qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
From: "Alex Bennée" <alex.bennee@linaro.org>
To: Richard Henderson <richard.henderson@linaro.org>
Cc: qemu-devel@nongnu.org
Subject: Re: [PATCH v3 00/25] risu cleanups and improvements
Date: Tue, 23 Jun 2020 14:30:42 +0100	[thread overview]
Message-ID: <87mu4t9af1.fsf@linaro.org> (raw)
In-Reply-To: <87v9ji88cj.fsf@linaro.org>


Alex Bennée <alex.bennee@linaro.org> writes:

> Richard Henderson <richard.henderson@linaro.org> writes:
>
>> Version 3 changes the --dump option to --fulldump and --diffdump,
>> after an off-hand suggestion by Alex.
>>
>> These are now mode options, similar to --master.  Which means that
>> dumping is an orthogonal apprentice type, which means that we can
>> dump from a socket.  I'm not sure that will be useful as such, but
>> I think it makes main be a bit cleaner.
>
> Hmm recording traces I ran into a difference, need to track down if its
> a master or apprentice bug (both are native):
>
>   ./builds/arm64/risu aarch64-all-v8dot0/insn_LDAPR__INC.risu.bin -t aarch64-all-v8dot0/insn_LDAPR__INC.risu.bin.trace
>
> fails with:
>
>   loading test image aarch64-all-v8dot0/insn_LDAPR__INC.risu.bin...
>   starting apprentice image at 0xffff8548c000
>   starting image
>   Mismatch reg after 4 checkpoints
>   master reginfo:
>   faulting insn 38bfc1f4
>   .
>   .
>   .
>   mismatch detail (master : apprentice):
>     X15    : 0000ffff9eba41dc vs 0000ffff8548c1dc

This is sort of bogus due to LDAPR being a v8.3 instruction so we fault
and x15 is a pointer into the memblock which isn't stable. If it had
executed we would have nixed the absolution address to and offset before
the OP_COMPARE.

I think for OP_SIGILL the only thing that can be truly valid would be to
compare PCs so we can check the apprentice also faults on the same SIGILL?

>
>>
>> If using old trace files with the new risu, you get
>>
>>   Unexpected magic number: 0x000078
>>
>> If for somehow you use different risu for master and apprentice on
>> sockets, the apprentice will hang waiting for data that the master
>> will never write.  This is less than helpful, but should be trivial
>> to avoid.
>>
>> While cleaning up the interface for reginfo_dump_mismatch(), I
>> noticed some bugs on the ppc64 side.
>>
>> The patches without reviews are:
>>
>> 0014-Merge-reginfo.c-into-risu.c.patch
>> 0015-Rearrange-reginfo-and-memblock-buffers.patch
>> 0016-Split-out-recv_register_info.patch
>> 0017-Add-magic-and-size-to-the-trace-header.patch
>> 0018-Compute-reginfo_size-based-on-the-reginfo.patch
>> 0019-aarch64-Reorg-sve-reginfo-to-save-space.patch
>> 0020-aarch64-Use-arch_init-to-configure-sve.patch
>> 0021-ppc64-Use-uint64_t-to-represent-double.patch
>> 0022-Standardize-reginfo_dump_mismatch-printing.patch
>> 0023-Add-fulldump-and-diffdup-options.patch
>> 0024-Remove-return-value-from-reginfo_dump.patch
>> 0025-ppc64-Clean-up-reginfo-handling.patch
>>
>> most of which are new, and those that aren't new have had
>> significant modifications.
>>
>>
>> r~
>>
>>
>> Richard Henderson (25):
>>   Use bool for tracing variables
>>   Unify master_fd and apprentice_fd to comm_fd
>>   Hoist trace file and socket opening
>>   Adjust tracefile open for write
>>   Use EXIT_FAILURE, EXIT_SUCCESS
>>   Make some risu.c symbols static
>>   Add enum RisuOp
>>   Add enum RisuResult
>>   Unify i/o functions and use RisuResult
>>   Pass non-OK result back through siglongjmp
>>   Always write for --master
>>   Simplify syncing with master
>>   Split RES_MISMATCH for registers and memory
>>   Merge reginfo.c into risu.c
>>   Rearrange reginfo and memblock buffers
>>   Split out recv_register_info
>>   Add magic and size to the trace header
>>   Compute reginfo_size based on the reginfo
>>   aarch64: Reorg sve reginfo to save space
>>   aarch64: Use arch_init to configure sve
>>   ppc64: Use uint64_t to represent double
>>   Standardize reginfo_dump_mismatch printing
>>   Add --fulldump and --diffdup options
>>   Remove return value from reginfo_dump
>>   ppc64: Clean up reginfo handling
>>
>>  Makefile               |   2 +-
>>  risu.h                 | 103 +++----
>>  risu_reginfo_aarch64.h |  16 +-
>>  risu_reginfo_ppc64.h   |   3 +-
>>  comms.c                |  34 +--
>>  reginfo.c              | 183 -----------
>>  risu.c                 | 676 ++++++++++++++++++++++++++++++-----------
>>  risu_aarch64.c         |   6 +-
>>  risu_arm.c             |   6 +-
>>  risu_i386.c            |   4 +-
>>  risu_m68k.c            |   4 +-
>>  risu_ppc64.c           |   4 +-
>>  risu_reginfo_aarch64.c | 212 +++++++------
>>  risu_reginfo_arm.c     |  32 +-
>>  risu_reginfo_i386.c    |  22 +-
>>  risu_reginfo_m68k.c    |  37 +--
>>  risu_reginfo_ppc64.c   | 183 +++++------
>>  17 files changed, 803 insertions(+), 724 deletions(-)
>>  delete mode 100644 reginfo.c


