This commit was manufactured by cvs2svn to create branch 'vserver'.
[linux-2.6.git] / arch / i386 / kernel / irq-xen.c
diff --git a/arch/i386/kernel/irq-xen.c b/arch/i386/kernel/irq-xen.c
new file mode 100644 (file)
index 0000000..b4a38e4
--- /dev/null
@@ -0,0 +1,306 @@
+/*
+ *     linux/arch/i386/kernel/irq.c
+ *
+ *     Copyright (C) 1992, 1998 Linus Torvalds, Ingo Molnar
+ *
+ * This file contains the lowest level x86-specific interrupt
+ * entry, irq-stacks and irq statistics code. All the remaining
+ * irq logic is done by the generic kernel/irq/ code and
+ * by the x86-specific irq controller code. (e.g. i8259.c and
+ * io_apic.c.)
+ */
+
+#include <asm/uaccess.h>
+#include <linux/module.h>
+#include <linux/seq_file.h>
+#include <linux/interrupt.h>
+#include <linux/kernel_stat.h>
+#include <linux/notifier.h>
+#include <linux/cpu.h>
+#include <linux/delay.h>
+
+DEFINE_PER_CPU(irq_cpustat_t, irq_stat) ____cacheline_internodealigned_in_smp;
+EXPORT_PER_CPU_SYMBOL(irq_stat);
+
+#ifndef CONFIG_X86_LOCAL_APIC
+/*
+ * 'what should we do if we get a hw irq event on an illegal vector'.
+ * each architecture has to answer this themselves.
+ */
+void ack_bad_irq(unsigned int irq)
+{
+       printk("unexpected IRQ trap at vector %02x\n", irq);
+}
+#endif
+
+#ifdef CONFIG_4KSTACKS
+/*
+ * per-CPU IRQ handling contexts (thread information and stack)
+ */
+union irq_ctx {
+       struct thread_info      tinfo;
+       u32                     stack[THREAD_SIZE/sizeof(u32)];
+};
+
+static union irq_ctx *hardirq_ctx[NR_CPUS];
+static union irq_ctx *softirq_ctx[NR_CPUS];
+#endif
+
+/*
+ * do_IRQ handles all normal device IRQ's (the special
+ * SMP cross-CPU interrupts have their own specific
+ * handlers).
+ */
+fastcall unsigned int do_IRQ(struct pt_regs *regs)
+{      
+       /* high bit used in ret_from_ code */
+       int irq = ~regs->orig_eax;
+#ifdef CONFIG_4KSTACKS
+       union irq_ctx *curctx, *irqctx;
+       u32 *isp;
+#endif
+
+       irq_enter();
+#ifdef CONFIG_DEBUG_STACKOVERFLOW
+       /* Debugging check for stack overflow: is there less than 1KB free? */
+       {
+               long esp;
+
+               __asm__ __volatile__("andl %%esp,%0" :
+                                       "=r" (esp) : "0" (THREAD_SIZE - 1));
+               if (unlikely(esp < (sizeof(struct thread_info) + STACK_WARN))) {
+                       printk("do_IRQ: stack overflow: %ld\n",
+                               esp - sizeof(struct thread_info));
+                       dump_stack();
+               }
+       }
+#endif
+
+#ifdef CONFIG_4KSTACKS
+
+       curctx = (union irq_ctx *) current_thread_info();
+       irqctx = hardirq_ctx[smp_processor_id()];
+
+       /*
+        * this is where we switch to the IRQ stack. However, if we are
+        * already using the IRQ stack (because we interrupted a hardirq
+        * handler) we can't do that and just have to keep using the
+        * current stack (which is the irq stack already after all)
+        */
+       if (curctx != irqctx) {
+               int arg1, arg2, ebx;
+
+               /* build the stack frame on the IRQ stack */
+               isp = (u32*) ((char*)irqctx + sizeof(*irqctx));
+               irqctx->tinfo.task = curctx->tinfo.task;
+               irqctx->tinfo.previous_esp = current_stack_pointer;
+
+               asm volatile(
+                       "       xchgl   %%ebx,%%esp      \n"
+                       "       call    __do_IRQ         \n"
+                       "       movl   %%ebx,%%esp      \n"
+                       : "=a" (arg1), "=d" (arg2), "=b" (ebx)
+                       :  "0" (irq),   "1" (regs),  "2" (isp)
+                       : "memory", "cc", "ecx"
+               );
+       } else
+#endif
+               __do_IRQ(irq, regs);
+
+       irq_exit();
+
+       return 1;
+}
+
+#ifdef CONFIG_4KSTACKS
+
+/*
+ * These should really be __section__(".bss.page_aligned") as well, but
+ * gcc's 3.0 and earlier don't handle that correctly.
