[PATCH 4/7] drm/xe/lrc: Add table with LRC layout

[Lucas De Marchi]

On Fri, Jul 04, 2025 at 11:41:17AM +0100, Tvrtko Ursulin wrote:
>
>On 03/07/2025 23:41, Lucas De Marchi wrote:
>>Add a table to document the LRC's BO layout to make it easier to
>>visualize how each region stacks on top of each other.
>>
>>Signed-off-by: Lucas De Marchi <lucas.demarchi at intel.com>
>>---
>>  drivers/gpu/drm/xe/xe_lrc.c | 21 +++++++++++++++++++++
>>  1 file changed, 21 insertions(+)
>>
>>diff --git a/drivers/gpu/drm/xe/xe_lrc.c b/drivers/gpu/drm/xe/xe_lrc.c
>>index 7cde827ddbec9..75f4678cb090a 100644
>>--- a/drivers/gpu/drm/xe/xe_lrc.c
>>+++ b/drivers/gpu/drm/xe/xe_lrc.c
>>@@ -42,6 +42,27 @@
>>  #define LRC_INDIRECT_RING_STATE_SIZE		SZ_4K
>>  #define LRC_WA_BB_SIZE				SZ_4K
>>+/*
>>+ * Layout of the LRC and associated data allocated as
>>+ * lrc->bo:
>>+ *
>>+ *   Region                       Size
>>+ *  +============================+=================+ <- __xe_lrc_ring_offset()
>>+ *  | Ring                       | ring_size, see  |
>>+ *  |                            | xe_lrc_init()   |
>>+ *  +============================+=================+ <- __xe_lrc_pphwsp_offset()
>>+ *  | PPHWSP (includes SW state) | 4K              |
>>+ *  +----------------------------+-----------------+ <- __xe_lrc_ring_offset()
>>+ *  | Engine Context Image       | n * 4K, see     |
>>+ *  |                            | xe_gt_lrc_size()|
>>+ *  +----------------------------+-----------------+ <- __xe_lrc_indirect_ring_offset()
>>+ *  | Indirect Ring State Page   | 0 or 4k, see    |
>>+ *  |                            | xe_gt_lrc_size()|
>>+ *  +============================+=================+ <- __xe_lrc_wa_bb_offset()
>>+ *  | WA BB Per Ctx              | 4k              |
>>+ *  +============================+=================+ <- xe_bo_size(lrc->bo)
>>+ */
>
>I assume '---' separates parts which are considered engine lrc state, 
>while '===' is for other data which happens to be in the same BO. So 

yes

>for indirect ctx I will use '==='. Just unsure if indirect ring state 
>should in that case also be '==='? Looks like LRC image points to it's 
>ggtt address which would suggest it is equally separate.

correct, indirect state would be another page, separate by ===, since
it's not really part of the LRC, just kept in the same bo.

>
>>+
>>  static struct xe_device *
>>  lrc_to_xe(struct xe_lrc *lrc)
>>  {
>>
>
>Modulo the above:
>
>Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin at igalia.com>

thanks
Lucas De Marchi

>
>Regards,
>
>Tvrtko
>