It will cost more time if compressed buffers are allocated on demand for
low-latency algorithms (like lz4) so EROFS uses per-CPU buffers to keep
compressed data if in-place decompression is unfulfilled. While it is kind
of wasteful of memory for a device with hundreds of CPUs, and only a small
number of CPUs concurrently decompress most of the time.
This patch renames it as 'global buffer pool' and makes it configurable.
This allows two or more CPUs to share a common buffer to reduce memory
occupation.
Signed-off-by: Chunhai Guo
Suggested-by: Gao Xiang
---
fs/erofs/Makefile | 2 +-
fs/erofs/decompressor.c | 6 +-
fs/erofs/internal.h | 14 ++--
fs/erofs/pcpubuf.c | 148
fs/erofs/super.c| 9 ++-
fs/erofs/utils.c| 148
6 files changed, 166 insertions(+), 161 deletions(-)
delete mode 100644 fs/erofs/pcpubuf.c
diff --git a/fs/erofs/Makefile b/fs/erofs/Makefile
index 994d0b9deddf..6cf1504c63e6 100644
--- a/fs/erofs/Makefile
+++ b/fs/erofs/Makefile
@@ -3,7 +3,7 @@
obj-$(CONFIG_EROFS_FS) += erofs.o
erofs-objs := super.o inode.o data.o namei.o dir.o utils.o sysfs.o
erofs-$(CONFIG_EROFS_FS_XATTR) += xattr.o
-erofs-$(CONFIG_EROFS_FS_ZIP) += decompressor.o zmap.o zdata.o pcpubuf.o
+erofs-$(CONFIG_EROFS_FS_ZIP) += decompressor.o zmap.o zdata.o
erofs-$(CONFIG_EROFS_FS_ZIP_LZMA) += decompressor_lzma.o
erofs-$(CONFIG_EROFS_FS_ZIP_DEFLATE) += decompressor_deflate.o
erofs-$(CONFIG_EROFS_FS_ONDEMAND) += fscache.o
diff --git a/fs/erofs/decompressor.c b/fs/erofs/decompressor.c
index d4cee95af14c..72a11c5eba8f 100644
--- a/fs/erofs/decompressor.c
+++ b/fs/erofs/decompressor.c
@@ -54,7 +54,7 @@ static int z_erofs_load_lz4_config(struct super_block *sb,
sbi->lz4.max_distance_pages = distance ?
DIV_ROUND_UP(distance, PAGE_SIZE) + 1 :
LZ4_MAX_DISTANCE_PAGES;
- return erofs_pcpubuf_growsize(sbi->lz4.max_pclusterblks);
+ return z_erofs_gbuf_growsize(sbi->lz4.max_pclusterblks);
}
/*
@@ -159,7 +159,7 @@ static void *z_erofs_lz4_handle_overlap(struct
z_erofs_lz4_decompress_ctx *ctx,
docopy:
/* Or copy compressed data which can be overlapped to per-CPU buffer */
in = rq->in;
- src = erofs_get_pcpubuf(ctx->inpages);
+ src = z_erofs_get_gbuf(ctx->inpages);
if (!src) {
DBG_BUGON(1);
kunmap_local(inpage);
@@ -260,7 +260,7 @@ static int z_erofs_lz4_decompress_mem(struct
z_erofs_lz4_decompress_ctx *ctx,
} else if (maptype == 1) {
vm_unmap_ram(src, ctx->inpages);
} else if (maptype == 2) {
- erofs_put_pcpubuf(src);
+ z_erofs_put_gbuf(src);
} else if (maptype != 3) {
DBG_BUGON(1);
return -EFAULT;
diff --git a/fs/erofs/internal.h b/fs/erofs/internal.h
index b0409badb017..574a7c5b3929 100644
--- a/fs/erofs/internal.h
+++ b/fs/erofs/internal.h
@@ -471,11 +471,11 @@ int erofs_try_to_free_all_cached_pages(struct
erofs_sb_info *sbi,
struct erofs_workgroup *egrp);
int z_erofs_map_blocks_iter(struct inode *inode, struct erofs_map_blocks *map,
int flags);
-void *erofs_get_pcpubuf(unsigned int requiredpages);
-void erofs_put_pcpubuf(void *ptr);
-int erofs_pcpubuf_growsize(unsigned int nrpages);
-void __init erofs_pcpubuf_init(void);
-void erofs_pcpubuf_exit(void);
+void *z_erofs_get_gbuf(unsigned int requiredpages);
+void z_erofs_put_gbuf(void *ptr);
+int z_erofs_gbuf_growsize(unsigned int nrpages);
+int __init z_erofs_gbuf_init(void);
+void z_erofs_gbuf_exit(void);
int erofs_init_managed_cache(struct super_block *sb);
int z_erofs_parse_cfgs(struct super_block *sb, struct erofs_super_block *dsb);
#else
@@ -485,8 +485,8 @@ static inline int erofs_init_shrinker(void) { return 0; }
static inline void erofs_exit_shrinker(void) {}
static inline int z_erofs_init_zip_subsystem(void) { return 0; }
static inline void z_erofs_exit_zip_subsystem(void) {}
-static inline void erofs_pcpubuf_init(void) {}
-static inline void erofs_pcpubuf_exit(void) {}
+static inline void z_erofs_gbuf_init(void) {}
+static inline void z_erofs_gbuf_exit(void) {}
static inline int erofs_init_managed_cache(struct super_block *sb) { return 0;
}
#endif /* !CONFIG_EROFS_FS_ZIP */
diff --git a/fs/erofs/pcpubuf.c b/fs/erofs/pcpubuf.c
deleted file mode 100644
index c7a4b1d77069..
--- a/fs/erofs/pcpubuf.c
+++ /dev/null
@@ -1,148 +0,0 @@
-// SPDX-License-Identifier: GPL-2.0-only
-/*
- * Copyright (C) Gao Xiang
- *
- * For low-latency decompression algorithms (e.g. lz4), reserve consecutive
- * per-CPU virtual memory (in pages) in advance to store such inplace I/O
- * data if inplace decompression is failed (due to unmet inplace margin for
- * example).
- */
-#include "internal.h"
-
-struct erofs_pcpubuf