[PATCH] ppc64: Optimize exception/syscall entry/exit

This rewrites the PPC64 exception entry/exit routines to make them
smaller and faster.

In particular we no longer save all of the registers for the common
exceptions - system calls, hardware interrupts and decrementer (timer)
interrupts - only the volatile registers.  The other registers are saved
and restored (if used) by the C functions we call.  This involved
changing the registers we use in early exception processing from r20-r23
to r9-r12, which ended up changing quite a lot of code in head.S. 
Overall this gives us about a 20% reduction in null syscall time. 

Some system calls need all the registers (e.g.  fork/clone/vfork and
[rt_]sigsuspend).  For these the syscall dispatch code calls a stub that
saves the nonvolatile registers before calling the real handler.

This also implements the force_successful_syscall_return() thing for
ppc64.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

arch/ppc64/kernel/asm-offsets.c

@@ -49,6 +49,7 @@ int main(void)
 	DEFINE(THREAD_SIZE, THREAD_SIZE);
 	DEFINE(TI_FLAGS, offsetof(struct thread_info, flags));
 	DEFINE(TI_PREEMPT, offsetof(struct thread_info, preempt_count));
+	DEFINE(TI_SC_NOERR, offsetof(struct thread_info, syscall_noerror));
 
 	/* task_struct->thread */
 	DEFINE(THREAD, offsetof(struct task_struct, thread));
@@ -100,7 +101,10 @@ int main(void)
 	DEFINE(PACALPPACA, offsetof(struct paca_struct, xLpPaca));
         DEFINE(LPPACA, offsetof(struct paca_struct, xLpPaca));
         DEFINE(PACAREGSAV, offsetof(struct paca_struct, xRegSav));
-        DEFINE(PACAEXC, offsetof(struct paca_struct, exception_stack));
+        DEFINE(PACA_EXGEN, offsetof(struct paca_struct, exgen));
+        DEFINE(PACA_EXMC, offsetof(struct paca_struct, exmc));
+        DEFINE(PACA_EXSLB, offsetof(struct paca_struct, exslb));
+        DEFINE(PACA_EXDSI, offsetof(struct paca_struct, exdsi));
         DEFINE(PACAGUARD, offsetof(struct paca_struct, guard));
         DEFINE(LPPACASRR0, offsetof(struct ItLpPaca, xSavedSrr0));
         DEFINE(LPPACASRR1, offsetof(struct ItLpPaca, xSavedSrr1));
@@ -137,6 +141,10 @@ int main(void)
 	DEFINE(GPR7, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[7]));
 	DEFINE(GPR8, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[8]));
 	DEFINE(GPR9, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[9]));
+	DEFINE(GPR10, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[10]));
+	DEFINE(GPR11, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[11]));
+	DEFINE(GPR12, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[12]));
+	DEFINE(GPR13, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[13]));
 	DEFINE(GPR20, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[20]));
 	DEFINE(GPR21, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[21]));
 	DEFINE(GPR22, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[22]));
@@ -155,7 +163,7 @@ int main(void)
 	DEFINE(_DSISR, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, dsisr));
 	DEFINE(ORIG_GPR3, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, orig_gpr3));
 	DEFINE(RESULT, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, result));
-	DEFINE(TRAP, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, trap));
+	DEFINE(_TRAP, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, trap));
 	DEFINE(SOFTE, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, softe));
 
 	/* These _only_ to be used with {PROM,RTAS}_FRAME_SIZE!!! */

arch/ppc64/kernel/misc.S

@@ -85,16 +85,17 @@ _GLOBAL(local_irq_restore)
 	cmpw	0,r3,r5
 	beqlr
 	/* are we enabling interrupts? */
-	cmpi	0,r3,0
+	cmpdi	0,r3,0
 	stb	r3,PACAPROCENABLED(r13)
 	beqlr
 	/* Check pending interrupts */
 	/*   A decrementer, IPI or PMC interrupt may have occurred
 	 *   while we were in the hypervisor (which enables) */
-	CHECKANYINT(r4,r5)
+	ld	r4,PACALPPACA+LPPACAANYINT(r13)
+	cmpdi	r4,0
 	beqlr
 
