interrupts in when in NT version
Hi! I have Laurent Pinchart's safedisc patch applied and working for a while now. (and it works perfectly). But the recent interrupt changes tend to cause cvs conflicts and to break it. This patch is supposed to work only if winver is nt40 so very probably this applies only to NT. What the patch relies on is that the interrupt handling code (which was INSTR_EmulateInstruction before the changes), return EXCEPTION_ACCESS_VIOLATION if interrupt 0x01 is called (if I understand it correctly). This was done with the next patches: Index: dlls/ntdll/signal_i386.c =================================================================== RCS file: /home/wine/wine/dlls/ntdll/signal_i386.c,v retrieving revision 1.46 diff -u -p -r1.46 signal_i386.c --- dlls/ntdll/signal_i386.c 12 Sep 2002 22:07:03 -0000 1.46 +++ dlls/ntdll/signal_i386.c 24 Oct 2002 10:13:36 -0000 @@ -729,6 +729,7 @@ static void do_segv( CONTEXT *context, i { EXCEPTION_RECORD rec; DWORD page_fault_code = EXCEPTION_ACCESS_VIOLATION; + DWORD priv_instr_code = EXCEPTION_PRIV_INSTRUCTION; #ifdef CR2_sig /* we want the page-fault case to be fast */ @@ -758,8 +759,8 @@ static void do_segv( CONTEXT *context, i case T_SEGNPFLT: /* Segment not present exception */ case T_PROTFLT: /* General protection fault */ case T_UNKNOWN: /* Unknown fault code */ - if (INSTR_EmulateInstruction( context )) return; - rec.ExceptionCode = EXCEPTION_PRIV_INSTRUCTION; + if (INSTR_EmulateInstruction( context, &priv_instr_code )) return; + rec.ExceptionCode = priv_instr_code; break; case T_PAGEFLT: /* Page fault */ #ifdef CR2_sig Index: memory/instr.c =================================================================== RCS file: /home/wine/wine/memory/instr.c,v retrieving revision 1.19 diff -u -p -r1.19 instr.c --- memory/instr.c 23 Oct 2002 22:24:10 -0000 1.19 +++ memory/instr.c 24 Oct 2002 10:14:08 -0000 @@ -397,7 +397,7 @@ static void INSTR_outport( WORD port, in * * Emulate a privileged instruction. Returns TRUE if emulation successful. */ -BOOL INSTR_EmulateInstruction( CONTEXT86 *context ) +BOOL INSTR_EmulateInstruction( CONTEXT86 *context, DWORD *exception_code ) { int prefix, segprefix, prefixlen, len, repX, long_op, long_addr; SEGPTR gpHandler; @@ -685,6 +685,29 @@ BOOL INSTR_EmulateInstruction( CONTEXT86 break; /* Unable to emulate it */ case 0xcd: /* int <XX> */ +<<<<<<< instr.c + if (long_op) + { + ERR("int 0x%02x from 32-bit code is not supported.\n", instr[1]); + if (instr[1] == 0x01) + *exception_code = EXCEPTION_ACCESS_VIOLATION; + break; /* Unable to emulate it */ + } + else + { + FARPROC16 addr = INT_GetPMHandler( instr[1] ); + WORD *stack = get_stack( context ); + /* Push the flags and return address on the stack */ + *(--stack) = LOWORD(context->EFlags); + *(--stack) = context->SegCs; + *(--stack) = LOWORD(context->Eip) + prefixlen + 2; + add_stack(context, -3 * sizeof(WORD)); + /* Jump to the interrupt handler */ + context->SegCs = HIWORD(addr); + context->Eip = LOWORD(addr); + } + return TRUE; +======= if(!Dosvm.EmulateInterruptPM && !DPMI_LoadDosSystem()) ERR("could not initialize interrupt handling\n"); else { @@ -693,6 +716,7 @@ BOOL INSTR_EmulateInstruction( CONTEXT86 return TRUE; } break; /* Unable to emulate it */ +>>>>>>> 1.19 case 0xcf: /* iret */ if (long_op) The second one is pretty ugly since I have not resolved the cvs conflict (I'm waiting for your input :), but the important part is this: + if (long_op) + { + ERR("int 0x%02x from 32-bit code is not supported.\n", instr[1]); + if (instr[1] == 0x01) + *exception_code = EXCEPTION_ACCESS_VIOLATION; + break; /* Unable to emulate it */ + } Here is the EXCEPTION_ACCESS_VIOLATION is returned if interrupt 0x01 is called. This was good when interrupts from 32-bit code were not supported. But now we need a special case for this. I think that on NT calling 0x01 is forbidden. But I have not tested this (and will not, because I dont have windows devel tools - I just want to run my Longman Dictionary :) I would like if this could be fixed properly and that would work with the safedisk patch, if thats possible. Thanks Zsolt
On Thu, Oct 24, 2002 at 12:26:15PM +0200, Zsolt Rizsanyi wrote:
But the recent interrupt changes tend to cause cvs conflicts and to break it. This patch is supposed to work only if winver is nt40 so very probably this applies only to NT. ... I would like if this could be fixed properly and that would work with the safedisk patch, if thats possible.
