2.5.29 NUMA-Q minimal workaround updates

@@ -38,7 +38,8 @@
 
 static void end_8259A_irq (unsigned int irq)
 {
-	if (!(irq_desc[irq].status & (IRQ_DISABLED|IRQ_INPROGRESS)))
+	if (!(irq_desc[irq].status & (IRQ_DISABLED|IRQ_INPROGRESS)) &&
+							irq_desc[irq].action)
 		enable_8259A_irq(irq);
 }
 

@@ -219,7 +219,7 @@
 #define IRQ_ALLOWED(cpu,allowed_mask) \
 		((1 << cpu) & (allowed_mask))
 
-#if CONFIG_SMP
+#if CONFIG_SMP && !CONFIG_MULTIQUAD
 static unsigned long move(int curr_cpu, unsigned long allowed_mask, unsigned long now, int direction)
 {
 	int search_idle = 1;

@@ -187,10 +187,6 @@
 #if CONFIG_SMP
 inline void synchronize_irq(unsigned int irq)
 {
-	/* is there anything to synchronize with? */
-	if (!irq_desc[irq].action)
-		return;
-
 	while (irq_desc[irq].status & IRQ_INPROGRESS)
 		cpu_relax();
 }

@@ -350,7 +346,7 @@
 	 * use the action we have.
 	 */
 	action = NULL;
-	if (!(status & (IRQ_DISABLED | IRQ_INPROGRESS))) {
+	if (likely(!(status & (IRQ_DISABLED | IRQ_INPROGRESS)))) {
 		action = desc->action;
 		status &= ~IRQ_PENDING; /* we commit to handling */
 		status |= IRQ_INPROGRESS; /* we are handling it */

@@ -363,7 +359,7 @@
 	   a different instance of this same irq, the other processor
 	   will take care of it.
 	 */
-	if (!action)
+	if (unlikely(!action))
 		goto out;
 
 	/*

@@ -381,12 +377,12 @@
 		handle_IRQ_event(irq, &regs, action);
 		spin_lock(&desc->lock);
 		
-		if (!(desc->status & IRQ_PENDING))
+		if (likely(!(desc->status & IRQ_PENDING)))
 			break;
 		desc->status &= ~IRQ_PENDING;
 	}
-	desc->status &= ~IRQ_INPROGRESS;
 out:
+	desc->status &= ~IRQ_INPROGRESS;
 	/*
 	 * The ->end() handler has to deal with interrupts which got
 	 * disabled while the handler was running.

@@ -566,7 +566,7 @@
 	struct call_data_struct data;
 	int cpus = num_online_cpus()-1;
 
-	if (!cpus)
+	if (cpus <= 0)
 		return 0;
 
 	data.func = func;

@@ -454,6 +454,8 @@
 	__asm__ __volatile__("lgdt %0": "=m" (cpu_gdt_descr[cpu]));
 	__asm__ __volatile__("lidt %0": "=m" (idt_descr));
 
+	printk(KERN_INFO "Loading GDT/IDT for CPU#%d\n", cpu);
+
 	/*
 	 * Delete NT
 	 */

@@ -474,6 +476,8 @@
 	load_TR_desc();
 	load_LDT(&init_mm.context);
 
+	printk(KERN_INFO "Loaded per-cpu LDT/TSS for CPU#%d\n", cpu);
+
 	/* Clear %fs and %gs. */
 	asm volatile ("xorl %eax, %eax; movl %eax, %fs; movl %eax, %gs");
 

@@ -491,4 +495,6 @@
 	clear_thread_flag(TIF_USEDFPU);
 	current->used_math = 0;
 	stts();
+
+	printk(KERN_INFO "Cleaned up FPU and debug regs for CPU#%d\n", cpu);
 }

@@ -54,7 +54,7 @@
 		
 #if CONFIG_X86_PAE
 	pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
-	set_pgd(pgd, __pgd(__pa(md_table) | _PAGE_PRESENT));
+	set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
 	if (pmd_table != pmd_offset(pgd, 0)) 
 		BUG();
 #else

@@ -856,7 +856,7 @@
 		 * Then release the RX buffers - jumbo buffers were
 		 * already released in ace_close().
 		 */
-		synchronize_irq();
+		synchronize_irq(dev->irq);
 
 		for (i = 0; i < RX_STD_RING_ENTRIES; i++) {
 			struct sk_buff *skb = ap->skb->rx_std_skbuff[i].skb;

