Fix sysenter and syscall features.

Those features were not configured properly and had a race-condition
when enabling back interrupts before a sysexit/sysret.

Configuration: The BSP is executing tss_init on behalf of APs. Thus any
writes in the MSRs in tss_init would end up in the BSP's MSRs leading
to:
	* BSP not using the correct kernel stack.
	* APs using undefined instruction/stack pointer and code segment
upon entering the kernel.
Fix: Let the cores writing in their own MSRs.

Race-condition: When restoring a process' EFLAGS before a sysexit/sysret
interrupts could be enabled while the core was still in the kernel. If
an interrupt is triggered at this point it would lead to a deadlock on
the BKL down the line.
Fix: Disable IF in saved EFLAGS and use sti before sysexit/sysret.

minix/kernel/arch/i386/arch_smp.c

@@ -218,6 +218,9 @@ static void ap_finish_booting(void)
 	/* inform the world of our presence. */
 	ap_cpu_ready = cpu;
 
+	/* Set up sysenter/syscall. */
+	setup_sysenter_syscall();
+
 	/*
 	 * Finish processor initialisation.  CPUs must be excluded from running.
 	 * lapic timer calibration locks and unlocks the BKL because of the
@@ -298,6 +301,7 @@ void smp_init (void)
 	ioapic_enabled = 0;
 
 	tss_init_all();
+	setup_sysenter_syscall();
 
 	/* 
 	 * we still run on the boot stack and we cannot use cpuid as its value

minix/kernel/arch/i386/include/arch_proto.h

@@ -249,6 +249,11 @@ reg_t read_ebp(void);
  */
 int tss_init(unsigned cpu, void * kernel_stack);
 
+/*
+ * Set up MSRs for sysenter or syscall for the current cpu.
+ * Assumes that tss_init has been called before for that cpu ! */
+void setup_sysenter_syscall(void);
+
 void int_gate_idt(unsigned vec_nr, vir_bytes offset, unsigned dpl_type);
 
 void __copy_msg_from_user_end(void);

minix/kernel/arch/i386/include/arch_smp.h

@@ -16,6 +16,7 @@
  */
 #define smp_single_cpu_fallback() do {		\
 	  tss_init(0, get_k_stack_top(0));	\
+	  setup_sysenter_syscall();		\
 	  bsp_cpu_id = 0;			\
 	  ncpus = 1;				\
 	  bsp_finish_booting();			\

minix/kernel/arch/i386/mpx.S

@@ -405,6 +405,7 @@ ENTRY(restore_user_context_sysenter)
 
         /* restore PSW */
         movl    PSWREG(%ebp), %edi      /* load desired PSW to EDI */
+        andl    $0xfffffdff, %edi	/* Clear IF to avoid receiving int. */
         push    %edi
         popf
 
@@ -421,6 +422,7 @@ ENTRY(restore_user_context_syscall)
 
         /* restore PSW (before we switch to user stack!) */
         movl    PSWREG(%ebp), %edi      /* load desired PSW to EDI */
+        andl    $0xfffffdff, %edi	/* Clear IF to avoid receiving int. */
         push    %edi
         popf
 
@@ -429,6 +431,7 @@ ENTRY(restore_user_context_syscall)
 	mov	AXREG(%ebp), %eax	/* trap return value */
 	mov	BXREG(%ebp), %ebx	/* secondary return value */
 
+	sti
 	sysret				/* jump to EIP in user */
 
 ENTRY(restore_user_context_int)

minix/kernel/arch/i386/protect.c

@@ -151,35 +151,10 @@ static struct gate_table_s gate_table_exceptions[] = {
 	{ NULL, 0, 0}
 };
 
-int tss_init(unsigned cpu, void * kernel_stack)
+void setup_sysenter_syscall(void)
 {
-	struct tss_s * t = &tss[cpu];
+	const int cpu = cpuid;
-	int index = TSS_INDEX(cpu);
+	struct tss_s const *const t = &tss[cpu];
-	struct segdesc_s *tssgdt;
-
-	tssgdt = &gdt[index];
-  
-	init_param_dataseg(tssgdt, (phys_bytes) t,
-			sizeof(struct tss_s), INTR_PRIVILEGE);
-	tssgdt->access = PRESENT | (INTR_PRIVILEGE << DPL_SHIFT) | TSS_TYPE;
-
-	/* Build TSS. */
-	memset(t, 0, sizeof(*t));
-	t->ds = t->es = t->fs = t->gs = t->ss0 = KERN_DS_SELECTOR;
-	t->cs = KERN_CS_SELECTOR;
-	t->iobase = sizeof(struct tss_s);	/* empty i/o permissions map */
-
-	/* 
-	 * make space for process pointer and cpu id and point to the first
-	 * usable word
-	 */
-	k_percpu_stacks[cpu] = t->sp0 = ((unsigned) kernel_stack) - X86_STACK_TOP_RESERVED;
-	/* 
-	 * set the cpu id at the top of the stack so we know on which cpu is
-	 * this stack in use when we trap to kernel
-	 */
-	*((reg_t *)(t->sp0 + 1 * sizeof(reg_t))) = cpu;
-
 	/* Set up Intel SYSENTER support if available. */
 	if(minix_feature_flags & MKF_I386_INTEL_SYSENTER) {
 	  ia32_msr_write(INTEL_MSR_SYSENTER_CS, 0, KERN_CS_SELECTOR);
@@ -211,7 +186,36 @@ int tss_init(unsigned cpu, void * kernel_stack)
 		set_star_cpu(7);
 		assert(CONFIG_MAX_CPUS <= 8);
   	}
+}
 
+int tss_init(unsigned cpu, void * kernel_stack)
+{
+	struct tss_s * t = &tss[cpu];
+	int index = TSS_INDEX(cpu);
+	struct segdesc_s *tssgdt;
+
+	tssgdt = &gdt[index];
+  
+	init_param_dataseg(tssgdt, (phys_bytes) t,
+			sizeof(struct tss_s), INTR_PRIVILEGE);
+	tssgdt->access = PRESENT | (INTR_PRIVILEGE << DPL_SHIFT) | TSS_TYPE;
+
+	/* Build TSS. */
+	memset(t, 0, sizeof(*t));
+	t->ds = t->es = t->fs = t->gs = t->ss0 = KERN_DS_SELECTOR;
+	t->cs = KERN_CS_SELECTOR;
+	t->iobase = sizeof(struct tss_s);	/* empty i/o permissions map */
+
+	/* 
+	 * make space for process pointer and cpu id and point to the first
+	 * usable word
+	 */
+	k_percpu_stacks[cpu] = t->sp0 = ((unsigned) kernel_stack) - X86_STACK_TOP_RESERVED;
+	/* 
+	 * set the cpu id at the top of the stack so we know on which cpu is
+	 * this stack in use when we trap to kernel
+	 */
+	*((reg_t *)(t->sp0 + 1 * sizeof(reg_t))) = cpu;
 	return SEG_SELECTOR(index);
 }
 

minix/kernel/main.c

@@ -42,6 +42,7 @@ void bsp_finish_booting(void)
   sprofiling = 0;      /* we're not profiling until instructed to */
 #endif /* SPROFILE */
 
+  setup_sysenter_syscall();
   cpu_identify();
 
   vm_running = 0;