OSDN Git Service

qmiga/qemu.git
19 months agocrypto: strengthen the check for key slots overlapping with LUKS header
Daniel P. Berrangé [Mon, 5 Sep 2022 12:57:01 +0000 (13:57 +0100)]
crypto: strengthen the check for key slots overlapping with LUKS header

The LUKS header data on disk is a fixed size, however, there's expected
to be a gap between the end of the header and the first key slot to get
alignment with the 2nd sector on 4k drives. This wasn't originally part
of the LUKS spec, but was always part of the reference implementation,
so it is worth validating this.

Reviewed-by: Richard W.M. Jones <rjones@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
19 months agocrypto: validate that LUKS payload doesn't overlap with header
Daniel P. Berrangé [Mon, 5 Sep 2022 12:50:03 +0000 (13:50 +0100)]
crypto: validate that LUKS payload doesn't overlap with header

We already validate that LUKS keyslots don't overlap with the
header, or with each other. This closes the remaining hole in
validation of LUKS file regions.

Reviewed-by: Richard W.M. Jones <rjones@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
19 months agocrypto: enforce that key material doesn't overlap with LUKS header
Daniel P. Berrangé [Tue, 10 May 2022 13:35:57 +0000 (14:35 +0100)]
crypto: enforce that key material doesn't overlap with LUKS header

We already check that key material doesn't overlap between key slots,
and that it doesn't overlap with the payload. We didn't check for
overlap with the LUKS header.

Reviewed-by: Richard W.M. Jones <rjones@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
19 months agocrypto: enforce that LUKS stripes is always a fixed value
Daniel P. Berrangé [Tue, 10 May 2022 13:27:33 +0000 (14:27 +0100)]
crypto: enforce that LUKS stripes is always a fixed value

Although the LUKS stripes are encoded in the keyslot header and so
potentially configurable, in pratice the cryptsetup impl mandates
this has the fixed value 4000. To avoid incompatibility apply the
same enforcement in QEMU too. This also caps the memory usage for
key material when QEMU tries to open a LUKS volume.

Reviewed-by: Richard W.M. Jones <rjones@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
19 months agocrypto: sanity check that LUKS header strings are NUL-terminated
Daniel P. Berrangé [Tue, 10 May 2022 13:17:43 +0000 (14:17 +0100)]
crypto: sanity check that LUKS header strings are NUL-terminated

The LUKS spec requires that header strings are NUL-terminated, and our
code relies on that. Protect against maliciously crafted headers by
adding validation.

Reviewed-by: Richard W.M. Jones <rjones@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
19 months agotests: avoid DOS line endings in PSK file
Daniel P. Berrangé [Wed, 28 Sep 2022 14:57:24 +0000 (15:57 +0100)]
tests: avoid DOS line endings in PSK file

Using FILE * APIs for writing the PSK file results in translation from
UNIX to DOS line endings on Windows. When the crypto PSK code later
loads the credentials the stray \r will result in failure to load the
PSK credentials into GNUTLS.

Rather than switching the FILE* APIs to open in binary format, just
switch to the more concise g_file_set_contents API.

Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
19 months agocrypto: check for and report errors setting PSK credentials
Daniel P. Berrangé [Wed, 28 Sep 2022 14:56:42 +0000 (15:56 +0100)]
crypto: check for and report errors setting PSK credentials

If setting credentials fails, the handshake will later fail to complete
with an obscure error message which is hard to diagnose.

Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
19 months agoscripts: check if .git exists before checking submodule status
Daniel P. Berrangé [Tue, 26 Apr 2022 13:06:49 +0000 (14:06 +0100)]
scripts: check if .git exists before checking submodule status

Currently we check status of each submodule, before actually checking
if we're in a git repo. These status commands will all fail, but we
are hiding their output so we don't see it currently.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
19 months agoseccomp: Get actual errno value from failed seccomp functions
Michal Privoznik [Wed, 26 Oct 2022 07:30:24 +0000 (09:30 +0200)]
seccomp: Get actual errno value from failed seccomp functions

Upon failure, a libseccomp API returns actual errno value very
rarely. Fortunately, after its commit 34bf78ab (contained in
2.5.0 release), the SCMP_FLTATR_API_SYSRAWRC attribute can be set
which makes subsequent APIs return true errno on failure.

This is especially critical when seccomp_load() fails, because
generic -ECANCELED says nothing.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
19 months agoio/channel-watch: Fix socket watch on Windows
Bin Meng [Thu, 6 Oct 2022 15:19:25 +0000 (23:19 +0800)]
io/channel-watch: Fix socket watch on Windows

Random failure was observed when running qtests on Windows due to
"Broken pipe" detected by qmp_fd_receive(). What happened is that
the qtest executable sends testing data over a socket to the QEMU
under test but no response is received. The errno of the recv()
call from the qtest executable indicates ETIMEOUT, due to the qmp
chardev's tcp_chr_read() is never called to receive testing data
hence no response is sent to the other side.

tcp_chr_read() is registered as the callback of the socket watch
GSource. The reason of the callback not being called by glib, is
that the source check fails to indicate the source is ready. There
are two socket watch sources created to monitor the same socket
event object from the char-socket backend in update_ioc_handlers().
During the source check phase, qio_channel_socket_source_check()
calls WSAEnumNetworkEvents() to discover occurrences of network
events for the indicated socket, clear internal network event records,
and reset the event object. Testing shows that if we don't reset the
event object by not passing the event handle to WSAEnumNetworkEvents()
the symptom goes away and qtest runs very stably.

It seems we don't need to call WSAEnumNetworkEvents() at all, as we
don't parse the result of WSANETWORKEVENTS returned from this API.
We use select() to poll the socket status. Fix this instability by
dropping the WSAEnumNetworkEvents() call.

Some side notes:

During the testing, I removed the following codes in update_ioc_handlers():

  remove_hup_source(s);
  s->hup_source = qio_channel_create_watch(s->ioc, G_IO_HUP);
  g_source_set_callback(s->hup_source, (GSourceFunc)tcp_chr_hup,
                        chr, NULL);
  g_source_attach(s->hup_source, chr->gcontext);

and such change also makes the symptom go away.

And if I moved the above codes to the beginning, before the call to
io_add_watch_poll(), the symptom also goes away.

It seems two sources watching on the same socket event object is
the key that leads to the instability. The order of adding a source
watch seems to also play a role but I can't explain why.
Hopefully a Windows and glib expert could explain this behavior.

Signed-off-by: Bin Meng <bin.meng@windriver.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
19 months agoio/channel-watch: Drop the unnecessary cast
Bin Meng [Thu, 6 Oct 2022 15:19:24 +0000 (23:19 +0800)]
io/channel-watch: Drop the unnecessary cast

There is no need to do a type cast on ssource->socket as it is
already declared as a SOCKET.

Suggested-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Bin Meng <bin.meng@windriver.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
19 months agoio/channel-watch: Drop a superfluous '#ifdef WIN32'
Bin Meng [Thu, 6 Oct 2022 15:19:23 +0000 (23:19 +0800)]
io/channel-watch: Drop a superfluous '#ifdef WIN32'

In the win32 version qio_channel_create_socket_watch() body there is
no need to do a '#ifdef WIN32'.

Signed-off-by: Bin Meng <bin.meng@windriver.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
19 months agoutil/qemu-sockets: Use g_get_tmp_dir() to get the directory for temporary files
Bin Meng [Tue, 27 Sep 2022 11:05:42 +0000 (19:05 +0800)]
util/qemu-sockets: Use g_get_tmp_dir() to get the directory for temporary files

Replace the existing logic to get the directory for temporary files
with g_get_tmp_dir(), which works for win32 too.

Signed-off-by: Bin Meng <bin.meng@windriver.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
19 months agocrypto/luks: Support creating LUKS image on Darwin
Jungmin Park [Wed, 24 Aug 2022 12:37:06 +0000 (21:37 +0900)]
crypto/luks: Support creating LUKS image on Darwin

When the user creates a LUKS-encrypted qcow2 image using the qemu-img
program, the passphrase is hashed using PBKDF2 with a dynamic
number of iterations. The number of iterations is determined by
measuring thread cpu time usage, such that it takes approximately
2 seconds to compute the hash.

Because Darwin doesn't implement getrusage(RUSAGE_THREAD), we get an
error message:
> qemu-img: test.qcow2: Unable to calculate thread CPU usage on this platform
for this command:
> qemu-img create --object secret,id=key,data=1234 -f qcow2 -o 'encrypt.format=luks,encrypt.key-secret=key' test.qcow2 100M

This patch implements qcrypto_pbkdf2_get_thread_cpu() for Darwin so that
the above command works.

