- Wine-devel - list.winehq.org

Re: [PATCH v6 0/9] MR6: Avoid performance degradation due to vDSO unmapping (BZ#52313)
by Jinoh Kang (＠iamahuman) 05 May '22

05 May '22

Fixed a bug where `free_vma_list` would inadvertently leave a hole in preload reserved ranges. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/6#note_698

1 0

[PATCH v5 0/9] MR6: Draft: Avoid performance degradation due to vDSO unmapping (BZ#52313)
by Jinoh Kang (＠iamahuman) 05 May '22

05 May '22

Commit f558741fabc116534fa598aa890ffed683a7153b removes vDSO if it conflicts with reserved ranges: > Remove the AT_SYSINFO and AT_SYSINFO_EHDR values if the sysinfo page > is in one of our reserved ranges. However, missing vDSO leads to performance issues on some syscalls (e.g. clock_gettime, gettimeofday) and may even lead to crash when run with some ancient C libraries that does not supply a custom signal restorer. vDSO pages can clash with reserved ranges especially in a 32-bit address space with address space layout randomization (ASLR) turned on. Recent versions of the Linux kernel introduced support for mremap()-ping vDSO pages, partly in an effort to support checkpoint restore in userspace (CRIU). Special programs that require specific memory layout constraints (such as Wine preloader) can take advantage of this support to modify the address space to meet its requirements. The following test script has been used to test each changes (use with `git rebase --exec=...`): ```sh set -e make -C ../wine64-build -j5 make -C ../wine32-build -j5 cd ../wine64-build export WINEPRELOADREMAPSTACK export WINEPRELOADREMAPVDSO for WINEPRELOADREMAPSTACK in skip never always force auto on-demand '' do for WINEPRELOADREMAPVDSO in skip never always force auto on-demand '' do ./loader/wine64 wineboot ./loader/wine wineboot done done ``` -- v5: loader: Switch stack if the old stack address is in reserved range. loader: Relocate sigpage on conflict with reserved ranges in ARM. loader: Relocate vDSO on conflict with reserved ranges. loader: Fix return type of get_auxiliary(). loader: Don't clobber existing memory mappings when reserving addresses. https://gitlab.winehq.org/wine/wine/-/merge_requests/6

2 9

[PATCH v2 0/9] MR39: winex11: PE conversion preparation
by Jacek Caban (＠jacek) 05 May '22

05 May '22

-- v2: win32u: Use KeUserModeCallback interface for CopyImage. winex11: Directly use win32u in more places. winex11: Directly use NtUserGetIconInfo. win32u: Move NtUserCallMsgFilter implementation from user32. https://gitlab.winehq.org/wine/wine/-/merge_requests/39

3 10

[tools 1/3] winetest/build-patterns: Use CSS to highlight the new failure modes.
by Francois Gouget 05 May '22

05 May '22

Signed-off-by: Francois Gouget <fgouget(a)codeweavers.com> --- winetest/build-patterns | 20 ++++++++++++-------- winetest/report.css | 4 ++++ 2 files changed, 16 insertions(+), 8 deletions(-) diff --git a/winetest/build-patterns b/winetest/build-patterns index aa6a6e495..b6e5531e1 100755 --- a/winetest/build-patterns +++ b/winetest/build-patterns @@ -999,7 +999,7 @@ sub write_newmodes_line($$) foreach my $status (sort cmpstatus keys %{$test->{newmodes}}) { my ($symbol, $class, $title, $_link, $attrs) = get_status_html($status, $test->{colors}); - print $html " $symbol"; + print $html " $symbol"; } print $html "\n"; } @@ -1016,7 +1016,7 @@ sub write_pattern_line($$$) my ($range_start, $range_end, $range_title); foreach my $build (@sortedbuilds) { - my ($symbol, $class, $title); + my ($symbol, $class, $title, $is_new_mode); my ($tag, $attrs) = ("span", ""); my $status = $testreport->{status}->{$build->{name}}; @@ -1066,11 +1066,7 @@ sub write_pattern_line($$$) $attrs .= sprintf " href='%s/%s/%s.html#%s'", $build->{name}, $reportdir, $link, $dll; } - if ($test->{newmodes}->{$status}) - { - $symbol = "$symbol"; - $has_newmode = 1; - } + $is_new_mode = $test->{newmodes}->{$status}; } if ($range_symbol eq $symbol) @@ -1091,7 +1087,15 @@ sub write_pattern_line($$$) } # Start a new pattern range - $class = " class='pat$class'" if ($class); + if ($is_new_mode) + { + $class = $class ? " class='pat$class newM'" : " class='newM'"; + $has_newmode = 1; + } + elsif ($class) + { + $class = " class='pat$class'"; + } print $html "<$tag$class$attrs"; if ($tag eq "a") { diff --git a/winetest/report.css b/winetest/report.css index 25f156d6f..4b2d126ad 100644 --- a/winetest/report.css +++ b/winetest/report.css @@ -134,6 +134,10 @@ div.pattern :hover { color: black; text-decoration: underline; } .patV { /* test blocked by the anti-virus */ background-color: #99e6ff; } +.newM { /* new failure mode */ + font-style: italic; + font-weight: bold; +} .commitR { color: #cc0000; -- 2.30.2

1 2

[PATCH 0/9] wined3d: Support non-zero colors in renderpass clears
by Stefan Dösinger 05 May '22