-	/* 
+	/*
 	 * Handle pending interrupts in interrupt context
 	 */
 	li	r0,0x5555
@@ -608,7 +609,7 @@ _GLOBAL(kernel_thread)
 _GLOBAL(sys_call_table32)
 	.llong .sys_restart_syscall	/* 0 */
 	.llong .sys_exit
-	.llong .sys_fork
+	.llong .ppc_fork
 	.llong .sys_read
 	.llong .sys_write
 	.llong .sys32_open		/* 5 */
@@ -678,7 +679,7 @@ _GLOBAL(sys_call_table32)
 	.llong .sys32_ssetmask
 	.llong .sys_setreuid	        /* 70 */
 	.llong .sys_setregid
-	.llong .sys32_sigsuspend
+	.llong .ppc32_sigsuspend
 	.llong .compat_sys_sigpending
 	.llong .sys32_sethostname
 	.llong .compat_sys_setrlimit	        /* 75 */
@@ -726,7 +727,7 @@ _GLOBAL(sys_call_table32)
 	.llong .sys32_ipc
 	.llong .sys_fsync
 	.llong .ppc32_sigreturn
-	.llong .sys_clone		/* 120 */
+	.llong .ppc_clone		/* 120 */
 	.llong .sys32_setdomainname
 	.llong .ppc64_newuname
 	.llong .sys_ni_syscall		/* old modify_ldt syscall */
@@ -784,7 +785,7 @@ _GLOBAL(sys_call_table32)
 	.llong .sys32_rt_sigpending     /* 175 */
 	.llong .sys32_rt_sigtimedwait
 	.llong .sys32_rt_sigqueueinfo
-	.llong .sys32_rt_sigsuspend
+	.llong .ppc32_rt_sigsuspend
 	.llong .sys32_pread64
 	.llong .sys32_pwrite64	        /* 180 */
 	.llong .sys_chown
@@ -795,7 +796,7 @@ _GLOBAL(sys_call_table32)
 	.llong .sys32_sendfile
 	.llong .sys_ni_syscall		/* reserved for streams1 */
 	.llong .sys_ni_syscall		/* reserved for streams2 */
-	.llong .sys_vfork
+	.llong .ppc_vfork
 	.llong .compat_sys_getrlimit	        /* 190 */
 	.llong .sys32_readahead
 	.llong .sys32_mmap2
@@ -880,7 +881,7 @@ _GLOBAL(sys_call_table32)
 _GLOBAL(sys_call_table)
 	.llong .sys_restart_syscall	/* 0 */
 	.llong .sys_exit
-	.llong .sys_fork
+	.llong .ppc_fork
 	.llong .sys_read
 	.llong .sys_write
 	.llong .sys_open		/* 5 */
@@ -998,7 +999,7 @@ _GLOBAL(sys_call_table)
 	.llong .sys_ipc
 	.llong .sys_fsync
 	.llong .sys_ni_syscall
-	.llong .sys_clone		/* 120 */
+	.llong .ppc_clone		/* 120 */
 	.llong .sys_setdomainname
 	.llong .ppc64_newuname
 	.llong .sys_ni_syscall		/* old modify_ldt syscall */
@@ -1056,7 +1057,7 @@ _GLOBAL(sys_call_table)
 	.llong .sys_rt_sigpending	/* 175 */
 	.llong .sys_rt_sigtimedwait
 	.llong .sys_rt_sigqueueinfo
-	.llong .sys_rt_sigsuspend
+	.llong .ppc64_rt_sigsuspend
 	.llong .sys_pread64
 	.llong .sys_pwrite64	        /* 180 */
 	.llong .sys_chown
@@ -1067,7 +1068,7 @@ _GLOBAL(sys_call_table)
 	.llong .sys_sendfile64
 	.llong .sys_ni_syscall		/* reserved for streams1 */
 	.llong .sys_ni_syscall		/* reserved for streams2 */
-	.llong .sys_vfork
+	.llong .ppc_vfork
 	.llong .sys_getrlimit	        /* 190 */
 	.llong .sys_readahead
 	.llong .sys_ni_syscall		/* 32bit only mmap2 */

