[vbox-dev] 回复: GDT entries syncing between qemu mode & raw mode bug

[Michal Necasek]

Unless you can provide a reproduction scenario, I don't think we'll look at the patch. "Android" is no better than "Linux". A preconfigured VM would be acceptable if you can't provide any real information. 

- Michal 

----- Original Message ----- 
From: manjian2006 at gmail.com 
To: michal.necasek at oracle.com, vbox-dev at virtualbox.org 
Sent: Thursday, December 24, 2015 11:29:43 AM GMT +01:00 Amsterdam / Berlin / Bern / Rome / Stockholm / Vienna 
Subject: 回复: [vbox-dev] GDT entries syncing between qemu mode & raw mode bug 

android x86 

来自我的华为手机 



-------- 原始邮件 -------- 
主题：Re: [vbox-dev] GDT entries syncing between qemu mode & raw mode bug 
发件人：Michal Necasek 
收件人：vbox-dev at virtualbox.org 
抄送： 





This doesn't look unreasonable, but one important thing is missing: How exactly do we reproduce the problem? If we can't reproduce it, we can't verify that the patch does anything useful. I'm sure you know that "linux" is not a particularly accurate description of the guest OS. 

Thanks, 
Michal 

----- Original Message ----- 
From: manjian2006 at gmail.com 
To: vbox-dev at virtualbox.org 
Sent: Thursday, December 24, 2015 4:23:56 AM GMT +01:00 Amsterdam / Berlin / Bern / Rome / Stockholm / Vienna 
Subject: [vbox-dev] GDT entries syncing between qemu mode & raw mode bug 

Hi friends, 
When I turn off the vt of my machine, my linux occasionally meets 
a coredump, which eip points to the following something like this: 
movl %ecx, %gs:0 
And that's the tls of linux conventions controlled by gdt. 
And the wired thing is that when a gdb attached into the 
will-coredump process, and enter 'continue', the process that should 
trigger coredump continues to run normally. 
I guess this is a syncing bug between raw mode and qemu mode, So I 
add patch like following: 
diff --git a/src/recompiler/VBoxRecompiler.c b/src/recompiler/VBoxRecompiler.c 
index d74eeb0..fdba032 100644 
--- a/src/recompiler/VBoxRecompiler.c 
+++ b/src/recompiler/VBoxRecompiler.c 
@@ -2063,6 +2063,7 @@ void remR3RecordCall(CPUX86State *env) 
#endif 
} 

+#define DEEP_SYNC_SEGMENT_REG /* add by linsh for VT */ 

/** 
* Syncs the internal REM state with the VM. 
@@ -2089,7 +2090,14 @@ REMR3DECL(int) REMR3State(PVM pVM, PVMCPU pVCpu) 
uint8_t u8TrapNo; 
uint32_t uCpl; 
int rc; 
+#ifdef DEEP_SYNC_SEGMENT_REG /* add by linsh for VT */ 

+ VBOXGDTR GDTR; 
+ RTGCPTR GCPtrGDT; 
+ unsigned iGDT; 
+ unsigned cGDTs; 
+ 
+#endif 
STAM_PROFILE_START(&pVM->rem.s.StatsState, a); 
Log2(("REMR3State:\n")); 

@@ -2406,7 +2414,52 @@ REMR3DECL(int) REMR3State(PVM pVM, PVMCPU pVCpu) 
SYNC_IN_SREG(&pVM->rem.s.Env, GS, &pVM->rem.s.Env.segs[R_GS], &pCtx->gs); 
/** @todo need to find a way to communicate potential GDT/LDT changes and thread switches. The selector might 
* be the same but not the base/limit. */ 
+#ifdef DEEP_SYNC_SEGMENT_REG /* add by linsh for VT */ 

+/* 
+*Check for the base address of the SegmentCache 
+*/ 
+#define DEEP_SYNC_SEG_REG(a_pRemSReg,a_SReg,a_pVBoxSReg,a_pGDTE,a_Sel) \ 
+ do \ 
+ { \ 
+ if (a_Sel == (((a_pVBoxSReg)->Sel) >> 3)/* && (a_Sel == 6 || a_Sel == 7) */) \ 
+ { \ 
+ if ((a_pRemSReg)->base != X86DESC_BASE(a_pGDTE) ) \ 
+ { \ 
+ Log2(("REMR3State: " #a_SReg " base had changed from %08x to %08x!\n", \ 
+ (a_pRemSReg)->base, X86DESC_BASE(a_pGDTE))); \ 
+ (a_pRemSReg)->base = X86DESC_BASE(a_pGDTE); \ 
+ (a_pRemSReg)->limit = X86DESC_LIMIT_G(a_pGDTE); \ 
+ } \ 
+ } \ 
+ } while (0) 
+ 
+ /* Get the GDTR */ 
+ CPUMGetGuestGDTR(pVCpu, &GDTR); 
+ if (GDTR.cbGdt < sizeof(X86DESC)) 
+ { 
+ Log2(("REMR3State: No GDT entries...\n")); 
+ return VINF_SUCCESS; 
+ } 
+ 
+ GCPtrGDT = GDTR.pGdt; 
+ cGDTs = ((unsigned)GDTR.cbGdt + 1) / sizeof(X86DESC); 
+ 
+ for (iGDT = 0; iGDT < cGDTs; iGDT++, GCPtrGDT += sizeof(X86DESC)) 
+ { 
+ X86DESC GDTE; 
+ int rc = PGMPhysSimpleReadGCPtr(pVCpu, &GDTE, GCPtrGDT, sizeof(GDTE)); 
+ if (RT_SUCCESS(rc)) 
+ { 
+ if (GDTE.Gen.u1Present) 
+ { 
+ Log2(("REMR3State: iGDT:%d,base:%08x,limit:%08x\n",iGDT,X86DESC_BASE(&GDTE),X86DESC_LIMIT_G(&GDTE))); 
+ DEEP_SYNC_SEG_REG(&pVM->rem.s.Env.segs[R_GS], GS, &pCtx->gs, &GDTE, iGDT ); 
+ } 
+ } 
+ } 
+ 
+#endif 

The coredump never ever happend again. 
-- 
Lin Zuojian 

_______________________________________________ 
vbox-dev mailing list 
vbox-dev at virtualbox.org 
https://www.virtualbox.org/mailman/listinfo/vbox-dev 

_______________________________________________ 
vbox-dev mailing list 
vbox-dev at virtualbox.org 
https://www.virtualbox.org/mailman/listinfo/vbox-dev 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.virtualbox.org/pipermail/vbox-dev/attachments/20151225/903898d5/attachment.html>