+ */
+static char softirq_stack[NR_CPUS * THREAD_SIZE]
+               __attribute__((__aligned__(THREAD_SIZE)));
+
+static char hardirq_stack[NR_CPUS * THREAD_SIZE]
+               __attribute__((__aligned__(THREAD_SIZE)));
+
+/*
+ * allocate per-cpu stacks for hardirq and for softirq processing
+ */
+void irq_ctx_init(int cpu)
+{
+       union irq_ctx *irqctx;
+
+       if (hardirq_ctx[cpu])
+               return;
+
+       irqctx = (union irq_ctx*) &hardirq_stack[cpu*THREAD_SIZE];
+       irqctx->tinfo.task              = NULL;
+       irqctx->tinfo.exec_domain       = NULL;
+       irqctx->tinfo.cpu               = cpu;
+       irqctx->tinfo.preempt_count     = HARDIRQ_OFFSET;
+       irqctx->tinfo.addr_limit        = MAKE_MM_SEG(0);
+
+       hardirq_ctx[cpu] = irqctx;
+
+       irqctx = (union irq_ctx*) &softirq_stack[cpu*THREAD_SIZE];
+       irqctx->tinfo.task              = NULL;
+       irqctx->tinfo.exec_domain       = NULL;
+       irqctx->tinfo.cpu               = cpu;
+       irqctx->tinfo.preempt_count     = SOFTIRQ_OFFSET;
+       irqctx->tinfo.addr_limit        = MAKE_MM_SEG(0);
+
+       softirq_ctx[cpu] = irqctx;
+
+       printk("CPU %u irqstacks, hard=%p soft=%p\n",
+               cpu,hardirq_ctx[cpu],softirq_ctx[cpu]);
+}
+
+void irq_ctx_exit(int cpu)
+{
+       hardirq_ctx[cpu] = NULL;
+}
+
+extern asmlinkage void __do_softirq(void);
+
+asmlinkage void do_softirq(void)
+{
+       unsigned long flags;
+       struct thread_info *curctx;
+       union irq_ctx *irqctx;
+       u32 *isp;
+
+       if (in_interrupt())
+               return;
+
+       local_irq_save(flags);
+
+       if (local_softirq_pending()) {
+               curctx = current_thread_info();
+               irqctx = softirq_ctx[smp_processor_id()];
+               irqctx->tinfo.task = curctx->task;
+               irqctx->tinfo.previous_esp = current_stack_pointer;
+
+               /* build the stack frame on the softirq stack */
+               isp = (u32*) ((char*)irqctx + sizeof(*irqctx));
+
+               asm volatile(
+                       "       xchgl   %%ebx,%%esp     \n"
+                       "       call    __do_softirq    \n"
+                       "       movl    %%ebx,%%esp     \n"
+                       : "=b"(isp)
+                       : "0"(isp)
+                       : "memory", "cc", "edx", "ecx", "eax"
+               );
+       }
+
+       local_irq_restore(flags);
+}
+
+EXPORT_SYMBOL(do_softirq);
+#endif
+
+/*
+ * Interrupt statistics:
+ */
+
+atomic_t irq_err_count;
+
+/*
+ * /proc/interrupts printing:
+ */
+
+int show_interrupts(struct seq_file *p, void *v)
+{
+       int i = *(loff_t *) v, j;
+       struct irqaction * action;
+       unsigned long flags;
+
+       if (i == 0) {
+               seq_printf(p, "           ");
+               for_each_online_cpu(j)
+                       seq_printf(p, "CPU%d       ",j);
+               seq_putc(p, '\n');
+       }
+
+       if (i < NR_IRQS) {
+               spin_lock_irqsave(&irq_desc[i].lock, flags);
+               action = irq_desc[i].action;
+               if (!action)
+                       goto skip;
+               seq_printf(p, "%3d: ",i);
+#ifndef CONFIG_SMP
+               seq_printf(p, "%10u ", kstat_irqs(i));
+#else
+               for_each_online_cpu(j)
+                       seq_printf(p, "%10u ", kstat_cpu(j).irqs[i]);
+#endif
+               seq_printf(p, " %14s", irq_desc[i].handler->typename);
+               seq_printf(p, "  %s", action->name);
+
+               for (action=action->next; action; action = action->next)
+                       seq_printf(p, ", %s", action->name);
+
+               seq_putc(p, '\n');
+skip:
+               spin_unlock_irqrestore(&irq_desc[i].lock, flags);
+       } else if (i == NR_IRQS) {
+               seq_printf(p, "NMI: ");
+               for_each_online_cpu(j)
+                       seq_printf(p, "%10u ", nmi_count(j));
+               seq_putc(p, '\n');
+#ifdef CONFIG_X86_LOCAL_APIC
+               seq_printf(p, "LOC: ");
+               for_each_online_cpu(j)
+                       seq_printf(p, "%10u ",
+                               per_cpu(irq_stat,j).apic_timer_irqs);
+               seq_putc(p, '\n');
+#endif
+               seq_printf(p, "ERR: %10u\n", atomic_read(&irq_err_count));
+#if defined(CONFIG_X86_IO_APIC)
+               seq_printf(p, "MIS: %10u\n", atomic_read(&irq_mis_count));
+#endif
+       }
+       return 0;
+}
+
+#ifdef CONFIG_HOTPLUG_CPU
+
+void fixup_irqs(cpumask_t map)
+{
+       unsigned int irq;
+       static int warned;
+
+       for (irq = 0; irq < NR_IRQS; irq++) {
+               cpumask_t mask;
+               if (irq == 2)
+                       continue;
+
+               cpus_and(mask, irq_affinity[irq], map);
+               if (any_online_cpu(mask) == NR_CPUS) {
+                       /*printk("Breaking affinity for irq %i\n", irq);*/
+                       mask = map;
+               }
+               if (irq_desc[irq].handler->set_affinity)
+                       irq_desc[irq].handler->set_affinity(irq, mask);
+               else if (irq_desc[irq].action && !(warned++))
+                       printk("Cannot set affinity for irq %i\n", irq);
+       }
+
+#if 0
+       barrier();
+       /* Ingo Molnar says: "after the IO-APIC masks have been redirected
+          [note the nop - the interrupt-enable boundary on x86 is two
+          instructions from sti] - to flush out pending hardirqs and
+          IPIs. After this point nothing is supposed to reach this CPU." */
+       __asm__ __volatile__("sti; nop; cli");
+       barrier();
+#else
+       /* That doesn't seem sufficient.  Give it 1ms. */
+       local_irq_enable();
+       mdelay(1);
+       local_irq_disable();
+#endif
+}
+#endif
+