arch/ppc64/kernel/pacaData.c

@@ -62,8 +62,6 @@ struct systemcfg *systemcfg;
 		.xDesc = 0xd397d9e2,	/* "LpRS" */			    \
 		.xSize = sizeof(struct ItLpRegSave)			    \
 	},								    \
-	.exception_sp =							    \
-		(&paca[number].exception_stack[0]) - EXC_FRAME_SIZE,	    \
 }
 
 struct paca_struct paca[] __page_aligned = {

arch/ppc64/kernel/process.c

@@ -219,6 +219,7 @@ struct task_struct *__switch_to(struct task_struct *prev,
 void show_regs(struct pt_regs * regs)
 {
 	int i;
+	unsigned long trap;
 
 	printk("NIP: %016lX XER: %016lX LR: %016lX\n",
 	       regs->nip, regs->xer, regs->link);
@@ -229,7 +230,8 @@ void show_regs(struct pt_regs * regs)
 	       regs->msr & MSR_FP ? 1 : 0,regs->msr&MSR_ME ? 1 : 0,
 	       regs->msr&MSR_IR ? 1 : 0,
 	       regs->msr&MSR_DR ? 1 : 0);
-	if (regs->trap == 0x300 || regs->trap == 0x380 || regs->trap == 0x600)
+	trap = TRAP(regs);
+	if (trap == 0x300 || trap == 0x380 || trap == 0x600)
 		printk("DAR: %016lx, DSISR: %016lx\n", regs->dar, regs->dsisr);
 	printk("TASK: %p[%d] '%s' THREAD: %p",
 	       current, current->pid, current->comm, current->thread_info);
@@ -244,6 +246,8 @@ void show_regs(struct pt_regs * regs)
 		}
 
 		printk("%016lX ", regs->gpr[i]);
+		if (i == 13 && !FULL_REGS(regs))
+			break;
 	}
 	printk("\n");
 	/*

arch/ppc64/kernel/signal.c

@@ -528,13 +528,13 @@ int do_signal(sigset_t *oldset, struct pt_regs *regs)
 		struct k_sigaction *ka = &current->sighand->action[signr-1];
 
 		/* Whee!  Actually deliver the signal.  */
-		if (regs->trap == 0x0C00)
+		if (TRAP(regs) == 0x0C00)
 			syscall_restart(regs, ka);
 		handle_signal(signr, ka, &info, oldset, regs);
 		return 1;
 	}
 
-	if (regs->trap == 0x0C00) {	/* System Call! */
+	if (TRAP(regs) == 0x0C00) {	/* System Call! */
 		if ((int)regs->result == -ERESTARTNOHAND ||
 		    (int)regs->result == -ERESTARTSYS ||
 		    (int)regs->result == -ERESTARTNOINTR) {

arch/ppc64/kernel/signal32.c

@@ -932,7 +932,7 @@ int do_signal32(sigset_t *oldset, struct pt_regs *regs)
 
 	ka = (signr == 0)? NULL: &current->sighand->action[signr-1];
 
-	if (regs->trap == 0x0C00		/* System Call! */
+	if (TRAP(regs) == 0x0C00		/* System Call! */
 	    && regs->ccr & 0x10000000		/* error signalled */
 	    && ((ret = regs->gpr[3]) == ERESTARTSYS
 		|| ret == ERESTARTNOHAND || ret == ERESTARTNOINTR

arch/ppc64/kernel/syscalls.c

@@ -237,5 +237,19 @@ asmlinkage time_t sys64_time(time_t __user * tloc)
 	return secs;
 }
 