Signed-off-by: Jungmin Park <pjm0616@gmail.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
19 months agoMerge tag 'pull-9p-20221024' of https://github.com/cschoenebeck/qemu into staging
Stefan Hajnoczi [Mon, 24 Oct 2022 18:27:12 +0000 (14:27 -0400)]
Merge tag 'pull-9p-20221024' of https://github.com/cschoenebeck/qemu into staging

9pfs: performance, Windows host prep, tests restructure

* Highlight of this PR is Linus Heckemann's GHashTable patch which
  brings massive general performance improvements of 9p server
  somewhere between factor 6 .. 12.

* Bin Meng's g_mkdir patch is a preparatory patch for upcoming
  Windows host support of 9p server.

* The rest of the patches in this PR are 9p test code restructuring
  and refactoring changes to improve readability and to ease
  maintenance of 9p test code on the long-term.

# -----BEGIN PGP SIGNATURE-----
#
# iQJLBAABCgA1FiEEltjREM96+AhPiFkBNMK1h2Wkc5UFAmNWbs8XHHFlbXVfb3Nz
# QGNydWRlYnl0ZS5jb20ACgkQNMK1h2Wkc5V4cw/8CqoSJqoJixlP8kAGDYWq3CgF
# SKd09rIzLSWyyufAoZr1TqLwRrvEQRlZJSpL4fGvRpQLv0IQCu4x59ohHRob25Tm
# Fe7IxYBNuBwLW4yu+Y7FaujeGoYAi9Qw5q4ijq3/aSSiIeuXySKB2JmW71CQ+Tbe
# uwivsnMtWzQ7qsNwrtXYbxDs7UGkdsiW2sEQUS26GMApAXZoB+38hwtTW2Y9MOrC
# 58JuZza/fUVPzo0V1D0ggRawb5O2VTF5fz8aGFG4FvoyIW6DDZFSfnyre9QxivOl
# 5McWwSQ/D04vdEK9ornGPYr9YRGuP8g07p1EW9OfKeie4I41e9pS3UminK5lVCgo
# SfBHzz96efM5XR+Wnl4yVKowivmTqjwUU8lDqW2eB/7YBRuYUzrpxYe//UPv4q1J
# zaQV3pgwFAVkVJCnkcLCa1JQbH581bXSsuRlDdYqoRYfyzXoxbywNjvn9BXE0PrG
# WRecS//GyN3GVZYxMwb3H052110pYsYIg2YZ2H4QiqCwpEHHvy+L/ZXm19vbDm7B
# GYJQPUK8/y0NGwZsUYcUSx1TWlU9ZPwrbqZfv7e7+B6FL4VNjdaqb8PvS9admWSq
# LOSzrVVIus+nb7tP99d1Fb6oRyCy3x8E48gTr5UtTJHC4SAw/OBJmem6GOc/D490
# H7Dq8Y27qsQ6fT7iPm8=
# =MxSG
# -----END PGP SIGNATURE-----
# gpg: Signature made Mon 24 Oct 2022 06:54:07 EDT
# gpg:                using RSA key 96D8D110CF7AF8084F88590134C2B58765A47395
# gpg:                issuer "qemu_oss@crudebyte.com"
# gpg: Good signature from "Christian Schoenebeck <qemu_oss@crudebyte.com>" [unknown]
# gpg: WARNING: The key's User ID is not certified with a trusted signature!
# gpg:          There is no indication that the signature belongs to the owner.
# Primary key fingerprint: ECAB 1A45 4014 1413 BA38  4926 30DB 47C3 A012 D5F4
#      Subkey fingerprint: 96D8 D110 CF7A F808 4F88  5901 34C2 B587 65A4 7395

* tag 'pull-9p-20221024' of https://github.com/cschoenebeck/qemu: (23 commits)
  tests/9p: remove unnecessary g_strdup() calls
  tests/9p: merge v9fs_tunlinkat() and do_unlinkat()
  tests/9p: merge v9fs_tlink() and do_hardlink()
  tests/9p: merge v9fs_tsymlink() and do_symlink()
  tests/9p: merge v9fs_tlcreate() and do_lcreate()
  tests/9p: merge v9fs_tmkdir() and do_mkdir()
  tests/9p: convert v9fs_tflush() to declarative arguments
  tests/9p: simplify callers of twrite()
  tests/9p: convert v9fs_twrite() to declarative arguments
  tests/9p: simplify callers of tlopen()
  tests/9p: convert v9fs_tlopen() to declarative arguments
  tests/9p: simplify callers of treaddir()
  tests/9p: convert v9fs_treaddir() to declarative arguments
  tests/9p: simplify callers of tgetattr()
  tests/9p: convert v9fs_tgetattr() to declarative arguments
  tests/9p: simplify callers of tattach()
  tests/9p: merge v9fs_tattach(), do_attach(), do_attach_rqid()
  tests/9p: merge v9fs_tversion() and do_version()
  tests/9p: simplify callers of twalk()
  tests/9p: merge *walk*() functions
  ...

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
19 months agoMerge tag 'for-upstream' of https://gitlab.com/bonzini/qemu into staging
Stefan Hajnoczi [Mon, 24 Oct 2022 18:27:06 +0000 (14:27 -0400)]
Merge tag 'for-upstream' of https://gitlab.com/bonzini/qemu into staging

* target/i386: new decoder bugfix
* target/i386: complete x86-v3 support for TCG

# -----BEGIN PGP SIGNATURE-----
#
# iQFHBAABCAAyFiEE8TM4V0tmI4mGbHaCv/vSX3jHroMFAmNTlqQUHHBib256aW5p
# QHJlZGhhdC5jb20ACgkQv/vSX3jHroOQNQf430MHbrtN9WKKiXv3684XxmcnoRqg
# PHmaGg2SKp7UB+hI2FMYgCZWOl5s3cGTHtwX8byFCttmE4kI7HJR7IouW6znm57j
# 7QVx2TJXIZgqSYcfYzfLu46yS6pNqJUA+mBv5In3Vqt4ZQT2szefVBg6BzmuF6lT
# HXbu/llc3iVfW4SNLJOABXzKNbPacmmpmLjoporfwOHwHjv4iikuXNUOZ84FFL11
# 2tkdcff282q00IRgHm1lSyiRiqh+kAxzSDanMjOZbphBiE9gNJjLGoV5F2X63e1O
# DQGg4wqBWP68O/r8Fj8tOUMCTW212DwWyv1+d/lQB+wwpJK+P4O14dCW
# =Fd+y
# -----END PGP SIGNATURE-----
# gpg: Signature made Sat 22 Oct 2022 03:07:16 EDT
# gpg:                using RSA key F13338574B662389866C7682BFFBD25F78C7AE83
# gpg:                issuer "pbonzini@redhat.com"
# gpg: Good signature from "Paolo Bonzini <bonzini@gnu.org>" [full]
# gpg:                 aka "Paolo Bonzini <pbonzini@redhat.com>" [full]
# Primary key fingerprint: 46F5 9FBD 57D6 12E7 BFD4  E2F7 7E15 100C CD36 69B1
#      Subkey fingerprint: F133 3857 4B66 2389 866C  7682 BFFB D25F 78C7 AE83

* tag 'for-upstream' of https://gitlab.com/bonzini/qemu:
  target/i386: implement FMA instructions
  target/i386: implement F16C instructions
  target/i386: introduce function to set rounding mode from FPCW or MXCSR bits
  target/i386: decode-new: avoid out-of-bounds access to xmm_regs[-1]

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
19 months agoMerge tag 'm68k-for-7.2-pull-request' of https://github.com/vivier/qemu-m68k into...
Stefan Hajnoczi [Mon, 24 Oct 2022 18:26:57 +0000 (14:26 -0400)]
Merge tag 'm68k-for-7.2-pull-request' of https://github.com/vivier/qemu-m68k into staging

