Merge to Fedora kernel-2.6.18-1.2255_FC5-vs2.0.2.2-rc9 patched with stable patch...
[linux-2.6.git] / arch / i386 / kernel / irq-xen.c
1 /*
2  *      linux/arch/i386/kernel/irq.c
3  *
4  *      Copyright (C) 1992, 1998 Linus Torvalds, Ingo Molnar
5  *
6  * This file contains the lowest level x86-specific interrupt
7  * entry, irq-stacks and irq statistics code. All the remaining
8  * irq logic is done by the generic kernel/irq/ code and
9  * by the x86-specific irq controller code. (e.g. i8259.c and
10  * io_apic.c.)
11  */
12
13 #include <asm/uaccess.h>
14 #include <linux/module.h>
15 #include <linux/seq_file.h>
16 #include <linux/interrupt.h>
17 #include <linux/kernel_stat.h>
18 #include <linux/notifier.h>
19 #include <linux/cpu.h>
20 #include <linux/delay.h>
21
22 DEFINE_PER_CPU(irq_cpustat_t, irq_stat) ____cacheline_internodealigned_in_smp;
23 EXPORT_PER_CPU_SYMBOL(irq_stat);
24
25 #ifndef CONFIG_X86_LOCAL_APIC
26 /*
27  * 'what should we do if we get a hw irq event on an illegal vector'.
28  * each architecture has to answer this themselves.
29  */
30 void ack_bad_irq(unsigned int irq)
31 {
32         printk("unexpected IRQ trap at vector %02x\n", irq);
33 }
34 #endif
35
36 #ifdef CONFIG_4KSTACKS
37 /*
38  * per-CPU IRQ handling contexts (thread information and stack)
39  */
40 union irq_ctx {
41         struct thread_info      tinfo;
42         u32                     stack[THREAD_SIZE/sizeof(u32)];
43 };
44
45 static union irq_ctx *hardirq_ctx[NR_CPUS] __read_mostly;
46 static union irq_ctx *softirq_ctx[NR_CPUS] __read_mostly;
47 #endif
48
49 /*
50  * do_IRQ handles all normal device IRQ's (the special
51  * SMP cross-CPU interrupts have their own specific
52  * handlers).
53  */
54 fastcall unsigned int do_IRQ(struct pt_regs *regs)
55 {       
56         /* high bit used in ret_from_ code */
57         int irq = ~regs->orig_eax;
58 #ifdef CONFIG_4KSTACKS
59         union irq_ctx *curctx, *irqctx;
60         u32 *isp;
61 #endif
62
63         if (unlikely((unsigned)irq >= NR_IRQS)) {
64                 printk(KERN_EMERG "%s: cannot handle IRQ %d\n",
65                                         __FUNCTION__, irq);
66                 BUG();
67         }
68
69         irq_enter();
70 #ifdef CONFIG_DEBUG_STACKOVERFLOW
71         /* Debugging check for stack overflow: is there less than 1KB free? */
72         {
73                 long esp;
74
75                 __asm__ __volatile__("andl %%esp,%0" :
76                                         "=r" (esp) : "0" (THREAD_SIZE - 1));
77                 if (unlikely(esp < (sizeof(struct thread_info) + STACK_WARN))) {
78                         printk("do_IRQ: stack overflow: %ld\n",
79                                 esp - sizeof(struct thread_info));
80                         dump_stack();
81                 }
82         }
83 #endif
84 #ifdef CONFIG_4KSTACKS
85
86         curctx = (union irq_ctx *) current_thread_info();
87         irqctx = hardirq_ctx[smp_processor_id()];
88
89         /*
90          * this is where we switch to the IRQ stack. However, if we are
91          * already using the IRQ stack (because we interrupted a hardirq
92          * handler) we can't do that and just have to keep using the
93          * current stack (which is the irq stack already after all)
94          */
95         if (curctx != irqctx) {
96                 int arg1, arg2, ebx;
97
98                 /* build the stack frame on the IRQ stack */
99                 isp = (u32*) ((char*)irqctx + sizeof(*irqctx));
100                 irqctx->tinfo.task = curctx->tinfo.task;
101                 irqctx->tinfo.previous_esp = current_stack_pointer;
102
103                 /*
104                  * Copy the softirq bits in preempt_count so that the
105                  * softirq checks work in the hardirq context.
