return false;
}
-/* page_size = 2^(reclamation_size + XE_PTE_SHIFT) */
-#define COMPUTE_RECLAIM_ADDRESS_MASK(page_size) \
-({ \
- BUILD_BUG_ON(!__builtin_constant_p(page_size)); \
- ilog2(page_size) - XE_PTE_SHIFT; \
-})
-
static int generate_reclaim_entry(struct xe_tile *tile,
struct xe_page_reclaim_list *prl,
u64 pte, struct xe_pt *xe_child)
{
struct xe_gt *gt = tile->primary_gt;
struct xe_guc_page_reclaim_entry *reclaim_entries = prl->entries;
- u64 phys_addr = pte & XE_PTE_ADDR_MASK;
+ bool is_2m = xe_child->level == 1 && (pte & XE_PDE_PS_2M);
+ bool is_64k = xe_child->level == 0 && ((pte & XE_PTE_PS64) || xe_child->is_compact);
+ u32 page_shift = is_2m ? ilog2(SZ_2M) : is_64k ? ilog2(SZ_64K) : ilog2(SZ_4K);
+ /* Physical address bits start at page shift: 2M->[51:21], 64K->[51:16], 4K->[51:12] */
+ u64 phys_addr = pte & XE_PAGE_ADDR_MASK(page_shift);
+ /* Page address is relative to 4K page regardless of entry level */
u64 phys_page = phys_addr >> XE_PTE_SHIFT;
int num_entries = prl->num_entries;
- u32 reclamation_size;
+ u32 reclamation_size = page_shift - XE_PTE_SHIFT;
xe_tile_assert(tile, xe_child->level <= MAX_HUGEPTE_LEVEL);
xe_tile_assert(tile, reclaim_entries);
* Page size is computed as 2^(reclamation_size + XE_PTE_SHIFT) bytes.
* Only 4K, 64K (level 0), and 2M pages are supported by hardware for page reclaim
*/
- if (xe_child->level == 0 && !(pte & XE_PTE_PS64)) {
- xe_gt_stats_incr(gt, XE_GT_STATS_ID_PRL_4K_ENTRY_COUNT, 1);
- reclamation_size = COMPUTE_RECLAIM_ADDRESS_MASK(SZ_4K); /* reclamation_size = 0 */
- xe_tile_assert(tile, phys_addr % SZ_4K == 0);
- } else if (xe_child->level == 0) {
- xe_gt_stats_incr(gt, XE_GT_STATS_ID_PRL_64K_ENTRY_COUNT, 1);
- reclamation_size = COMPUTE_RECLAIM_ADDRESS_MASK(SZ_64K); /* reclamation_size = 4 */
- xe_tile_assert(tile, phys_addr % SZ_64K == 0);
- } else if (xe_child->level == 1 && pte & XE_PDE_PS_2M) {
+ if (is_2m) {
xe_gt_stats_incr(gt, XE_GT_STATS_ID_PRL_2M_ENTRY_COUNT, 1);
- reclamation_size = COMPUTE_RECLAIM_ADDRESS_MASK(SZ_2M); /* reclamation_size = 9 */
- xe_tile_assert(tile, phys_addr % SZ_2M == 0);
+ } else if (is_64k) {
+ xe_gt_stats_incr(gt, XE_GT_STATS_ID_PRL_64K_ENTRY_COUNT, 1);
+ } else if (xe_child->level == 0) {
+ xe_gt_stats_incr(gt, XE_GT_STATS_ID_PRL_4K_ENTRY_COUNT, 1);
} else {
xe_page_reclaim_list_abort(tile->primary_gt, prl,
"unsupported PTE level=%u pte=%#llx",
return 0;
}
+static int add_pte_to_prl(struct xe_tile *tile, struct xe_page_reclaim_list *prl,
+ struct xe_pt *xe_child, u64 pte, u64 addr)
+{
+ /*
+ * In rare scenarios, pte may not be written yet due to racy conditions.
+ * In such cases, invalidate the PRL and fallback to full PPC invalidation.
