digitalmars.D - Idea: Implement Concurrent GC on the windows Platform
- Benjamin Thaut (29/29) Jan 06 2019 Hi,
- Andre Pany (7/13) Jan 06 2019 That sounds really great, could you create an issue for this,
- Radu (4/10) Jan 06 2019 Rainer did some experimenting [1] in the past, should be nice to
- rikki cattermole (92/92) Jan 13 2019 It does not look like this is a viable option.
Hi, I recently came accross this article: https://blogs.msdn.microsoft.com/vcblog/2018/09/26/step-back-going-back-in-c-time/ which mentions the following "new" kernel API of Windows. https://docs.microsoft.com/de-de/windows/desktop/api/processsnapshot/nf-processsnapshot-psscapturesnapshot It has the following properties (See "under the hood" chapter in the first link.) <quote> * Create a ‘snapshot’ which looks suspiciously like the child process of an existing process that has no threads running. * Mark the processes memory, it’s page tables (Wikipedia), as copy-on-write (Wikipedia). That means that whenever a table is written to, the table is copied. </quote> Now this looks a lot like a fork. At least it does what you would need from a fork for implementing the concurrent GC. If I remember correctly the main problem of the concurrent GC on windows was that there is no equivalent of fork. The only problem is that there are no threads running after the "fork" but for that CreateRemoteThread could be used. Unfortunately I currently don't have the time to try this out myself. But in case someone has time and is interrested and this does work out I could see the concurrent GC become a viable replacement option for the current gc. Main downside is, that this API has been added with Windows 8.1. But as windows 7 is going out of support starting in 2020 I think this is not that big of a downside. Kind Regards Benjamin Thaut
Jan 06 2019
Andre Pany <andre s-e-a-p.de> On Sunday, 6 January 2019 at 18:32:58 UTC, Benjamin Thaut wrote:Hi, I recently came accross this article: https://blogs.msdn.microsoft.com/vcblog/2018/09/26/step-back-going-back-in-c-time/ which mentions the following "new" kernel API of Windows. https://docs.microsoft.com/de-de/windows/desktop/api/processsnapshot/nf-processsnapshot-psscapturesnapshot [...]That sounds really great, could you create an issue for this, therefore the idea do not get lost. That is definitely an idea for which a donation round make sense... Kind regards Andre
Jan 06 2019
On Sunday, 6 January 2019 at 18:32:58 UTC, Benjamin Thaut wrote:Hi, I recently came accross this article: https://blogs.msdn.microsoft.com/vcblog/2018/09/26/step-back-going-back-in-c-time/ which mentions the following "new" kernel API of Windows. https://docs.microsoft.com/de-de/windows/desktop/api/processsnapshot/nf-processsnapshot-psscapturesnapshot [...]Rainer did some experimenting [1] in the past, should be nice to see if that function can improve it. 1 - http://rainers.github.io/visuald/druntime/concurrentgc.html
Jan 06 2019
It does not look like this is a viable option.
I could not get a remote thread created. Code below for anyone else that
wants to have a play. On my machine it'll crash with error 5 trying to
create a thread.
---
import std.stdio;
import core.sys.windows.basetsd;
import core.sys.windows.windows;
import core.thread;
extern(Windows) {
alias HPSS = HANDLE;
enum PSS_CAPTURE_FLAGS {
PSS_CAPTURE_NONE = 0x00000000,
PSS_CAPTURE_VA_CLONE = 0x00000001,
PSS_CAPTURE_RESERVED_00000002 = 0x00000002,
PSS_CAPTURE_HANDLES = 0x00000004,
PSS_CAPTURE_HANDLE_NAME_INFORMATION = 0x00000008,
PSS_CAPTURE_HANDLE_BASIC_INFORMATION = 0x00000010,
PSS_CAPTURE_HANDLE_TYPE_SPECIFIC_INFORMATION = 0x00000020,
PSS_CAPTURE_HANDLE_TRACE = 0x00000040,
PSS_CAPTURE_THREADS = 0x00000080,
PSS_CAPTURE_THREAD_CONTEXT = 0x00000100,
PSS_CAPTURE_THREAD_CONTEXT_EXTENDED = 0x00000200,
PSS_CAPTURE_RESERVED_00000400 = 0x00000400,
PSS_CAPTURE_VA_SPACE = 0x00000800,
PSS_CAPTURE_VA_SPACE_SECTION_INFORMATION = 0x00001000,
PSS_CAPTURE_IPT_TRACE = 0x00002000,
PSS_CREATE_BREAKAWAY_OPTIONAL = 0x04000000,
PSS_CREATE_BREAKAWAY = 0x08000000,
PSS_CREATE_FORCE_BREAKAWAY = 0x10000000,
PSS_CREATE_USE_VM_ALLOCATIONS = 0x20000000,
PSS_CREATE_MEASURE_PERFORMANCE = 0x40000000,
PSS_CREATE_RELEASE_SECTION = 0x80000000
}
enum PSS_QUERY_INFORMATION_CLASS {
PSS_QUERY_PROCESS_INFORMATION = 0,
PSS_QUERY_VA_CLONE_INFORMATION = 1,
PSS_QUERY_AUXILIARY_PAGES_INFORMATION = 2,
PSS_QUERY_VA_SPACE_INFORMATION = 3,
PSS_QUERY_HANDLE_INFORMATION = 4,
PSS_QUERY_THREAD_INFORMATION = 5,
PSS_QUERY_HANDLE_TRACE_INFORMATION = 6,
PSS_QUERY_PERFORMANCE_COUNTERS = 7
}
struct PSS_VA_CLONE_INFORMATION {
HANDLE VaCloneHandle;
}
DWORD GetProcessId(HANDLE);
DWORD PssCaptureSnapshot(HANDLE, PSS_CAPTURE_FLAGS, DWORD, HPSS*);
DWORD PssQuerySnapshot(HPSS, PSS_QUERY_INFORMATION_CLASS, void*,
DWORD);
DWORD PssFreeSnapshot(HANDLE, HPSS);
}
__gshared Thread someThread;
void main() {
writeln("Hello from host");
HPSS snapshotRef;
DWORD status = PssCaptureSnapshot(GetCurrentProcess(),
PSS_CAPTURE_FLAGS.PSS_CAPTURE_VA_CLONE,
CONTEXT_ALL, &snapshotRef);
writeln("status 1: ", status);
assert(status == 0);
readln;
PSS_VA_CLONE_INFORMATION cloneInfo;
status = PssQuerySnapshot(snapshotRef,
PSS_QUERY_INFORMATION_CLASS.PSS_QUERY_VA_CLONE_INFORMATION, &cloneInfo,
PSS_VA_CLONE_INFORMATION.sizeof);
writeln("status 2: ", status);
assert(status == 0);
writeln(GetProcessId(cloneInfo.VaCloneHandle));
readln;
HANDLE remoteThreadRef =
CreateRemoteThread(cloneInfo.VaCloneHandle, null, 0, &threadFuncInit,
null, CREATE_SUSPENDED, null);
writeln("status 3: ", GetLastError());
assert(remoteThreadRef !is null);
readln;
status = PssFreeSnapshot(GetCurrentProcess(), snapshotRef);
writeln("status 4: ", status);
}
void threadFunc() {
writeln("Hi from thread!");
}
extern(Windows) {
uint threadFuncInit(void*) {
writeln("WHAT???");
stdout.flush;
someThread = new Thread(&threadFunc);
someThread.start;
return 0;
}
}
Jan 13 2019