From 735c466465eba51deaee3012d8403c10fc7c8c03 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Christian=20K=C3=B6nig?= <christian.koenig@amd.com>
Date: Mon, 7 Nov 2022 20:40:11 +0100
Subject: [PATCH] drm/ttm: optimize pool allocations a bit v2
Mime-version: 1.0
Content-type: text/plain; charset=UTF-8
Content-transfer-encoding: 8bit
Git-commit: 735c466465eba51deaee3012d8403c10fc7c8c03
Patch-mainline: v6.2-rc1
References: git-fixes

If we got a page pool use it as much as possible.

If we can't get more pages from the pool allocate as much as possible.

Only if that still doesn't work reduce the order and try again.

V2: minor cleanups

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221107195808.1873-1-christian.koenig@amd.com
Acked-by: Takashi Iwai <tiwai@suse.de>

---
 drivers/gpu/drm/ttm/ttm_pool.c | 82 ++++++++++++++++++++++++----------
 1 file changed, 58 insertions(+), 24 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c
index 21b61631f73a..9f6764bf3b15 100644
--- a/drivers/gpu/drm/ttm/ttm_pool.c
+++ b/drivers/gpu/drm/ttm/ttm_pool.c
@@ -344,6 +344,28 @@ static unsigned int ttm_pool_page_order(struct ttm_pool *pool, struct page *p)
 	return p->private;
 }
 
+/* Called when we got a page, either from a pool or newly allocated */
+static int ttm_pool_page_allocated(struct ttm_pool *pool, unsigned int order,
+				   struct page *p, dma_addr_t **dma_addr,
+				   unsigned long *num_pages,
+				   struct page ***pages)
+{
+	unsigned int i;
+	int r;
+
+	if (*dma_addr) {
+		r = ttm_pool_map(pool, order, p, dma_addr);
+		if (r)
+			return r;
+	}
+
+	*num_pages -= 1 << order;
+	for (i = 1 << order; i; --i, ++(*pages), ++p)
+		**pages = p;
+
+	return 0;
+}
+
 /**
  * ttm_pool_alloc - Fill a ttm_tt object
  *
@@ -385,17 +407,47 @@ int ttm_pool_alloc(struct ttm_pool *pool, struct ttm_tt *tt,
 	for (order = min_t(unsigned int, MAX_ORDER - 1, __fls(num_pages));
 	     num_pages;
 	     order = min_t(unsigned int, order, __fls(num_pages))) {
-		bool apply_caching = false;
 		struct ttm_pool_type *pt;
 
 		pt = ttm_pool_select_type(pool, tt->caching, order);
 		p = pt ? ttm_pool_type_take(pt) : NULL;
 		if (p) {
-			apply_caching = true;
-		} else {
-			p = ttm_pool_alloc_page(pool, gfp_flags, order);
-			if (p && PageHighMem(p))
-				apply_caching = true;
+			r = ttm_pool_apply_caching(caching, pages,
+						   tt->caching);
+			if (r)
+				goto error_free_page;
+
+			do {
+				r = ttm_pool_page_allocated(pool, order, p,
+							    &dma_addr,
+							    &num_pages,
+							    &pages);
+				if (r)
+					goto error_free_page;
+
+				if (num_pages < (1 << order))
+					break;
+
+				p = ttm_pool_type_take(pt);
+			} while (p);
+			caching = pages;
+		}
+
+		while (num_pages >= (1 << order) &&
+		       (p = ttm_pool_alloc_page(pool, gfp_flags, order))) {
+
+			if (PageHighMem(p)) {
+				r = ttm_pool_apply_caching(caching, pages,
+							   tt->caching);
+				if (r)
+					goto error_free_page;
+			}
+			r = ttm_pool_page_allocated(pool, order, p, &dma_addr,
+						    &num_pages, &pages);
+			if (r)
+				goto error_free_page;
+			if (PageHighMem(p))
+				caching = pages;
 		}
 
 		if (!p) {
@@ -406,24 +458,6 @@ int ttm_pool_alloc(struct ttm_pool *pool, struct ttm_tt *tt,
 			r = -ENOMEM;
 			goto error_free_all;
 		}
-
-		if (apply_caching) {
-			r = ttm_pool_apply_caching(caching, pages,
-						   tt->caching);
-			if (r)
-				goto error_free_page;
-			caching = pages + (1 << order);
-		}
-
-		if (dma_addr) {
-			r = ttm_pool_map(pool, order, p, &dma_addr);
-			if (r)
-				goto error_free_page;
-		}
-
-		num_pages -= 1 << order;
-		for (i = 1 << order; i; --i)
-			*(pages++) = p++;
 	}
 
 	r = ttm_pool_apply_caching(caching, pages, tt->caching);
-- 
2.35.3