106                  */
107                 irqctx->tinfo.preempt_count =
108                         (irqctx->tinfo.preempt_count & ~SOFTIRQ_MASK) |
109                          (curctx->tinfo.preempt_count & SOFTIRQ_MASK);
110
111                 asm volatile(
112                         "       xchgl   %%ebx,%%esp      \n"
113                         "       call    __do_IRQ         \n"
114                         "       movl   %%ebx,%%esp      \n"
115                         : "=a" (arg1), "=d" (arg2), "=b" (ebx)
116                         :  "0" (irq),   "1" (regs),  "2" (isp)
117                         : "memory", "cc", "ecx"
118                 );
119         } else
120 #endif
121                 __do_IRQ(irq, regs);
122
123         irq_exit();
124
125         return 1;
126 }
127
128 #ifdef CONFIG_4KSTACKS
129
130 /*
131  * These should really be __section__(".bss.page_aligned") as well, but
132  * gcc's 3.0 and earlier don't handle that correctly.
133  */
134 static char softirq_stack[NR_CPUS * THREAD_SIZE]
135                 __attribute__((__aligned__(THREAD_SIZE)));
136
137 static char hardirq_stack[NR_CPUS * THREAD_SIZE]
138                 __attribute__((__aligned__(THREAD_SIZE)));
139
140 /*
141  * allocate per-cpu stacks for hardirq and for softirq processing
142  */
143 void irq_ctx_init(int cpu)
144 {
145         union irq_ctx *irqctx;
146
147         if (hardirq_ctx[cpu])
148                 return;
149
150         irqctx = (union irq_ctx*) &hardirq_stack[cpu*THREAD_SIZE];
151         irqctx->tinfo.task              = NULL;
152         irqctx->tinfo.exec_domain       = NULL;
153         irqctx->tinfo.cpu               = cpu;
154         irqctx->tinfo.preempt_count     = HARDIRQ_OFFSET;
155         irqctx->tinfo.addr_limit        = MAKE_MM_SEG(0);
156
157         hardirq_ctx[cpu] = irqctx;
158
159         irqctx = (union irq_ctx*) &softirq_stack[cpu*THREAD_SIZE];
160         irqctx->tinfo.task              = NULL;
161         irqctx->tinfo.exec_domain       = NULL;
162         irqctx->tinfo.cpu               = cpu;
163         irqctx->tinfo.preempt_count     = 0;
164         irqctx->tinfo.addr_limit        = MAKE_MM_SEG(0);
165
166         softirq_ctx[cpu] = irqctx;
167
168         printk("CPU %u irqstacks, hard=%p soft=%p\n",
169                 cpu,hardirq_ctx[cpu],softirq_ctx[cpu]);
170 }
171
172 void irq_ctx_exit(int cpu)
173 {
174         hardirq_ctx[cpu] = NULL;
175 }
176
177 extern asmlinkage void __do_softirq(void);
178
179 asmlinkage void do_softirq(void)
180 {
181         unsigned long flags;
182         struct thread_info *curctx;
183         union irq_ctx *irqctx;
184         u32 *isp;
185
186         if (in_interrupt())
187                 return;
188
189         local_irq_save(flags);
190
191         if (local_softirq_pending()) {
192                 curctx = current_thread_info();
193                 irqctx = softirq_ctx[smp_processor_id()];
194                 irqctx->tinfo.task = curctx->task;
195                 irqctx->tinfo.previous_esp = current_stack_pointer;
196
197                 /* build the stack frame on the softirq stack */
198                 isp = (u32*) ((char*)irqctx + sizeof(*irqctx));
199
200                 asm volatile(
201                         "       xchgl   %%ebx,%%esp     \n"
202                         "       call    __do_softirq    \n"
203                         "       movl    %%ebx,%%esp     \n"
204                         : "=b"(isp)
205                         : "0"(isp)
206                         : "memory", "cc", "edx", "ecx", "eax"
207                 );
208                 /*
209                  * Shouldnt happen, we returned above if in_interrupt():
210                  */
211                 WARN_ON_ONCE(softirq_count());