Pull request m68k branch 20221024

Update rng seed boot parameter

# -----BEGIN PGP SIGNATURE-----
#
# iQJGBAABCAAwFiEEzS913cjjpNwuT1Fz8ww4vT8vvjwFAmNWU7YSHGxhdXJlbnRA
# dml2aWVyLmV1AAoJEPMMOL0/L748C/kQAJHbVrDGCCfw+imsOAOqXBL3K0WpUadP
# T+lCIHpqCPsppJ2ikQlZ95ky0RZREPiRwmwJHQsgaTiVJgKDGJ3hsDuGNAyu3sn4
# C/QFVyH2E7dcrDjSNk+n1nGVw+h6gnH9G6qRYrJZHlQXN9WsaMrcofhjDdPUC9bK
# o2IexKWBs80wJ/k/O9be8hPARkdwbVYRbR3+khtcUVv7sgeoKfUmpUiliIhi1G3G
# jphHEMgWtjyK8HmTyeOujmbOO4RZGhScnCJpUnYitBmun6YWedP+DrOz5kXU5xgP
# vIJOalpPqQt+Fpe6OF5fBXwLl0QNPlLjhBP+YXe6inKWHk3RYZRYbFNz4+HZt6WY
# i5wO0pz77vzL6+7nnESS7Q9g/8ASQcUgOsswUBY2uhQoGUG8LuUl2O5SyJWGB41o
# MjNq+h7PcFD8MalLMMvf5jKIuZ0lJrkirUXbTXlGu19ynyf0lASY773cENch+ApS
# fu2De777qwYDx3CvR+K5oObiTqUTplQT9Mq+yUlD5mgJs1bCP+XPFGlL+JemgR25
# EjV5kLDY2f5i0hb7ZQ8IEL3O4siW9GoHXBBFkLY3WPO36Nd71cfhzVv2jUaoSyom
# xsFsJ/TGBapdiDj2Y5DjZQWrUIsm3TgcwATLhEyH2+Bt4hLd+yI6s1m16vVUn74O
# UgMzybUqLwk0
# =B+rc
# -----END PGP SIGNATURE-----
# gpg: Signature made Mon 24 Oct 2022 04:58:30 EDT
# gpg:                using RSA key CD2F75DDC8E3A4DC2E4F5173F30C38BD3F2FBE3C
# gpg:                issuer "laurent@vivier.eu"
# gpg: Good signature from "Laurent Vivier <lvivier@redhat.com>" [full]
# gpg:                 aka "Laurent Vivier <laurent@vivier.eu>" [full]
# gpg:                 aka "Laurent Vivier (Red Hat) <lvivier@redhat.com>" [full]
# Primary key fingerprint: CD2F 75DD C8E3 A4DC 2E4F  5173 F30C 38BD 3F2F BE3C

* tag 'm68k-for-7.2-pull-request' of https://github.com/vivier/qemu-m68k:
  m68k: write bootinfo as rom section and re-randomize on reboot
  m68k: rework BI_VIRT_RNG_SEED as BI_RNG_SEED

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
19 months agotests/9p: remove unnecessary g_strdup() calls
Christian Schoenebeck [Tue, 4 Oct 2022 20:54:30 +0000 (22:54 +0200)]
tests/9p: remove unnecessary g_strdup() calls

This is a leftover from before the recent function merge and
refactoring patches:

As these functions do not return control to the caller in
between, it is not necessary to duplicate strings passed to them.

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <0f80141cde3904ed0591354059da49d1d60bcdbc.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: merge v9fs_tunlinkat() and do_unlinkat()
Christian Schoenebeck [Tue, 4 Oct 2022 20:54:16 +0000 (22:54 +0200)]
tests/9p: merge v9fs_tunlinkat() and do_unlinkat()

As with previous patches, unify those 2 functions into a single function
v9fs_tunlinkat() by using a declarative function arguments approach.

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <1dea593edd464908d92501933c068388c01f1744.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: merge v9fs_tlink() and do_hardlink()
Christian Schoenebeck [Tue, 4 Oct 2022 20:54:14 +0000 (22:54 +0200)]
tests/9p: merge v9fs_tlink() and do_hardlink()

As with previous patches, unify those 2 functions into a single function
v9fs_tlink() by using a declarative function arguments approach.

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <cb4d42203e1e4e6027df4924bbe4bdbc002f668b.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: merge v9fs_tsymlink() and do_symlink()
Christian Schoenebeck [Tue, 4 Oct 2022 20:54:11 +0000 (22:54 +0200)]
tests/9p: merge v9fs_tsymlink() and do_symlink()

As with previous patches, unify those 2 functions into a single function
v9fs_tsymlink() by using a declarative function arguments approach.

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <563f3ad04fe596ce0ae1e2654d1d08237f18c830.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: merge v9fs_tlcreate() and do_lcreate()
Christian Schoenebeck [Tue, 4 Oct 2022 20:54:09 +0000 (22:54 +0200)]
tests/9p: merge v9fs_tlcreate() and do_lcreate()

As with previous patches, unify those 2 functions into a single function
v9fs_tlcreate() by using a declarative function arguments approach.

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <4c01b2caa5f5b54a2020fc92701deadd2abf0571.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: merge v9fs_tmkdir() and do_mkdir()
Christian Schoenebeck [Tue, 4 Oct 2022 20:54:06 +0000 (22:54 +0200)]
tests/9p: merge v9fs_tmkdir() and do_mkdir()

As with previous patches, unify those 2 functions into a single function
v9fs_tmkdir() by using a declarative function arguments approach.

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <b87b2c972921df980440ff5b2d3e6bb8163d6551.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: convert v9fs_tflush() to declarative arguments
Christian Schoenebeck [Tue, 4 Oct 2022 20:54:03 +0000 (22:54 +0200)]
tests/9p: convert v9fs_tflush() to declarative arguments

Use declarative function arguments for function v9fs_tflush().

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <91b7b154298c500d100b05137146c2905c3acdec.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: simplify callers of twrite()
Christian Schoenebeck [Tue, 4 Oct 2022 20:54:00 +0000 (22:54 +0200)]
tests/9p: simplify callers of twrite()

Now as twrite() is using a declarative approach, simplify the
code of callers of this function.

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <7f280ec6a1f9d8afed46567a796562c4dc28afa9.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: convert v9fs_twrite() to declarative arguments
Christian Schoenebeck [Tue, 4 Oct 2022 20:53:58 +0000 (22:53 +0200)]
tests/9p: convert v9fs_twrite() to declarative arguments

Use declarative function arguments for function v9fs_twrite().

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <be0326e2d9ab66f68c06b1766ddf103849d570b4.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: simplify callers of tlopen()
Christian Schoenebeck [Tue, 4 Oct 2022 20:53:55 +0000 (22:53 +0200)]
tests/9p: simplify callers of tlopen()

Now as tlopen() is using a declarative approach, simplify the
code of callers of this function.

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <f74b6153e079fc7a340e5cb575ee32e0fe1e0ae6.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: convert v9fs_tlopen() to declarative arguments
Christian Schoenebeck [Tue, 4 Oct 2022 20:53:52 +0000 (22:53 +0200)]
tests/9p: convert v9fs_tlopen() to declarative arguments

Use declarative function arguments for function v9fs_tlopen().

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <765ab515353c56f88f0a163631f626a44e9565d6.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: simplify callers of treaddir()
Christian Schoenebeck [Tue, 4 Oct 2022 20:53:49 +0000 (22:53 +0200)]
tests/9p: simplify callers of treaddir()

Now as treaddir() is using a declarative approach, simplify the
code of callers of this function.

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <7cec6f2c7011a481806c34908893b7282702a7a6.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: convert v9fs_treaddir() to declarative arguments
Christian Schoenebeck [Tue, 4 Oct 2022 20:53:47 +0000 (22:53 +0200)]
tests/9p: convert v9fs_treaddir() to declarative arguments

Use declarative function arguments for function v9fs_treaddir().

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <a66aae4ceb19ec12d245b8c7f33a639584c8e272.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: simplify callers of tgetattr()
Christian Schoenebeck [Tue, 4 Oct 2022 20:53:44 +0000 (22:53 +0200)]
tests/9p: simplify callers of tgetattr()

Now as tgetattr() is using a declarative approach, simplify the
code of callers of this function.

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <60c6a083f320b86f3172951445df7bbc895932e2.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: convert v9fs_tgetattr() to declarative arguments
Christian Schoenebeck [Tue, 4 Oct 2022 20:53:41 +0000 (22:53 +0200)]
tests/9p: convert v9fs_tgetattr() to declarative arguments

Use declarative function arguments for function v9fs_tgetattr().

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <d340a91be96fbfecfb8dacdd7558223b3c0d0e2c.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: simplify callers of tattach()
Christian Schoenebeck [Tue, 4 Oct 2022 20:53:38 +0000 (22:53 +0200)]
tests/9p: simplify callers of tattach()

Now as tattach() is using a declarative approach, simplify the
code of callers of this function.

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <9b50e5b89a0072e84a9191d18c19a53546a28bba.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: merge v9fs_tattach(), do_attach(), do_attach_rqid()
Christian Schoenebeck [Tue, 4 Oct 2022 20:53:36 +0000 (22:53 +0200)]
tests/9p: merge v9fs_tattach(), do_attach(), do_attach_rqid()

As with previous patches, unify those 3 functions into a single function
v9fs_tattach() by using a declarative function arguments approach.

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <a6756b30bf2a1b25729c5bbabd1c9534a8f20d6f.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: merge v9fs_tversion() and do_version()
Christian Schoenebeck [Tue, 4 Oct 2022 20:53:33 +0000 (22:53 +0200)]
tests/9p: merge v9fs_tversion() and do_version()

As with previous patches, unify functions v9fs_tversion() and do_version()
into a single function v9fs_tversion() by using a declarative function
arguments approach.

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <2d253491aaffd267ec295f056dda47456692cd0c.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: simplify callers of twalk()
Christian Schoenebeck [Tue, 4 Oct 2022 20:53:29 +0000 (22:53 +0200)]
tests/9p: simplify callers of twalk()

Now as twalk() is using a declarative approach, simplify the
code of callers of this function.

