lkml.org 
[lkml]   [2020]   [Dec]   [3]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v6 1/3] lib: zstd: Add kernel-specific API
On Thu, Dec 03, 2020 at 03:59:21AM +0000, Nick Terrell wrote:
> On Dec 2, 2020, at 7:14 PM, Michał Mirosław <mirq-linux@rere.qmqm.pl> wrote:
> > On Thu, Dec 03, 2020 at 01:42:03AM +0000, Nick Terrell wrote:
> >> On Dec 2, 2020, at 5:16 PM, Michał Mirosław <mirq-linux@rere.qmqm.pl> wrote:
> >>> On Wed, Dec 02, 2020 at 12:32:40PM -0800, Nick Terrell wrote:
> >>>> From: Nick Terrell <terrelln@fb.com>
> >>>>
> >>>> This patch:
> >>>> - Moves `include/linux/zstd.h` -> `lib/zstd/zstd.h`
> >>>> - Adds a new API in `include/linux/zstd.h` that is functionally
> >>>> equivalent to the in-use subset of the current API. Functions are
> >>>> renamed to avoid symbol collisions with zstd, to make it clear it is
> >>>> not the upstream zstd API, and to follow the kernel style guide.
> >>>> - Updates all callers to use the new API.
> >>>>
> >>>> There are no functional changes in this patch. Since there are no
> >>>> functional change, I felt it was okay to update all the callers in a
> >>>> single patch, since once the API is approved, the callers are
> >>>> mechanically changed.
> >>> [...]
> >>>> --- a/lib/decompress_unzstd.c
> >>>> +++ b/lib/decompress_unzstd.c
> >>> [...]
> >>>> static int INIT handle_zstd_error(size_t ret, void (*error)(char *x))
> >>>> {
> >>>> - const int err = ZSTD_getErrorCode(ret);
> >>>> -
> >>>> - if (!ZSTD_isError(ret))
> >>>> + if (!zstd_is_error(ret))
> >>>> return 0;
> >>>>
> >>>> - switch (err) {
> >>>> - case ZSTD_error_memory_allocation:
> >>>> - error("ZSTD decompressor ran out of memory");
> >>>> - break;
> >>>> - case ZSTD_error_prefix_unknown:
> >>>> - error("Input is not in the ZSTD format (wrong magic bytes)");
> >>>> - break;
> >>>> - case ZSTD_error_dstSize_tooSmall:
> >>>> - case ZSTD_error_corruption_detected:
> >>>> - case ZSTD_error_checksum_wrong:
> >>>> - error("ZSTD-compressed data is corrupt");
> >>>> - break;
> >>>> - default:
> >>>> - error("ZSTD-compressed data is probably corrupt");
> >>>> - break;
> >>>> - }
> >>>> + error("ZSTD decompression failed");
> >>>> return -1;
> >>>> }
> >>>
> >>> This looses diagnostics specificity - is this intended? At least the
> >>> out-of-memory condition seems useful to distinguish.
> >>
> >> Good point. The zstd API no longer exposes the error code enum,
> >> but it does expose zstd_get_error_name() which can be used here.
> >> I was thinking that the string needed to be static for some reason, but
> >> that is not the case. I will make that change.
> >>
> >>>> +size_t zstd_compress_stream(zstd_cstream *cstream,
> >>>> + struct zstd_out_buffer *output, struct zstd_in_buffer *input)
> >>>> +{
> >>>> + ZSTD_outBuffer o;
> >>>> + ZSTD_inBuffer i;
> >>>> + size_t ret;
> >>>> +
> >>>> + memcpy(&o, output, sizeof(o));
> >>>> + memcpy(&i, input, sizeof(i));
> >>>> + ret = ZSTD_compressStream(cstream, &o, &i);
> >>>> + memcpy(output, &o, sizeof(o));
> >>>> + memcpy(input, &i, sizeof(i));
> >>>> + return ret;
> >>>> +}
> >>>
> >>> Is all this copying necessary? How is it different from type-punning by
> >>> direct pointer cast?
> >>
> >> If breaking strict aliasing and type-punning by pointer casing is okay, then
> >> we can do that here. These memcpys will be negligible for performance, but
> >> type-punning would be more succinct if allowed.
> >
> > Ah, this might break LTO builds due to strict aliasing violation.
> > So I would suggest to just #define the ZSTD names to kernel ones
> > for the library code. Unless there is a cleaner solution...
>
> I don’t want to do that because I want in the 3rd series of the patchset I update
> to zstd-1.4.6. And I’m using zstd-1.4.6 as-is in upstream. This allows us to keep
> the kernel version up to date, since the patch to update to a new version can be
> generated automatically (and manually tested), so it doesn’t fall years behind
> upstream again.
>
> The alternative would be to make upstream zstd’s header public and
> #define zstd_in_buffer ZSTD_inBuffer. But that would make zstd’s header
> public, which would somewhat defeat the purpose of having a kernel wrapper.

I thought the problem was API style spill-over from the library to other parts
of the kernel. A header-only wrapper can stop this. I'm not sure symbol
visibility (namespace pollution) was a concern.

Best Regards
Michał Mirosław

\
 
 \ /
  Last update: 2020-12-03 06:05    [W:0.107 / U:0.216 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site