[PATCH v6 5/5] drm/xe: Refactor default device atomic settings

[Nirmoy Das <nirmoy.das@intel.com>]

The default behavior of device atomics depends on the
VM type and buffer allocation types. Device atomics are
expected to function with all types of allocations for
traditional applications/APIs. Additionally, in compute/SVM
API scenarios with fault mode or LR mode VMs, device atomics
must work with single-region allocations. In all other cases
device atomics should be disabled by default also on platforms
where we know device atomics doesn't on work on particular
allocations types.

v3: fault mode requires LR mode so only check for LR mode
    to determine compute API(Jose).
    Handle SMEM+LMEM BO's migration to LMEM where device
    atomics is expected to work. (Brian).
v2: Fix platform checks to correct atomics behaviour on PVC.

Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
Acked-by: Michal Mrozek <michal.mrozek@intel.com>
---
 drivers/gpu/drm/xe/xe_pt.c | 37 ++++++++++++++++++++++++++++++++++---
 drivers/gpu/drm/xe/xe_vm.c |  2 +-
 2 files changed, 35 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
index 8d3765d3351e..87975e45622a 100644
--- a/drivers/gpu/drm/xe/xe_pt.c
+++ b/drivers/gpu/drm/xe/xe_pt.c
@@ -619,9 +619,40 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,
 	struct xe_pt *pt = xe_vma_vm(vma)->pt_root[tile->id];
 	int ret;
 
-	if ((vma->gpuva.flags & XE_VMA_ATOMIC_PTE_BIT) &&
-	    (is_devmem || !IS_DGFX(xe)))
-		xe_walk.default_pte |= XE_USM_PPGTT_PTE_AE;
+	/**
+	 * Default atomic expectations for different allocation scenarios are as follows:
+	 *
+	 * 1. Traditional API: When the VM is not in LR mode:
+	 *    - Device atomics are expected to function with all allocations.
+	 *
+	 * 2. Compute/SVM API: When the VM is in LR mode:
+	 *    - Device atomics are the default behavior when the bo is placed in a single region.
+	 *    - In all other cases device atomics will be disabled with AE=0 until an application
+	 *      request differently using a ioctl like madvise.
+	 */
+	if (vma->gpuva.flags & XE_VMA_ATOMIC_PTE_BIT) {
+		if (xe_vm_in_lr_mode(xe_vma_vm(vma))) {
+			if (bo && xe_bo_has_single_placement(bo))
+				xe_walk.default_pte |= XE_USM_PPGTT_PTE_AE;
+			/**
+			 * If a SMEM+LMEM allocation is backed by SMEM, a device
+			 * atomics will cause a gpu page fault and which then
+			 * gets migrated to LMEM, bind such allocations with
+			 * device atomics enabled.
+			 */
+			else if (is_devmem && !xe_bo_has_single_placement(bo))
+				xe_walk.default_pte |= XE_USM_PPGTT_PTE_AE;
+		} else {
+			xe_walk.default_pte |= XE_USM_PPGTT_PTE_AE;
+		}
+
+		/**
+		 * Unset AE if the platform(PVC) doesn't support it on an
+		 * allocation
+		 */
+		if (!xe->info.has_device_atomics_on_smem && !is_devmem)
+			xe_walk.default_pte &= ~XE_USM_PPGTT_PTE_AE;
+	}
 
 	if (is_devmem) {
 		xe_walk.default_pte |= XE_PPGTT_PTE_DM;
diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
index f1357e2a3b10..d17192c8b7de 100644
--- a/drivers/gpu/drm/xe/xe_vm.c
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -888,7 +888,7 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm,
 	for_each_tile(tile, vm->xe, id)
 		vma->tile_mask |= 0x1 << id;
 
-	if (GRAPHICS_VER(vm->xe) >= 20 || vm->xe->info.platform == XE_PVC)
+	if (vm->xe->info.has_atomic_enable_pte_bit)
 		vma->gpuva.flags |= XE_VMA_ATOMIC_PTE_BIT;
 
 	vma->pat_index = pat_index;
-- 
2.42.0