drm/xe/migrate: fix MI_ARB_ON_OFF usage
authorMatthew Auld <matthew.auld@intel.com>
Wed, 25 Oct 2023 17:39:39 +0000 (18:39 +0100)
committerRodrigo Vivi <rodrigo.vivi@intel.com>
Thu, 21 Dec 2023 16:43:31 +0000 (11:43 -0500)
Spec says: "This is a privileged command; it will not be effective (will
be converted to a no-op) if executed from within a non-privileged batch
buffer." However here it looks like we are just emitting it inside some
bb which was jumped to via the ppGTT, which should be considered
a non-privileged address space.

It looks like we just need some way of preventing things like the
emit_pte() and later copy/clear being preempted in-between so rather
just emit directly in the ring for migration jobs.

Bspec: 45716
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
drivers/gpu/drm/xe/xe_migrate.c
drivers/gpu/drm/xe/xe_ring_ops.c

index 731beb622fe876224ab9b5856452a1ebf169e370..67b71244b1f2ec629aa5e0afcf20c67f0d689df8 100644 (file)
@@ -406,12 +406,6 @@ struct xe_migrate *xe_migrate_init(struct xe_tile *tile)
        return m;
 }
 
-static void emit_arb_clear(struct xe_bb *bb)
-{
-       /* 1 dword */
-       bb->cs[bb->len++] = MI_ARB_ON_OFF | MI_ARB_DISABLE;
-}
-
 static u64 xe_migrate_res_sizes(struct xe_res_cursor *cur)
 {
        /*
@@ -745,10 +739,6 @@ struct dma_fence *xe_migrate_copy(struct xe_migrate *m,
                        goto err_sync;
                }
 
-               /* Preemption is enabled again by the ring ops. */
-               if (!src_is_vram || !dst_is_vram)
-                       emit_arb_clear(bb);
-
                if (!src_is_vram)
                        emit_pte(m, bb, src_L0_pt, src_is_vram, &src_it, src_L0,
                                 src_bo);
@@ -994,7 +984,6 @@ struct dma_fence *xe_migrate_clear(struct xe_migrate *m,
 
                /* Preemption is enabled again by the ring ops. */
                if (!clear_vram) {
-                       emit_arb_clear(bb);
                        emit_pte(m, bb, clear_L0_pt, clear_vram, &src_it, clear_L0,
                                 bo);
                } else {
@@ -1285,9 +1274,6 @@ xe_migrate_update_pgtables(struct xe_migrate *m,
                                VM_SA_UPDATE_UNIT_SIZE;
                }
 
-               /* Preemption is enabled again by the ring ops. */
-               emit_arb_clear(bb);
-
                /* Map our PT's to gtt */
                bb->cs[bb->len++] = MI_STORE_DATA_IMM | MI_SDI_NUM_QW(num_updates);
                bb->cs[bb->len++] = ppgtt_ofs * XE_PAGE_SIZE + page_ofs;
@@ -1316,8 +1302,6 @@ xe_migrate_update_pgtables(struct xe_migrate *m,
                bb->cs[bb->len++] = MI_BATCH_BUFFER_END;
                update_idx = bb->len;
 
-               /* Preemption is enabled again by the ring ops. */
-               emit_arb_clear(bb);
                for (i = 0; i < num_updates; i++)
                        write_pgtable(tile, bb, 0, &updates[i], pt_update);
        }
index 58676f4b989f363cd9961a105117695ec5c6da9c..59e0aa2d6a4c5214165aec3a476af11db21f21b6 100644 (file)
@@ -355,6 +355,8 @@ static void emit_migration_job_gen12(struct xe_sched_job *job,
        i = emit_store_imm_ggtt(xe_lrc_start_seqno_ggtt_addr(lrc),
                                seqno, dw, i);
 
+       dw[i++] = MI_ARB_ON_OFF | MI_ARB_DISABLE; /* Enabled again below */
+
        i = emit_bb_start(job->batch_addr[0], BIT(8), dw, i);
 
        /* XXX: Do we need this? Leaving for now. */