June 2022 - Wine-Devel - list.winehq.org

[PATCH v4 4/9] d3dx10: Add ID3DX10ThreadPump:GetWorkItemCount implementation.

by Piotr Caban

From: Piotr Caban <piotr(a)codeweavers.com> Signed-off-by: Piotr Caban <piotr(a)codeweavers.com> --- dlls/d3dx10_43/async.c | 18 ++++++++++++++++-- 1 file changed, 16 insertions(+), 2 deletions(-) diff --git a/dlls/d3dx10_43/async.c b/dlls/d3dx10_43/async.c index a3d1014986b..33dd3ceca9a 100644 --- a/dlls/d3dx10_43/async.c +++ b/dlls/d3dx10_43/async.c @@ -636,6 +636,8 @@ struct thread_pump ID3DX10ThreadPump ID3DX10ThreadPump_iface; LONG refcount; + LONG processing_count; + SRWLOCK io_lock; CONDITION_VARIABLE io_cv; unsigned int io_count; @@ -758,6 +760,7 @@ static HRESULT WINAPI thread_pump_AddWorkItem(ID3DX10ThreadPump *iface, ID3DX10D if (object) *object = NULL; + InterlockedIncrement(&thread_pump->processing_count); AcquireSRWLockExclusive(&thread_pump->io_lock); ++thread_pump->io_count; list_add_tail(&thread_pump->io_queue, &work_item->entry); @@ -768,8 +771,15 @@ static HRESULT WINAPI thread_pump_AddWorkItem(ID3DX10ThreadPump *iface, ID3DX10D static UINT WINAPI thread_pump_GetWorkItemCount(ID3DX10ThreadPump *iface) { - FIXME("iface %p stub!\n", iface); - return 0; + struct thread_pump *thread_pump = impl_from_ID3DX10ThreadPump(iface); + UINT ret; + + TRACE("iface %p.\n", iface); + + AcquireSRWLockExclusive(&thread_pump->device_lock); + ret = thread_pump->processing_count + thread_pump->device_count; + ReleaseSRWLockExclusive(&thread_pump->device_lock); + return ret; } static HRESULT WINAPI thread_pump_WaitForAllItems(ID3DX10ThreadPump *iface) @@ -842,6 +852,7 @@ static DWORD WINAPI io_thread(void *arg) if (work_item->result) *work_item->result = hr; work_item_free(work_item, FALSE); + InterlockedDecrement(&thread_pump->processing_count); continue; } @@ -894,6 +905,7 @@ static DWORD WINAPI proc_thread(void *arg) if (work_item->result) *work_item->result = hr; work_item_free(work_item, FALSE); + InterlockedDecrement(&thread_pump->processing_count); continue; } @@ -908,6 +920,7 @@ static DWORD WINAPI proc_thread(void *arg) if (work_item->result) *work_item->result = hr; work_item_free(work_item, FALSE); + InterlockedDecrement(&thread_pump->processing_count); continue; } @@ -921,6 +934,7 @@ static DWORD WINAPI proc_thread(void *arg) list_add_tail(&thread_pump->device_queue, &work_item->entry); ++thread_pump->device_count; + InterlockedDecrement(&thread_pump->processing_count); ReleaseSRWLockExclusive(&thread_pump->device_lock); } return 0; -- GitLab https://gitlab.winehq.org/wine/wine/-/merge_requests/272

June 26, 2022

[PATCH v4 3/9] d3dx10: Add ID3DX10ThreadPump:AddWorkItem implementation.

