[PATCH] ia64: Preemption patch against ~2.5.60

Latest premption patch.

arch/ia64/Kconfig

@@ -424,6 +424,18 @@ config SMP
 
 	  If you don't know what to do here, say N.
 
+config PREEMPT
+	bool "Preemptible Kernel"
+        help
+          This option reduces the latency of the kernel when reacting to
+          real-time or interactive events by allowing a low priority process to
+          be preempted even if it is in kernel mode executing a system call.
+          This allows applications to run more reliably even when the system is
+          under load.
+
+          Say Y here if you are building a kernel for a desktop, embedded
+          or real-time system.  Say N if you are unsure.
+
 config IA32_SUPPORT
 	bool "Support running of Linux/x86 binaries"
 	help
@@ -875,6 +887,12 @@ config DEBUG_SPINLOCK
 	  best used in conjunction with the NMI watchdog so that spinlock
 	  deadlocks are also debuggable.
 
+config DEBUG_SPINLOCK_SLEEP
+	  bool "Sleep-inside-spinlock checking"
+	  help
+	    If you say Y here, various routines which may sleep will become very
+	    noisy if they are called with a spinlock held.        
+
 config IA64_DEBUG_CMPXCHG
 	bool "Turn on compare-and-exchange bug checking (slow!)"
 	depends on DEBUG_KERNEL

arch/ia64/hp/sim/simserial.c

@@ -63,7 +63,6 @@ extern void ia64_ssc_connect_irq (long intr, long irq);
 
 static char *serial_name = "SimSerial driver";
 static char *serial_version = "0.6";
-static spinlock_t serial_lock = SPIN_LOCK_UNLOCKED;
 
 /*
  * This has been extracted from asm/serial.h. We need one eventually but
@@ -235,14 +234,14 @@ static void rs_put_char(struct tty_struct *tty, unsigned char ch)
 
 	if (!tty || !info->xmit.buf) return;
 
-	spin_lock_irqsave(&serial_lock, flags);
+	local_irq_save(flags);
 	if (CIRC_SPACE(info->xmit.head, info->xmit.tail, SERIAL_XMIT_SIZE) == 0) {
-		spin_unlock_irqrestore(&serial_lock, flags);
+		local_irq_restore(flags);
 		return;
 	}
 	info->xmit.buf[info->xmit.head] = ch;
 	info->xmit.head = (info->xmit.head + 1) & (SERIAL_XMIT_SIZE-1);
-	spin_unlock_irqrestore(&serial_lock, flags);
+	local_irq_restore(flags);
 }
 
 static _INLINE_ void transmit_chars(struct async_struct *info, int *intr_done)
@@ -250,7 +249,8 @@ static _INLINE_ void transmit_chars(struct async_struct *info, int *intr_done)
 	int count;
 	unsigned long flags;
 
-	spin_lock_irqsave(&serial_lock, flags);
+
+	local_irq_save(flags);
 
 	if (info->x_char) {
 		char c = info->x_char;
@@ -293,7 +293,7 @@ static _INLINE_ void transmit_chars(struct async_struct *info, int *intr_done)
 		info->xmit.tail += count;
 	}
 out:
-	spin_unlock_irqrestore(&serial_lock, flags);
+	local_irq_restore(flags);
 }
 
 static void rs_flush_chars(struct tty_struct *tty)
@@ -334,7 +334,7 @@ static int rs_write(struct tty_struct * tty, int from_user,
 				break;
 			}
 