-- 
Alex Bennée


  reply	other threads:[~2020-06-23 13:31 UTC|newest]

Thread overview: 49+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-05-22  2:34 [PATCH v3 00/25] risu cleanups and improvements Richard Henderson
2020-05-22  2:34 ` [PATCH v3 01/25] Use bool for tracing variables Richard Henderson
2020-05-22  2:34 ` [PATCH v3 02/25] Unify master_fd and apprentice_fd to comm_fd Richard Henderson
2020-05-22  2:34 ` [PATCH v3 03/25] Hoist trace file and socket opening Richard Henderson
2020-05-22  2:34 ` [PATCH v3 04/25] Adjust tracefile open for write Richard Henderson
2020-05-22  2:34 ` [PATCH v3 05/25] Use EXIT_FAILURE, EXIT_SUCCESS Richard Henderson
2020-05-22  2:34 ` [PATCH v3 06/25] Make some risu.c symbols static Richard Henderson
2020-05-22  2:34 ` [PATCH v3 07/25] Add enum RisuOp Richard Henderson
2020-05-22  2:34 ` [PATCH v3 08/25] Add enum RisuResult Richard Henderson
2020-05-22  2:34 ` [PATCH v3 09/25] Unify i/o functions and use RisuResult Richard Henderson
2020-05-22  2:34 ` [PATCH v3 10/25] Pass non-OK result back through siglongjmp Richard Henderson
2020-05-22  2:34 ` [PATCH v3 11/25] Always write for --master Richard Henderson
2020-05-22  2:34 ` [PATCH v3 12/25] Simplify syncing with master Richard Henderson
2020-05-22  2:34 ` [PATCH v3 13/25] Split RES_MISMATCH for registers and memory Richard Henderson
2020-05-22  2:34 ` [PATCH v3 14/25] Merge reginfo.c into risu.c Richard Henderson
2020-06-23  8:33   ` Alex Bennée
2020-05-22  2:34 ` [PATCH v3 15/25] Rearrange reginfo and memblock buffers Richard Henderson
2020-06-23  8:47   ` Alex Bennée
2020-05-22  2:34 ` [PATCH v3 16/25] Split out recv_register_info Richard Henderson
2020-06-23 10:38   ` Alex Bennée
2020-05-22  2:34 ` [PATCH v3 17/25] Add magic and size to the trace header Richard Henderson
2020-06-23 14:52   ` Alex Bennée
2020-05-22  2:34 ` [PATCH v3 18/25] Compute reginfo_size based on the reginfo Richard Henderson
2020-06-23 14:54   ` Alex Bennée
2020-05-22  2:34 ` [PATCH v3 19/25] aarch64: Reorg sve reginfo to save space Richard Henderson
2020-06-23 16:32   ` Alex Bennée
2020-05-22  2:34 ` [PATCH v3 20/25] aarch64: Use arch_init to configure sve Richard Henderson
2020-06-23 16:54   ` Alex Bennée
2020-05-22  2:34 ` [PATCH v3 21/25] ppc64: Use uint64_t to represent double Richard Henderson
2020-06-23 16:58   ` Alex Bennée
2020-05-22  2:34 ` [PATCH v3 22/25] Standardize reginfo_dump_mismatch printing Richard Henderson
2020-06-23 17:03   ` Alex Bennée
2020-05-22  2:34 ` [PATCH v3 23/25] Add --fulldump and --diffdup options Richard Henderson
2020-05-22  2:34 ` [PATCH v3 24/25] Remove return value from reginfo_dump Richard Henderson
2020-06-23 17:38   ` Alex Bennée
2020-05-22  2:34 ` [PATCH v3 25/25] ppc64: Clean up reginfo handling Richard Henderson
2020-05-22  4:10   ` Richard Henderson
2020-06-23 17:45   ` Alex Bennée
2020-06-23 18:06     ` Richard Henderson
2020-06-22 19:17 ` [PATCH v3 00/25] risu cleanups and improvements Peter Maydell
2020-06-23  8:23   ` Alex Bennée
2020-06-23  9:00 ` Alex Bennée
2020-06-23 13:30   ` Alex Bennée [this message]
2020-06-23 20:26   ` Richard Henderson
2020-06-23 14:44 ` [PATCH] risu: don't do a full register compare for OP_SIGILL Alex Bennée
2020-06-23 15:23   ` Richard Henderson
2020-06-23 15:54   ` Peter Maydell
2020-06-23 16:17     ` Alex Bennée
2020-06-23 19:55       ` Richard Henderson

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=87mu4t9af1.fsf@linaro.org \
    --to=alex.bennee@linaro.org \
    --cc=qemu-devel@nongnu.org \
    --cc=richard.henderson@linaro.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).