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <8b9d3c656ad43b6c953d6bdacd8d9f4c8e599b2a.1664917004.git.qemu_oss@crudebyte.com>

19 months agotests/9p: merge *walk*() functions
Christian Schoenebeck [Tue, 4 Oct 2022 20:53:23 +0000 (22:53 +0200)]
tests/9p: merge *walk*() functions

Introduce declarative function calls.

There are currently 4 different functions for sending a 9p 'Twalk'
request: v9fs_twalk(), do_walk(), do_walk_rqids() and
do_walk_expect_error(). They are all doing the same thing, just in a
slightly different way and with slightly different function arguments.

Merge those 4 functions into a single function by using a struct for
function call arguments and use designated initializers when calling
this function to turn usage into a declarative approach, which is
better readable and easier to maintain.

Also move private functions genfid(), split() and split_free() from
virtio-9p-test.c to virtio-9p-client.c.

Based-on: <E1odrya-0004Fv-97@lizzy.crudebyte.com>
Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <607969dbfbc63c1be008df9131133711b046e979.1664917004.git.qemu_oss@crudebyte.com>

19 months ago9pfs: use GHashTable for fid table
Linus Heckemann [Tue, 4 Oct 2022 10:41:21 +0000 (12:41 +0200)]
9pfs: use GHashTable for fid table

The previous implementation would iterate over the fid table for
lookup operations, resulting in an operation with O(n) complexity on
the number of open files and poor cache locality -- for every open,
stat, read, write, etc operation.

This change uses a hashtable for this instead, significantly improving
the performance of the 9p filesystem. The runtime of NixOS's simple
installer test, which copies ~122k files totalling ~1.8GiB from 9p,
decreased by a factor of about 10.

Signed-off-by: Linus Heckemann <git@sphalerite.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Greg Kurz <groug@kaod.org>
[CS: - Retain BUG_ON(f->clunked) in get_fid().
     - Add TODO comment in clunk_fid(). ]
Message-Id: <20221004104121.713689-1-git@sphalerite.org>
[CS: - Drop unnecessary goto and out: label. ]
Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
19 months agotests/9p: split virtio-9p-test.c into tests and 9p client part
Christian Schoenebeck [Thu, 29 Sep 2022 11:41:06 +0000 (13:41 +0200)]
tests/9p: split virtio-9p-test.c into tests and 9p client part

This patch is pure refactoring, it does not change behaviour.

virtio-9p-test.c grew to 1657 lines. Let's split this file up between
actual 9p test cases vs. 9p test client, to make it easier to
concentrate on the actual 9p tests.

Move the 9p test client code to a new unit virtio-9p-client.c, which
are basically all functions and types prefixed with v9fs_* already.

Note that some client wrapper functions (do_*) are preserved in
virtio-9p-test.c, simply because these wrapper functions are going to
be wiped with subsequent patches anyway.

As the global QGuestAllocator variable is moved to virtio-9p-client.c,
add a new function v9fs_set_allocator() to be used by virtio-9p-test.c
instead of fiddling with a global variable across units and libraries.

Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Message-Id: <E1odrya-0004Fv-97@lizzy.crudebyte.com>

19 months agofsdev/virtfs-proxy-helper: Use g_mkdir()
Bin Meng [Tue, 27 Sep 2022 11:06:04 +0000 (19:06 +0800)]
fsdev/virtfs-proxy-helper: Use g_mkdir()

Use g_mkdir() to create a directory on all platforms.

Signed-off-by: Bin Meng <bin.meng@windriver.com>
Reviewed-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Message-Id: <20220927110632.1973965-27-bmeng.cn@gmail.com>
Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
19 months agom68k: write bootinfo as rom section and re-randomize on reboot
Jason A. Donenfeld [Sun, 23 Oct 2022 19:13:41 +0000 (21:13 +0200)]
m68k: write bootinfo as rom section and re-randomize on reboot

Rather than poking directly into RAM, add the bootinfo block as a proper
ROM, so that it's restored when rebooting the system. This way, if the
guest corrupts any of the bootinfo items, but then tries to reboot,
it'll still be restored back to normal as expected.

Then, since the RNG seed needs to be fresh on each boot, regenerate the
RNG seed in the ROM when reseting the CPU.

Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Message-Id: <20221023191340.36238-1-Jason@zx2c4.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
19 months agotarget/i386: implement FMA instructions
Paolo Bonzini [Wed, 19 Oct 2022 11:22:06 +0000 (13:22 +0200)]
target/i386: implement FMA instructions

The only issue with FMA instructions is that there are _a lot_ of them (30
opcodes, each of which comes in up to 4 versions depending on VEX.W and
VEX.L; a total of 96 possibilities).  However, they can be implement with
only 6 helpers, two for scalar operations and four for packed operations.
(Scalar versions do not do any merging; they only affect the bottom 32
or 64 bits of the output operand.  Therefore, there is no separate XMM
and YMM of the scalar helpers).

First, we can reduce the number of helpers to one third by passing four
operands (one output and three inputs); the reordering of which operands
go to the multiply and which go to the add is done in emit.c.

Second, the different instructions also dispatch to the same softfloat
function, so the flags for float32_muladd and float64_muladd are passed
in the helper as int arguments, with a little extra complication to
handle FMADDSUB and FMSUBADD.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agom68k: rework BI_VIRT_RNG_SEED as BI_RNG_SEED
Jason A. Donenfeld [Mon, 26 Sep 2022 11:38:59 +0000 (13:38 +0200)]
m68k: rework BI_VIRT_RNG_SEED as BI_RNG_SEED

Following a change on the kernel side (see link), pass BI_RNG_SEED
instead of BI_VIRT_RNG_SEED. This should have no impact on
compatibility, as there will simply be no effect if it's an old kernel,
which is how things have always been. We then use this as an opportunity
to add this to q800, since now we can, which is a nice improvement.

Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Laurent Vivier <laurent@vivier.eu>
Link: https://lore.kernel.org/lkml/20220923170340.4099226-3-Jason@zx2c4.com/
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Message-Id: <20220926113900.1256630-1-Jason@zx2c4.com>
[lv: s/^I/         /g]
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
19 months agoMerge tag 'pull-target-arm-20221020' of https://git.linaro.org/people/pmaydell/qemu...
Stefan Hajnoczi [Thu, 20 Oct 2022 18:36:12 +0000 (14:36 -0400)]
Merge tag 'pull-target-arm-20221020' of https://git.linaro.org/people/pmaydell/qemu-arm into staging

target-arm queue:
 * Switch to TARGET_TB_PCREL
 * More pagetable-walk refactoring preparatory to HAFDBS
 * update the cortex-a15 MIDR to latest rev
 * hw/char/pl011: fix baud rate calculation
 * hw/ide/microdrive: Use device_cold_reset() for self-resets

# -----BEGIN PGP SIGNATURE-----
#
# iQJNBAABCAA3FiEE4aXFk81BneKOgxXPPCUl7RQ2DN4FAmNRPQ4ZHHBldGVyLm1h
# eWRlbGxAbGluYXJvLm9yZwAKCRA8JSXtFDYM3l/BD/0ZAUmL9FtLHUtf9Y7S1sb/
# nuKcbaXim88AWRgSpiCJaSCIiyxNPydrRwfC1wZ7kQEXBKczR9x3ZsE8jdU6lway
# muqaLdzpBKVYZSh0Q03+3iufbm7s3hGxvCvNj0s5986ITmW86ndZM9CGm4bkyB6B
# KKN7v7lGPeYCF/sR+Ax/++eXK2BOALoPFF7bWycmJNcO4aYXE12onDNIDJU9CyCu
# y+nqPXTM5xAJ950iCOmdGKnMwOX1sBfZYUvCuG4tS63SYnjnKThyfLDttggapTB0
# T4I1SZ22+Bh7UuGEn9JAIeVZK4UUkYu78ufnX7FT2NDOFJJV1BJtKRe8UT/CsZha
# n2eAuVcddBwEStAhgUIvpbDm6WjtH3xiQWPphFX9bZl4AD1uI3sx/xe1b7PsPBTc
# e8kYRVW8zkRlCbC2BaJnPhcDiujf6AE2Sq3A8hgG1izS6juvHzzk9zn69TTbodIl
# ETtt+jn9E/OtlLC7bLJGgudGerqBELubwwVAHqYo207fvhLLiMMAhm0NbZmnJCqU
# hEgg6a9w4Ss3CYx5oNagWZ2Nfw5ru5KiwjZJ1wltjbBEQzp4y42a4XFWRvcXFFiz
# 3yGyvLmtHBywsIRpMc8IpXLC1SdD0fgT8/5VbWCHRziFjQLNCiuihsKkdiZOcJ22
# MoetN7iJxSDPCbHUp8ldsg==
# =NEl6
# -----END PGP SIGNATURE-----
# gpg: Signature made Thu 20 Oct 2022 08:20:30 EDT
# gpg:                using RSA key E1A5C593CD419DE28E8315CF3C2525ED14360CDE
# gpg:                issuer "peter.maydell@linaro.org"
# gpg: Good signature from "Peter Maydell <peter.maydell@linaro.org>" [full]
# gpg:                 aka "Peter Maydell <pmaydell@gmail.com>" [full]
# gpg:                 aka "Peter Maydell <pmaydell@chiark.greenend.org.uk>" [full]
# gpg:                 aka "Peter Maydell <peter@archaic.org.uk>" [unknown]
# Primary key fingerprint: E1A5 C593 CD41 9DE2 8E83  15CF 3C25 25ED 1436 0CDE