-			spin_lock_irqsave(&serial_lock, flags);
+			local_irq_save(flags);
 			{
 				c1 = CIRC_SPACE_TO_END(info->xmit.head, info->xmit.tail,
 						       SERIAL_XMIT_SIZE);
@@ -344,7 +344,7 @@ static int rs_write(struct tty_struct * tty, int from_user,
 				info->xmit.head = ((info->xmit.head + c) &
 						   (SERIAL_XMIT_SIZE-1));
 			}
-			spin_unlock_irqrestore(&serial_lock, flags);
+			local_irq_restore(flags);
 
 			buf += c;
 			count -= c;
@@ -352,7 +352,7 @@ static int rs_write(struct tty_struct * tty, int from_user,
 		}
 		up(&tmp_buf_sem);
 	} else {
-		spin_lock_irqsave(&serial_lock, flags);
+		local_irq_save(flags);
 		while (1) {
 			c = CIRC_SPACE_TO_END(info->xmit.head, info->xmit.tail, SERIAL_XMIT_SIZE);
 			if (count < c)
@@ -367,7 +367,7 @@ static int rs_write(struct tty_struct * tty, int from_user,
 			count -= c;
 			ret += c;
 		}
-		spin_unlock_irqrestore(&serial_lock, flags);
+		local_irq_restore(flags);
 	}
 	/*
 	 * Hey, we transmit directly from here in our case
@@ -398,9 +398,9 @@ static void rs_flush_buffer(struct tty_struct *tty)
 	struct async_struct *info = (struct async_struct *)tty->driver_data;
 	unsigned long flags;
 
-	spin_lock_irqsave(&serial_lock, flags);
+	local_irq_save(flags);
 	info->xmit.head = info->xmit.tail = 0;
-	spin_unlock_irqrestore(&serial_lock, flags);
+	local_irq_restore(flags);
 
 	wake_up_interruptible(&tty->write_wait);
 
@@ -573,7 +573,7 @@ static void shutdown(struct async_struct * info)
 	       state->irq);
 #endif
 
-	spin_lock_irqsave(&serial_lock, flags);
+	local_irq_save(flags);
 	{
 		/*
 		 * First unlink the serial port from the IRQ chain...
@@ -611,7 +611,7 @@ static void shutdown(struct async_struct * info)
 
 		info->flags &= ~ASYNC_INITIALIZED;
 	}
-	spin_unlock_irqrestore(&serial_lock, flags);
+	local_irq_restore(flags);
 }
 
 /*
@@ -634,13 +634,13 @@ static void rs_close(struct tty_struct *tty, struct file * filp)
 
 	state = info->state;
 
-	spin_lock_irqsave(&serial_lock, flags);
+	local_irq_save(flags);
 	if (tty_hung_up_p(filp)) {
 #ifdef SIMSERIAL_DEBUG
 		printk("rs_close: hung_up\n");
 #endif
 		MOD_DEC_USE_COUNT;
-		spin_unlock_irqrestore(&serial_lock, flags);
+		local_irq_restore(flags);
 		return;
 	}
 #ifdef SIMSERIAL_DEBUG
@@ -665,11 +665,11 @@ static void rs_close(struct tty_struct *tty, struct file * filp)
 	}
 	if (state->count) {
 		MOD_DEC_USE_COUNT;
-		spin_unlock_irqrestore(&serial_lock, flags);
+		local_irq_restore(flags);
 		return;
 	}
 	info->flags |= ASYNC_CLOSING;
-	spin_unlock_irqrestore(&serial_lock, flags);
+	local_irq_restore(flags);
 
 	/*
 	 * Now we wait for the transmit buffer to clear; and we notify
@@ -776,7 +776,7 @@ startup(struct async_struct *info)
 	if (!page)
 		return -ENOMEM;
 