by Piotr Caban

From: Piotr Caban <piotr(a)codeweavers.com> Signed-off-by: Piotr Caban <piotr(a)codeweavers.com> --- dlls/d3dx10_43/async.c | 248 ++++++++++++++++++++++++++++++++++++++++- 1 file changed, 245 insertions(+), 3 deletions(-) diff --git a/dlls/d3dx10_43/async.c b/dlls/d3dx10_43/async.c index 7913f634c21..a3d1014986b 100644 --- a/dlls/d3dx10_43/async.c +++ b/dlls/d3dx10_43/async.c @@ -23,6 +23,7 @@ #include "dxhelpers.h" #include "wine/debug.h" +#include "wine/list.h" WINE_DEFAULT_DEBUG_CHANNEL(d3dx); @@ -610,10 +611,47 @@ HRESULT WINAPI D3DX10PreprocessShaderFromMemory(const char *data, SIZE_T data_si return E_NOTIMPL; } +struct work_item +{ + struct list entry; + + ID3DX10DataLoader *loader; + ID3DX10DataProcessor *processor; + HRESULT *result; + void **object; +}; + +static inline void work_item_free(struct work_item *work_item, BOOL cancel) +{ + ID3DX10DataLoader_Destroy(work_item->loader); + ID3DX10DataProcessor_Destroy(work_item->processor); + if (cancel && work_item->result) + *work_item->result = S_FALSE; + free(work_item); +} + +#define THREAD_PUMP_EXITING UINT_MAX struct thread_pump { ID3DX10ThreadPump ID3DX10ThreadPump_iface; LONG refcount; + + SRWLOCK io_lock; + CONDITION_VARIABLE io_cv; + unsigned int io_count; + struct list io_queue; + + SRWLOCK proc_lock; + CONDITION_VARIABLE proc_cv; + unsigned int proc_count; + struct list proc_queue; + + SRWLOCK device_lock; + unsigned int device_count; + struct list device_queue; + + unsigned int thread_count; + HANDLE threads[1]; }; static inline struct thread_pump *impl_from_ID3DX10ThreadPump(ID3DX10ThreadPump *iface) @@ -652,11 +690,49 @@ static ULONG WINAPI thread_pump_Release(ID3DX10ThreadPump *iface) { struct thread_pump *thread_pump = impl_from_ID3DX10ThreadPump(iface); ULONG refcount = InterlockedDecrement(&thread_pump->refcount); + struct work_item *item, *next; + struct list list; + unsigned int i; TRACE("%p decreasing refcount to %lu.\n", iface, refcount); if (!refcount) + { + AcquireSRWLockExclusive(&thread_pump->io_lock); + thread_pump->io_count = THREAD_PUMP_EXITING; + ReleaseSRWLockExclusive(&thread_pump->io_lock); + WakeAllConditionVariable(&thread_pump->io_cv); + + AcquireSRWLockExclusive(&thread_pump->proc_lock); + thread_pump->proc_count = THREAD_PUMP_EXITING; + ReleaseSRWLockExclusive(&thread_pump->proc_lock); + WakeAllConditionVariable(&thread_pump->proc_cv); + + AcquireSRWLockExclusive(&thread_pump->device_lock); + thread_pump->device_count = THREAD_PUMP_EXITING; + ReleaseSRWLockExclusive(&thread_pump->device_lock); + + for (i = 0; i < thread_pump->thread_count; ++i) + { + if (!thread_pump->threads[i]) + continue; + + WaitForSingleObject(thread_pump->threads[i], INFINITE); + CloseHandle(thread_pump->threads[i]); + } + + list_init(&list); + list_move_tail(&list, &thread_pump->io_queue); + list_move_tail(&list, &thread_pump->proc_queue); + list_move_tail(&list, &thread_pump->device_queue); + LIST_FOR_EACH_ENTRY_SAFE(item, next, &list, struct work_item, entry) + { + list_remove(&item->entry); + work_item_free(item, TRUE); + } + free(thread_pump); + } return refcount; } @@ -664,9 +740,30 @@ static ULONG WINAPI thread_pump_Release(ID3DX10ThreadPump *iface) static HRESULT WINAPI thread_pump_AddWorkItem(ID3DX10ThreadPump *iface, ID3DX10DataLoader *loader, ID3DX10DataProcessor *processor, HRESULT *result, void **object) { - FIXME("iface %p, loader %p, processor %p, result %p, object %p stub!\n", + struct thread_pump *thread_pump = impl_from_ID3DX10ThreadPump(iface); + struct work_item *work_item; + + TRACE("iface %p, loader %p, processor %p, result %p, object %p.\n", iface, loader, processor, result, object); - return E_NOTIMPL; + + work_item = malloc(sizeof(*work_item)); + if (!work_item) + return E_OUTOFMEMORY; + + work_item->loader = loader; + work_item->processor = processor; + work_item->result = result; + work_item->object = object; + + if (object) + *object = NULL; + + AcquireSRWLockExclusive(&thread_pump->io_lock); + ++thread_pump->io_count; + list_add_tail(&thread_pump->io_queue, &work_item->entry); + ReleaseSRWLockExclusive(&thread_pump->io_lock); + WakeConditionVariable(&thread_pump->io_cv); + return S_OK; } static UINT WINAPI thread_pump_GetWorkItemCount(ID3DX10ThreadPump *iface) @@ -714,20 +811,165 @@ static const ID3DX10ThreadPumpVtbl thread_pump_vtbl = thread_pump_GetQueueStatus }; +static DWORD WINAPI io_thread(void *arg) +{ + struct thread_pump *thread_pump = arg; + struct work_item *work_item; + HRESULT hr; + + TRACE("%p thread started.