* tag 'pull-target-arm-20221020' of https://git.linaro.org/people/pmaydell/qemu-arm: (24 commits)
  hw/ide/microdrive: Use device_cold_reset() for self-resets
  target/arm: Enable TARGET_TB_PCREL
  target/arm: Introduce gen_pc_plus_diff for aarch32
  target/arm: Introduce gen_pc_plus_diff for aarch64
  target/arm: Change gen_jmp* to work on displacements
  target/arm: Remove gen_exception_internal_insn pc argument
  target/arm: Change gen_exception_insn* to work on displacements
  target/arm: Change gen_*set_pc_im to gen_*update_pc
  target/arm: Change gen_goto_tb to work on displacements
  target/arm: Introduce curr_insn_len
  target/arm: Use bool consistently for get_phys_addr subroutines
  target/arm: Split out get_phys_addr_twostage
  target/arm: Use softmmu tlbs for page table walking
  target/arm: Move be test for regime into S1TranslateResult
  target/arm: Plumb debug into S1Translate
  target/arm: Split out S1Translate type
  target/arm: Restrict tlb flush from vttbr_write to vmid change
  target/arm: Move ARMMMUIdx_Stage2 to a real tlb mmu_idx
  target/arm: Add ARMMMUIdx_Phys_{S,NS}
  target/arm: Use probe_access_full for BTI
  ...

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
19 months agotarget/i386: implement F16C instructions
Paolo Bonzini [Wed, 19 Oct 2022 11:22:06 +0000 (13:22 +0200)]
target/i386: implement F16C instructions

F16C only consists of two instructions, which are a bit peculiar
nevertheless.

First, they access only the low half of an YMM or XMM register for the
packed-half operand; the exact size still depends on the VEX.L flag.
This is similar to the existing avx_movx flag, but not exactly because
avx_movx is hardcoded to affect operand 2.  To this end I added a "ph"
format name; it's possible to reuse this approach for the VPMOVSX and
VPMOVZX instructions, though that would also require adding two more
formats for the low-quarter and low-eighth of an operand.

Second, VCVTPS2PH is somewhat weird because it *stores* the result of
the instruction into memory rather than loading it.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: introduce function to set rounding mode from FPCW or MXCSR bits
Paolo Bonzini [Wed, 19 Oct 2022 12:01:36 +0000 (14:01 +0200)]
target/i386: introduce function to set rounding mode from FPCW or MXCSR bits

VROUND, FSTCW and STMXCSR all have to perform the same conversion from
x86 rounding modes to softfloat constants.  Since the ISA is consistent
on the meaning of the two-bit rounding modes, extract the common code
into a wrapper for set_float_rounding_mode.

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: decode-new: avoid out-of-bounds access to xmm_regs[-1]
Paolo Bonzini [Wed, 19 Oct 2022 12:32:04 +0000 (14:32 +0200)]
target/i386: decode-new: avoid out-of-bounds access to xmm_regs[-1]

If the destination is a memory register, op->n is -1.  Going through
tcg_gen_gvec_dup_imm path is both useless (the value has been stored
by the gen_* function already) and wrong because of the out-of-bounds
access.

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agohw/ide/microdrive: Use device_cold_reset() for self-resets
Peter Maydell [Thu, 13 Oct 2022 17:40:42 +0000 (18:40 +0100)]
hw/ide/microdrive: Use device_cold_reset() for self-resets

Currently the microdrive code uses device_legacy_reset() to reset
itself, and has its reset method call reset on the IDE bus as the
last thing it does.  Switch to using device_cold_reset().

The only concrete microdrive device is the TYPE_DSCM1XXXX; it is not
command-line pluggable, so it is used only by the old pxa2xx Arm
boards 'akita', 'borzoi', 'spitz', 'terrier' and 'tosa'.

You might think that this would result in the IDE bus being
reset automatically, but it does not, because the IDEBus type
does not set the BusClass::reset method. Instead the controller
must explicitly call ide_bus_reset(). We therefore leave that
call in md_reset().

Note also that because the PCMCIA card device is a direct subclass of
TYPE_DEVICE and we don't model the PCMCIA controller-to-card
interface as a qbus, PCMCIA cards are not on any qbus and so they
don't get reset when the system is reset.  The reset only happens via
the dscm1xxxx_attach() and dscm1xxxx_detach() functions during
machine creation.

Because our aim here is merely to try to get rid of calls to the
device_legacy_reset() function, we leave these other dubious
reset-related issues alone.  (They all stem from this code being
absolutely ancient.)

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-id: 20221013174042.1602926-1-peter.maydell@linaro.org

19 months agotarget/arm: Enable TARGET_TB_PCREL
Richard Henderson [Thu, 20 Oct 2022 03:06:41 +0000 (13:06 +1000)]
target/arm: Enable TARGET_TB_PCREL

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221020030641.2066807-10-richard.henderson@linaro.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Introduce gen_pc_plus_diff for aarch32
Richard Henderson [Thu, 20 Oct 2022 03:06:40 +0000 (13:06 +1000)]
target/arm: Introduce gen_pc_plus_diff for aarch32

In preparation for TARGET_TB_PCREL, reduce reliance on absolute values.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221020030641.2066807-9-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Introduce gen_pc_plus_diff for aarch64
Richard Henderson [Thu, 20 Oct 2022 03:06:39 +0000 (13:06 +1000)]
target/arm: Introduce gen_pc_plus_diff for aarch64

In preparation for TARGET_TB_PCREL, reduce reliance on absolute values.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221020030641.2066807-8-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Change gen_jmp* to work on displacements
Richard Henderson [Thu, 20 Oct 2022 03:06:38 +0000 (13:06 +1000)]
target/arm: Change gen_jmp* to work on displacements

In preparation for TARGET_TB_PCREL, reduce reliance on absolute values.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221020030641.2066807-7-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Remove gen_exception_internal_insn pc argument
Richard Henderson [Thu, 20 Oct 2022 03:06:37 +0000 (13:06 +1000)]
target/arm: Remove gen_exception_internal_insn pc argument

In preparation for TARGET_TB_PCREL, reduce reliance on absolute values.
Since we always pass dc->pc_curr, fold the arithmetic to zero displacement.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221020030641.2066807-6-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Change gen_exception_insn* to work on displacements
Richard Henderson [Thu, 20 Oct 2022 03:06:36 +0000 (13:06 +1000)]
target/arm: Change gen_exception_insn* to work on displacements

In preparation for TARGET_TB_PCREL, reduce reliance on absolute values.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221020030641.2066807-5-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Change gen_*set_pc_im to gen_*update_pc
Richard Henderson [Thu, 20 Oct 2022 03:06:35 +0000 (13:06 +1000)]
target/arm: Change gen_*set_pc_im to gen_*update_pc

In preparation for TARGET_TB_PCREL, reduce reliance on
absolute values by passing in pc difference.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221020030641.2066807-4-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Change gen_goto_tb to work on displacements
Richard Henderson [Thu, 20 Oct 2022 03:06:34 +0000 (13:06 +1000)]
target/arm: Change gen_goto_tb to work on displacements

In preparation for TARGET_TB_PCREL, reduce reliance on absolute values.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221020030641.2066807-3-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Introduce curr_insn_len
Richard Henderson [Thu, 20 Oct 2022 03:06:33 +0000 (13:06 +1000)]
target/arm: Introduce curr_insn_len

A simple helper to retrieve the length of the current insn.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221020030641.2066807-2-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Use bool consistently for get_phys_addr subroutines
Richard Henderson [Tue, 11 Oct 2022 03:18:59 +0000 (20:18 -0700)]
target/arm: Use bool consistently for get_phys_addr subroutines

The return type of the functions is already bool, but in a few
instances we used an integer type with the return statement.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221011031911.2408754-13-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Split out get_phys_addr_twostage
Richard Henderson [Tue, 11 Oct 2022 03:18:58 +0000 (20:18 -0700)]
target/arm: Split out get_phys_addr_twostage

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221011031911.2408754-12-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Use softmmu tlbs for page table walking
Richard Henderson [Tue, 11 Oct 2022 03:18:57 +0000 (20:18 -0700)]
target/arm: Use softmmu tlbs for page table walking

