zlib icon indicating copy to clipboard operation
zlib copied to clipboard

Published 20 hours ago • verified publisher icon madler

Add support for IBM Z hardware-accelerated deflate

Open iii-i opened this issue 6 years ago • 34 comments

Note: this PR is based on https://github.com/madler/zlib/pull/750 and https://github.com/iii-i/zlib/releases/tag/crc32vx-v6 in order to simplify integration into distributions, which normally want all three changes.

IBM Z mainframes starting from version z15 provide DFLTCC instruction, which implements deflate algorithm in hardware with estimated compression and decompression performance orders of magnitude faster than the current zlib and ratio comparable with that of level 1.

This patch adds DFLTCC support to zlib. It can be enabled using the following build commands:

$ ./configure --dfltcc
$ make

When built like this, zlib would compress in hardware on level 1, and in software on all other levels. Decompression will always happen in hardware. In order to enable DFLTCC compression for levels 1-6 (i.e., to make it used by default) one could either configure with --dfltcc-level-mask=0x7e or export DFLTCC_LEVEL_MASK=0x7e at run time.

Two DFLTCC compression calls produce the same results only when they both are made on machines of the same generation, and when the respective buffers have the same offset relative to the start of the page. Therefore care should be taken when using hardware compression when reproducible results are desired. One such use case - reproducible software builds - is handled explicitly: when the SOURCE_DATE_EPOCH environment variable is set, the hardware compression is disabled.

DFLTCC does not support every single zlib feature, in particular:

* `inflate(Z_BLOCK)` and `inflate(Z_TREES)`
* `inflateMark()`
* `inflatePrime()`
* `inflateSyncPoint()`

When used, these functions will either switch to software, or, in case this is not possible, gracefully fail.

This patch tries to add DFLTCC support in the least intrusive way. All SystemZ-specific code is placed into a separate file, but unfortunately there is still a noticeable amount of changes in the main zlib code. Below is the summary of these changes.

DFLTCC takes as arguments a parameter block, an input buffer, an output buffer and a window. Since DFLTCC requires parameter block to be doubleword-aligned, and it's reasonable to allocate it alongside deflate and inflate states, The ZALLOC_STATE(), ZFREE_STATE() and ZCOPY_STATE() macros are introduced in order to encapsulate the allocation details. The same is true for window, for which the ZALLOC_WINDOW() and TRY_FREE_WINDOW() macros are introduced.

Software and hardware window formats do not match, therefore, deflateSetDictionary(), deflateGetDictionary(), inflateSetDictionary() and inflateGetDictionary() need special handling, which is triggered using the new DEFLATE_SET_DICTIONARY_HOOK(), DEFLATE_GET_DICTIONARY_HOOK(), INFLATE_SET_DICTIONARY_HOOK() and INFLATE_GET_DICTIONARY_HOOK() macros.

deflateResetKeep() and inflateResetKeep() now update the DFLTCC parameter block, which is allocated alongside zlib state, using the new DEFLATE_RESET_KEEP_HOOK() and INFLATE_RESET_KEEP_HOOK() macros.

The new DEFLATE_PARAMS_HOOK() macro switches between the hardware and the software deflate implementations when the deflateParams() arguments demand this.

The new INFLATE_PRIME_HOOK(), INFLATE_MARK_HOOK() and INFLATE_SYNC_POINT_HOOK() macros make the respective unsupported calls gracefully fail.

The algorithm implemented in the hardware has different compression ratio than the one implemented in software. In order for deflateBound() to return the correct results for the hardware implementation, the new DEFLATE_BOUND_ADJUST_COMPLEN() and DEFLATE_NEED_CONSERVATIVE_BOUND() macros are introduced.

Actual compression and decompression are handled by the new DEFLATE_HOOK() and INFLATE_TYPEDO_HOOK() macros. Since inflation with DFLTCC manages the window on its own, calling updatewindow() is suppressed using the new INFLATE_NEED_UPDATEWINDOW() macro.

In addition to the compression, DFLTCC computes the CRC-32 and Adler-32 checksums, therefore, whenever it's used, the software checksumming is suppressed using the new DEFLATE_NEED_CHECKSUM() and INFLATE_NEED_CHECKSUM() macros.

DFLTCC will refuse to write an End-of-block Symbol if there is no input data, thus in some cases it is necessary to do this manually. In order to achieve this, send_bits(), bi_reverse(), bi_windup() and flush_pending() are promoted from local to ZLIB_INTERNAL. Furthermore, since the block and the stream termination must be handled in software as well, enum block_state is moved to deflate.h.