and on 28 May 2002 18:58:51 -0700, Alexandre wrote:
Laurent Pinchart <laurent.pinchart(a)skynet.be> writes:
There are two patches in there. The first one sets a fault handler for the SharedUserData. This looks fine to me, and will probably be applied when memory/emulate.c will be applied.
It still isn't clear to me why you need a fault handler at all. You could simply put the right data into it at allocation time.
The second one has been written by Alexandre to fix the process suspension at creation time. Once again I'd like to hear from him to know what he doesn't like about the code, as he's the one who wrote it :-)
The main problem is that it isn't generic enough, it only applies to the create process request but other requests have the same problem and should be fixed too. I'll work on a more general solution.
I looks like the more general solution was never done due to time/lack of nagging reasons? Ciao Jörg -- Joerg Mayer <jmayer(a)loplof.de> I found out that "pro" means "instead of" (as in proconsul). Now I know what proactive means.
Joerg Mayer <jmayer(a)loplof.de> writes:
and on 28 May 2002 18:58:51 -0700, Alexandre wrote:
The main problem is that it isn't generic enough, it only applies to the create process request but other requests have the same problem and should be fixed too. I'll work on a more general solution.
I looks like the more general solution was never done due to time/lack of nagging reasons?
Well it's part of the "finalize server protocol" task that's supposed to be done for 1.0. So it will get done eventually... -- Alexandre Julliard julliard(a)winehq.com
On Friday 25 October 2002 13:46, Joerg Mayer wrote:
On Thu, Oct 24, 2002 at 12:26:15PM +0200, Zsolt Rizsanyi wrote:
But the recent interrupt changes tend to cause cvs conflicts and to break it. This patch is supposed to work only if winver is nt40 so very probably this applies only to NT.
...
I would like if this could be fixed properly and that would work with the safedisk patch, if thats possible.
and on 28 May 2002 18:58:51 -0700, Alexandre wrote:
Laurent Pinchart <laurent.pinchart(a)skynet.be> writes:
There are two patches in there. The first one sets a fault handler for the SharedUserData. This looks fine to me, and will probably be applied when memory/emulate.c will be applied.
It still isn't clear to me why you need a fault handler at all. You could simply put the right data into it at allocation time.
The second one has been written by Alexandre to fix the process suspension at creation time. Once again I'd like to hear from him to know what he doesn't like about the code, as he's the one who wrote it :-)
The main problem is that it isn't generic enough, it only applies to the create process request but other requests have the same problem and should be fixed too. I'll work on a more general solution.
I looks like the more general solution was never done due to time/lack of nagging reasons?
Yes. Alexandre has not yet did the proper fix for the problems on thread suspension. I dont really know the reasons, but maybe it was the lack of nagging :) I think I'm the only one who uses this safedisc patch, and I was not too loud about that. But this (thread suspension) has nothing to do with interrupts. As much as I know the int 0x01 is called by the program to test if it has a sane environment (check to see if the program is not debugged). Could somebody answer/test if calling interrupt int 0x01 is allowed from a win32 program in WinNT environment? I would pretty like to know that! Thanks Zsolt
Rizsanyi Zsolt a écrit :
Could somebody answer/test if calling interrupt int 0x01 is allowed from a win32 program in WinNT environment? I would pretty like to know that!
I don't think it is it was allowed on Win9x, and is used in this context IIRC as a ring 3 to ring 0 transition code (how to create a cheap call gate) A+
participants (5)
-
Alexandre Julliard -
Eric Pouech -
Joerg Mayer -
Rizsanyi Zsolt -
Zsolt Rizsanyi