+void do_show_syscall(unsigned long r3, unsigned long r4, unsigned long r5,
+		     unsigned long r6, unsigned long r7, unsigned long r8,
+		     struct pt_regs *regs)
+{
+	printk("syscall %ld(%lx, %lx, %lx, %lx, %lx, %lx) regs=%p current=%p"
+	       " cpu=%d\n", regs->gpr[0], r3, r4, r5, r6, r7, r8, regs,
+	       current, smp_processor_id());
+}
+
+void do_show_syscall_exit(unsigned long r3)
+{
+	printk(" -> %lx, current=%p cpu=%d\n", r3, current, smp_processor_id());
+}
+
 /* Only exists on P-series. */
 cond_syscall(ppc_rtas);

arch/ppc64/kernel/traps.c

@@ -441,8 +441,22 @@ void KernelFPUnavailableException(struct pt_regs *regs)
 	die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
 }
 
-void KernelAltivecUnavailableException(struct pt_regs *regs)
+void AltivecUnavailableException(struct pt_regs *regs)
 {
+#ifndef CONFIG_ALTIVEC
+	if (user_mode(regs)) {
+		/* A user program has executed an altivec instruction,
+		   but this kernel doesn't support altivec. */
+		siginfo_t info;
+
+		memset(&info, 0, sizeof(info));
+		info.si_signo = SIGILL;
+		info.si_code = ILL_ILLOPC;
+		info.si_addr = (void *) regs->nip;
+		_exception(SIGILL, &info, regs);
+		return;
+	}
+#endif
 	printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
 			  "%lx at %lx\n", regs->trap, regs->nip);
 	die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);

arch/ppc64/mm/fault.c

@@ -80,36 +80,45 @@ static int store_updates_sp(struct pt_regs *regs)
  *  - DSISR for a non-SLB data access fault,
  *  - SRR1 & 0x08000000 for a non-SLB instruction access fault
  *  - 0 any SLB fault.
+ * The return value is 0 if the fault was handled, or the signal
+ * number if this is a kernel fault that can't be handled here.
  */
-void do_page_fault(struct pt_regs *regs, unsigned long address,
-		   unsigned long error_code)
+int do_page_fault(struct pt_regs *regs, unsigned long address,
+		  unsigned long error_code)
 {
 	struct vm_area_struct * vma;
 	struct mm_struct *mm = current->mm;
 	siginfo_t info;
 	unsigned long code = SEGV_MAPERR;
 	unsigned long is_write = error_code & 0x02000000;
+	unsigned long trap = TRAP(regs);
 
-	if (regs->trap == 0x300 || regs->trap == 0x380) {
+	if (trap == 0x300 || trap == 0x380) {
 		if (debugger_fault_handler(regs))
-			return;
+			return 0;
 	}
 
 	/* On a kernel SLB miss we can only check for a valid exception entry */
-	if (!user_mode(regs) && (regs->trap == 0x380)) {
-		bad_page_fault(regs, address, SIGSEGV);
-		return;
-	}
+	if (!user_mode(regs) && (trap == 0x380 || address >= TASK_SIZE))
+		return SIGSEGV;
 
 	if (error_code & 0x00400000) {
 		if (debugger_dabr_match(regs))
-			return;
+			return 0;
 	}
 
 	if (in_atomic() || mm == NULL) {
-		bad_page_fault(regs, address, SIGSEGV);
-		return;
+		if (!user_mode(regs))
+			return SIGSEGV;
+		/* in_atomic() in user mode is really bad,
+		   as is current->mm == NULL. */
+		printk(KERN_EMERG "Page fault in user mode with"
+		       "in_atomic() = %d mm = %p\n", in_atomic(), mm);
+		printk(KERN_EMERG "NIP = %lx  MSR = %lx\n",
+		       regs->nip, regs->msr);
+		die("Weird page fault", regs, SIGSEGV);
 	}
+
 	down_read(&mm->mmap_sem);
 	vma = find_vma(mm, address);
 	if (!vma)
@@ -195,7 +204,7 @@ void do_page_fault(struct pt_regs *regs, unsigned long address,
 	}
 
