[PATCH libaio 03/11] harness: Add KERNEL_RW_POINTER for various 32-bit architectures

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Gbp-Pq: Name 0003-harness-Add-KERNEL_RW_POINTER-for-various-32-bit-arc.patch

[PATCH libaio 02/11] syscall: Fix ia64 definitions to match exactly the kernel

Otherwise we get preprocessor warnings due to the redefinitions.

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Gbp-Pq: Name 0002-syscall-Fix-ia64-definitions-to-match-exactly-the-ke.patch

[PATCH libaio 01/11] syscall: Fix ARM definitions to match exactly the kernel

Otherwise we get preprocessor warnings due to the redefinitions.

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Gbp-Pq: Name 0001-syscall-Fix-ARM-definitions-to-match-exactly-the-ker.patch

[PATCH libaio] Fix test issue with gcc-11

Test 3.t exercises io_submit() with invalid addresses,
and one test attempted to pass in "-1" as an invalid
pointer. But gcc-11 realizes you cannot take an offset
from an invalid pointer, making it impossible to
compile or run this test:

> sh# make CC=gcc-11 partcheck
> make[1]: Entering directory '/alt/public_software/libaio/src'
> make[1]: Nothing to be done for 'all'.
> make[1]: Leaving directory '/alt/public_software/libaio/src'
> make[1]: Entering directory '/alt/public_software/libaio/harness'
> gcc-11 -Wall -Werror -I../src -g -O2 -DTEST_NAME=\"cases/2.t\" -o cases/2.p main.c ../src/libaio.a -lpthread
> gcc-11 -Wall -Werror -I../src -g -O2 -DTEST_NAME=\"cases/3.t\" -o cases/3.p main.c ../src/libaio.a -lpthread
> In file included from main.c:24:
> cases/3.t: In function ‘test_main’:
> cases/3.t:18:19: error: ‘attempt_io_submit’ accessing 8 bytes in a region of size 0 [-Werror=stringop-overflow=]
>    18 |         status |= attempt_io_submit( io_ctx, 1, (void *)-1, -EFAULT);
>       |                   ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> cases/3.t:18:19: note: referencing argument 3 of type ‘struct iocb **’
> In file included from cases/3.t:5,
>                  from main.c:24:
> cases/aio_setup.h:15:5: note: in a call to function ‘attempt_io_submit’
>    15 | int attempt_io_submit(io_context_t ctx, long nr, struct iocb *ios[], int expect)
>       |     ^~~~~~~~~~~~~~~~~
> cc1: all warnings being treated as errors
> make[1]: *** [Makefile:19: cases/3.p] Error 1
> make[1]: Leaving directory '/alt/public_software/libaio/harness'
> make: *** [Makefile:23: partcheck] Error 2

The fix is to tell gcc to relax it's stringop overflow
testing for this one test.

Changes since v1:
* First version just skipped the test in question

Signed-off-by: Lee Duncan <leeman.duncan@gmail.com>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0001-Fix-test-issue-with-gcc-11.patch

[PATCH libaio 28/28] harness: Skip the test if io_pgetevents() is not implemented

Either the kernel or the syscall wrapper will return -ENOSYS when this
syscall is not implemented. So we should cope with this in the test
suite and SKIP the test case.

This is currently the case for alpha and ia64, even though they have
already been wired up very recently in the kernel mainline.

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0028-harness-Skip-the-test-if-io_pgetevents-is-not-implem.patch

[PATCH libaio 27/28] harness: Print better error messages on error conditions in 22.t

These should help diagnose problems when dealing with error failures.
In particular this helped distinguish the problem with io_pgetevents()
not being implemented and it failing due to the sigset_t layout bug.

Signed-off-by: Guillem Jover <guillem@hadrons.org>
[JEM: fix up 80 columns violations]
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0027-harness-Print-better-error-messages-on-error-conditi.patch

[PATCH libaio 26/28] harness: Fix PROT_WRITE mmap check

This partially reverts commit d7f5065448efb49b2a26e728ff735e12ea05b62e.

