On Mon, Jul 08, 2019 at 06:43:23AM -0700, Guenter Roeck wrote:
> With CONFIG_BTRFS_FS=y and CONFIG_CRYPTO_CRC32C=m, we get:
>
> fs/btrfs/super.o: In function `btrfs_mount_root':
> fs/btrfs/super.c:1557: undefined reference to `crc32c_impl'
> fs/btrfs/super.o: In function `btrfs_print_mod_info':
> fs/btrfs/super.c:2348: undefined reference to `crc32c_impl'
> fs/btrfs/extent-tree.o: In function `btrfs_crc32c':
> fs/btrfs/ctree.h:2609: undefined reference to `crc32c'
> fs/btrfs/ctree.h:2609: undefined reference to `crc32c'
> fs/btrfs/ctree.h:2609: undefined reference to `crc32c'
> fs/btrfs/dir-item.o: In function `btrfs_name_hash':
> fs/btrfs/ctree.h:2619: undefined reference to `crc32c'
> fs/btrfs/ctree.h:2619: undefined reference to `crc32c'
>
> and more.
>
> Note that the comment in the offending commit "The module dependency on
> crc32c is preserved via MODULE_SOFTDEP("pre: crc32c"), which was previously
> provided by LIBCRC32C config option doing the same." is wrong, since it
> permits CONFIG_BTRFS_FS=y in combination with CONFIG_CRYPTO_CRC32C=m.
>
Meh, that should have been CONFIG_LIBCRC32C=m. Sorry for the confusion.
Anyway, I'll select CONFIG_LIBCRC32C=y in conjunction with CONFIG_BTRFS_FS=y
in my own build/boot tests going forward. That won't fix the problem, but
it will avoid loss of test coverage.
Guenter
> Cc: David Sterba <dste...@suse.com>
> Fixes: d5178578bcd4 ("btrfs: directly call into crypto framework for
> checksumming")
> Signed-off-by: Guenter Roeck <li...@roeck-us.net>
> ---
> fs/btrfs/Kconfig | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/fs/btrfs/Kconfig b/fs/btrfs/Kconfig
> index 212b4a854f2c..4c80c70597f9 100644
> --- a/fs/btrfs/Kconfig
> +++ b/fs/btrfs/Kconfig
> @@ -2,6 +2,7 @@
>
> config BTRFS_FS
> tristate "Btrfs filesystem support"
> + select LIBCRC32C
> select CRYPTO
> select CRYPTO_CRC32C
> select ZLIB_INFLATE
> --
> 2.7.4
>