So far, limit the change to S1_ptw_translate, arm_ldl_ptw, and
arm_ldq_ptw.  Use probe_access_full to find the host address,
and if so use a host load.  If the probe fails, we've got our
fault info already.  On the off chance that page tables are not
in RAM, continue to use the address_space_ld* functions.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221011031911.2408754-11-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Move be test for regime into S1TranslateResult
Richard Henderson [Tue, 11 Oct 2022 03:18:56 +0000 (20:18 -0700)]
target/arm: Move be test for regime into S1TranslateResult

Hoist this test out of arm_ld[lq]_ptw into S1_ptw_translate.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221011031911.2408754-10-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Plumb debug into S1Translate
Richard Henderson [Tue, 11 Oct 2022 03:18:55 +0000 (20:18 -0700)]
target/arm: Plumb debug into S1Translate

Before using softmmu page tables for the ptw, plumb down
a debug parameter so that we can query page table entries
from gdbstub without modifying cpu state.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221011031911.2408754-9-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Split out S1Translate type
Richard Henderson [Tue, 11 Oct 2022 03:18:54 +0000 (20:18 -0700)]
target/arm: Split out S1Translate type

Consolidate most of the inputs and outputs of S1_ptw_translate
into a single structure.  Plumb this through arm_ld*_ptw from
the controlling get_phys_addr_* routine.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 20221011031911.2408754-8-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Restrict tlb flush from vttbr_write to vmid change
Richard Henderson [Tue, 11 Oct 2022 03:18:53 +0000 (20:18 -0700)]
target/arm: Restrict tlb flush from vttbr_write to vmid change

Compare only the VMID field when considering whether we need to flush.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 20221011031911.2408754-7-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Move ARMMMUIdx_Stage2 to a real tlb mmu_idx
Richard Henderson [Tue, 11 Oct 2022 03:18:52 +0000 (20:18 -0700)]
target/arm: Move ARMMMUIdx_Stage2 to a real tlb mmu_idx

We had been marking this ARM_MMU_IDX_NOTLB, move it to a real tlb.
Flush the tlb when invalidating stage 1+2 translations.  Re-use
alle1_tlbmask() for other instances of EL1&0 + Stage2.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 20221011031911.2408754-6-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Add ARMMMUIdx_Phys_{S,NS}
Richard Henderson [Tue, 11 Oct 2022 03:18:51 +0000 (20:18 -0700)]
target/arm: Add ARMMMUIdx_Phys_{S,NS}

Not yet used, but add mmu indexes for 1-1 mapping
to physical addresses.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221011031911.2408754-5-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Use probe_access_full for BTI
Richard Henderson [Tue, 11 Oct 2022 03:18:50 +0000 (20:18 -0700)]
target/arm: Use probe_access_full for BTI

Add a field to TARGET_PAGE_ENTRY_EXTRA to hold the guarded bit.
In is_guarded_page, use probe_access_full instead of just guessing
that the tlb entry is still present.  Also handles the FIXME about
executing from device memory.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221011031911.2408754-4-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Use probe_access_full for MTE
Richard Henderson [Tue, 11 Oct 2022 03:18:49 +0000 (20:18 -0700)]
target/arm: Use probe_access_full for MTE

The CPUTLBEntryFull structure now stores the original pte attributes, as
well as the physical address.  Therefore, we no longer need a separate
bit in MemTxAttrs, nor do we need to walk the tree of memory regions.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221011031911.2408754-3-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: Enable TARGET_PAGE_ENTRY_EXTRA
Richard Henderson [Tue, 11 Oct 2022 03:18:48 +0000 (20:18 -0700)]
target/arm: Enable TARGET_PAGE_ENTRY_EXTRA

Copy attrs and shareability, into the TLB.  This will eventually
be used by S1_ptw_translate to report stage1 translation failures,
and by do_ats_write to fill in PAR_EL1.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20221011031911.2408754-2-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agotarget/arm: update the cortex-a15 MIDR to latest rev
Alex Bennée [Mon, 10 Oct 2022 15:32:25 +0000 (16:32 +0100)]
target/arm: update the cortex-a15 MIDR to latest rev

QEMU doesn't model micro-architectural details which includes most
chip errata. The ARM_ERRATA_798181 work around in the Linux
kernel (see erratum_a15_798181_init) currently detects QEMU's
cortex-a15 as broken and triggers additional expensive TLB flushes as
a result.

Change the MIDR to report what the latest silicon would (r4p0). We
explicitly set the IMPDEF revidr bits to 0 because we don't need to
set anything other than the silicon revision to indicate these flushes
are not needed. This cuts about 5s from my Debian kernel boot with the
latest 6.0rc1 kernel (29s->24s).

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Tested-by: Anders Roxell <anders.roxell@linaro.org>
Message-id: 20221010153225.506394-1-alex.bennee@linaro.org
Cc: Arnd Bergmann <arnd@linaro.org>
Cc: Anders Roxell <anders.roxell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Tested-by: Anders Roxell <anders.roxell@linaro.org>
Message-Id: <20220906172257.2776521-1-alex.bennee@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agohw/char/pl011: fix baud rate calculation
Baruch Siach [Thu, 6 Oct 2022 10:19:48 +0000 (13:19 +0300)]
hw/char/pl011: fix baud rate calculation

The PL011 TRM says that "UARTIBRD = 0 is invalid and UARTFBRD is ignored
when this is the case". But the code looks at FBRD for the invalid case.
Fix this.

Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Message-id: 1408f62a2e45665816527d4845ffde650957d5ab.1665051588.git.baruchs-c@neureality.ai
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
19 months agoMerge tag 'for-upstream' of https://gitlab.com/bonzini/qemu into staging
Stefan Hajnoczi [Tue, 18 Oct 2022 15:14:31 +0000 (11:14 -0400)]
Merge tag 'for-upstream' of https://gitlab.com/bonzini/qemu into staging

* configure: don't enable firmware for targets that are not built
* configure: don't use strings(1)
* scsi, target/i386: switch from device_legacy_reset() to device_cold_reset()
* target/i386: AVX support for TCG
* target/i386: fix SynIC SINT assertion failure on guest reset
* target/i386: Use atomic operations for pte updates and other cleanups
* tests/tcg: extend SSE tests to AVX
* virtio-scsi: send "REPORTED LUNS CHANGED" sense data upon disk hotplug events

# -----BEGIN PGP SIGNATURE-----
#
# iQFIBAABCAAyFiEE8TM4V0tmI4mGbHaCv/vSX3jHroMFAmNOlOcUHHBib256aW5p
# QHJlZGhhdC5jb20ACgkQv/vSX3jHroNuvwgAj/Z5pI9KU33XiWKFR3bZf2lHh21P
# xmTzNtPmnP1WHDY1DNug/UB+BLg3c+carpTf5n3B8aKI4X3FfxGSJvYlXy4BONFD
# XqYMH3OZB5GaR8Wza9trNYjDs/9hOZus/0R6Hqdl/T38PlMjf8mmayULJIGdcFcJ
# WJvITVntbcCwwbpyJbRC5BNigG8ZXTNRoKBgtFVGz6Ox+n0YydwKX5qU5J7xRfCU
# lW41LjZ0Fk5lonH16+xuS4WD5EyrNt8cMKCGsxnyxhI7nehe/OGnYr9l+xZJclrh
# inQlSwJv0IpUJcrGCI4Xugwux4Z7ZXv3JQ37FzsdZcv/ZXpGonXMeXNJ9A==
# =o6x7
# -----END PGP SIGNATURE-----
# gpg: Signature made Tue 18 Oct 2022 07:58:31 EDT
# gpg:                using RSA key F13338574B662389866C7682BFFBD25F78C7AE83
# gpg:                issuer "pbonzini@redhat.com"
# gpg: Good signature from "Paolo Bonzini <bonzini@gnu.org>" [full]
# gpg:                 aka "Paolo Bonzini <pbonzini@redhat.com>" [full]
# Primary key fingerprint: 46F5 9FBD 57D6 12E7 BFD4  E2F7 7E15 100C CD36 69B1
#      Subkey fingerprint: F133 3857 4B66 2389 866C  7682 BFFB D25F 78C7 AE83

* tag 'for-upstream' of https://gitlab.com/bonzini/qemu: (53 commits)
  target/i386: remove old SSE decoder
  target/i386: move 3DNow to the new decoder
  tests/tcg: extend SSE tests to AVX
  target/i386: Enable AVX cpuid bits when using TCG
  target/i386: implement VLDMXCSR/VSTMXCSR
  target/i386: implement XSAVE and XRSTOR of AVX registers
  target/i386: reimplement 0x0f 0x28-0x2f, add AVX
  target/i386: reimplement 0x0f 0x10-0x17, add AVX
  target/i386: reimplement 0x0f 0xc2, 0xc4-0xc6, add AVX
  target/i386: reimplement 0x0f 0x38, add AVX
  target/i386: Use tcg gvec ops for pmovmskb
  target/i386: reimplement 0x0f 0x3a, add AVX
  target/i386: clarify (un)signedness of immediates from 0F3Ah opcodes
  target/i386: reimplement 0x0f 0xd0-0xd7, 0xe0-0xe7, 0xf0-0xf7, add AVX
  target/i386: reimplement 0x0f 0x70-0x77, add AVX
  target/i386: reimplement 0x0f 0x78-0x7f, add AVX
  target/i386: reimplement 0x0f 0x50-0x5f, add AVX
  target/i386: reimplement 0x0f 0xd8-0xdf, 0xe8-0xef, 0xf8-0xff, add AVX
  target/i386: reimplement 0x0f 0x60-0x6f, add AVX
  target/i386: Introduce 256-bit vector helpers
  ...

