[Intel-gfx] [PATCH v5 3/9] drm/i915/gt: Clear compress metadata for Flat-ccs objects
Thomas Hellström (Intel)
thomas_os at shipmail.org
Thu Mar 24 16:14:53 UTC 2022
Hi, Ram
On 3/21/22 23:44, Ramalingam C wrote:
> Xe-HP and latest devices support Flat CCS which reserved a portion of
> the device memory to store compression metadata, during the clearing of
> device memory buffer object we also need to clear the associated
> CCS buffer.
>
> XY_CTRL_SURF_COPY_BLT is a BLT cmd used for reading and writing the
> ccs surface of a lmem memory. So on Flat-CCS capable platform we use
> XY_CTRL_SURF_COPY_BLT to clear the CCS meta data.
>
> v2: Fixed issues with platform naming [Lucas]
> v3: Rebased [Ram]
> Used the round_up funcs [Bob]
> v4: Fixed ccs blk calculation [Ram]
> Added Kdoc on flat-ccs.
> v5: GENMASK is used [Matt]
> mocs fix [Matt]
> Comments Fix [Matt]
> Flush address programming [Ram]
> v6: FLUSH_DW is fixed
> Few coding style fix
> v7: Adopting the XY_FAST_COLOR_BLT (Thomas]
> v8: XY_CTRL_SURF_COPY_BLT for ccs clearing.
> v9: emit_copy_ccs is used.
>
> Signed-off-by: Ramalingam C <ramalingam.c at intel.com>
> Signed-off-by: Ayaz A Siddiqui <ayaz.siddiqui at intel.com>
> ---
> drivers/gpu/drm/i915/gt/intel_gpu_commands.h | 15 ++
> drivers/gpu/drm/i915/gt/intel_migrate.c | 164 ++++++++++++++++++-
> 2 files changed, 175 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gt/intel_gpu_commands.h b/drivers/gpu/drm/i915/gt/intel_gpu_commands.h
> index 925e55b6a94f..6b4eb7927ec7 100644
> --- a/drivers/gpu/drm/i915/gt/intel_gpu_commands.h
> +++ b/drivers/gpu/drm/i915/gt/intel_gpu_commands.h
> @@ -153,8 +153,10 @@
> #define MI_FLUSH_DW_PROTECTED_MEM_EN (1 << 22)
> #define MI_FLUSH_DW_STORE_INDEX (1<<21)
> #define MI_INVALIDATE_TLB (1<<18)
> +#define MI_FLUSH_DW_CCS (1<<16)
> #define MI_FLUSH_DW_OP_STOREDW (1<<14)
> #define MI_FLUSH_DW_OP_MASK (3<<14)
> +#define MI_FLUSH_DW_LLC (1<<9)
> #define MI_FLUSH_DW_NOTIFY (1<<8)
> #define MI_INVALIDATE_BSD (1<<7)
> #define MI_FLUSH_DW_USE_GTT (1<<2)
> @@ -203,6 +205,19 @@
> #define GFX_OP_DRAWRECT_INFO ((0x3<<29)|(0x1d<<24)|(0x80<<16)|(0x3))
> #define GFX_OP_DRAWRECT_INFO_I965 ((0x7900<<16)|0x2)
>
> +#define XY_CTRL_SURF_INSTR_SIZE 5
> +#define MI_FLUSH_DW_SIZE 3
> +#define XY_CTRL_SURF_COPY_BLT ((2 << 29) | (0x48 << 22) | 3)
> +#define SRC_ACCESS_TYPE_SHIFT 21
> +#define DST_ACCESS_TYPE_SHIFT 20
> +#define CCS_SIZE_MASK GENMASK(17, 8)
> +#define XY_CTRL_SURF_MOCS_MASK GENMASK(31, 25)
> +#define NUM_CCS_BYTES_PER_BLOCK 256
> +#define NUM_BYTES_PER_CCS_BYTE 256
> +#define NUM_CCS_BLKS_PER_XFER 1024
> +#define INDIRECT_ACCESS 0
> +#define DIRECT_ACCESS 1
> +
> #define COLOR_BLT_CMD (2 << 29 | 0x40 << 22 | (5 - 2))
> #define XY_COLOR_BLT_CMD (2 << 29 | 0x50 << 22)
> #define XY_FAST_COLOR_BLT_CMD (2 << 29 | 0x44 << 22)
> diff --git a/drivers/gpu/drm/i915/gt/intel_migrate.c b/drivers/gpu/drm/i915/gt/intel_migrate.c
> index b656685a486d..39a5f8ae664d 100644
> --- a/drivers/gpu/drm/i915/gt/intel_migrate.c
> +++ b/drivers/gpu/drm/i915/gt/intel_migrate.c
> @@ -16,7 +16,8 @@ struct insert_pte_data {
> };
>
> #define CHUNK_SZ SZ_8M /* ~1ms at 8GiB/s preemption delay */
> -
> +#define GET_CCS_BYTES(i915, size) (HAS_FLAT_CCS(i915) ? \
> + DIV_ROUND_UP(size, NUM_BYTES_PER_CCS_BYTE) : 0)
> static bool engine_supports_migration(struct intel_engine_cs *engine)
> {
> if (!engine)
> @@ -467,6 +468,145 @@ 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 exhaust 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.
