On Mon, 3 Nov 2025 19:07:24 +0800 Kuan-Wei Chiu <[email protected]> wrote:
> +Cc David > > Hi Guan-Chun, > > If we need to respin this series, please Cc David when sending the next > version. > > On Mon, Nov 03, 2025 at 11:24:35AM +0100, Andy Shevchenko wrote: > > On Fri, Oct 31, 2025 at 09:09:47PM -0700, Andrew Morton wrote: > > > On Wed, 29 Oct 2025 18:17:25 +0800 Guan-Chun Wu > > > <[email protected]> wrote: > > > > > > > This series introduces a generic Base64 encoder/decoder to the kernel > > > > library, eliminating duplicated implementations and delivering > > > > significant > > > > performance improvements. > > > > > > > > The Base64 API has been extended to support multiple variants (Standard, > > > > URL-safe, and IMAP) as defined in RFC 4648 and RFC 3501. The API now > > > > takes > > > > a variant parameter and an option to control padding. As part of this > > > > series, users are migrated to the new interface while preserving their > > > > specific formats: fscrypt now uses BASE64_URLSAFE, Ceph uses > > > > BASE64_IMAP, > > > > and NVMe is updated to BASE64_STD. > > > > > > > > On the encoder side, the implementation processes input in 3-byte > > > > blocks, > > > > mapping 24 bits directly to 4 output symbols. This avoids bit-by-bit > > > > streaming and reduces loop overhead, achieving about a 2.7x speedup > > > > compared > > > > to previous implementations. > > > > > > > > On the decoder side, replace strchr() lookups with per-variant reverse > > > > tables > > > > and process input in 4-character groups. Each group is mapped to > > > > numeric values > > > > and combined into 3 bytes. Padded and unpadded forms are validated > > > > explicitly, > > > > rejecting invalid '=' usage and enforcing tail rules. > > > > > > Looks like wonderful work, thanks. And it's good to gain a selftest > > > for this code. > > > > > > > This improves throughput by ~43-52x. > > > > > > Well that isn't a thing we see every day. > > > > I agree with the judgement, the problem is that this broke drastically a > > build: > > > > lib/base64.c:35:17: error: initializer overrides prior initialization of > > this subobject [-Werror,-Winitializer-overrides] > > 35 | [BASE64_STD] = BASE64_REV_INIT('+', '/'), > > | ^~~~~~~~~~~~~~~~~~~~~~~~~ > > lib/base64.c:26:11: note: expanded from macro 'BASE64_REV_INIT' > > 26 | ['A'] = 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, > > \ > > | ^ > > lib/base64.c:35:17: note: previous initialization is here > > 35 | [BASE64_STD] = BASE64_REV_INIT('+', '/'), > > | ^~~~~~~~~~~~~~~~~~~~~~~~~ > > lib/base64.c:25:16: note: expanded from macro 'BASE64_REV_INIT' > > 25 | [0 ... 255] = -1, \ > > | ^~ > > ... > > fatal error: too many errors emitted, stopping now [-ferror-limit=] > > 20 errors generated. > > > Since I didn't notice this build failure, I guess this happens during a > W=1 build? Sorry for that. Maybe I should add W=1 compilation testing > to my checklist before sending patches in the future. I also got an > email from the kernel test robot with a duplicate initialization > warning from the sparse tool [1], pointing to the same code. > > This implementation was based on David's previous suggestion [2] to > first default all entries to -1 and then set the values for the 64 > character entries. This was to avoid expanding the large 256 * 3 table > and improve code readability. > > Hi David, > > Since I believe many people test and care about W=1 builds, Last time I tried a W=1 build it failed horribly because of 'type-limits'. The kernel does that all the time - usually for its own error tests inside #define and inline functions. Certainly some of the changes I've seen to stop W=1 warnings are really a bad idea - but that is a bit of a digression. Warnings can be temporarily disabled using #pragma. That might be the best thing to do here with this over-zealous warning. This compiles on gcc and clang (even though the warnings have different names): #pragma GCC diagnostic push #pragma GCC diagnostic ignored "-Woverride-init" int x[16] = { [0 ... 15] = -1, [5] = 5}; #pragma GCC diagnostic pop > I think we need to find another way to avoid this warning? > Perhaps we could consider what you suggested: > > #define BASE64_REV_INIT(val_plus, val_comma, val_minus, val_slash, val_under) > { \ > [ 0 ... '+'-1 ] = -1, \ > [ '+' ] = val_plus, val_comma, val_minus, -1, val_slash, \ > [ '0' ] = 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, \ > [ '9'+1 ... 'A'-1 ] = -1, \ > [ 'A' ] = 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, \ > 23, 24, 25, 26, 27, 28, 28, 30, 31, 32, 33, 34, 35, \ > [ 'Z'+1 ... '_'-1 ] = -1, \ > [ '_' ] = val_under, \ > [ '_'+1 ... 'a'-1 ] = -1, \ > [ 'a' ] = 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, \ > 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, \ > [ 'z'+1 ... 255 ] = -1 \ > } I just checked, neither gcc nor clang allow empty ranges (eg [ 6 ... 5 ] = -1). Which means the coder has to know which characters are adjacent as well as getting the order right. Basically avoiding the warning sucks. > Or should we just expand the 256 * 3 table as it was before? That has much the same issue - IIRC it relies on three big sequential lists. The #pragma may be best - but doesn't solve sparse (unless it processes them as well). David > > [1]: > https://lore.kernel.org/oe-kbuild-all/[email protected]/ > [2]: https://lore.kernel.org/lkml/20250928195736.71bec9ae@pumpkin/ > > Regards, > Kuan-Wei