\n", thread_pump); + + for (;;) + { + AcquireSRWLockExclusive(&thread_pump->io_lock); + + while (!thread_pump->io_count) + SleepConditionVariableSRW(&thread_pump->io_cv, &thread_pump->io_lock, INFINITE, 0); + + if (thread_pump->io_count == THREAD_PUMP_EXITING) + { + ReleaseSRWLockExclusive(&thread_pump->io_lock); + return 0; + } + + thread_pump->io_count--; + work_item = LIST_ENTRY(list_head(&thread_pump->io_queue), struct work_item, entry); + list_remove(&work_item->entry); + ReleaseSRWLockExclusive(&thread_pump->io_lock); + + if (FAILED(hr = ID3DX10DataLoader_Load(work_item->loader))) + { + if (work_item->result) + *work_item->result = hr; + work_item_free(work_item, FALSE); + continue; + } + + AcquireSRWLockExclusive(&thread_pump->proc_lock); + if (thread_pump->proc_count == THREAD_PUMP_EXITING) + { + ReleaseSRWLockExclusive(&thread_pump->proc_lock); + work_item_free(work_item, TRUE); + return 0; + } + + list_add_tail(&thread_pump->proc_queue, &work_item->entry); + ++thread_pump->proc_count; + ReleaseSRWLockExclusive(&thread_pump->proc_lock); + WakeConditionVariable(&thread_pump->proc_cv); + } + return 0; +} + +static DWORD WINAPI proc_thread(void *arg) +{ + struct thread_pump *thread_pump = arg; + struct work_item *work_item; + SIZE_T size; + void *data; + HRESULT hr; + + TRACE("%p thread started.\n", thread_pump); + + for (;;) + { + AcquireSRWLockExclusive(&thread_pump->proc_lock); + + while (!thread_pump->proc_count) + SleepConditionVariableSRW(&thread_pump->proc_cv, &thread_pump->proc_lock, INFINITE, 0); + + if (thread_pump->proc_count == THREAD_PUMP_EXITING) + { + ReleaseSRWLockExclusive(&thread_pump->proc_lock); + return 0; + } + + thread_pump->proc_count--; + work_item = LIST_ENTRY(list_head(&thread_pump->proc_queue), struct work_item, entry); + list_remove(&work_item->entry); + ReleaseSRWLockExclusive(&thread_pump->proc_lock); + + if (FAILED(hr = ID3DX10DataLoader_Decompress(work_item->loader, &data, &size))) + { + if (work_item->result) + *work_item->result = hr; + work_item_free(work_item, FALSE); + continue; + } + + if (thread_pump->device_count == THREAD_PUMP_EXITING) + { + work_item_free(work_item, TRUE); + return 0; + } + + if (FAILED(hr = ID3DX10DataProcessor_Process(work_item->processor, data, size))) + { + if (work_item->result) + *work_item->result = hr; + work_item_free(work_item, FALSE); + continue; + } + + AcquireSRWLockExclusive(&thread_pump->device_lock); + if (thread_pump->device_count == THREAD_PUMP_EXITING) + { + ReleaseSRWLockExclusive(&thread_pump->device_lock); + work_item_free(work_item, TRUE); + return 0; + } + + list_add_tail(&thread_pump->device_queue, &work_item->entry); + ++thread_pump->device_count; + ReleaseSRWLockExclusive(&thread_pump->device_lock); + } + return 0; +} + HRESULT WINAPI D3DX10CreateThreadPump(UINT io_threads, UINT proc_threads, ID3DX10ThreadPump **pump) { struct thread_pump *object; + int i; TRACE("io_threads %u, proc_threads %u, pump %p.\n", io_threads, proc_threads, pump); if (io_threads >= 1024 || proc_threads >= 1024) return E_FAIL; - if (!(object = calloc(1, sizeof(*object)))) + if (!io_threads) + io_threads = 1; + if (!proc_threads) + { + SYSTEM_INFO info; + + GetSystemInfo(&info); + proc_threads = info.dwNumberOfProcessors; + } + + if (!(object = calloc(1, FIELD_OFFSET(struct thread_pump, threads[io_threads + proc_threads])))) return E_OUTOFMEMORY; object->ID3DX10ThreadPump_iface.lpVtbl = &thread_pump_vtbl; object->refcount = 1; + InitializeSRWLock(&object->io_lock); + InitializeConditionVariable(&object->io_cv); + list_init(&object->io_queue); + InitializeSRWLock(&object->proc_lock); + InitializeConditionVariable(&object->proc_cv); + list_init(&object->proc_queue); + InitializeSRWLock(&object->device_lock); + list_init(&object->device_queue); + object->thread_count = io_threads + proc_threads; + + for (i = 0; i < object->thread_count; ++i) + { + object->threads[i] = CreateThread(NULL, 0, i < io_threads ? io_thread : proc_thread, object, 0, NULL); + if (!object->threads[i]) + { + ID3DX10ThreadPump_Release(&object->ID3DX10ThreadPump_iface); + return E_FAIL; + } + } *pump = &object->ID3DX10ThreadPump_iface; return S_OK; -- GitLab https://gitlab.winehq.org/wine/wine/-/merge_requests/272