> + */
> +
> +static inline u32 *i915_flush_dw(u32 *cmd, u32 flags)
> +{
> + *cmd++ = MI_FLUSH_DW | flags;
> + *cmd++ = 0;
> + *cmd++ = 0;
> +
> + 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_BYTES(i915, size))
> + return 0;
> +
> + /*
> + * XY_CTRL_SURF_COPY_BLT transfers CCS in 256 byte
> + * blocks. one XY_CTRL_SURF_COPY_BLT command can
> + * transfer upto 1024 blocks.
> + */
> + num_blks = DIV_ROUND_UP(GET_CCS_BYTES(i915, size),
> + NUM_CCS_BYTES_PER_BLOCK);
> + num_cmds = DIV_ROUND_UP(num_blks, NUM_CCS_BLKS_PER_XFER);
> + total_size = XY_CTRL_SURF_INSTR_SIZE * num_cmds;
> +
> + /*
> + * Adding 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, u8 dst_mem_access,
> + int src_mocs, int dst_mocs,
> + u32 ccs_blocks)
> +{
> + /*
> + * The XY_CTRL_SURF_COPY_BLT instruction is used to copy the 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 need to add
> + * another instruction to the same batch buffer.
> + *
> + * 1024 blocks of 256 bytes of CCS represent a total 256KB of CCS.
> + *
> + * 256 KB of CCS represents 256 * 256 KB = 64 MB of LMEM.
> + */
> + do {
> + int blks_per_copy;
Considering that the main surface copy only ever is CHUNK_SZ,
We will never do more than one iteration here, right? Perhaps consider
merging with the below function, with a check that blks_per_copy never
exceeds NUM_CCS_BLKS_PER_XFER.
> +
> + blks_per_copy = ccs_blocks >= NUM_CCS_BLKS_PER_XFER ?
> + NUM_CCS_BLKS_PER_XFER : ccs_blocks;
> + *cmd++ = XY_CTRL_SURF_COPY_BLT |
> + src_mem_access << SRC_ACCESS_TYPE_SHIFT |
> + dst_mem_access << DST_ACCESS_TYPE_SHIFT |
> + FIELD_PREP(CCS_SIZE_MASK, blks_per_copy - 1);
> + *cmd++ = lower_32_bits(src_addr);
> + *cmd++ = (upper_32_bits(src_addr) & 0xFFFF) |
> + FIELD_PREP(XY_CTRL_SURF_MOCS_MASK, src_mocs);
> + *cmd++ = lower_32_bits(dst_addr);
> + *cmd++ = (upper_32_bits(dst_addr) & 0xFFFF) |
> + FIELD_PREP(XY_CTRL_SURF_MOCS_MASK, dst_mocs);
> + src_addr += SZ_64M;
> + dst_addr += SZ_64M;
> + ccs_blocks -= blks_per_copy;
> + } while (ccs_blocks > 0);
> +
> + return cmd;
> +}
> +
> +static int emit_copy_ccs(struct i915_request *rq,
> + u32 dst_offset, u8 dst_access,
> + u32 src_offset, u8 src_access, int size)
> +{
> + struct drm_i915_private *i915 = rq->engine->i915;
> + int mocs = rq->engine->gt->mocs.uc_index << 1;
> + u32 num_ccs_blks, ccs_ring_size;
> + u32 *cs;
> +
> + src_offset += (u64)rq->engine->instance << 32;
> + dst_offset += (u64)rq->engine->instance << 32;
These are nops, considering that the offsets are 32-bit. Better to pass
the instance as a separate parameter to the above function, like is done
elsewhere.