-	spin_lock_irqsave(&serial_lock, flags);
+	local_irq_save(flags);
 
 	if (info->flags & ASYNC_INITIALIZED) {
 		free_page(page);
@@ -857,11 +857,11 @@ startup(struct async_struct *info)
 	}
 
 	info->flags |= ASYNC_INITIALIZED;
-	spin_unlock_irqrestore(&serial_lock, flags);
+	local_irq_restore(flags);
 	return 0;
 
 errout:
-	spin_unlock_irqrestore(&serial_lock, flags);
+	local_irq_restore(flags);
 	return retval;
 }
 

arch/ia64/ia32/ia32_support.c

@@ -93,7 +93,7 @@ ia32_load_state (struct task_struct *t)
 {
 	unsigned long eflag, fsr, fcr, fir, fdr, csd, ssd, tssd;
 	struct pt_regs *regs = ia64_task_regs(t);
-	int nr = smp_processor_id();	/* LDT and TSS depend on CPU number: */
+	int nr = get_cpu();	/* LDT and TSS depend on CPU number: */
 
 	eflag = t->thread.eflag;
 	fsr = t->thread.fsr;
@@ -119,6 +119,7 @@ ia32_load_state (struct task_struct *t)
 
 	regs->r17 = (_TSS(nr) << 48) | (_LDT(nr) << 32) | (__u32) regs->r17;
 	regs->r30 = load_desc(_LDT(nr));				/* LDTD */
+	put_cpu();
 }
 
 /*

arch/ia64/kernel/entry.S

@@ -586,10 +586,21 @@ GLOBAL_ENTRY(ia64_leave_kernel)
 	// work.need_resched etc. mustn't get changed by this CPU before it returns to
 	// user- or fsys-mode:
 (pUStk)	cmp.eq.unc p6,p0=r0,r0			// p6 <- pUStk
+#ifdef CONFIG_PREEMPT
+	rsm psr.i				// disable interrupts
+	adds r17=TI_FLAGS+IA64_TASK_SIZE,r13
+(pKStk) adds r20=TI_PRE_COUNT+IA64_TASK_SIZE,r13
+	;;
+(pKStk) ld4  r21=[r20]			// preempt_count ->r21
+	;;
+(pKStk)	cmp4.eq	p6,p0=r21,r0		// p6 <- preempt_count == 0
+	;;
+#else // CONFIG_PREEMPT
 (pUStk)	rsm psr.i
 	;;
 (pUStk)	adds r17=TI_FLAGS+IA64_TASK_SIZE,r13
 	;;
+#endif // CONFIG_PREEMPT
 .work_processed:
 (p6)	ld4 r18=[r17]				// load current_thread_info()->flags
 	adds r2=PT(R8)+16,r12
@@ -810,15 +821,27 @@ skip_rbs_switch:
 .work_pending:
 	tbit.z p6,p0=r18,TIF_NEED_RESCHED		// current_thread_info()->need_resched==0?
 (p6)	br.cond.sptk.few .notify
+#ifdef CONFIG_PREEMPT
+(pKStk)	dep r21=-1,r0,PREEMPT_ACTIVE_BIT,1
+	;;
+(pKStk) st4 [r20]=r21
+	ssm  psr.i		// enable interrupts
+#endif
+	
 #if __GNUC__ < 3
 	br.call.spnt.many rp=invoke_schedule
 #else
 	br.call.spnt.many rp=schedule
 #endif
 .ret9:	cmp.eq p6,p0=r0,r0				// p6 <- 1
-	rsm psr.i
+	rsm psr.i		// disable interrupts
 	;;
 	adds r17=TI_FLAGS+IA64_TASK_SIZE,r13
+#if CONFIG_PREEMPT
+(pKStk)	adds r20=TI_PRE_COUNT+IA64_TASK_SIZE,r13
+	;;
+(pKStk)	st4 [r20]=r0	// preempt_count() <- 0
+#endif
 	br.cond.sptk.many .work_processed		// re-check
 