June 26, 2022

[PATCH v4 2/9] d3dx10: Add D3DX10CreateThreadPump stub.

by Piotr Caban

From: Piotr Caban <piotr(a)codeweavers.com> Signed-off-by: Piotr Caban <piotr(a)codeweavers.com> --- dlls/d3dx10_43/async.c | 123 ++++++++++++++++++++++++++++++++++ dlls/d3dx10_43/d3dx10_43.spec | 2 +- include/d3dx10core.h | 1 + 3 files changed, 125 insertions(+), 1 deletion(-) diff --git a/dlls/d3dx10_43/async.c b/dlls/d3dx10_43/async.c index bb1cf30a217..7913f634c21 100644 --- a/dlls/d3dx10_43/async.c +++ b/dlls/d3dx10_43/async.c @@ -609,3 +609,126 @@ HRESULT WINAPI D3DX10PreprocessShaderFromMemory(const char *data, SIZE_T data_si return E_NOTIMPL; } + +struct thread_pump +{ + ID3DX10ThreadPump ID3DX10ThreadPump_iface; + LONG refcount; +}; + +static inline struct thread_pump *impl_from_ID3DX10ThreadPump(ID3DX10ThreadPump *iface) +{ + return CONTAINING_RECORD(iface, struct thread_pump, ID3DX10ThreadPump_iface); +} + +static HRESULT WINAPI thread_pump_QueryInterface(ID3DX10ThreadPump *iface, REFIID riid, void **out) +{ + TRACE("iface %p, riid %s, out %p.\n", iface, debugstr_guid(riid), out); + + if (IsEqualGUID(riid, &IID_ID3DX10ThreadPump) + || IsEqualGUID(riid, &IID_IUnknown)) + { + ID3DX10ThreadPump_AddRef(iface); + *out = iface; + return S_OK; + } + + WARN("%s not implemented, returning E_NOINTERFACE.\n", debugstr_guid(riid)); + *out = NULL; + return E_NOINTERFACE; +} + +static ULONG WINAPI thread_pump_AddRef(ID3DX10ThreadPump *iface) +{ + struct thread_pump *thread_pump = impl_from_ID3DX10ThreadPump(iface); + ULONG refcount = InterlockedIncrement(&thread_pump->refcount); + + TRACE("%p increasing refcount to %lu.\n", iface, refcount); + + return refcount; +} + +static ULONG WINAPI thread_pump_Release(ID3DX10ThreadPump *iface) +{ + struct thread_pump *thread_pump = impl_from_ID3DX10ThreadPump(iface); + ULONG refcount = InterlockedDecrement(&thread_pump->refcount); + + TRACE("%p decreasing refcount to %lu.\n", iface, refcount); + + if (!refcount) + free(thread_pump); + + return refcount; +} + +static HRESULT WINAPI thread_pump_AddWorkItem(ID3DX10ThreadPump *iface, ID3DX10DataLoader *loader, + ID3DX10DataProcessor *processor, HRESULT *result, void **object) +{ + FIXME("iface %p, loader %p, processor %p, result %p, object %p stub!\n", + iface, loader, processor, result, object); + return E_NOTIMPL; +} + +static UINT WINAPI thread_pump_GetWorkItemCount(ID3DX10ThreadPump *iface) +{ + FIXME("iface %p stub!\n", iface); + return 0; +} + +static HRESULT WINAPI thread_pump_WaitForAllItems(ID3DX10ThreadPump *iface) +{ + FIXME("iface %p stub!\n", iface); + return E_NOTIMPL; +} + +static HRESULT WINAPI thread_pump_ProcessDeviceWorkItems(ID3DX10ThreadPump *iface, UINT count) +{ + FIXME("iface %p, count %u stub!\n", iface, count); + return E_NOTIMPL; +} + +static HRESULT WINAPI thread_pump_PurgeAllItems(ID3DX10ThreadPump *iface) +{ + FIXME("iface %p stub!\n", iface); + return E_NOTIMPL; +} + +static HRESULT WINAPI thread_pump_GetQueueStatus(ID3DX10ThreadPump *iface, + UINT *io_queue, UINT *process_queue, UINT *device_queue) +{ + FIXME("iface %p, io_queue %p, process_queue %p, device_queue %p stub!\n", + iface, io_queue, process_queue, device_queue); + return E_NOTIMPL; +} + +static const ID3DX10ThreadPumpVtbl thread_pump_vtbl = +{ + thread_pump_QueryInterface, + thread_pump_AddRef, + thread_pump_Release, + thread_pump_AddWorkItem, + thread_pump_GetWorkItemCount, + thread_pump_WaitForAllItems, + thread_pump_ProcessDeviceWorkItems, + thread_pump_PurgeAllItems, + thread_pump_GetQueueStatus +}; + +HRESULT WINAPI D3DX10CreateThreadPump(UINT io_threads, UINT proc_threads, ID3DX10ThreadPump **pump) +{ + struct thread_pump *object; + + TRACE("io_threads %u, proc_threads %u, pump %p.\n", io_threads, proc_threads, pump); + + if (io_threads >= 1024 || proc_threads >= 1024) + return E_FAIL; + + if (!(object = calloc(1, sizeof(*object)))) + return E_OUTOFMEMORY; + + object->ID3DX10ThreadPump_iface.lpVtbl = &thread_pump_vtbl; + object->refcount = 1; + + *pump = &object->ID3DX10ThreadPump_iface; + return S_OK; +} diff --git a/dlls/d3dx10_43/d3dx10_43.spec b/dlls/d3dx10_43/d3dx10_43.spec index 95160a067c5..2359c7c6f02 100644 --- a/dlls/d3dx10_43/d3dx10_43.spec +++ b/dlls/d3dx10_43/d3dx10_43.spec @@ -1,4 +1,4 @@ -@ stub D3DX10CreateThreadPump(long long ptr) +@ stdcall D3DX10CreateThreadPump(long long ptr) @ stdcall D3DX10CheckVersion(long long) @ stub D3DX10CompileFromFileA(str ptr ptr str str long long ptr ptr ptr ptr) @ stub D3DX10CompileFromFileW(wstr ptr ptr str str long long ptr ptr ptr ptr) diff --git a/include/d3dx10core.h b/include/d3dx10core.h index a9ba7854e90..cca9052cc13 100644 --- a/include/d3dx10core.h +++ b/include/d3dx10core.h @@ -298,3 +298,4 @@ HRESULT WINAPI D3DX10CreateFontW(ID3D10Device *device, INT height, UINT width, U UINT miplevels, BOOL italic, UINT charset, UINT precision, UINT quality, UINT pitchandfamily, const WCHAR *facename, ID3DX10Font **font); HRESULT WINAPI D3DX10CreateSprite(ID3D10Device *device, UINT size, ID3DX10Sprite **sprite); +HRESULT WINAPI D3DX10CreateThreadPump(UINT io_threads, UINT proc_threads, ID3DX10ThreadPump **pump); -- GitLab https://gitlab.winehq.org/wine/wine/-/merge_requests/272