@@ -2877,7 +2877,7 @@
 		}
 	} else {
 		while (test_and_set_bit(0, &ap->jumbo_refill_busy));
-		synchronize_irq();
+		synchronize_irq(dev->irq);
 		ace_set_rxtx_parms(dev, 0);
 		if (ap->jumbo) {
 			struct cmd cmd;

@@ -755,7 +755,7 @@
 
 	/* synchronize with the interrupt handler and kill it */
 	dev->rx_info.up = 0;
-	synchronize_irq();
+	synchronize_irq(dev->irq);
 
 	/* touch the pci bus... */
 	readl(dev->base + IMR);

@@ -1296,11 +1296,11 @@
 	readl(dev->base + IER);
 
 	dev->rx_info.up = 0;
-	synchronize_irq();
+	synchronize_irq(dev->irq);
 
 	ns83820_do_reset(dev, CR_RST);
 
-	synchronize_irq();
+	synchronize_irq(dev->irq);
 
 	dev->IMR_cache &= ~(ISR_TXURN | ISR_TXIDLE | ISR_TXERR | ISR_TXDESC | ISR_TXOK);
 	ns83820_cleanup_rx(dev);

@@ -2544,7 +2544,7 @@
 	spin_unlock_irqrestore(&sc->sbm_lock, flags);
 	
 	/* Make sure there is no irq-handler running on a different CPU. */
-	synchronize_irq();
+	synchronize_irq(dev->irq);
 	
 	free_irq(dev->irq, dev);
 	

@@ -1699,7 +1699,7 @@
 {
 	atomic_inc(&adapter->irq_sem);
 	E1000_WRITE_REG(&adapter->hw, IMC, ~0);
-	synchronize_irq();
+	synchronize_irq(adapter->pdev->irq);
 }
 
 /**

@@ -1381,7 +1381,7 @@
 	don = esp->dma_ports_p(esp);
 	if(don) {
 		esp->dma_ints_off(esp);
-		synchronize_irq();
+		synchronize_irq(esp->irq);
 	}
 	if(esp->issue_SC) {
 		Scsi_Cmnd **prev, *this;

@@ -24,7 +24,7 @@
  * The largest-sized BIO which this code will assemble, in bytes.  Set this
  * to PAGE_CACHE_SIZE if your drivers are broken.
  */
-#define MPAGE_BIO_MAX_SIZE BIO_MAX_SIZE
+#define MPAGE_BIO_MAX_SIZE PAGE_CACHE_SIZE
 
 /*
  * I/O completion handler for multipage BIOs.

@@ -108,7 +108,11 @@
 
 #define APIC_BASE (fix_to_virt(FIX_APIC_BASE))
 
+#ifndef CONFIG_MULTIQUAD
 #define MAX_IO_APICS 8
+#else
+#define MAX_IO_APICS 1024
+#endif /* CONFIG_MULTIQUAD */
 
 /*
  * the local APIC register structure, memory mapped. Not terribly well

@@ -527,6 +527,19 @@
 	do_initcalls();
 }
 
+static void do_pre_smp_initcalls(void)
+{
+#if CONFIG_SMP
+	extern int migration_init(void);
+#endif
+	extern int spawn_ksoftirqd(void);
+
+#if CONFIG_SMP
+	migration_init();
+#endif
+	spawn_ksoftirqd();
+}
+
 extern void prepare_namespace(void);
 
 static int init(void * unused)

@@ -536,6 +549,9 @@
 	lock_kernel();
 	/* Sets up cpus_possible() */
 	smp_prepare_cpus(max_cpus);
+
+	do_pre_smp_initcalls();
+
 	smp_init();
 	do_basic_setup();
 

120 lines not on this page · raw message

@@ -410,11 +410,9 @@
 
 static struct notifier_block cpu_nfb = { &cpu_callback, NULL, 0 };
 
-static __init int spawn_ksoftirqd(void)
+__init int spawn_ksoftirqd(void)
 {
 	cpu_callback(&cpu_nfb, CPU_ONLINE, (void *)smp_processor_id());
 	register_cpu_notifier(&cpu_nfb);
 	return 0;
 }
-
-__initcall(spawn_ksoftirqd);