Re: [PATCH v4 0/6] lib/base64: add generic encoder/decoder, migrate users

[David Laight <david.laight.linux@gmail.com>]

On Mon, 3 Nov 2025 19:07:24 +0800
Kuan-Wei Chiu <visitorckw@gmail.com> 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 <409411716@gms.tku.edu.tw> 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/202511021343.107utehN-lkp@intel.com/
> [2]: https://lore.kernel.org/lkml/20250928195736.71bec9ae@pumpkin/
> 
> Regards,
> Kuan-Wei