
allocated.
- And CCS data can be copied in and out of CCS region through
- XY_CTRL_SURF_COPY_BLT. CPU can't access the CCS data
directly.
- When we exaust the lmem, if the object's placements support
smem, then we can
- directly decompress the compressed lmem object into smem and
start using it
- from smem itself.
- But when we need to swapout the compressed lmem object into a
smem region
- though objects' placement doesn't support smem, then we copy
the lmem content
- as it is into smem region along with ccs data (using
XY_CTRL_SURF_COPY_BLT).
- When the object is referred, lmem content will be swaped in
along with
- restoration of the CCS data (using XY_CTRL_SURF_COPY_BLT) at
corresponding
- location.
- Flat-CCS Modifiers for different compression formats
- I915_FORMAT_MOD_F_TILED_DG2_RC_CCS - used to indicate the
buffers of Flat CCS
- render compression formats. Though the general layout is same
as
- I915_FORMAT_MOD_Y_TILED_GEN12_RC_CCS, new hashing/compression
algorithm is
- used. Render compression uses 128 byte compression blocks
- I915_FORMAT_MOD_F_TILED_DG2_MC_CCS -used to indicate the
buffers of Flat CCS
- media compression formats. Though the general layout is same
as
- I915_FORMAT_MOD_Y_TILED_GEN12_MC_CCS, new hashing/compression
algorithm is
- used. Media compression uses 256 byte compression blocks.
- I915_FORMAT_MOD_F_TILED_DG2_RC_CCS_CC - used to indicate the
buffers of Flat
- CCS clear color render compression formats. Unified
compression format for
- clear color render compression. The genral layout is a tiled
layout using
- 4Kb tiles i.e Tile4 layout.
- */
-static inline u32 *i915_flush_dw(u32 *cmd, u64 dst, u32 flags) -{
/* Mask the 3 LSB to use the PPGTT address space */*cmd++ = MI_FLUSH_DW | flags;*cmd++ = lower_32_bits(dst);*cmd++ = upper_32_bits(dst);return cmd;-}
-static u32 calc_ctrl_surf_instr_size(struct drm_i915_private *i915, int size) -{
u32 num_cmds, num_blks, total_size;if (!GET_CCS_SIZE(i915, size))return 0;/** XY_CTRL_SURF_COPY_BLT transfers CCS in 256 byte* blocks. one XY_CTRL_SURF_COPY_BLT command can* trnasfer upto 1024 blocks.*/num_blks = GET_CCS_SIZE(i915, size);num_cmds = (num_blks + (NUM_CCS_BLKS_PER_XFER - 1)) >>10;
total_size = (XY_CTRL_SURF_INSTR_SIZE) * num_cmds;/** We need to add a flush before and after* XY_CTRL_SURF_COPY_BLT*/total_size += 2 * MI_FLUSH_DW_SIZE;return total_size;-}
-static u32 *_i915_ctrl_surf_copy_blt(u32 *cmd, u64 src_addr, u64 dst_addr,
u8 src_mem_access, u8dst_mem_access,
int src_mocs, int dst_mocs,u16 num_ccs_blocks)-{
int i = num_ccs_blocks;/** The XY_CTRL_SURF_COPY_BLT instruction is used to copythe CCS
* data in and out of the CCS region.** We can copy at most 1024 blocks of 256 bytes using one* XY_CTRL_SURF_COPY_BLT instruction.** In case we need to copy more than 1024 blocks, we needto add
* another instruction to the same batch buffer.** 1024 blocks of 256 bytes of CCS represent a total256KB of CCS.
** 256 KB of CCS represents 256 * 256 KB = 64 MB of LMEM.*/do {/** We use logical AND with 1023 since the sizefield
* takes values which is in the range of 0 - 1023*/*cmd++ = ((XY_CTRL_SURF_COPY_BLT) |(src_mem_access <<SRC_ACCESS_TYPE_SHIFT) |
(dst_mem_access <<DST_ACCESS_TYPE_SHIFT) |
(((i - 1) & 1023) << CCS_SIZE_SHIFT));*cmd++ = lower_32_bits(src_addr);*cmd++ = ((upper_32_bits(src_addr) & 0xFFFF) |(src_mocs << XY_CTRL_SURF_MOCS_SHIFT));*cmd++ = lower_32_bits(dst_addr);*cmd++ = ((upper_32_bits(dst_addr) & 0xFFFF) |(dst_mocs << XY_CTRL_SURF_MOCS_SHIFT));src_addr += SZ_64M;dst_addr += SZ_64M;i -= NUM_CCS_BLKS_PER_XFER;} while (i > 0);return cmd;-}
static int emit_clear(struct i915_request *rq, u64 offset, int size,