+ */
+ if (!pte) {
+ xe_page_reclaim_list_abort(tile->primary_gt, prl,
+ "found zero pte at addr=%#llx", addr);
+ return -EINVAL;
+ }
+
+ /* Ensure it is a defined page */
+ xe_tile_assert(tile, xe_child->level == 0 ||
+ (pte & (XE_PDE_PS_2M | XE_PDPE_PS_1G)));
+
+ /* Account for NULL terminated entry on end (-1) */
+ if (prl->num_entries >= XE_PAGE_RECLAIM_MAX_ENTRIES - 1) {
+ xe_page_reclaim_list_abort(tile->primary_gt, prl,
+ "overflow while adding pte=%#llx", pte);
+ return -ENOSPC;
+ }
+
+ return generate_reclaim_entry(tile, prl, pte, xe_child);
+}
+
+static bool add_compact_pt_prl(struct xe_tile *tile, struct xe_page_reclaim_list *prl,
+ struct xe_device *xe, struct xe_pt *compact_pt, u64 addr)
+{
+ struct iosys_map *map = &compact_pt->bo->vmap;
+
+ for (pgoff_t i = 0; i < SZ_2M / SZ_64K && xe_page_reclaim_list_valid(prl); i++) {
+ u64 pte = xe_map_rd(xe, map, i * sizeof(u64), u64);
+
+ if (add_pte_to_prl(tile, prl, compact_pt, pte, addr + i * SZ_64K))
+ break;
+ }
+
+ return xe_page_reclaim_list_valid(prl);
+}
+
static int xe_pt_stage_unbind_entry(struct xe_ptw *parent, pgoff_t offset,
unsigned int level, u64 addr, u64 next,
struct xe_ptw **child,
struct xe_pt *xe_child = container_of(*child, typeof(*xe_child), base);
struct xe_pt_stage_unbind_walk *xe_walk =
container_of(walk, typeof(*xe_walk), base);
- struct xe_device *xe = tile_to_xe(xe_walk->tile);
+ struct xe_page_reclaim_list *prl = xe_walk->prl;
+ struct xe_tile *tile = xe_walk->tile;
+ struct xe_device *xe = tile_to_xe(tile);
pgoff_t first = xe_pt_offset(addr, xe_child->level, walk);
bool killed;
XE_WARN_ON(!*child);
XE_WARN_ON(!level);
/* Check for leaf node */
- if (xe_walk->prl && xe_page_reclaim_list_valid(xe_walk->prl) &&
+ if (prl && xe_page_reclaim_list_valid(prl) &&
xe_child->level <= MAX_HUGEPTE_LEVEL) {
struct iosys_map *leaf_map = &xe_child->bo->vmap;
pgoff_t count = xe_pt_num_entries(addr, next, xe_child->level, walk);
for (pgoff_t i = 0; i < count; i++) {
u64 pte;
- int ret;
/*
* If not a leaf pt, skip unless non-leaf pt is interleaved between
u64 pt_size = 1ULL << walk->shifts[xe_child->level];
bool edge_pt = (i == 0 && !IS_ALIGNED(addr, pt_size)) ||
(i == count - 1 && !IS_ALIGNED(next, pt_size));
-
- if (!edge_pt) {
- xe_page_reclaim_list_abort(xe_walk->tile->primary_gt,
- xe_walk->prl,
+ struct xe_pt *child_pt =
+ container_of(xe_child->base.children[first + i],
+ struct xe_pt, base);
+
+ /* Compact PTs always fill a full 2M-aligned slot, never an edge. */
+ XE_WARN_ON(child_pt->is_compact && edge_pt);
+ if (edge_pt)
+ continue;
+
+ /* Walker never descends into compact PTs, descend now */
+ if (child_pt->is_compact) {
+ if (!add_compact_pt_prl(tile, prl, xe, child_pt,
+ addr + (u64)i * pt_size))
+ break;
+ } else {
+ xe_page_reclaim_list_abort(tile->primary_gt,
+ prl,
"PT is skipped by walk at level=%u offset=%lu",
xe_child->level, first + i);
break;
pte = xe_map_rd(xe, leaf_map, (first + i) * sizeof(u64), u64);
- /*
- * In rare scenarios, pte may not be written yet due to racy conditions.
- * In such cases, invalidate the PRL and fallback to full PPC invalidation.
- */
- if (!pte) {
- xe_page_reclaim_list_abort(xe_walk->tile->primary_gt, xe_walk->prl,
- "found zero pte at addr=%#llx", addr);
+ if (add_pte_to_prl(tile, prl, xe_child, pte, addr))
break;
- }
-
- /* Ensure it is a defined page */
- xe_tile_assert(xe_walk->tile, xe_child->level == 0 ||
- (pte & (XE_PDE_PS_2M | XE_PDPE_PS_1G)));
/* An entry should be added for 64KB but contigious 4K have XE_PTE_PS64 */
if (pte & XE_PTE_PS64)
i += 15; /* Skip other 15 consecutive 4K pages in the 64K page */
-
- /* Account for NULL terminated entry on end (-1) */
- if (xe_walk->prl->num_entries < XE_PAGE_RECLAIM_MAX_ENTRIES - 1) {
- ret = generate_reclaim_entry(xe_walk->tile, xe_walk->prl,
- pte, xe_child);
- if (ret)
- break;
- } else {
- /* overflow, mark as invalid */
- xe_page_reclaim_list_abort(xe_walk->tile->primary_gt, xe_walk->prl,
- "overflow while adding pte=%#llx",
- pte);
- break;
- }
}
}
* Verify if any PTE are potentially dropped at non-leaf levels, either from being
* killed or the page walk covers the region.
*/
- if (xe_walk->prl && xe_page_reclaim_list_valid(xe_walk->prl) &&
+ if (prl && xe_page_reclaim_list_valid(prl) &&
xe_child->level > MAX_HUGEPTE_LEVEL && xe_child->num_live) {
bool covered = xe_pt_covers(addr, next, xe_child->level, &xe_walk->base);
* we need to invalidate the PRL.
*/
if (killed || covered)
- xe_page_reclaim_list_abort(xe_walk->tile->primary_gt, xe_walk->prl,
+ xe_page_reclaim_list_abort(tile->primary_gt, prl,
"kill at level=%u addr=%#llx next=%#llx num_live=%u",
level, addr, next, xe_child->num_live);
}
projects / thirdparty / linux.git / commitdiff