05 May '22

This patchset implements delayed clears in the Vulkan renderer with arbitrary colors. The Vulkan side of it is fairly simple - most of the patches extend the CPU fallback path to support those colors. The final patch adds a d3d9 test to exercise the fallback code. The Vk renderer is functional enough in d3d9 to run this particular test. I don't see a way to reliably test this codepath through d3d10/11 without quirky assumptions about CopyResource. Dynamic and staging resources can't be written by the GPU, and default resources can't be read by the CPU. Stefan Dösinger (9): wined3d: Separate mapping from data writing in surface_cpu_blt_colour_fill. wined3d: Prepare wined3d_format_convert_from_float for > 32bpp formats. wined3d: Add support for sysmem-clearing float32 formats. wined3d: Support CPU clear of float16 formats. wined3d: Add 16 bit per channel UNORM formats to wined3d_format_convert_from_float. wined3d: Add a memory_colour_fill for clearing the entire level to zero. wined3d: Prepare sysmem LOCATION_CLEARED handling for non-zero colors. wined3d: Store clear colors in subresources. d3d9/tests: Extend color_fill_test. dlls/d3d9/tests/visual.c | 132 ++++++++++++++++++++++------ dlls/wined3d/context_vk.c | 57 ++++++++++-- dlls/wined3d/device.c | 7 +- dlls/wined3d/resource.c | 112 +++++++++++++++++++++++ dlls/wined3d/surface.c | 139 +---------------------------- dlls/wined3d/texture.c | 52 +++++++---- dlls/wined3d/utils.c | 156 +++++++++++++++++++++++++-------- dlls/wined3d/wined3d_private.h | 91 +++++++++++++++++-- 8 files changed, 514 insertions(+), 232 deletions(-) -- 2.35.1

4 19

[PATCH 0/9] MR39: winex11: PE conversion preparation
by Jacek Caban (＠jacek) 05 May '22

05 May '22

-- https://gitlab.winehq.org/wine/wine/-/merge_requests/39

3 10

[PATCH v3 0/6] MR22: winegstreamer: Implement more sample attributes and timestamps.
by Rémi Bernon 05 May '22

05 May '22

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=45988 Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=47084 Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=49715 Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=52183 -- v3: winegstreamer: Support dynamic wg_transform output format change. winegstreamer: Track caps changes and keep them with the output buffers. winegstreamer: Remove width/height from caps when format uses 0. winegstreamer: Only resize buffer when read is incomplete. winegstreamer: Support the MFSampleExtension_CleanPoint sample attribute. winegstreamer: Add timestamp and duration to struct wg_sample. https://gitlab.winehq.org/wine/wine/-/merge_requests/22

3 12

[PATCH] mountmgr.sys: Avoid passing a stack pointer to the run_loop thread.
by Rémi Bernon 05 May '22

05 May '22

Signed-off-by: Rémi Bernon <rbernon(a)codeweavers.com> --- dlls/mountmgr.sys/mountmgr.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/dlls/mountmgr.sys/mountmgr.c b/dlls/mountmgr.sys/mountmgr.c index 1f9bd56a384..66034621b44 100644 --- a/dlls/mountmgr.sys/mountmgr.c +++ b/dlls/mountmgr.sys/mountmgr.c @@ -609,7 +609,8 @@ static DWORD WINAPI device_op_thread( void *arg ) static DWORD WINAPI run_loop_thread( void *arg ) { - return MOUNTMGR_CALL( run_loop, arg ); + struct run_loop_params params = {.op_thread = arg, .op_apc = device_op}; + return MOUNTMGR_CALL( run_loop, &params ); } @@ -624,7 +625,7 @@ NTSTATUS WINAPI DriverEntry( DRIVER_OBJECT *driver, UNICODE_STRING *path ) DEVICE_OBJECT *device; HKEY devicemap_key; NTSTATUS status; - struct run_loop_params params; + HANDLE thread; TRACE( "%s\n", debugstr_w(path->Buffer) ); @@ -655,9 +656,8 @@ NTSTATUS WINAPI DriverEntry( DRIVER_OBJECT *driver, UNICODE_STRING *path ) RtlInitUnicodeString( &nameW, L"\\Driver\\Harddisk" ); status = IoCreateDriver( &nameW, harddisk_driver_entry ); - params.op_thread = CreateThread( NULL, 0, device_op_thread, NULL, 0, NULL ); - params.op_apc = device_op; - CloseHandle( CreateThread( NULL, 0, run_loop_thread, &params, 0, NULL )); + thread = CreateThread( NULL, 0, device_op_thread, NULL, 0, NULL ); + CloseHandle( CreateThread( NULL, 0, run_loop_thread, thread, 0, NULL )); #ifdef _WIN64 /* create a symlink so that the Wine port overrides key can be edited with 32-bit reg or regedit */ -- 2.35.1

1 0

[PATCH 0/8] MR38: kernel32/tests: Add some more heap tests.
by Rémi Bernon 05 May '22

05 May '22

-- https://gitlab.winehq.org/wine/wine/-/merge_requests/38

1 8

Re: [PATCH v2 0/1] MR13: kernelbase: Add support for progress callback in CopyFileEx.
by Alistair Leslie-Hughes (＠alesliehughes) 05 May '22

05 May '22

On 5/5/22 17:40, Nikolay Sivov (@nsivov) wrote: > Thank you. It's actually possible to test access mode through the callback, and it does look like DELETE is used internally, so that's good. > > My only concern now is whether we need stream-switch call at all? Have you tested if it's used for regular files, or on filesystems where streams are not supported. Maybe it could be dropped. > I've only tested on NTFS and ext4. On Windows, I can confirm that the callback is called multiple times, as parts of the file are copied. Beyond that nothing besides confirming the progress bar on the application in question works. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/13#note_630

1 0