June 26, 2022

[PATCH v4 1/9] d3dx10/tests: Fix texture leak in check_resource_data.

by Piotr Caban

From: Piotr Caban <piotr(a)codeweavers.com> Signed-off-by: Piotr Caban <piotr(a)codeweavers.com> --- dlls/d3dx10_43/tests/d3dx10.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/dlls/d3dx10_43/tests/d3dx10.c b/dlls/d3dx10_43/tests/d3dx10.c index 1c28a62b700..d1c772197b9 100644 --- a/dlls/d3dx10_43/tests/d3dx10.c +++ b/dlls/d3dx10_43/tests/d3dx10.c @@ -1237,7 +1237,7 @@ static void check_resource_data(ID3D10Resource *resource, const struct test_imag ok_(__FILE__, line)(hr == S_OK, "Map failed, hr %#x.\n", hr); if (hr != S_OK) { - ID3D10Texture2D_Unmap(readback, 0); + ID3D10Texture2D_Release(readback); return; } @@ -1253,6 +1253,7 @@ static void check_resource_data(ID3D10Resource *resource, const struct test_imag } ID3D10Texture2D_Unmap(readback, 0); + ID3D10Texture2D_Release(readback); } static void test_D3DX10UnsetAllDeviceObjects(void) @@ -2062,7 +2063,7 @@ static void test_D3DX10CreateAsyncTextureProcessor(void) CoUninitialize(); - ID3D10Device_Release(device); + ok(!ID3D10Device_Release(device), "Unexpected refcount.\n"); } static void test_get_image_info(void) @@ -2417,7 +2418,7 @@ static void test_create_texture(void) CoUninitialize(); - ID3D10Device_Release(device); + ok(!ID3D10Device_Release(device), "Unexpected refcount.\n"); } #define check_rect(rect, left, top, right, bottom) _check_rect(__LINE__, rect, left, top, right, bottom) -- GitLab https://gitlab.winehq.org/wine/wine/-/merge_requests/272