> +
> + ccs_ring_size = calc_ctrl_surf_instr_size(i915, size);
> + WARN_ON(!ccs_ring_size);
> +
> + cs = intel_ring_begin(rq, round_up(ccs_ring_size, 2));
> + if (IS_ERR(cs))
> + return PTR_ERR(cs);
> +
> + num_ccs_blks = DIV_ROUND_UP(GET_CCS_BYTES(i915, size),
> + NUM_CCS_BYTES_PER_BLOCK);
> +
> + cs = i915_flush_dw(cs, MI_FLUSH_DW_LLC | MI_FLUSH_DW_CCS);
> + cs = _i915_ctrl_surf_copy_blt(cs, src_offset, dst_offset,
> + src_access, dst_access,
> + mocs, mocs, num_ccs_blks);
> + cs = i915_flush_dw(cs, MI_FLUSH_DW_LLC | MI_FLUSH_DW_CCS);
> + if (ccs_ring_size & 1)
> + *cs++ = MI_NOOP;
> +
> + intel_ring_advance(rq, cs);
> +
> + return 0;
> +}
> +
> static int emit_copy(struct i915_request *rq,
> u32 dst_offset, u32 src_offset, int size)
> {
> @@ -616,7 +756,7 @@ static int emit_clear(struct i915_request *rq, u64 offset, int size,
> u32 value, bool is_lmem)
> {
> struct drm_i915_private *i915 = rq->engine->i915;
> - int mocs = rq->engine->gt->mocs.uc_index << 1;
> + u32 mocs = rq->engine->gt->mocs.uc_index << 1;
> const int ver = GRAPHICS_VER(i915);
> int ring_sz;
> u32 *cs;
> @@ -632,7 +772,7 @@ static int emit_clear(struct i915_request *rq, u64 offset, int size,
> else
> ring_sz = 6;
>
> - cs = intel_ring_begin(rq, ring_sz);
> + cs = intel_ring_begin(rq, round_up(ring_sz, 2));
ring_sz is always even in this function? If not, then this is a bugfix
and should be in a separate patch.
> if (IS_ERR(cs))
> return PTR_ERR(cs);
>
> @@ -690,6 +830,7 @@ intel_context_migrate_clear(struct intel_context *ce,
> u32 value,
> struct i915_request **out)
> {
> + struct drm_i915_private *i915 = ce->engine->i915;
> struct sgt_dma it = sg_sgt(sg);
> struct i915_request *rq;
> u32 offset;
> @@ -701,7 +842,7 @@ intel_context_migrate_clear(struct intel_context *ce,
> GEM_BUG_ON(ce->ring->size < SZ_64K);
>
> offset = 0;
> - if (HAS_64K_PAGES(ce->engine->i915) && is_lmem)
> + if (HAS_64K_PAGES(i915) && is_lmem)
> offset = CHUNK_SZ;
>
> do {
> @@ -743,6 +884,21 @@ intel_context_migrate_clear(struct intel_context *ce,
> goto out_rq;
>
> err = emit_clear(rq, offset, len, value, is_lmem);
> + if (err)
> + goto out_rq;
> +
> + if (HAS_FLAT_CCS(i915) && is_lmem && !value) {
> + /*
> + * copy the content of memory into corresponding
> + * ccs surface
> + */
> + err = emit_copy_ccs(rq, offset, INDIRECT_ACCESS, offset,
> + DIRECT_ACCESS, len);
> + if (err)
> + goto out_rq;
> + }
> +
> + err = rq->engine->emit_flush(rq, EMIT_INVALIDATE);
>
> /* Arbitration is re-enabled between requests. */
> out_rq:
More information about the Intel-gfx
mailing list