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
19 months agoMerge tag 'pull-ppc-20221017' of https://gitlab.com/danielhb/qemu into staging
Stefan Hajnoczi [Tue, 18 Oct 2022 15:13:57 +0000 (11:13 -0400)]
Merge tag 'pull-ppc-20221017' of https://gitlab.com/danielhb/qemu into staging

ppc patch queue for 2022-10-18:

This queue contains improvements in the e500 and ppc4xx boards, changes
in the maintainership of the project, a new QMP/HMP command and bug
fixes:

- Cedric is stepping back from qemu-ppc maintainership;
- ppc4xx_sdram: QOMification and clean ups;
- e500: add new types of flash and clean ups;
- QMP/HMP: introduce dumpdtb command;
- spapr_pci, booke doorbell interrupt and xvcmp* bit fixes;

The 'dumpdtb' implementation is also making changes to RISC-V files that
were acked by Alistair Francis and are being included in this queue.

# -----BEGIN PGP SIGNATURE-----
#
# iHUEABYKAB0WIQQX6/+ZI9AYAK8oOBk82cqW3gMxZAUCY02qEgAKCRA82cqW3gMx
# ZIadAQCYY9f+NFrSJBm3z4JjUaP+GmbgEjibjZW05diyKwbqzQEAjE1KXFCcd40D
# 3Brs2Dm4YruaJCwb68vswVQAYteXaQ8=
# =hl94
# -----END PGP SIGNATURE-----
# gpg: Signature made Mon 17 Oct 2022 15:16:34 EDT
# gpg:                using EDDSA key 17EBFF9923D01800AF2838193CD9CA96DE033164
# gpg: Good signature from "Daniel Henrique Barboza <danielhb413@gmail.com>" [unknown]
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg:          There is no indication that the signature belongs to the owner.
# Primary key fingerprint: 17EB FF99 23D0 1800 AF28  3819 3CD9 CA96 DE03 3164

* tag 'pull-ppc-20221017' of https://gitlab.com/danielhb/qemu: (38 commits)
  hw/riscv: set machine->fdt in spike_board_init()
  hw/riscv: set machine->fdt in sifive_u_machine_init()
  hw/ppc: set machine->fdt in spapr machine
  hw/ppc: set machine->fdt in pnv_reset()
  hw/ppc: set machine->fdt in pegasos2_machine_reset()
  hw/ppc: set machine->fdt in xilinx_load_device_tree()
  hw/ppc: set machine->fdt in sam460ex_load_device_tree()
  hw/ppc: set machine->fdt in bamboo_load_device_tree()
  hw/nios2: set machine->fdt in nios2_load_dtb()
  qmp/hmp, device_tree.c: introduce dumpdtb
  hw/ppc/spapr_pci.c: Use device_cold_reset() rather than device_legacy_reset()
  target/ppc: Fix xvcmp* clearing FI bit
  hw/ppc/e500: Remove if statement which is now always true
  hw/ppc/mpc8544ds: Add platform bus
  hw/ppc/mpc8544ds: Rename wrongly named method
  hw/ppc/e500: Reduce usage of sysbus API
  docs/system/ppc/ppce500: Add heading for networking chapter
  hw/gpio/meson: Introduce dedicated config switch for hw/gpio/mpc8xxx
  hw/ppc/meson: Allow e500 boards to be enabled separately
  ppc440_uc.c: Remove unneeded parenthesis
  ...

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
19 months agotarget/i386: remove old SSE decoder
Paolo Bonzini [Fri, 2 Sep 2022 14:30:15 +0000 (16:30 +0200)]
target/i386: remove old SSE decoder

With all SSE (and AVX!) instructions now implemented in disas_insn_new,
it's possible to remove gen_sse, as well as the helpers for instructions
that now use gvec.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: move 3DNow to the new decoder
Paolo Bonzini [Mon, 5 Sep 2022 21:27:53 +0000 (23:27 +0200)]
target/i386: move 3DNow to the new decoder

This adds another kind of weirdness when you thought you had seen it all:
an opcode byte that comes _after_ the address, not before.  It's not
worth adding a new X86_SPECIAL_* constant for it, but it's actually
not unlike VCMP; so, forgive me for exploiting the similarity and just
deciding to dispatch to the right gen_helper_* call in a single code
generation function.

In fact, the old decoder had a bug where s->rip_offset should have
been set to 1 for 3DNow! instructions, and it's fixed now.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotests/tcg: extend SSE tests to AVX
Paolo Bonzini [Tue, 20 Sep 2022 16:00:03 +0000 (18:00 +0200)]
tests/tcg: extend SSE tests to AVX

Extracted from a patch by Paul Brook <paul@nowt.org>.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: Enable AVX cpuid bits when using TCG
Paul Brook [Sun, 24 Apr 2022 22:02:01 +0000 (23:02 +0100)]
target/i386: Enable AVX cpuid bits when using TCG

Include AVX, AVX2 and VAES in the guest cpuid features supported by TCG.

Signed-off-by: Paul Brook <paul@nowt.org>
Message-Id: <20220424220204.2493824-40-paul@nowt.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: implement VLDMXCSR/VSTMXCSR
Paolo Bonzini [Sun, 11 Sep 2022 11:22:32 +0000 (13:22 +0200)]
target/i386: implement VLDMXCSR/VSTMXCSR

These are exactly the same as the non-VEX version, but one has to be careful
that only VEX.L=0 is allowed.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: implement XSAVE and XRSTOR of AVX registers
Paolo Bonzini [Tue, 6 Sep 2022 17:18:26 +0000 (19:18 +0200)]
target/i386: implement XSAVE and XRSTOR of AVX registers

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: reimplement 0x0f 0x28-0x2f, add AVX
Paolo Bonzini [Tue, 6 Sep 2022 16:44:02 +0000 (18:44 +0200)]
target/i386: reimplement 0x0f 0x28-0x2f, add AVX

Here the code is a bit uglier due to the truncation and extension
of registers to and from 32-bit.  There is also a mistake in the
manual with respect to the size of the memory operand of CVTPS2PI
and CVTTPS2PI, reported by Ricky Zhou.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: reimplement 0x0f 0x10-0x17, add AVX
Paolo Bonzini [Sat, 17 Sep 2022 21:22:36 +0000 (23:22 +0200)]
target/i386: reimplement 0x0f 0x10-0x17, add AVX

These are mostly moves, and yet are a total pain.  The main issue
is that:

1) some instructions are selected by mod==11 (register operand)
vs. mod=00/01/10 (memory operand)

2) stores to memory are two-operand operations, while the 3-register
and load-from-memory versions operate on the entire contents of the
destination; this makes it easier to separate the gen_* function for
the store case

3) it's inefficient to load into xmm_T0 only to move the value out
again, so the gen_* function for the load case is separated too

The manual also has various mistakes in the operands here, for example
the store case of MOVHPS operates on a 128-bit source (albeit discarding
the bottom 64 bits) and therefore should be Mq,Vdq rather than Mq,Vq.
Likewise for the destination and source of MOVHLPS.

VUNPCK?PS and VUNPCK?PD are the same as VUNPCK?DQ and VUNPCK?QDQ,
but encoded as prefixes rather than separate operands.  The helpers
can be reused however.

For MOVSLDUP, MOVSHDUP and MOVDDUP I chose to reimplement them as
helpers.  I named the helper for MOVDDUP "movdldup" in preparation
for possible future introduction of MOVDHDUP and to clarify the
similarity with MOVSLDUP.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: reimplement 0x0f 0xc2, 0xc4-0xc6, add AVX
Paolo Bonzini [Tue, 6 Sep 2022 08:34:11 +0000 (10:34 +0200)]
target/i386: reimplement 0x0f 0xc2, 0xc4-0xc6, add AVX

Nothing special going on here, for once.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: reimplement 0x0f 0x38, add AVX
Paolo Bonzini [Wed, 14 Sep 2022 16:52:44 +0000 (18:52 +0200)]
target/i386: reimplement 0x0f 0x38, add AVX

There are several special cases here:

1) extending moves have different widths for the helpers vs. for the
memory loads, and the width for memory loads depends on VEX.L too.
This is represented by X86_SPECIAL_AVXExtMov.