 	up_read(&mm->mmap_sem);
-	return;
+	return 0;
 
 bad_area:
 	up_read(&mm->mmap_sem);
@@ -207,11 +216,10 @@ void do_page_fault(struct pt_regs *regs, unsigned long address,
 		info.si_code = code;
 		info.si_addr = (void *) address;
 		force_sig_info(SIGSEGV, &info, current);
-		return;
+		return 0;
 	}
 
-	bad_page_fault(regs, address, SIGSEGV);
-	return;
+	return SIGSEGV;
 
 /*
  * We ran out of memory, or some other thing happened to us that made
@@ -227,18 +235,19 @@ void do_page_fault(struct pt_regs *regs, unsigned long address,
 	printk("VM: killing process %s\n", current->comm);
 	if (user_mode(regs))
 		do_exit(SIGKILL);
-	bad_page_fault(regs, address, SIGKILL);
-	return;
+	return SIGKILL;
 
 do_sigbus:
 	up_read(&mm->mmap_sem);
-	info.si_signo = SIGBUS;
-	info.si_errno = 0;
-	info.si_code = BUS_ADRERR;
-	info.si_addr = (void *)address;
-	force_sig_info (SIGBUS, &info, current);
-	if (!user_mode(regs))
-		bad_page_fault(regs, address, SIGBUS);
+	if (user_mode(regs)) {
+		info.si_signo = SIGBUS;
+		info.si_errno = 0;
+		info.si_code = BUS_ADRERR;
+		info.si_addr = (void *)address;
+		force_sig_info(SIGBUS, &info, current);
+		return 0;
+	}
+	return SIGBUS;
 }
 
 /*

arch/ppc64/xmon/xmon.c

@@ -44,9 +44,6 @@ static int xmon_owner;
 static int xmon_gate;
 #endif /* CONFIG_SMP */
 
-#define TRAP(regs)	((regs)->trap)
-#define FULL_REGS(regs)	1
-
 static unsigned long in_xmon = 0;
 
 static unsigned long adrs;

include/asm-ppc64/paca.h

@@ -136,23 +136,21 @@ struct paca_struct {
 	u8  rsvd6[0x500 - 0x8];
 
 /*=====================================================================================
- * CACHE_LINE_31 0x0F00 - 0x0F7F Exception stack
+ * CACHE_LINE_31-32 0x0F00 - 0x0FFF Exception register save areas
  *=====================================================================================
  */
-	u8 exception_stack[N_EXC_STACK*EXC_FRAME_SIZE];
+	u64 exgen[8];		/* used for most interrupts/exceptions */
+	u64 exmc[8];		/* used for machine checks */
+	u64 exslb[8];		/* used for SLB/segment table misses
+				 * on the linear mapping */
+	u64 exdsi[8];		/* used for linear mapping hash table misses */
 
 /*=====================================================================================
- * CACHE_LINE_32 0x0F80 - 0x0FFF Reserved
+ * Page 2 used as a stack when we detect a bad kernel stack pointer,
+ * and early in SMP boots before relocation is enabled.
  *=====================================================================================
  */
-	u8 rsvd7[0x80];                  /* Give the stack some rope ... */
-
-/*=====================================================================================
- * Page 2 Reserved for guard page.  Also used as a stack early in SMP boots before
- *        relocation is enabled.
- *=====================================================================================
- */
-	u8 guard[0x1000];               /* ... and then hang 'em         */ 
+	u8 guard[0x1000];
 };
 