June 26, 2022

[PATCH v4 0/9] MR272: d3dx10: Add D3DX10CreateThreadPump implementation.

by Piotr Caban (＠piotr)

The GetWorkItemCount implementation may look strange. It's done this way because of how WaitForAllItems is implemented. Alternative solution is to introduce separate counter for GetWorkItemCount. -- v4: d3dx10/tests: Add D3DX10CreateThreadPump tests. d3dx10: Add ID3DX10ThreadPump:PurgeAllItems implementation. d3dx10: Add ID3DX10ThreadPump:GetQueueStatus implementation. d3dx10: Add ID3DX10ThreadPump:WaitForAllItems implementation. d3dx10: Add ID3DX10ThreadPump:ProcessDeviceWorkItems implementation. d3dx10: Add ID3DX10ThreadPump:GetWorkItemCount implementation. d3dx10: Add ID3DX10ThreadPump:AddWorkItem implementation. d3dx10: Add D3DX10CreateThreadPump stub. d3dx10/tests: Fix texture leak in check_resource_data. https://gitlab.winehq.org/wine/wine/-/merge_requests/272

June 26, 2022

Re: [PATCH 0/6] MR265: kernelbase: Implement PrefetchVirtualMemory.

by Jinoh Kang (＠iamahuman)

On Wed Jun 22 22:26:03 2022 +0000, Etaash Mathamsetty wrote: > what exactly is this for? > the if(!once++) It causes the log to be emitted only once. `if (!once++)` is a common pattern throughout the wine codebase. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/265#note_2723

June 26, 2022

[PATCH 2/2] ntdll: Preserve EFLAGS across syscall on x64.

by Jinoh Kang

From: Jinoh Kang <jinoh.kang.kr(a)gmail.com> Signed-off-by: Jinoh Kang <jinoh.kang.kr(a)gmail.com> --- dlls/ntdll/tests/exception.c | 1 - dlls/ntdll/unix/signal_x86_64.c | 10 ++++++---- 2 files changed, 6 insertions(+), 5 deletions(-) diff --git a/dlls/ntdll/tests/exception.c b/dlls/ntdll/tests/exception.c index 006da0336be..9f4477a4da1 100644 --- a/dlls/ntdll/tests/exception.c +++ b/dlls/ntdll/tests/exception.c @@ -5010,7 +5010,6 @@ static void test_syscall_clobbered_regs(void) ok(status == STATUS_INVALID_PARAMETER_1, "Got unexpected status %#lx.\n", status); ok((BYTE *)regs.rcx > (BYTE *)pNtWaitForMultipleObjects && (BYTE *)regs.rcx < (BYTE *)pNtWaitForMultipleObjects + 0x20, "Got unexpected rcx %s, pNtWaitForMultipleObjects %p.\n", wine_dbgstr_longlong(regs.rcx), pNtWaitForMultipleObjects); - todo_wine ok((regs.eflags & 0x400) != 0, "Expected direction flag to be set in EFLAGS (%#x).\n", regs.eflags); context.ContextFlags = CONTEXT_CONTROL; diff --git a/dlls/ntdll/unix/signal_x86_64.c b/dlls/ntdll/unix/signal_x86_64.c index 6c87e347eac..87ff148a628 100644 --- a/dlls/ntdll/unix/signal_x86_64.c +++ b/dlls/ntdll/unix/signal_x86_64.c @@ -3415,13 +3415,15 @@ __ASM_GLOBAL_FUNC( __wine_syscall_dispatcher, "movq 0x28(%rcx),%rdi\n\t" "movq 0x20(%rcx),%rsi\n\t" "movq 0x08(%rcx),%rbx\n\t" + "leaq 0x70(%rcx),%rsp\n\t" "testl $0x3,%edx\n\t" /* CONTEXT_CONTROL | CONTEXT_INTEGER */ "jnz 1f\n\t" - "movq 0x88(%rcx),%rsp\n\t" - "movq 0x70(%rcx),%rcx\n\t" /* frame->rip */ + "pop %rcx\n\t" /* frame->rip */ + "add $8,%rsp\n\t" + "popfq\n\t" + "pop %rsp\n\t" "jmpq *%rcx\n\t" - "1:\tleaq 0x70(%rcx),%rsp\n\t" - "testl $0x2,%edx\n\t" /* CONTEXT_INTEGER */ + "1:\ttestl $0x2,%edx\n\t" /* CONTEXT_INTEGER */ "jnz 1f\n\t" "movq (%rsp),%rcx\n\t" /* frame->rip */ "iretq\n" -- GitLab https://gitlab.winehq.org/wine/wine/-/merge_requests/314