 .notify:

arch/ia64/kernel/irq.c

@@ -340,12 +340,14 @@ unsigned int do_IRQ(unsigned long irq, struct pt_regs *regs)
 	 * 0 return value means that this irq is already being
 	 * handled by some other CPU. (or is disabled)
 	 */
-	int cpu = smp_processor_id();
+	int cpu;
 	irq_desc_t *desc = irq_desc(irq);
 	struct irqaction * action;
 	unsigned int status;
 
 	irq_enter();
+	cpu = smp_processor_id();
+
 	kstat_cpu(cpu).irqs[irq]++;
 
 	if (desc->status & IRQ_PER_CPU) {

arch/ia64/kernel/palinfo.c

@@ -894,11 +894,13 @@ palinfo_read_entry(char *page, char **start, off_t off, int count, int *eof, voi
 	 * in SMP mode, we may need to call another CPU to get correct
 	 * information. PAL, by definition, is processor specific
 	 */
-	if (f->req_cpu == smp_processor_id())
+	if (f->req_cpu == get_cpu())
 		len = (*palinfo_entries[f->func_id].proc_read)(page);
 	else
 		len = palinfo_handle_smp(f, page);
 
+	put_cpu();
+
 	if (len <= off+count) *eof = 1;
 
 	*start = page + off;

arch/ia64/kernel/smp.c

@@ -90,7 +90,7 @@ stop_this_cpu (void)
 void
 handle_IPI (int irq, void *dev_id, struct pt_regs *regs)
 {
-	int this_cpu = smp_processor_id();
+	int this_cpu = get_cpu();
 	unsigned long *pending_ipis = &__get_cpu_var(ipi_operation);
 	unsigned long ops;
 
@@ -146,8 +146,12 @@ handle_IPI (int irq, void *dev_id, struct pt_regs *regs)
 		} while (ops);
 		mb();	/* Order data access and bit testing. */
 	}
+	put_cpu();
 }
 
+/*
+ * Called with preeemption disabled 
+ */
 static inline void
 send_IPI_single (int dest_cpu, int op)
 {
@@ -155,6 +159,9 @@ send_IPI_single (int dest_cpu, int op)
 	platform_send_ipi(dest_cpu, IA64_IPI_VECTOR, IA64_IPI_DM_INT, 0);
 }
 
+/*
+ * Called with preeemption disabled 
+ */
 static inline void
 send_IPI_allbutself (int op)
 {
@@ -166,6 +173,9 @@ send_IPI_allbutself (int op)
 	}
 }
 
+/*
+ * Called with preeemption disabled 
+ */
 static inline void
 send_IPI_all (int op)
 {
@@ -176,12 +186,18 @@ send_IPI_all (int op)
 			send_IPI_single(i, op);
 }
 
+/*
+ * Called with preeemption disabled 
+ */
 static inline void
 send_IPI_self (int op)
 {
 	send_IPI_single(smp_processor_id(), op);
 }
 
+/*
+ * Called with preeemption disabled 
+ */
 void
 smp_send_reschedule (int cpu)
 {
@@ -197,12 +213,15 @@ void
 smp_send_reschedule_all (void)
 {
 	int i;
+	int cpu = get_cpu(); /* disable preemption */
 
 	for (i = 0; i < NR_CPUS; i++)
-		if (cpu_online(i) && i != smp_processor_id())
+		if (cpu_online(i) && i != cpu)
 			smp_send_reschedule(i);
+	put_cpu();
 }
 
+
 void
 smp_flush_tlb_all (void)
 {
@@ -247,9 +266,11 @@ smp_call_function_single (int cpuid, void (*func) (void *info), void *info, int
 {
 	struct call_data_struct data;
 	int cpus = 1;
+	int me = get_cpu(); /* prevent preemption and reschedule on another processor */
 
-	if (cpuid == smp_processor_id()) {
+	if (cpuid == me) {
 		printk("%s: trying to call self\n", __FUNCTION__);
+		put_cpu();
 		return -EBUSY;
 	}
 
