[PATCH] i386 tlb flush optimization

[Manfred Spraul]

the i386 tlb flush is still paranoid about lock-ups, but this is
superflous:
e.g. the Alpha port uses a normal smp_call_function() in it's tlb flush
code.

My attached patch removes most of these paranoid checks.

Could you please test it?
I'm preparing a follow-up patch that optimizes flush_tlb_page() for
multi-threaded applications, and that patch will be incompatible with
the current, anti-lock-up flush_tlb_others() code.

The patch is tested on 2.3.40 i386 SMP.

--
	Manfred
------------ próxima parte ------------
--- 2.3/arch/i386/kernel/smp.c	Fri Jan 21 12:59:23 2000
+++ build-2.3/arch/i386/kernel/smp.c	Wed Jan 26 18:52:24 2000
@@ -281,56 +281,52 @@
 #endif
 }
 
-/*
- * This is fraught with deadlocks. Probably the situation is not that
- * bad as in the early days of SMP, so we might ease some of the
- * paranoia here.
- */
+#define TLB_PARANOIA 1
+
 static void flush_tlb_others(unsigned int cpumask)
 {
-	int cpu = smp_processor_id();
-	int stuck;
-	unsigned long flags;
+#ifdef TLB_PARANOIA
+	if(in_interrupt()) {
+		printk("tlb flush from interrupt: %d,%d",
+			local_bh_count[smp_processor_id()],
+			local_irq_count[smp_processor_id()]);
+	}
+	if(cpumask & (1<<smp_processor_id())) {
+		printk("flush_tlb_others: bad cpumask!");
+		cpumask &= ~(1<<smp_processor_id());	
+		local_flush_tlb();
+	}
+	{
+		int flags;
+
+		save_flags(flags);
+		if(flags != 1) {
+static int limit=10;
+			if(limit > 0) {
+				limit--;
+				printk("flush_tlb_others: possible lock-up, broken!(%d)",
+						flags);
+			}
+			sti();
+		}
+	}
+#endif
+	cpumask &= cpu_online_map;
 
 	/*
 	 * it's important that we do not generate any APIC traffic
 	 * until the AP CPUs have booted up!
 	 */
-	cpumask &= cpu_online_map;
+
 	if (cpumask) {
 		atomic_set_mask(cpumask, &smp_invalidate_needed);
-
-		/*
-		 * Processors spinning on some lock with IRQs disabled
-		 * will see this IRQ late. The smp_invalidate_needed
-		 * map will ensure they don't do a spurious flush tlb
-		 * or miss one.
-		 */
-	
-		__save_flags(flags);
-		__cli();
-
 		send_IPI_allbutself(INVALIDATE_TLB_VECTOR);
 
-		/*
-		 * Spin waiting for completion
-		 */
-
-		stuck = 50000000;
 		while (smp_invalidate_needed) {
-			/*
-			 * Take care of "crossing" invalidates
+			/* FIXME: lockup-detection, print backtrace on
+			 * lock-up
 			 */
-			if (test_bit(cpu, &smp_invalidate_needed))
-				do_flush_tlb_local();
-
-			--stuck;
-			if (!stuck) {
-				printk("stuck on TLB IPI wait (CPU#%d)\n",cpu);
-				break;
-			}
 		}
-		__restore_flags(flags);
 	}
 }
 
--- 2.3/arch/i386/kernel/irq.c	Fri Jan 21 12:59:23 2000
+++ build-2.3/arch/i386/kernel/irq.c	Wed Jan 26 17:46:35 2000
@@ -192,20 +192,6 @@
 atomic_t global_bh_count;
 atomic_t global_bh_lock;
 
-/*
- * "global_cli()" is a special case, in that it can hold the
- * interrupts disabled for a longish time, and also because
- * we may be doing TLB invalidates when holding the global
- * IRQ lock for historical reasons. Thus we may need to check
- * SMP invalidate events specially by hand here (but not in
- * any normal spinlocks)
- */
-static inline void check_smp_invalidate(int cpu)
-{
-	if (test_bit(cpu, &smp_invalidate_needed))
-		do_flush_tlb_local();
-}
-
 static void show(char * str)
 {
 	int i;
@@ -294,7 +280,6 @@
 			__sti();
 			SYNC_OTHER_CORES(cpu);
 			__cli();
-			check_smp_invalidate(cpu);
 			if (atomic_read(&global_irq_count))
 				continue;
 			if (global_irq_lock)
@@ -346,7 +331,6 @@
 		/* Uhhuh.. Somebody else got it. Wait.. */
 		do {
 			do {
-				check_smp_invalidate(cpu);
 			} while (test_bit(0,&global_irq_lock));
 		} while (test_and_set_bit(0,&global_irq_lock));		
 	}