June 26, 2022

[PATCH 1/2] ntdll/tests: Test for preservation of EFLAGS across syscall on x64.

by Jinoh Kang

From: Jinoh Kang <jinoh.kang.kr(a)gmail.com> Signed-off-by: Jinoh Kang <jinoh.kang.kr(a)gmail.com> --- dlls/ntdll/tests/exception.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+) diff --git a/dlls/ntdll/tests/exception.c b/dlls/ntdll/tests/exception.c index 820e435bc1b..006da0336be 100644 --- a/dlls/ntdll/tests/exception.c +++ b/dlls/ntdll/tests/exception.c @@ -4949,6 +4949,7 @@ static void test_syscall_clobbered_regs(void) struct regs { UINT64 rcx; + UINT32 eflags; }; static const BYTE code[] = { @@ -4959,6 +4960,7 @@ static void test_syscall_clobbered_regs(void) 0x48, 0x83, 0xe8, 0x08, /* subq $8,%rax */ 0x48, 0x89, 0x20, /* movq %rsp,0(%rax) */ 0x48, 0x89, 0xc4, /* movq %rax,%rsp */ + 0xfd, /* std */ 0x41, 0x50, /* push %r8 */ 0x53, 0x55, 0x57, 0x56, 0x41, 0x54, 0x41, 0x55, 0x41, 0x56, 0x41, 0x57, /* push %rbx, %rbp, %rdi, %rsi, %r12, %r13, %r14, %r15 */ @@ -4967,12 +4969,17 @@ static void test_syscall_clobbered_regs(void) /* pop %r15, %r14, %r13, %r12, %rsi, %rdi, %rbp, %rbx */ 0x41, 0x58, /* pop %r8 */ 0x49, 0x89, 0x48, 0x00, /* mov %rcx,(%r8) */ + 0x9c, /* pushfq */ + 0x59, /* pop %rcx */ + 0xfc, /* cld */ + 0x41, 0x89, 0x48, 0x08, /* mov %ecx,0x8(%r8) */ 0x5c, /* pop %rsp */ 0xc3, /* ret */ }; NTSTATUS (WINAPI *func)(void *arg1, void *arg2, struct regs *, void *call_addr); NTSTATUS (WINAPI *pNtCancelTimer)(HANDLE, BOOLEAN *); + NTSTATUS (WINAPI *pNtWaitForMultipleObjects)(DWORD, const HANDLE *, BOOLEAN, BOOLEAN, const LARGE_INTEGER *); HMODULE hntdll = GetModuleHandleA("ntdll.dll"); struct regs regs; CONTEXT context; @@ -4985,6 +4992,7 @@ static void test_syscall_clobbered_regs(void) memset(&regs, 0, sizeof(regs)); status = func((HANDLE)0xdeadbeef, NULL, &regs, pNtCancelTimer); ok(status == STATUS_INVALID_HANDLE, "Got unexpected status %#lx.\n", status); + ok((regs.eflags & 0x400) != 0, "Expected direction flag to be set in EFLAGS (%#x).\n", regs.eflags); /* After the syscall instruction rcx contains the address of the instruction next after syscall. */ ok((BYTE *)regs.rcx > (BYTE *)pNtCancelTimer && (BYTE *)regs.rcx < (BYTE *)pNtCancelTimer + 0x20, @@ -4994,28 +5002,42 @@ static void test_syscall_clobbered_regs(void) ok(status == STATUS_ACCESS_VIOLATION, "Got unexpected status %#lx.\n", status); ok((BYTE *)regs.rcx > (BYTE *)pNtCancelTimer && (BYTE *)regs.rcx < (BYTE *)pNtCancelTimer + 0x20, "Got unexpected rcx %s, pNtCancelTimer %p.\n", wine_dbgstr_longlong(regs.rcx), pNtCancelTimer); + ok((regs.eflags & 0x400) != 0, "Expected direction flag to be set in EFLAGS (%#x).\n", regs.eflags); + + pNtWaitForMultipleObjects = (void *)GetProcAddress(hntdll, "NtWaitForMultipleObjects"); + ok(!!pNtWaitForMultipleObjects, "NtWaitForMultipleObjects not found.\n"); + status = func((DWORD)0, (HANDLE *)NULL, &regs, pNtWaitForMultipleObjects); + ok(status == STATUS_INVALID_PARAMETER_1, "Got unexpected status %#lx.\n", status); + ok((BYTE *)regs.rcx > (BYTE *)pNtWaitForMultipleObjects && (BYTE *)regs.rcx < (BYTE *)pNtWaitForMultipleObjects + 0x20, + "Got unexpected rcx %s, pNtWaitForMultipleObjects %p.