nbdkit/0008-cache-Fix-misleading-LRU-diagram-and-comment.patch
Richard W.M. Jones b4d42fa29b More efficient cache and cow filters.
Add nbdkit-cow-filter cow-on-read option.
Add nbdkit-cache-filter cache-on-read=/PATH.
Add nbdkit-cache-filter cache-min-block-size option.
Add nbdkit-delay-filter delay-open and delay-close options.
Reduce verbosity of debugging from virt-v2v.
Miscellaneous bugfixes
resolves: rhbz#1950632
2021-07-30 18:40:23 +01:00

80 lines
3.5 KiB
Diff

From b436ca6c69ef7d8d826be609820027f10134274d Mon Sep 17 00:00:00 2001
From: "Richard W.M. Jones" <rjones@redhat.com>
Date: Tue, 27 Jul 2021 21:28:48 +0100
Subject: [PATCH] cache: Fix misleading LRU diagram and comment
Only comment changes.
(cherry picked from commit 7b33c86e0910d941dc34bdb481d61806f31cdcef)
---
filters/cache/lru.c | 32 ++++++++++++++++++++------------
1 file changed, 20 insertions(+), 12 deletions(-)
diff --git a/filters/cache/lru.c b/filters/cache/lru.c
index 1c3c3e10..716b4984 100644
--- a/filters/cache/lru.c
+++ b/filters/cache/lru.c
@@ -53,12 +53,14 @@
/* LRU bitmaps. These bitmaps implement a simple, fast LRU structure.
*
- * bm[0] bm[1] blocks not in either bitmap
- * ┌─────────┬──────────────────┬─────────────────────────────┐
- * │ │ │ │
- * └─────────┴──────────────────┴─────────────────────────────┘
- * ↑ c1 bits set
- * c0 bits set
+ * bm[0]
+ * ┌───────────────────────┐
+ * │ X XX X XXX │ c0 bits set
+ * └───────────────────────┘
+ * bm[1]
+ * ┌───────────────────────┐
+ * │ X XX X X │ c1 bits set
+ * └───────────────────────┘
*
* The LRU structure keeps track of the [approx] last N distinct
* blocks which have been most recently accessed. It can answer in
@@ -69,8 +71,7 @@
*
* When a new block is accessed, we set the corresponding bit in bm[0]
* and increment c0 (c0 counts the number of bits set in bm[0]). If
- * c0 == N/2 then we swap the two bitmaps, clear bm[0], and reset c0
- * to 0.
+ * c0 == N/2 then we move bm[1] <- bm[0], clear bm[0] and set c0 <- 0.
*
* To check if a block has been accessed within the previous N
* distinct accesses, we simply have to check both bitmaps. If it is
@@ -78,9 +79,11 @@
* reclaimed.
*
* You'll note that in fact we only keep track of between N/2 and N
- * recently accessed blocks. We could make the estimate more accurate
- * by having more bitmaps, but as this is only a heuristic we choose
- * to keep the implementation simple and memory usage low instead.
+ * recently accessed blocks because the same block can appear in both
+ * bitmaps. bm[1] is a last chance to hold on to blocks which are
+ * soon to be reclaimed. We could make the estimate more accurate by
+ * having more bitmaps, but as this is only a heuristic we choose to
+ * keep the implementation simple and memory usage low instead.
*/
static struct bitmap bm[2];
static unsigned c0 = 0, c1 = 0;
@@ -129,7 +132,12 @@ lru_set_recently_accessed (uint64_t blknum)
bitmap_set_blk (&bm[0], blknum, true);
c0++;
- /* If we've reached N/2 then we need to swap over the bitmaps. */
+ /* If we've reached N/2 then we need to swap over the bitmaps. Note
+ * the purpose of swapping here is to ensure that we do not have to
+ * copy the dynamically allocated bm->bitmap field (the pointers are
+ * swapped instead). The bm[0].bitmap field is immediately zeroed
+ * after the swap.
+ */
if (c0 >= N/2) {
struct bitmap tmp;
--
2.31.1