2) some instructions, such as variable-width shifts, select the vector element
size via REX.W.

3) VSIB instructions (VGATHERxPy, VPGATHERxy) are also part of this group,
and they have (among other things) two output operands.

3) the macros for 4-operand blends (which are under 0x0f 0x3a) have to be
extended to support 2-operand blends.  The 2-operand variant actually
came a few years earlier, but it is clearer to implement them in the
opposite order.

X86_TYPE_WM, introduced earlier for unaligned loads, is reused for helpers
that accept a Reg* but have a M argument.

These three-byte opcodes also include AVX new instructions, for which
the helpers were originally implemented by Paul Brook <paul@nowt.org>.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: Use tcg gvec ops for pmovmskb
Richard Henderson [Wed, 7 Sep 2022 07:25:06 +0000 (09:25 +0200)]
target/i386: Use tcg gvec ops for pmovmskb

As pmovmskb is used by strlen et al, this is the third
highest overhead sse operation at %0.8.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
[Reorganize to generate code for any vector size. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: reimplement 0x0f 0x3a, add AVX
Paolo Bonzini [Tue, 6 Sep 2022 08:34:11 +0000 (10:34 +0200)]
target/i386: reimplement 0x0f 0x3a, add AVX

The more complicated operations here are insertions and extractions.
Otherwise, there are just more entries than usual because the PS/PD/SS/SD
variations are encoded in the opcode rater than in the prefixes.

These three-byte opcodes also include AVX new instructions, whose
implementation in the helpers was originally done by Paul Brook
<paul@nowt.org>.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: clarify (un)signedness of immediates from 0F3Ah opcodes
Paolo Bonzini [Tue, 20 Sep 2022 15:48:14 +0000 (17:48 +0200)]
target/i386: clarify (un)signedness of immediates from 0F3Ah opcodes

Three-byte opcodes from the 0F3Ah area all have an immediate byte which
is usually unsigned.  Clarify in the helper code that it is unsigned;
the new decoder treats immediates as signed by default, and seeing
an intN_t in the prototype might give the wrong impression that one
can use decode->immediate directly.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: reimplement 0x0f 0xd0-0xd7, 0xe0-0xe7, 0xf0-0xf7, add AVX
Paolo Bonzini [Thu, 1 Sep 2022 12:27:55 +0000 (14:27 +0200)]
target/i386: reimplement 0x0f 0xd0-0xd7, 0xe0-0xe7, 0xf0-0xf7, add AVX

The more complicated ones here are d6-d7, e6-e7, f7.  The others
are trivial.

For LDDQU, using gen_load_sse directly might corrupt the register if
the second part of the load fails.  Therefore, add a custom X86_TYPE_WM
value; like X86_TYPE_W it does call gen_load(), but it also rejects a
value of 11 in the ModRM field like X86_TYPE_M.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: reimplement 0x0f 0x70-0x77, add AVX
Paolo Bonzini [Fri, 2 Sep 2022 16:19:06 +0000 (18:19 +0200)]
target/i386: reimplement 0x0f 0x70-0x77, add AVX

This includes shifts by immediate, which use bits 3-5 of the ModRM byte
as an opcode extension.  With the exception of 128-bit shifts, they are
implemented using gvec.

This also covers VZEROALL and VZEROUPPER, which use the same opcode
as EMMS.  If we were wanting to optimize out gen_clear_ymmh then this
would be one of the starting points.  The implementation of the VZEROALL
and VZEROUPPER helpers is by Paul Brook.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: reimplement 0x0f 0x78-0x7f, add AVX
Paolo Bonzini [Thu, 1 Sep 2022 12:27:55 +0000 (14:27 +0200)]
target/i386: reimplement 0x0f 0x78-0x7f, add AVX

These are a mixed batch, including the first two horizontal
(66 and F2 only) operations, more moves, and SSE4a extract/insert.

Because SSE4a is pretty rare, I chose to leave the helper as they are,
but it is possible to unify them by loading index and length from the
source XMM register and generating deposit or extract TCG ops.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: reimplement 0x0f 0x50-0x5f, add AVX
Paolo Bonzini [Thu, 1 Sep 2022 12:27:55 +0000 (14:27 +0200)]
target/i386: reimplement 0x0f 0x50-0x5f, add AVX

These are mostly floating-point SSE operations.  The odd ones out
are MOVMSK and CVTxx2yy, the others are straightforward.

Unary operations are a bit special in AVX because they have 2 operands
for PD/PS operands (VEX.vvvv must be 1111b), and 3 operands for SD/SS.
They are handled using X86_OP_GROUP3 for compactness.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: reimplement 0x0f 0xd8-0xdf, 0xe8-0xef, 0xf8-0xff, add AVX
Paolo Bonzini [Mon, 5 Sep 2022 13:39:36 +0000 (15:39 +0200)]
target/i386: reimplement 0x0f 0xd8-0xdf, 0xe8-0xef, 0xf8-0xff, add AVX

These are more simple integer instructions present in both MMX and SSE/AVX,
with no holes that were later occupied by newer instructions.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: reimplement 0x0f 0x60-0x6f, add AVX
Paolo Bonzini [Tue, 20 Sep 2022 09:42:45 +0000 (05:42 -0400)]
target/i386: reimplement 0x0f 0x60-0x6f, add AVX

These are both MMX and SSE/AVX instructions, except for vmovdqu.  In both
cases the inputs and output is in s->ptr{0,1,2}, so the only difference
between MMX, SSE, and AVX is which helper to call.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: Introduce 256-bit vector helpers
Paolo Bonzini [Thu, 1 Sep 2022 07:04:17 +0000 (09:04 +0200)]
target/i386: Introduce 256-bit vector helpers

The new implementation of SSE will cover AVX from the get go, because
all the work for the helper functions is already done.  We just need to
build them.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: implement additional AVX comparison operators
Paolo Bonzini [Thu, 25 Aug 2022 21:54:58 +0000 (23:54 +0200)]
target/i386: implement additional AVX comparison operators

The new implementation of SSE will cover AVX from the get go, so include
the 24 extra comparison operators that are only available with the VEX
prefix.

Based on a patch by Paul Brook <paul@nowt.org>.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: provide 3-operand versions of unary scalar helpers
Paolo Bonzini [Fri, 9 Sep 2022 09:08:18 +0000 (11:08 +0200)]
target/i386: provide 3-operand versions of unary scalar helpers

Compared to Paul's implementation, the new decoder will use a different approach
to implement AVX's merging of dst with src1 on scalar operations.  Adjust the
old SSE decoder to be compatible with new-style helpers.

The affected instructions are CVTSx2Sx, ROUNDSx, RSQRTSx, SQRTSx, RCPSx.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: support operand merging in binary scalar helpers
Paolo Bonzini [Fri, 9 Sep 2022 09:08:18 +0000 (11:08 +0200)]
target/i386: support operand merging in binary scalar helpers

Compared to Paul's implementation, the new decoder will use a different approach
to implement AVX's merging of dst with src1 on scalar operations.  Adjust the
helpers to provide this functionality.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: extend helpers to support VEX.V 3- and 4- operand encodings
Paolo Bonzini [Fri, 26 Aug 2022 22:00:50 +0000 (00:00 +0200)]
target/i386: extend helpers to support VEX.V 3- and 4- operand encodings

Add to the helpers all the operands that are needed to implement AVX.

Extracted from a patch by Paul Brook <paul@nowt.org>.

Message-Id: <20220424220204.2493824-26-paul@nowt.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: Prepare ops_sse_header.h for 256 bit AVX
Paul Brook [Sun, 24 Apr 2022 22:01:44 +0000 (23:01 +0100)]
target/i386: Prepare ops_sse_header.h for 256 bit AVX

Adjust all #ifdefs to match the ones in ops_sse.h.

Signed-off-by: Paul Brook <paul@nowt.org>
Message-Id: <20220424220204.2493824-23-paul@nowt.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: move scalar 0F 38 and 0F 3A instruction to new decoder
Paolo Bonzini [Wed, 24 Aug 2022 16:01:41 +0000 (18:01 +0200)]
target/i386: move scalar 0F 38 and 0F 3A instruction to new decoder

Because these are the only VEX instructions that QEMU supports, the
new decoder is entered on the first byte of a valid VEX prefix, and VEX
decoding only needs to be done in decode-new.c.inc.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
19 months agotarget/i386: validate SSE prefixes directly in the decoding table
Paolo Bonzini [Mon, 5 Sep 2022 13:30:02 +0000 (15:30 +0200)]
target/i386: validate SSE prefixes directly in the decoding table

Many SSE and AVX instructions are only valid with specific prefixes
(none, 66, F3, F2).  Introduce a direct way to encode this in the
decoding table to avoid using decode groups too much.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>