pseh2_64 relied on inline-asm clobbers across a C/asm funclet jump, which is not a reliable ABI boundary.
On AMD64 gcc path, this can corrupt non-volatile state during exception filtering/unwind.
This change makes the trampoline ABI-safe by explicitly preserving and restoring non-volatile registers and keeping unwind metadata/prologue consistent.
Changes:
- Save/restore rbx, rdi, rsi, r12-r15 in `__seh2_global_filter_func`.
- Adjust stack allocation/unwind annotation to match the new prologue/epilogue.
- Keep filter return in eax and jump back through `__seh2_global_filter_func_exit`.
Note: This is exception-only path: extra stack usage is acceptable for correctness.
How to test: Build and run ReactOS with gcc 13/15; you should no longer see stack exhaustion.
EMS support isn't platform-specific, so we can compile it everywhere
(as this is already done for the kernel and the SAC driver).
The only platform-specific code currently existing is the retrieval of
the system GUID, which is done on BIOS-based PC by reading the SMBIOS
table in the low-MB ROM region, but should be done differently on other
platforms. Add a compile-time warning for this.
A 32-bit ASM file can turn into an empty binary file in certain cases (compiled with MSVC).
MASM/ML produces a ".text$mn" section instead of the ".text", which makes obj2bin generate an empty binary file.
The problem didn't occur with our 16-bit artifacts (16-bit stubs, bootsectors...), because their object files all have similar sections pattern: the ".text" section and empty ".text$mn" section, which didn't matter since it is empty.
On the contrary, the .text$mn section(s) aren't empty for 32-bit stubs.
One visible case is the setup LiveCD error dialog, where the message box button can appear with a wrong bold/garbled look instead of the normal default-pushbutton appearance.
## Root cause
`win32ss/user/user32/controls/button.c` stores button state with `GetWindowLongPtrW` / `SetWindowLongPtrW`, but the control extra-byte layout was still defined as if the first slot were only `sizeof(LONG)`
On 64-bit builds, that causes the stored state to overlap the cached font, image, and UI-state fields.
- [SYSSETUP] Fix DST setting checkbox
Fix DST setting checkbox. The checkbox should be disabled and unchecked on timezones DST not applied, the checkbox should be enabled and checked on timezones DST applied. JIRA issue: CORE-20406
CORE-17977
- Fix a bug when subsequent port detection attempts fail because of
broken FIFO management. If FIFO logic is enabled the hardware will
not decode the legacy 0x30 and 0x32 I/O ports.
Therefore care should be taken when accessing these ports.
- Fix hang on boot when a serial device was not connected to the COM1 port.
Unlike 16550, a call to `CpDoesPortExist()` for the 8251 will succeed
even when if the user has not plug the serial port into PC-98 machine.
(I do not know how to put the 8251 into loopback mode.)
Fix this by checking for CTS prior to the LSR loop.
- Fix hang on boot when the chosen baud rate becomes too large due to
`DEFAULT_DEBUG_BAUD_RATE` being set to 115200.
- Some NP21/W workarounds were being incorrectly applied,
fix them and consolidate them into one check.
- Reuse exsisting code from the NS16550 driver to avoid code duplication.
- Rename HW definitions for better naming.
Co-authored-by: Hermès Bélusca-Maïto <[email protected]>
- Read the interface metric from the registry.
- Assign an automatic metric to an interface if a registry value is not available.
- Determine the automatic metric from the link speed. The metric for the loopback interface is always 1.
- Identify 127.x.x.x addresses as on-link addresses.
- Modify the neighbor cache to map all 127.x.x.x addresses to the loopback address (127.0.0.1).
- Create a single entry for all 127.x.x.x addresses instead of one entry per address. This entry belongs to the loopback address.
CORE-7484
RtlConvertUlongToLargeInteger and RtlConvertLongToLargeInteger are
obsolete NT routines that assign a 32-bit integer to LARGE_INTEGER.QuadPart.
Replace all driver/DLL usages with inline .QuadPart assignments or
standard C/C++ casts. This removes unnecessary function calls.
CORE-19438
- [USER32_WINETEST] Sync to wine-10.0.
- [USER32_WINETEST] Define out some function definitions we can't build yet.
- [USER32_WINETEST] Skip parts of test that crash on ReactOS or Windows
- [USER32_WINETEST] Pass behaviors from WS03-Win10 1607
- psdk/mmsystem.h: Add a #ifndef guard around SC_SCREENSAVE, MS public headers have the same guard
- psdk/ole2.h: Add guards around CreateDataAdviseHolder that are similar to the guards used in MS public headers
- psdk/winuser.h: Add missing defines, use DWORD and LONG casts to fix test failures on x64. These same casts exist in MS public headers
- wine/test.h: Add defines for winetest_platform_is_wine, flaky, and flaky_wine. The flaky definitions depend on flaky_if, which is a stub that does nothing.
- Fix incorrect IDE controller ownership leading to strange and inconsistent behavior during boot.
- Fix a mismatch between the type and flags of the drive.
- Remove obsolete definitions.
- Use SAL2 annotations.
CORE-17977
The UEFI boot menu countdown was broken for two reasons:
- StallExecutionProcessor and UefiHwIdle were empty stubs, so the countdown
had no actual delay and no proper idle, it ran instantly without waiting
KbHit used ReadKeyStroke to poll for input, which consumed the keystroke
before GetCh could read it.
This fixes both: StallExecutionProcessor and UefiHwIdle now have real
implementations using BootServices->Stall and a timer event.
KbHit is changed to use CheckEvent(WaitForKey) which polls availability
without consuming the key.
CORE-11954
Supplements commit cac013abef (PR #8104).
CORE-5680
The `EnableHexNumpad` string value is specified in the
`HKEY_CURRENT_USER\Control Panel\Input Method` registry key.
When enabled, the user can enter codepoints in hexadecimal form, using:
- `Alt .xxx` ('dot' prefix), for current ANSI codepage (ACP) codepoints;
- `Alt +xxx` ('plus' prefix), for Unicode codepoints.
These two modes complement the known `Alt xxx` and `Alt 0xxx` decimal forms
for entering codepoints in the current OEM or ACP codepage, respectively.
ROSTESTS-416
More and more winetests we import from Wine hardcode the `C:\Windows`
path and thus, are brittle (read: unreliable) whenever they are run
on any other configuration where Windows (or ReactOS) is **NOT**
installed in this path.
Dedicated to Carl Bialorucki ;)
Timo Kreuzer