212         }
213
214         local_irq_restore(flags);
215 }
216
217 EXPORT_SYMBOL(do_softirq);
218 #endif
219
220 /*
221  * Interrupt statistics:
222  */
223
224 atomic_t irq_err_count;
225
226 /*
227  * /proc/interrupts printing:
228  */
229
230 int show_interrupts(struct seq_file *p, void *v)
231 {
232         int i = *(loff_t *) v, j;
233         struct irqaction * action;
234         unsigned long flags;
235
236         if (i == 0) {
237                 seq_printf(p, "           ");
238                 for_each_online_cpu(j)
239                         seq_printf(p, "CPU%-8d       ",j);
240                 seq_putc(p, '\n');
241         }
242
243         if (i < NR_IRQS) {
244                 spin_lock_irqsave(&irq_desc[i].lock, flags);
245                 action = irq_desc[i].action;
246                 if (!action)
247                         goto skip;
248                 seq_printf(p, "%3d: ",i);
249 #ifndef CONFIG_SMP
250                 seq_printf(p, "%10u ", kstat_irqs(i));
251 #else
252                 for_each_online_cpu(j)
253                         seq_printf(p, "%10u ", kstat_cpu(j).irqs[i]);
254 #endif
255                 seq_printf(p, " %14s", irq_desc[i].chip->typename);
256                 seq_printf(p, "  %s", action->name);
257
258                 for (action=action->next; action; action = action->next)
259                         seq_printf(p, ", %s", action->name);
260
261                 seq_putc(p, '\n');
262 skip:
263                 spin_unlock_irqrestore(&irq_desc[i].lock, flags);
264         } else if (i == NR_IRQS) {
265                 seq_printf(p, "NMI: ");
266                 for_each_online_cpu(j)
267                         seq_printf(p, "%10u ", nmi_count(j));
268                 seq_putc(p, '\n');
269 #ifdef CONFIG_X86_LOCAL_APIC
270                 seq_printf(p, "LOC: ");
271                 for_each_online_cpu(j)
272                         seq_printf(p, "%10u ",
273                                 per_cpu(irq_stat,j).apic_timer_irqs);
274                 seq_putc(p, '\n');
275 #endif
276                 seq_printf(p, "ERR: %10u\n", atomic_read(&irq_err_count));
277 #if defined(CONFIG_X86_IO_APIC)
278                 seq_printf(p, "MIS: %10u\n", atomic_read(&irq_mis_count));
279 #endif
280         }
281         return 0;
282 }
283
284 #ifdef CONFIG_HOTPLUG_CPU
285
286 void fixup_irqs(cpumask_t map)
287 {
288         unsigned int irq;
289         static int warned;
290
291         for (irq = 0; irq < NR_IRQS; irq++) {
292                 cpumask_t mask;
293                 if (irq == 2)
294                         continue;
295
296                 cpus_and(mask, irq_desc[irq].affinity, map);
297                 if (any_online_cpu(mask) == NR_CPUS) {
298                         /*printk("Breaking affinity for irq %i\n", irq);*/
299                         mask = map;
300                 }
301                 if (irq_desc[irq].chip->set_affinity)
302                         irq_desc[irq].chip->set_affinity(irq, mask);
303                 else if (irq_desc[irq].action && !(warned++))
304                         printk("Cannot set affinity for irq %i\n", irq);
305         }
306
307 #if 0
308         barrier();
309         /* Ingo Molnar says: "after the IO-APIC masks have been redirected
310            [note the nop - the interrupt-enable boundary on x86 is two
311            instructions from sti] - to flush out pending hardirqs and
312            IPIs. After this point nothing is supposed to reach this CPU." */
313         __asm__ __volatile__("sti; nop; cli");
314         barrier();
315 #else
316         /* That doesn't seem sufficient.  Give it 1ms. */
317         local_irq_enable();
318         mdelay(1);
319         local_irq_disable();
320 #endif
321 }
322 #endif
323