drm/i915/gtt: Avoid overflowing the WC stash

An interesting issue cropped with making the pagetables be allocated and
freed concurrently (i.e. removing their grandeous struct_mutex guard)
was that we would overflow the page stash. This happens when we have
multiple allocators grabbing WC pages such that we fill the vm's local
page stash and then when we free another page, the page stash is already
full and we overflow.

The fix is quite simple: to check for a full page stash before adding
another. This results in us keeping a vm local page stash around for
much longer, which is both a blessing and a curse.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: NMatthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190529093407.31697-1-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/i915_gem_gtt.c

@@ -341,11 +341,11 @@ static struct page *stash_pop_page(struct pagestash *stash)
 
 static void stash_push_pagevec(struct pagestash *stash, struct pagevec *pvec)
 {
-	int nr;
+	unsigned int nr;
 
 	spin_lock_nested(&stash->lock, SINGLE_DEPTH_NESTING);
 
-	nr = min_t(int, pvec->nr, pagevec_space(&stash->pvec));
+	nr = min_t(typeof(nr), pvec->nr, pagevec_space(&stash->pvec));
 	memcpy(stash->pvec.pages + stash->pvec.nr,
 	       pvec->pages + pvec->nr - nr,
 	       sizeof(pvec->pages[0]) * nr);
@@ -399,7 +399,8 @@ static struct page *vm_alloc_page(struct i915_address_space *vm, gfp_t gfp)
 		page = stack.pages[--stack.nr];
 
 		/* Merge spare WC pages to the global stash */
-		stash_push_pagevec(&vm->i915->mm.wc_stash, &stack);
+		if (stack.nr)
+			stash_push_pagevec(&vm->i915->mm.wc_stash, &stack);
 
 		/* Push any surplus WC pages onto the local VM stash */
 		if (stack.nr)
@@ -469,8 +470,10 @@ static void vm_free_page(struct i915_address_space *vm, struct page *page)
 	 */
 	might_sleep();
 	spin_lock(&vm->free_pages.lock);
-	if (!pagevec_add(&vm->free_pages.pvec, page))
+	while (!pagevec_space(&vm->free_pages.pvec))
 		vm_free_pages_release(vm, false);
+	GEM_BUG_ON(pagevec_count(&vm->free_pages.pvec) >= PAGEVEC_SIZE);
+	pagevec_add(&vm->free_pages.pvec, page);
 	spin_unlock(&vm->free_pages.lock);
 }