From: Ben Gras <ben@minix3.org>
Date: Sun, 6 Jan 2013 18:18:41 +0000 (+0000)
Subject: kernel: trap-dependent state restore, trace fixes
X-Git-Tag: v3.2.1~135
X-Git-Url: http://zhaoyanbai.com/repos/%22http:/www.isc.org/icons/doc/syslog.conf.5.html?a=commitdiff_plain;h=604046faf3601a19517f26236bc3bad09a740bdc;p=minix.git

kernel: trap-dependent state restore, trace fixes

	. restore state depends on how saving of state was done;
	  also remember trap style in sig context
	. actually set and restore TRACEBIT with new trap styles;
	  have to remove it once process enters kernel though, done
	  in debug trap exception handler
	. introduce MF_STEP that makes arch-specific code
	  turn on trace bit instead of setting TRACEBIT directly,
	  a bit more arch-friendly and avoids keeping precious
	  state in per-process PSW arch-dependently
---

diff --git a/include/arch/i386/include/signal.h b/include/arch/i386/include/signal.h
index cf07f6277..c3436460a 100644
--- a/include/arch/i386/include/signal.h
+++ b/include/arch/i386/include/signal.h
@@ -25,6 +25,7 @@ struct sigframe {		/* stack frame created for signalled process */
 };
 
 struct sigcontext {
+  int trap_style;		/* how should context be restored? KTS_* */
   int sc_flags;			/* sigstack state to restore (including
 				 * MF_FPU_INITIALIZED)
 				 */
diff --git a/kernel/arch/arm/arch_system.c b/kernel/arch/arm/arch_system.c
index 7e1922495..8852cd123 100644
--- a/kernel/arch/arm/arch_system.c
+++ b/kernel/arch/arm/arch_system.c
@@ -48,7 +48,8 @@ void arch_proc_reset(struct proc *pr)
         	pr->p_reg.psr = INIT_PSR;
 }
 
-void arch_proc_setcontext(struct proc *p, struct stackframe_s *state, int isuser)
+void arch_proc_setcontext(struct proc *p, struct stackframe_s *state,
+	int isuser, int trapstyle)
 {
 }
 
diff --git a/kernel/arch/i386/arch_system.c b/kernel/arch/i386/arch_system.c
index 15ffa5cd5..14ce73e06 100644
--- a/kernel/arch/i386/arch_system.c
+++ b/kernel/arch/i386/arch_system.c
@@ -182,7 +182,7 @@ void arch_proc_reset(struct proc *pr)
 	pr->p_reg.ds = USER_DS_SELECTOR;
 
 	/* set full context and make sure it gets restored */
-	arch_proc_setcontext(pr, &reg, 0);
+	arch_proc_setcontext(pr, &reg, 0, KTS_FULLCONTEXT);
 }
 
 void arch_set_secondary_ipc_return(struct proc *p, u32_t val)
@@ -511,14 +511,21 @@ struct proc * arch_finish_switch_to_user(void)
 	*((reg_t *)stk) = (reg_t) p;
 
 	/* make sure IF is on in FLAGS so that interrupts won't be disabled
-	 * once p's context is restored. this should not be possible.
+	 * once p's context is restored.
 	 */
-        assert(p->p_reg.psw & (1L << 9));
+        p->p_reg.psw |= IF_MASK;
+
+	/* Set TRACEBIT state properly. */
+	if(p->p_misc_flags & MF_STEP)
+        	p->p_reg.psw |= TRACEBIT;
+	else
+        	p->p_reg.psw &= ~TRACEBIT;
 
 	return p;
 }
 
-void arch_proc_setcontext(struct proc *p, struct stackframe_s *state, int isuser)
+void arch_proc_setcontext(struct proc *p, struct stackframe_s *state,
+	int isuser, int trap_style)
 {
 	if(isuser) {
 		/* Restore user bits of psw from sc, maintain system bits
@@ -555,7 +562,7 @@ void arch_proc_setcontext(struct proc *p, struct stackframe_s *state, int isuser
 	}
 	if(p->p_seg.p_kern_trap_style == KTS_NONE)
 		printf("WARNINIG: setting full context of out-of-kernel process\n");
-	p->p_seg.p_kern_trap_style = KTS_FULLCONTEXT;
+	p->p_seg.p_kern_trap_style = trap_style;
 }
 