Since the first call to dfltcc_inflate() already needs the window, and it might be not allocated yet, inflate_ensure_window() is factored out of updatewindow() and made ZLIB_INTERNAL.

iii-i avatar Mar 15 '19 10:03 iii-i

Gentle ping.

iii-i avatar Apr 05 '19 09:04 iii-i

  • Fixed a bug when Z_SYNC_FLUSH usage led to incomplete EOBS write.
  • Fixed a "goto fail" bug in dfltcc_deflate_get_dictionary().
  • Replaced getenv() with secure_getenv().
  • Added sys/sdt.h feature test.

iii-i avatar May 06 '19 09:05 iii-i

  • Fixed incorrect usage of STFLE instruction (reported against gzip here, fixed in gzip here, corresponding zlib-ng pull request here).
  • Simplified hooks by removing #undef statements.

iii-i avatar Jun 04 '19 10:06 iii-i

  • Fixed 31-bit build:
    • Added machine mode hint for STFLE.
    • Adjusted offset calculations.
  • Fixed sys/sdt.h feature test.
  • Added an entry to contrib/README.contrib.

iii-i avatar Jul 08 '19 13:07 iii-i

Only issue with this PR is that looks like it is is controlled by CFLSGS injection instead proper autoconf ---with{,oout}-foo.

kloczek avatar Aug 23 '19 05:08 kloczek

  • Added support for switching between software and hardware compression.
  • Added --dfltcc configure flag (the old way of building it still works).

iii-i avatar May 11 '20 10:05 iii-i

  • Fix missing EOBS in raw streams.
  • Parse environment variables and facility bits only once.

iii-i avatar Aug 05 '20 16:08 iii-i

  • Fix a number of issues related to switching compression levels.
  • Fix building with clang.
  • Fix sys/sdt.h detection.

iii-i avatar Aug 24 '20 21:08 iii-i

  • Make inflateSyncPoint() gracefully fail instead of returning an incorrect result.

iii-i avatar Oct 13 '20 11:10 iii-i

  • Add --dfltcc-level-mask=... to configure as a shorthand for CFLAGS=-DDFLTCC_LEVEL_MASK=....
  • Add a short README.

iii-i avatar Nov 23 '20 16:11 iii-i

  • Fix compressBound().

iii-i avatar Oct 20 '21 00:10 iii-i

Is there any update on merging this PR? This is already used in production systems and there has been no bug reports so far.

danyspin97 avatar Mar 31 '22 07:03 danyspin97

Hello Ilya, Zlib has released a new 1.2.12 version a few days ago, and this PR starts to have a lot of conflicts.

Is it possible for you to check and correct them? They are way out of our scope and we could really use your help in this because you understand it the best.

This new release brings a new CVE which has to be fixed ASAP, so could you give me an estimate on how long it could take you to fix these conflicts?

Thank you for your effort.

ljavorsk avatar Apr 05 '22 12:04 ljavorsk

@ljavorsk I have a preview version here: https://github.com/iii-i/zlib/releases/tag/dfltcc-20220405 So far it's looking good, but I need to run more tests and also fuzz it a little. Hopefully I'll be done by the end of the day.

iii-i avatar Apr 05 '22 13:04 iii-i

@iii-i Great news thank you :)

Will you update this PR or create a new one with the fixed patches after you declare it's ready?

ljavorsk avatar Apr 05 '22 14:04 ljavorsk

Yes, that's the plan.

iii-i avatar Apr 05 '22 14:04 iii-i

  • Rebased.

iii-i avatar Apr 05 '22 18:04 iii-i

  • Fix updating strm.adler with inflate().

iii-i avatar Sep 21 '22 09:09 iii-i

Hi @iii-i ,

Could you please also rebase the patch on top of the new zlib-1.2.13 version?

Thank you so much :)

ljavorsk avatar Oct 17 '22 09:10 ljavorsk

  • Rebased on top of zlib 1.2.13.

iii-i avatar Oct 17 '22 22:10 iii-i

@madler: What do you think?

Neustradamus avatar Oct 26 '22 04:10 Neustradamus

  • Removed unused endptr variable in init_globals().

iii-i avatar Nov 22 '22 09:11 iii-i

  • Support inflate with small window.

iii-i avatar Dec 15 '22 13:12 iii-i

  • Do not update strm.adler for raw streams (https://bugzilla.redhat.com/show_bug.cgi?id=2155328).

iii-i avatar Dec 23 '22 13:12 iii-i

Could you please rebase this patch on top of the https://github.com/madler/zlib/pull/750 PR as we agreed?

There are a few conflicting changes and the patch doesn't apply cleanly, so I would have to rewrite it every single time you rebase it.

Thank you so much @iii-i :)

ljavorsk avatar Jan 09 '23 10:01 ljavorsk

  • Rebased on top of devel and https://github.com/madler/zlib/pull/750.

iii-i avatar Jan 09 '23 15:01 iii-i

Hello! I am zlib maintainer at SUSE and we are experiencing crashes on Firefox on s390x machines. After some investigation, we have find out that they are related to zlib, possibly incorrect usage from Firefox. However, since it is working with the bundled lib and the other architectures, I would also consider this a bug in the hardware accelerated implementation of deflate.

This is the bug with the description of the issue: https://bugzilla.suse.com/show_bug.cgi?id=1210593

danyspin97 avatar Apr 19 '23 09:04 danyspin97

  • Fix deflateBound() before deflateInit().

iii-i avatar Apr 28 '23 12:04 iii-i

Hi,

The change called "add DFLTCC support for using inflate() with a small window" (diff: https://github.com/madler/zlib/compare/26f2c0a4e17e5558d779797d713aa37ebaeef390..28a49d89b4329ca349e1b10e860e6b3f7ebbfa9e) has brought one COMPILER_WARNING zlib-1.2.11/inflate.c:1507:14: warning[-Wunused-variable]: unused variable 'wsize'

This could be easily fixed by removing the variable.

ljavorsk avatar May 22 '23 05:05 ljavorsk

  • Remove an unused variable.

iii-i avatar May 26 '23 09:05 iii-i