\n", wine_dbgstr_longlong(regs.rcx), pNtWaitForMultipleObjects); + todo_wine + ok((regs.eflags & 0x400) != 0, "Expected direction flag to be set in EFLAGS (%#x).\n", regs.eflags); context.ContextFlags = CONTEXT_CONTROL; status = func(GetCurrentThread(), &context, &regs, pNtGetContextThread); ok(status == STATUS_SUCCESS, "Got unexpected status %#lx.\n", status); ok((BYTE *)regs.rcx > (BYTE *)pNtGetContextThread && (BYTE *)regs.rcx < (BYTE *)pNtGetContextThread + 0x20, "Got unexpected rcx %s, pNtGetContextThread %p.\n", wine_dbgstr_longlong(regs.rcx), pNtGetContextThread); + ok((regs.eflags & 0x400) != 0, "Expected direction flag to be set in EFLAGS (%#x).\n", regs.eflags); status = func(GetCurrentThread(), &context, &regs, pNtSetContextThread); ok(status == STATUS_SUCCESS, "Got unexpected status %#lx.\n", status); ok((BYTE *)regs.rcx > (BYTE *)pNtGetContextThread && (BYTE *)regs.rcx < (BYTE *)pNtGetContextThread + 0x20, "Got unexpected rcx %s, pNtGetContextThread %p.\n", wine_dbgstr_longlong(regs.rcx), pNtGetContextThread); + ok((regs.eflags & 0x400) != 0, "Expected direction flag to be set in EFLAGS (%#x).\n", regs.eflags); context.ContextFlags = CONTEXT_INTEGER; status = func(GetCurrentThread(), &context, &regs, pNtGetContextThread); ok(status == STATUS_SUCCESS, "Got unexpected status %#lx.\n", status); ok((BYTE *)regs.rcx > (BYTE *)pNtGetContextThread && (BYTE *)regs.rcx < (BYTE *)pNtGetContextThread + 0x20, "Got unexpected rcx %s, pNtGetContextThread %p.\n", wine_dbgstr_longlong(regs.rcx), pNtGetContextThread); + ok((regs.eflags & 0x400) != 0, "Expected direction flag to be set in EFLAGS (%#x).\n", regs.eflags); status = func(GetCurrentThread(), &context, &regs, pNtSetContextThread); ok(status == STATUS_SUCCESS, "Got unexpected status %#lx.\n", status); ok((BYTE *)regs.rcx > (BYTE *)pNtSetContextThread && (BYTE *)regs.rcx < (BYTE *)pNtSetContextThread + 0x20, "Got unexpected rcx %s, pNtSetContextThread %p.\n", wine_dbgstr_longlong(regs.rcx), pNtSetContextThread); + ok((regs.eflags & 0x400) != 0, "Expected direction flag to be set in EFLAGS (%#x).\n", regs.eflags); } #elif defined(__arm__) -- GitLab https://gitlab.winehq.org/wine/wine/-/merge_requests/314

June 26, 2022

[PATCH 0/2] MR314: ntdll/tests: Preserve EFLAGS across syscall on x64.

by Jinoh Kang (＠iamahuman)

Signed-off-by: Jinoh Kang <jinoh.kang.kr(a)gmail.com> -- https://gitlab.winehq.org/wine/wine/-/merge_requests/314

June 26, 2022

Re: [PATCH v3 0/9] MR272: d3dx10: Add D3DX10CreateThreadPump implementation.

by Piotr Caban (＠piotr)

On Wed Jun 22 18:22:04 2022 +0000, Josh Simmons wrote: > Unsolicited nit, but since these are separately locked queues they > should probably be padded out to be on their own cache lines to avoid > false sharing. > (No idea whether this API is hot enough for that to be truly important, though) I don't think it will need additional optimizations. The fact that I have used separate locks is just an implementation detail. We can try to optimize it further in future if needed. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/272#note_2719

June 26, 2022