@@ -276,6 +297,7 @@ smp_call_function_single (int cpuid, void (*func) (void *info), void *info, int
 	call_data = NULL;
 
 	spin_unlock_bh(&call_lock);
+	put_cpu();
 	return 0;
 }
 

arch/ia64/mm/fault.c

@@ -55,7 +55,7 @@ ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *re
 	/*
 	 * If we're in an interrupt or have no user context, we must not take the fault..
 	 */
-	if (in_interrupt() || !mm)
+	if (in_atomic() || !mm)
 		goto no_context;
 
 	down_read(&mm->mmap_sem);

arch/ia64/mm/tlb.c

@@ -81,9 +81,13 @@ wrap_mmu_context (struct mm_struct *mm)
 	}
 	read_unlock(&tasklist_lock);
 	/* can't call flush_tlb_all() here because of race condition with O(1) scheduler [EF] */
-	for (i = 0; i < NR_CPUS; ++i)
-		if (i != smp_processor_id())
-			per_cpu(ia64_need_tlb_flush, i) = 1;
+	{
+		int cpu = get_cpu(); /* prevent preemption/migration */
+		for (i = 0; i < NR_CPUS; ++i)
+			if (i != cpu)
+				per_cpu(ia64_need_tlb_flush, i) = 1;
+		put_cpu();
+	}
 	local_flush_tlb_all();
 }
 

include/asm-ia64/hardirq.h

@@ -32,18 +32,18 @@
  *
  * - bits 0-7 are the preemption count (max preemption depth: 256)
  * - bits 8-15 are the softirq count (max # of softirqs: 256)
- * - bits 16-31 are the hardirq count (max # of hardirqs: 65536)
+ * - bits 16-29 are the hardirq count (max # of hardirqs: 16384)
  *
  * - (bit 63 is the PREEMPT_ACTIVE flag---not currently implemented.)
  *
  * PREEMPT_MASK: 0x000000ff
  * SOFTIRQ_MASK: 0x0000ff00
- * HARDIRQ_MASK: 0xffff0000
+ * HARDIRQ_MASK: 0x3fff0000
  */
 
 #define PREEMPT_BITS	8
 #define SOFTIRQ_BITS	8
-#define HARDIRQ_BITS	16
+#define HARDIRQ_BITS	14
 
 #define PREEMPT_SHIFT	0
 #define SOFTIRQ_SHIFT	(PREEMPT_SHIFT + PREEMPT_BITS)
@@ -83,13 +83,13 @@
 #define hardirq_trylock()	(!in_interrupt())
 #define hardirq_endlock()	do { } while (0)
 
-#define in_atomic()		(preempt_count() != 0)
 #define irq_enter()		(preempt_count() += HARDIRQ_OFFSET)
 
 #if CONFIG_PREEMPT
-# error CONFIG_PREEMT currently not supported.
+# define in_atomic()		((preempt_count() & ~PREEMPT_ACTIVE) != kernel_locked())
 # define IRQ_EXIT_OFFSET (HARDIRQ_OFFSET-1)
 #else
+# define in_atomic()		(preempt_count() != 0)
 # define IRQ_EXIT_OFFSET HARDIRQ_OFFSET
 #endif
 

include/asm-ia64/system.h

@@ -206,7 +206,7 @@ extern void ia64_load_extra (struct task_struct *task);
 
 #ifdef CONFIG_PERFMON
   DECLARE_PER_CPU(unsigned long, pfm_syst_info);
-# define PERFMON_IS_SYSWIDE() (get_cpu_var(pfm_syst_info) & 0x1)
+# define PERFMON_IS_SYSWIDE() (__get_cpu_var(pfm_syst_info) & 0x1)
 #else
 # define PERFMON_IS_SYSWIDE() (0)
 #endif

include/asm-ia64/thread_info.h

@@ -15,7 +15,8 @@
 #define TI_ADDR_LIMIT	0x10
 #define TI_PRE_COUNT	0x18
 
-#define PREEMPT_ACTIVE	0x4000000
+#define PREEMPT_ACTIVE_BIT 30
+#define PREEMPT_ACTIVE	(1 << PREEMPT_ACTIVE_BIT)
 
 #ifndef __ASSEMBLY__