 #endif /* _PPC64_PACA_H */

include/asm-ppc64/ppc_asm.h

@@ -28,6 +28,9 @@
 #define REST_8GPRS(n, base)	REST_4GPRS(n, base); REST_4GPRS(n+4, base)
 #define REST_10GPRS(n, base)	REST_8GPRS(n, base); REST_2GPRS(n+8, base)
 
+#define SAVE_NVGPRS(base)	SAVE_8GPRS(14, base); SAVE_10GPRS(22, base)
+#define REST_NVGPRS(base)	REST_8GPRS(14, base); REST_10GPRS(22, base)
+
 #define SAVE_FPR(n, base)	stfd	n,THREAD_FPR0+8*(n)(base)
 #define SAVE_2FPRS(n, base)	SAVE_FPR(n, base); SAVE_FPR(n+1, base)
 #define SAVE_4FPRS(n, base)	SAVE_2FPRS(n, base); SAVE_2FPRS(n+2, base)
@@ -54,11 +57,6 @@
 #define REST_16VRS(n,b,base)	REST_8VRS(n,b,base); REST_8VRS(n+8,b,base)
 #define REST_32VRS(n,b,base)	REST_16VRS(n,b,base); REST_16VRS(n+16,b,base)
 
-#define CHECKANYINT(ra,rb)			\
-	mfspr	rb,SPRG3;		/* Get Paca address */\
-	ld	ra,PACALPPACA+LPPACAANYINT(rb); /* Get pending interrupt flags */\
-	cmpldi	0,ra,0;
-
 /* Macros to adjust thread priority for Iseries hardware multithreading */
 #define HMT_LOW		or 1,1,1
 #define HMT_MEDIUM	or 2,2,2

include/asm-ppc64/processor.h

@@ -543,8 +543,7 @@ struct thread_struct {
 	double		fpr[32];	/* Complete floating point set */
 	unsigned long	fpscr;		/* Floating point status (plus pad) */
 	unsigned long	fpexc_mode;	/* Floating-point exception mode */
-	unsigned long	saved_msr;	/* Save MSR across signal handlers */
-	unsigned long	saved_softe;	/* Ditto for Soft Enable/Disable */
+	unsigned long	pad[3];		/* was saved_msr, saved_softe */
 #ifdef CONFIG_ALTIVEC
 	/* Complete AltiVec register set */
 	vector128	vr[32] __attribute((aligned(16)));

include/asm-ppc64/ptrace.h

@@ -71,6 +71,18 @@ struct pt_regs32 {
 #define instruction_pointer(regs) ((regs)->nip)
 #define user_mode(regs) ((((regs)->msr) >> MSR_PR_LG) & 0x1)
 
+#define force_successful_syscall_return()   \
+		(current_thread_info()->syscall_noerror = 1)
+
+/*
+ * We use the least-significant bit of the trap field to indicate
+ * whether we have saved the full set of registers, or only a
+ * partial set.  A 1 there means the partial set.
+ */
+#define FULL_REGS(regs)		(((regs)->trap & 1) == 0)
+#define TRAP(regs)		((regs)->trap & ~0xF)
+#define CHECK_FULL_REGS(regs)	BUG_ON(regs->trap & 1)
+
 /*
  * Offsets used by 'ptrace' system call interface.
  */

include/asm-ppc64/thread_info.h

@@ -26,6 +26,8 @@ struct thread_info {
 	int		cpu;			/* cpu we're on */
 	int		preempt_count;
 	struct restart_block restart_block;
+	/* set by force_successful_syscall_return */
+	unsigned char	syscall_noerror;
 };
 
 /*
@@ -84,8 +86,6 @@ static inline struct thread_info *current_thread_info(void)
 
 /*
  * thread information flag bit numbers
- * N.B. If TIF_SIGPENDING or TIF_NEED_RESCHED are changed
- * to be >= 4, code in entry.S will need to be changed.
  */
 #define TIF_SYSCALL_TRACE	0	/* syscall trace active */
 #define TIF_NOTIFY_RESUME	1	/* resumption notification requested */