 void restore_user_context(struct proc *p)
@@ -666,4 +673,3 @@ static void ser_init(void)
 	outb(COM1_LCR, lcr);
 }
 #endif
-
diff --git a/kernel/arch/i386/exception.c b/kernel/arch/i386/exception.c
index 3a4064d8d..08e73e419 100644
--- a/kernel/arch/i386/exception.c
+++ b/kernel/arch/i386/exception.c
@@ -228,6 +228,26 @@ void exception_handler(int is_nested, struct exception_frame * frame)
 		frame->eip = (reg_t) __frstor_failure;
 		return;
 	}
+
+  	if(frame->vector == DEBUG_VECTOR
+		&& (saved_proc->p_reg.psw & TRACEBIT)
+		&& (saved_proc->p_seg.p_kern_trap_style == KTS_NONE)) {
+		/* Getting a debug trap in the kernel is legitimate
+		 * if a traced process entered the kernel using sysenter
+		 * or syscall; the trap flag is not cleared then.
+		 *
+		 * It triggers on the first kernel entry so the trap
+		 * style is still KTS_NONE.
+		 */
+
+		frame->eflags &= ~TRACEBIT;
+
+		return;
+
+		/* If control passes, this case is not recognized as legitimate
+		 * and we panic later on after all.
+		 */
+	}
   }
 
   if(frame->vector == PAGE_FAULT_VECTOR) {
diff --git a/kernel/arch/i386/mpx.S b/kernel/arch/i386/mpx.S
index 75305e7aa..68f7a0b1d 100644
--- a/kernel/arch/i386/mpx.S
+++ b/kernel/arch/i386/mpx.S
@@ -240,6 +240,11 @@ syscall_sysenter_common:
 	mov	%esi, SPREG(%ebp)	/* esi is return esp */
 	mov	%edx, PCREG(%ebp)	/* edx is return eip */
 
+	/* save PSW */
+	pushf
+	pop	%edx
+	mov	%edx, PSWREG(%ebp)
+
 	/* check for call type; do_ipc? */
 	cmp	$IPCVEC_UM, %edi
 	jz	ipc_entry_common
@@ -396,6 +401,7 @@ ENTRY(restore_user_context_sysenter)
 	mov	SPREG(%ebp), %ecx	/* sysexit restores ESP using ECX */
 	mov	AXREG(%ebp), %eax	/* trap return value */
 	mov	BXREG(%ebp), %ebx	/* secondary return value */
+	movl	PSWREG(%ebp), %edi	/* load desired PSW to EDI */
 	sti				/* enable interrupts */
 	sysexit				/* jump to EIP in user */
 
@@ -410,6 +416,7 @@ ENTRY(restore_user_context_syscall)
 	mov	SPREG(%ebp), %esp	/* restore ESP directly */
 	mov	AXREG(%ebp), %eax	/* trap return value */
 	mov	BXREG(%ebp), %ebx	/* secondary return value */
+	movl	PSWREG(%ebp), %edi	/* load desired PSW to EDI */
 	sysret				/* jump to EIP in user */
 
 ENTRY(restore_user_context_int)
diff --git a/kernel/arch/i386/usermapped_glo_ipc.S b/kernel/arch/i386/usermapped_glo_ipc.S
index 72d9bb8a9..cb3b55b34 100644
--- a/kernel/arch/i386/usermapped_glo_ipc.S
+++ b/kernel/arch/i386/usermapped_glo_ipc.S
@@ -35,6 +35,8 @@ ENTRY(usermapped_ ## name ## _sysenter)					;\
 	SETARGS			/* call-specific register setup */	;\
 	sysenter		/* disappear into kernel */		;\
 0:									;\
+	push	%edi		/* kernel has desired PSW in %edi */	;\
+	popf			/* set PSW kernel wants us to have */	;\
 	mov	%ebx, %ecx	/* return w. state mangled; save %ebx */;\
 	pop	%edi							;\
 	pop	%esi							;\
@@ -59,6 +61,8 @@ ENTRY(usermapped_ ## name ## _syscall)					;\
 	SETARGS			/* call-specific register setup */	;\
 	movl	%ecx, %edx	/* %ecx is clobbered by SYSCALL */	;\
 	syscall			/* disappear into kernel */		;\
+	push	%edi		/* kernel has desired PSW in %edi */	;\
+	popf			/* set PSW kernel wants us to have */	;\
 	mov	%ebx, %ecx	/* return w. state mangled; save %ebx */;\
 	pop	%edi							;\
 	pop	%esi							;\
diff --git a/kernel/proc.h b/kernel/proc.h
index f645d3ea5..439396937 100644
--- a/kernel/proc.h
+++ b/kernel/proc.h
@@ -252,6 +252,7 @@ struct proc {
 				    message from this sender but could not
 				    because of VM modifying the sender's address
 				    space*/
+#define MF_STEP		 0x40000 /* Single-step process */
 
