[PATCH] irqpoll

Anyone reporting a stuck IRQ should try these options. Its effectiveness
varies we've found in the Fedora case. Quite a few systems with misdescribed
IRQ routing just work when you use irqpoll. It also fixes up the VIA systems
although thats now fixed with the VIA quirk (which we could just make default
as its what Redmond OS does but Linus didn't like it historically).

A small number of systems have jammed IRQ sources or misdescribes that cause
an IRQ that we have no handler registered anywhere for. In those cases it
doesn't help.

Signed-off-by: Alan Cox <number6@the-village.bc.nu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

authored by Alan Cox and committed by Linus Torvalds 200803df 21fe3471

+131 -6
+15
Documentation/kernel-parameters.txt
··· 437 437 Format: {"of[f]" | "sk[ipmbr]"} 438 438 See comment in arch/i386/boot/edd.S 439 439 440 + edd [EDD] 441 + Format: {"of[f]" | "sk[ipmbr]"} 442 + See comment in arch/i386/boot/edd.S 443 + 440 444 eicon= [HW,ISDN] 441 445 Format: <id>,<membase>,<irq> 442 446 ··· 625 621 626 622 ips= [HW,SCSI] Adaptec / IBM ServeRAID controller 627 623 See header of drivers/scsi/ips.c. 624 + 625 + irqfixup [HW] 626 + When an interrupt is not handled search all handlers 627 + for it. Intended to get systems with badly broken 628 + firmware running. 629 + 630 + irqpoll [HW] 631 + When an interrupt is not handled search all handlers 632 + for it. Also check all handlers each timer 633 + interrupt. Intended to get systems with badly broken 634 + firmware running. 628 635 629 636 isapnp= [ISAPNP] 630 637 Format: <RDP>, <reset>, <pci_scan>, <verbosity>
+1 -1
arch/ppc64/kernel/irq.c
··· 245 245 246 246 spin_lock(&desc->lock); 247 247 if (!noirqdebug) 248 - note_interrupt(irq, desc, action_ret); 248 + note_interrupt(irq, desc, action_ret, regs); 249 249 if (likely(!(desc->status & IRQ_PENDING))) 250 250 break; 251 251 desc->status &= ~IRQ_PENDING;
+3 -2
include/linux/irq.h
··· 85 85 extern int noirqdebug_setup(char *str); 86 86 87 87 extern fastcall int handle_IRQ_event(unsigned int irq, struct pt_regs *regs, 88 - struct irqaction *action); 88 + struct irqaction *action); 89 89 extern fastcall unsigned int __do_IRQ(unsigned int irq, struct pt_regs *regs); 90 - extern void note_interrupt(unsigned int irq, irq_desc_t *desc, int action_ret); 90 + extern void note_interrupt(unsigned int irq, irq_desc_t *desc, 91 + int action_ret, struct pt_regs *regs); 91 92 extern int can_request_irq(unsigned int irq, unsigned long irqflags); 92 93 93 94 extern void init_irq_proc(void);
+1 -1
kernel/irq/handle.c
··· 172 172 173 173 spin_lock(&desc->lock); 174 174 if (!noirqdebug) 175 - note_interrupt(irq, desc, action_ret); 175 + note_interrupt(irq, desc, action_ret, regs); 176 176 if (likely(!(desc->status & IRQ_PENDING))) 177 177 break; 178 178 desc->status &= ~IRQ_PENDING;
+111 -2
kernel/irq/spurious.c
··· 11 11 #include <linux/kallsyms.h> 12 12 #include <linux/interrupt.h> 13 13 14 + static int irqfixup; 15 + 16 + /* 17 + * Recovery handler for misrouted interrupts. 