[Intel-xe] [RFC v2 4/6] drm/xe/xe2: Updates on XY_CTRL_SURF_COPY_BLT

[Ghimiray, Himal Prasad]

On 24-11-2023 16:41, Thomas Hellström wrote:
> On Tue, 2023-11-21 at 15:39 +0530, Himal Prasad Ghimiray wrote:
>> - The XY_CTRL_SURF_COPY_BLT instruction operationg on ccs data
>> expects
>> size in pages of main memory for which CCS data should be copied.
>> - The bitfield representing copy size in XY_CTRL_SURF_COPY_BLT has
>> shifted one bit higher in the instruction.
>>
>> Cc: Thomas Hellström <thomas.hellstrom at linux.intel.com>
>> Signed-off-by: Himal Prasad Ghimiray
>> <himal.prasad.ghimiray at intel.com>
>> ---
>>   drivers/gpu/drm/xe/regs/xe_gpu_commands.h |  1 +
>>   drivers/gpu/drm/xe/xe_migrate.c           | 23 +++++++++++++++++----
>> --
>>   2 files changed, 18 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/xe/regs/xe_gpu_commands.h
>> b/drivers/gpu/drm/xe/regs/xe_gpu_commands.h
>> index 7f74592f99ce..47459841aa69 100644
>> --- a/drivers/gpu/drm/xe/regs/xe_gpu_commands.h
>> +++ b/drivers/gpu/drm/xe/regs/xe_gpu_commands.h
>> @@ -13,6 +13,7 @@
>>   #define   DST_ACCESS_TYPE_SHIFT                20
>>   #define   CCS_SIZE_MASK                        0x3FF
>>   #define   CCS_SIZE_SHIFT               8
>> +#define   XE2_CCS_SIZE_SHIFT           9
>>   #define   XY_CTRL_SURF_MOCS_MASK       GENMASK(31, 26)
>>   #define   XE2_XY_CTRL_SURF_MOCS_INDEX_MASK     GENMASK(31, 28)
>>   #define   NUM_CCS_BYTES_PER_BLOCK      256
>> diff --git a/drivers/gpu/drm/xe/xe_migrate.c
>> b/drivers/gpu/drm/xe/xe_migrate.c
>> index ed2f3f5109f3..06706fad67aa 100644
>> --- a/drivers/gpu/drm/xe/xe_migrate.c
>> +++ b/drivers/gpu/drm/xe/xe_migrate.c
>> @@ -523,21 +523,31 @@ static void emit_copy_ccs(struct xe_gt *gt,
>> struct xe_bb *bb,
>>          struct xe_device *xe = gt_to_xe(gt);
>>          u32 *cs = bb->cs + bb->len;
>>          u32 num_ccs_blks;
>> +       u32 num_ccs_pages;
>> +       u32 ccs_copy_size;
>>          u32 mocs;
>>   
>> -       num_ccs_blks = DIV_ROUND_UP(xe_device_ccs_bytes(gt_to_xe(gt),
>> size),
>> -                                   NUM_CCS_BYTES_PER_BLOCK);
>> -       xe_gt_assert(gt, num_ccs_blks <= NUM_CCS_BLKS_PER_XFER);
>> +       if (GRAPHICS_VERx100(xe) >= 2000) {
>> +               num_ccs_pages =
>> DIV_ROUND_UP(xe_device_ccs_bytes(gt_to_xe(gt), size),
>> +                                            XE_PAGE_SIZE);
>> +               xe_gt_assert(gt, num_ccs_pages <= 1024);
>>   
>> -       if (GRAPHICS_VERx100(xe) >= 2000)
>> +               ccs_copy_size = ((num_ccs_pages - 1) & CCS_SIZE_MASK)
>> << XE2_CCS_SIZE_SHIFT;
>>                  mocs = FIELD_PREP(XE2_XY_CTRL_SURF_MOCS_INDEX_MASK,
>> gt->mocs.uc_index);
>> -       else
>> +
>> +       } else {
>> +               num_ccs_blks =
>> DIV_ROUND_UP(xe_device_ccs_bytes(gt_to_xe(gt), size),
>> +                                           NUM_CCS_BYTES_PER_BLOCK);
>> +               xe_gt_assert(gt, num_ccs_blks <=
>> NUM_CCS_BLKS_PER_XFER);
>> +
>> +               ccs_copy_size = ((num_ccs_blks - 1) & CCS_SIZE_MASK)
>> << CCS_SIZE_SHIFT;
>>                  mocs = FIELD_PREP(XY_CTRL_SURF_MOCS_MASK, gt-
>>> mocs.uc_index);
>> +       }
>>   
>>          *cs++ = XY_CTRL_SURF_COPY_BLT |
>>                  (src_is_indirect ? 0x0 : 0x1) <<
>> SRC_ACCESS_TYPE_SHIFT |
>>                  (dst_is_indirect ? 0x0 : 0x1) <<
>> DST_ACCESS_TYPE_SHIFT |
>> -               ((num_ccs_blks - 1) & CCS_SIZE_MASK) <<
>> CCS_SIZE_SHIFT;
>> +               ccs_copy_size;
>>          *cs++ = lower_32_bits(src_ofs);
>>          *cs++ = upper_32_bits(src_ofs) | mocs;
>>          *cs++ = lower_32_bits(dst_ofs);
>> @@ -992,6 +1002,7 @@ struct dma_fence *xe_migrate_clear(struct
>> xe_migrate *m,
>>   
>>                  emit_clear(gt, bb, clear_L0_ofs, clear_L0,
>> XE_PAGE_SIZE,
>>                             clear_vram);
>> +
> Unrelated change. Please remove, Otherwise LGTM.
> Reviewed-by: Thomas Hellström <thomas.hellstrom at linux.intel.com>
Thanks for the review. Will remove unrelated change in next patch.
>
>
>>                  if (xe_device_has_flat_ccs(xe) && clear_vram) {
>>                          emit_copy_ccs(gt, bb, clear_L0_ofs, true,
>>                                        m->cleared_vram_ofs, false,
>> clear_L0);