 /* Magic process table addresses. */
 #define BEG_PROC_ADDR (&proc[0])
diff --git a/kernel/proto.h b/kernel/proto.h
index c0065dd89..fc0207b39 100644
--- a/kernel/proto.h
+++ b/kernel/proto.h
@@ -59,7 +59,8 @@ void enqueue(struct proc *rp);
 void dequeue(struct proc *rp);
 void switch_to_user(void);
 void arch_proc_reset(struct proc *rp);
-void arch_proc_setcontext(struct proc *rp, struct stackframe_s *state, int user);
+void arch_proc_setcontext(struct proc *rp, struct stackframe_s *state,
+	int user, int restorestyle);
 struct proc * arch_finish_switch_to_user(void);
 struct proc *endpoint_lookup(endpoint_t ep);
 #if DEBUG_ENABLE_IPC_WARNINGS
diff --git a/kernel/system/do_fork.c b/kernel/system/do_fork.c
index dfad1db37..f30927025 100644
--- a/kernel/system/do_fork.c
+++ b/kernel/system/do_fork.c
@@ -71,11 +71,8 @@ int do_fork(struct proc * caller, message * m_ptr)
   rpc->p_user_time = 0;		/* set all the accounting times to 0 */
   rpc->p_sys_time = 0;
 
-#if defined(__i386__)
-  rpc->p_reg.psw &= ~TRACEBIT;		/* clear trace bit */
-#endif
   rpc->p_misc_flags &=
-	~(MF_VIRT_TIMER | MF_PROF_TIMER | MF_SC_TRACE | MF_SPROF_SEEN);
+	~(MF_VIRT_TIMER | MF_PROF_TIMER | MF_SC_TRACE | MF_SPROF_SEEN | MF_STEP);
   rpc->p_virt_left = 0;		/* disable, clear the process-virtual timers */
   rpc->p_prof_left = 0;
 
diff --git a/kernel/system/do_sigreturn.c b/kernel/system/do_sigreturn.c
index b3603179c..606314b18 100644
--- a/kernel/system/do_sigreturn.c
+++ b/kernel/system/do_sigreturn.c
@@ -51,7 +51,7 @@ int do_sigreturn(struct proc * caller, message * m_ptr)
 #endif
 
   /* Restore the registers. */
-  arch_proc_setcontext(rp, &sc.sc_regs, 1);
+  arch_proc_setcontext(rp, &sc.sc_regs, 1, sc.trap_style);
 #if defined(__i386__)
   if(sc.sc_flags & MF_FPU_INITIALIZED)
   {
diff --git a/kernel/system/do_sigsend.c b/kernel/system/do_sigsend.c
index a8107a4df..bbf21149d 100644
--- a/kernel/system/do_sigsend.c
+++ b/kernel/system/do_sigsend.c
@@ -43,13 +43,21 @@ int do_sigsend(struct proc * caller, message * m_ptr)
 
   /* Copy the registers to the sigcontext structure. */
   memcpy(&sc.sc_regs, (char *) &rp->p_reg, sizeof(sigregs));
-  #if defined(__i386__)
+
+#if defined(__i386__)
+  sc.trap_style = rp->p_seg.p_kern_trap_style;
+
+  if(sc.trap_style == KTS_NONE) {
+  	printf("do_sigsend: sigsend an unsaved process\n");
+	return EINVAL;
+  }
+
     if(proc_used_fpu(rp)) {
 	    /* save the FPU context before saving it to the sig context */
 	    save_fpu(rp);
 	    memcpy(&sc.sc_fpu_state, rp->p_seg.fpu_state, FPU_XFP_SIZE);
     }
-  #endif
+#endif
 
   /* Finish the sigcontext initialization. */
   sc.sc_mask = smsg.sm_mask;
diff --git a/kernel/system/do_trace.c b/kernel/system/do_trace.c
index 83f30c5b6..24bfcc588 100644
--- a/kernel/system/do_trace.c
+++ b/kernel/system/do_trace.c
@@ -87,10 +87,8 @@ int do_trace(struct proc * caller, message * m_ptr)
   switch (tr_request) {
   case T_STOP:			/* stop process */
 	RTS_SET(rp, RTS_P_STOP);
-#if defined(__i386__)
-	rp->p_reg.psw &= ~TRACEBIT;	/* clear trace bit */
-#endif
-	rp->p_misc_flags &= ~MF_SC_TRACE;	/* clear syscall trace flag */
+	/* clear syscall trace and single step flags */
+	rp->p_misc_flags &= ~(MF_SC_TRACE | MF_STEP);
 	return(OK);
 
   case T_GETINS:		/* return value from instruction space */
@@ -170,9 +168,7 @@ int do_trace(struct proc * caller, message * m_ptr)
 	break;
 
   case T_STEP:			/* set trace bit */
-#if defined(__i386__)
-	rp->p_reg.psw |= TRACEBIT;
-#endif
+	rp->p_misc_flags |= MF_STEP;
 	RTS_UNSET(rp, RTS_P_STOP);
 	m_ptr->CTL_DATA = 0;
 	break;