18 + */ 19 + 20 + static int misrouted_irq(int irq, struct pt_regs *regs) 21 + { 22 + int i; 23 + irq_desc_t *desc; 24 + int ok = 0; 25 + int work = 0; /* Did we do work for a real IRQ */ 26 + 27 + for(i = 1; i < NR_IRQS; i++) { 28 + struct irqaction *action; 29 + 30 + if (i == irq) /* Already tried */ 31 + continue; 32 + desc = &irq_desc[i]; 33 + spin_lock(&desc->lock); 34 + action = desc->action; 35 + /* Already running on another processor */ 36 + if (desc->status & IRQ_INPROGRESS) { 37 + /* 38 + * Already running: If it is shared get the other 39 + * CPU to go looking for our mystery interrupt too 40 + */ 41 + if (desc->action && (desc->action->flags & SA_SHIRQ)) 42 + desc->status |= IRQ_PENDING; 43 + spin_unlock(&desc->lock); 44 + continue; 45 + } 46 + /* Honour the normal IRQ locking */ 47 + desc->status |= IRQ_INPROGRESS; 48 + spin_unlock(&desc->lock); 49 + while (action) { 50 + /* Only shared IRQ handlers are safe to call */ 51 + if (action->flags & SA_SHIRQ) { 52 + if (action->handler(i, action->dev_id, regs) == 53 + IRQ_HANDLED) 54 + ok = 1; 55 + } 56 + action = action->next; 57 + } 58 + local_irq_disable(); 59 + /* Now clean up the flags */ 60 + spin_lock(&desc->lock); 61 + action = desc->action; 62 + 63 + /* 64 + * While we were looking for a fixup someone queued a real 65 + * IRQ clashing with our walk 66 + */ 67 + 68 + while ((desc->status & IRQ_PENDING) && action) { 69 + /* 70 + * Perform real IRQ processing for the IRQ we deferred 71 + */ 72 + work = 1; 73 + spin_unlock(&desc->lock); 74 + handle_IRQ_event(i, regs, action); 75 + spin_lock(&desc->lock); 76 + desc->status &= ~IRQ_PENDING; 77 + } 78 + desc->status &= ~IRQ_INPROGRESS; 79 + /* 80 + * If we did actual work for the real IRQ line we must let the 81 + * IRQ controller clean up too 82 + */ 83 + if(work) 84 + desc->handler->end(i); 85 + spin_unlock(&desc->lock); 86 + } 87 + /* So the caller can adjust the irq error counts */ 88 + return ok; 89 + } 90 + 14 91 /* 15 92 * If 99,900 of the previous 100,000 interrupts have not been handled 16 93 * then assume that the IRQ is stuck in some manner. Drop a diagnostic ··· 108 31 printk(KERN_ERR "irq event %d: bogus return value %x\n", 109 32 irq, action_ret); 110 33 } else { 111 - printk(KERN_ERR "irq %d: nobody cared!\n", irq); 34 + printk(KERN_ERR "irq %d: nobody cared (try booting with " 35 + "the \"irqpoll\" option)\n", irq); 112 36 } 113 37 dump_stack(); 114 38 printk(KERN_ERR "handlers:\n"); ··· 133 55 } 134 56 } 135 57 136 - void note_interrupt(unsigned int irq, irq_desc_t *desc, irqreturn_t action_ret) 58 + void note_interrupt(unsigned int irq, irq_desc_t *desc, irqreturn_t action_ret, 59 + struct pt_regs *regs) 137 60 { 138 61 if (action_ret != IRQ_HANDLED) { 139 62 desc->irqs_unhandled++; 140 63 if (action_ret != IRQ_NONE) 141 64 report_bad_irq(irq, desc, action_ret); 65 + } 66 + 67 + if (unlikely(irqfixup)) { 68 + /* Don't punish working computers */ 69 + if ((irqfixup == 2 && irq == 0) || action_ret == IRQ_NONE) { 70 + int ok = misrouted_irq(irq, regs); 71 + if (action_ret == IRQ_NONE) 72 + desc->irqs_unhandled -= ok; 73 + } 142 74 } 143 75 144 76 desc->irq_count++; ··· 182 94 183 95 __setup("noirqdebug", noirqdebug_setup); 184 96 97 + static int __init irqfixup_setup(char *str) 98 + { 99 + irqfixup = 1; 100 + printk(KERN_WARNING "Misrouted IRQ fixup support enabled.\n"); 101 + printk(KERN_WARNING "This may impact system performance.\n"); 102 + return 1; 103 + } 104 + 105 + __setup("irqfixup", irqfixup_setup); 106 + 107 + static int __init irqpoll_setup(char *str) 108 + { 109 + irqfixup = 2; 110 + printk(KERN_WARNING "Misrouted IRQ fixup and polling support " 111 + "enabled\n"); 112 + printk(KERN_WARNING "This may significantly impact system " 113 + "performance\n"); 114 + return 1; 115 + } 116 + 117 + __setup("irqpoll", irqpoll_setup);