[PATCH v4 3/5] erofs-utils: mkfs: add --worker=# parameter
Gao Xiang
hsiangkao at linux.alibaba.com
Thu Feb 29 20:55:22 AEDT 2024
On 2024/2/29 00:16, Yifan Zhao wrote:
> This patch introduces a --worker=# parameter for the incoming
> multi-threaded compression support. It also introduces a segment size
> used in multi-threaded compression, which has the default value 16MB
> and cannot be modified.
>
> It also introduces a concept called `segment size` to split large files
> for multi-threading, which has the default value 16MB for now.
>
> Signed-off-by: Yifan Zhao <zhaoyifan at sjtu.edu.cn>
> ---
> include/erofs/config.h | 4 ++++
> lib/config.c | 4 ++++
> mkfs/main.c | 38 ++++++++++++++++++++++++++++++++++++++
> 3 files changed, 46 insertions(+)
>
> diff --git a/include/erofs/config.h b/include/erofs/config.h
> index 73e3ac2..d2f91ff 100644
> --- a/include/erofs/config.h
> +++ b/include/erofs/config.h
> @@ -75,6 +75,10 @@ struct erofs_configure {
> char c_force_chunkformat;
> /* < 0, xattr disabled and INT_MAX, always use inline xattrs */
> int c_inline_xattr_tolerance;
> +#ifdef EROFS_MT_ENABLED
> + u64 c_segment_size;
> + u32 c_mt_workers;
> +#endif
>
> u32 c_pclusterblks_max, c_pclusterblks_def, c_pclusterblks_packed;
> u32 c_max_decompressed_extent_bytes;
> diff --git a/lib/config.c b/lib/config.c
> index 947a183..2530274 100644
> --- a/lib/config.c
> +++ b/lib/config.c
> @@ -38,6 +38,10 @@ void erofs_init_configure(void)
> cfg.c_pclusterblks_max = 1;
> cfg.c_pclusterblks_def = 1;
> cfg.c_max_decompressed_extent_bytes = -1;
> +#ifdef EROFS_MT_ENABLED
> + cfg.c_segment_size = 16ULL * 1024 * 1024;
> + cfg.c_mt_workers = 1;
> +#endif
>
> erofs_stdout_tty = isatty(STDOUT_FILENO);
> }
> diff --git a/mkfs/main.c b/mkfs/main.c
> index 258c1ce..ce9c28b 100644
> --- a/mkfs/main.c
> +++ b/mkfs/main.c
> @@ -74,6 +74,9 @@ static struct option long_options[] = {
> {"ungzip", optional_argument, NULL, 517},
> #endif
> {"offset", required_argument, NULL, 518},
> +#ifdef EROFS_MT_ENABLED
> + {"workers", required_argument, NULL, 519},
> +#endif
> {0, 0, 0, 0},
> };
>
> @@ -179,6 +182,9 @@ static void usage(int argc, char **argv)
> " --product-out=X X=product_out directory\n"
> " --fs-config-file=X X=fs_config file\n"
> " --block-list-file=X X=block_list file\n"
> +#endif
> +#ifdef EROFS_MT_ENABLED
> + " --workers=# set the number of worker threads to # (default=1)\n"
> #endif
> );
> }
> @@ -408,6 +414,13 @@ static void erofs_rebuild_cleanup(void)
> rebuild_src_count = 0;
> }
>
> +#ifdef EROFS_MT_ENABLED
> +static u32 mkfs_max_worker_num() {
static unsigned int erofs_mkfs_max_worker_num()
{
return erofs_get_available_processors() ? : 16;
}
> + u32 ncpu = erofs_get_available_processors();
> + return ncpu ? ncpu : 16;
> +}
> +#endif
> +
> static int mkfs_parse_options_cfg(int argc, char *argv[])
> {
> char *endptr;
> @@ -650,6 +663,21 @@ static int mkfs_parse_options_cfg(int argc, char *argv[])
> return -EINVAL;
> }
> break;
> +#ifdef EROFS_MT_ENABLED
> + case 519:
> + cfg.c_mt_workers = strtoul(optarg, &endptr, 0);
> + if (errno || *endptr != '\0') {
> + erofs_err("invalid worker number %s", optarg);
> + return -EINVAL;
> + }
> + if (cfg.c_mt_workers > mkfs_max_worker_num()) {
> + erofs_warn(
> + "worker number %s is too large, setting to %ud",
> + optarg, mkfs_max_worker_num());
let's not break erofs_{err,warn,...} print line, it means:
cfg.c_mt_workers = mkfs_max_worker_num();
erofs_warn("worker number %s is too large, reseting to %ud",
optarg, cfg.c_mt_workers);
> + cfg.c_mt_workers = mkfs_max_worker_num();
> + }
> + break;
> +#endif
> case 'V':
> version();
> exit(0);
> @@ -803,6 +831,16 @@ static int mkfs_parse_options_cfg(int argc, char *argv[])
> }
> cfg.c_pclusterblks_packed = pclustersize_packed >> sbi.blkszbits;
> }
> +
> +#ifdef EROFS_MT_ENABLED
> + if (cfg.c_mt_workers > 1 &&
> + (cfg.c_dedupe || cfg.c_fragments || cfg.c_ztailpacking)) {
> + cfg.c_mt_workers = 1;
> + erofs_warn("Please note that dedupe/fragments/ztailpacking"
> + "is NOT supported in multi-threaded mode now, using worker=1.");
> + }
> +#endif
This part would be better to go with the next patch.
Thanks,
Gao Xiang
More information about the Linux-erofs
mailing list