The actual problem in the original code was that read() was being used
to assert whether the buffer was readable, but the kernel was instead
reading from the file descriptor and then writing into the buffer, so
no EFAULT was being generated (on architectures that do so).

We needed to use a write() so that the kernel would read from the
buffer.

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0026-harness-Fix-PROT_WRITE-mmap-check.patch

[PATCH libaio 25/28] harness: fix read into PROT_WRITE mmap test

This test has been broken forever.  Fix it up to perform an
aio_read using the result of a regular read as the expected
return code.

Reported-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0025-harness-fix-read-into-PROT_WRITE-mmap-test.patch

[PATCH libaio 24/28] harness: skip 22.p if async_poll isn't supported

Use the new skip error code instead of failing the test.  Also
add in a Local variables: section for emacs.

Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0024-harness-skip-22.p-if-async_poll-isn-t-supported.patch

[PATCH libaio 23/28] harness: Handle -ENOTSUP from io_submit() with RWF_NOWAIT

On filesystems such as tmpfs the syscall might return -ENOTSUP instead
of EINVAL when it does not support the RWF_NOWAIT flag.

Signed-off-by: Guillem Jover <guillem@hadrons.org>
[JEM: skip the test instead of returning success]
[JEM: make the error message differentiate between kernel and fs support]
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0023-harness-Handle-ENOTSUP-from-io_submit-with-RWF_NOWAI.patch

[PATCH libaio 22/28] harness: Add fallback code for filesystems not supporting O_DIRECT

When running the harness on a filesystem such as a tmpfs, which do not
support O_DIRECT, fallback to calls without the flag.

Signed-off-by: Guillem Jover <guillem@hadrons.org>
[JEM: change from duplicating the open call to using F_SETFL]
[JEM: 18 and 21 require O_DIRECT-skip if not present]
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0022-harness-Add-fallback-code-for-filesystems-not-suppor.patch

[PATCH libaio 21/28] harness: add support for skipping tests

Skipped tests will not cause the test harness to return failure.  An
exit status of "3" was chosen for skipped tests.  This doesn't
conflict with any of the current tests.

Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0021-harness-add-support-for-skipping-tests.patch

[PATCH libaio 20/28] harness: Make the test exit with a code matching the pass/fail state

This way we can use the exit code to check whether the tests passed or
failed, and fail the package build.

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0020-harness-Make-the-test-exit-with-a-code-matching-the-.patch

[PATCH libaio 19/28] harness: Make RISC-V use SYS_eventfd2 instead of unavailable SYS_eventfd

This is a recent architecture and as such does not provide legacy
support for SYS_eventfd. Declare that we need to use the new syscall.

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0019-harness-Make-RISC-V-use-SYS_eventfd2-instead-of-unav.patch

[PATCH libaio 18/28] harness: Use run-time _SC_PAGE_SIZE instead of build-time PAGESIZE

The getconf(1) command is inherently not cross-compilation friendly.
In addition PAGESIZE depends on the specific system, even within a
specific arch, so using a hard-coded value is never safe.

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0018-harness-Use-run-time-_SC_PAGE_SIZE-instead-of-build-.patch

[PATCH libaio 17/28] harness: Use destination strncpy() expression for sizeof() argument

Even though this is the same size, as the sizeof() is derived from the
source expression, recent gcc versions will emit a warning, which is
turned into an error by -Werror:

  error: argument to ‘sizeof’ in ‘strncpy’ call is the same expression
    as the source; did you mean to use the size of the destination?
    [-Werror=sizeof-pointer-memaccess]

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0017-harness-Use-destination-strncpy-expression-for-sizeo.patch

[PATCH libaio 16/28] Use ctx consistently for io_context_t instead of ctx_id

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0016-Use-ctx-consistently-for-io_context_t-instead-of-ctx.patch

[PATCH libaio 15/28] man: Escape verbatim \n in order to make it through roff

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0015-man-Escape-verbatim-n-in-order-to-make-it-through-ro.patch

[PATCH libaio 14/28] man: Fold short lines

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0014-man-Fold-short-lines.patch

[PATCH libaio 13/28] man: Fix markup

- Remove unnecessary macro argument quoting.
- Variables, pathnames in italics.
- Keywords in bold.
- Man page references in bold, followed by the man page number.
- Fix TP/TQ macro usage.

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0013-man-Fix-markup.patch

[PATCH libaio 12/28] man: Fix title header

- Update year.
- Balance double quotes.
- Remove version from source argument as recommended in man-pages(7).
- Fix all sections numbers to 3.

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0012-man-Fix-title-header.patch

[PATCH libaio 11/28] man: Fix typos

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0011-man-Fix-typos.patch

[PATCH libaio 10/28] man: Add "None" to empty sections

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0010-man-Add-None-to-empty-sections.patch

[PATCH libaio 09/28] man: Remove spurious text

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0009-man-Remove-spurious-text.patch

[PATCH libaio 08/28] man: Remove spurious spaces

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0008-man-Remove-spurious-spaces.patch

[PATCH libaio 07/28] man: Fix period formatting

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0007-man-Fix-period-formatting.patch

[PATCH libaio 06/28] man: Fix casing

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0006-man-Fix-casing.patch

[PATCH libaio 05/28] man: End sentences with a period

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0005-man-End-sentences-with-a-period.patch

[PATCH libaio 04/28] man: Refer to libaio.h instead of libio.h

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0004-man-Refer-to-libaio.h-instead-of-libio.h.patch

[PATCH libaio 03/28] man: Use the correct troff macro for comments

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0003-man-Use-the-correct-troff-macro-for-comments.patch

[PATCH libaio 02/28] man: Add missing space in man page references

Signed-off-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0002-man-Add-missing-space-in-man-page-references.patch

[PATCH libaio 01/28] harness: allow running tests against the installed library

A user can now specify "LIBAIO=/path/to/libaio.so" in order to run
the test harness against the installed library.

Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0001-harness-allow-running-tests-against-the-installed-li.patch

[PATCH libaio] Link against libgcc to avoid unresolved symbols

We need to link agaisnt -lgcc, on at least hppa, PPC and ARC.
That's because in some corner-cases like compilation with -Os
on ARC and PPC so-called millicode (basically function prologue and
epilogue) implemented in libgcc.a is used. So we end up with
GLOBAL UNDEFINED symbol in libaio.so and then on linkage of the final
applicaiton LD fails to proceed saying:
--------------------------->8----------------------
hidden symbol '__ld_r13_to_r15_ret' in .../libgcc.a(_millicodethunk_ret.o) is referenced by DSO
--------------------------->8----------------------

Also it looks like in general it is not the best idea to use either
"-nostartfiles" or "-nostdlib" when linking shared libs because
default construtor/destructor functions won't be executed, see
"5.2. Library constructor and destructor functions" in [1]

So let's stop passing "-nostdlib" and "-nostartfiles" and get required
stuff built-in libaio.

Initial patch taken from Debian [2].

Fixes build failures in Buildroot like blktrace [3], lvm2 [4].

[1] http://tldp.org/HOWTO/Program-Library-HOWTO/miscellaneous.html
[2] https://sources.debian.org/patches/libaio/0.3.111-1/01_link_libs.patch/
[3] http://autobuild.buildroot.net/results/17461209755038a30118d76acb4f43469a22a139/
[4] http://autobuild.buildroot.net/results/a5dfc87f94b97135e5cc84f6a876114891ed9dd9/

Signed-off-by: Guillem Jover <guillem@debian.org>
Signed-off-by: Alexey Brodkin <abrodkin@synopsys.com>
Gbp-Pq: Topic upstream
Gbp-Pq: Name 0001-Link-against-libgcc-to-avoid-unresolved-symbols.patch

libaio (0.3.112-12+rpi1) bookworm-staging; urgency=medium

  * Add armhf to list of architectures where tests are not expected to pass.

[dgit import unpatched libaio 0.3.112-12+rpi1]

Import libaio_0.3.112-12+rpi1.debian.tar.xz

[dgit import tarball libaio 0.3.112-12+rpi1 libaio_0.3.112-12+rpi1.debian.tar.xz]

Import libaio_0.3.112.orig.tar.xz

[dgit import orig libaio_0.3.112.orig.tar.xz]