commit | author | age
|
a07526
|
1 |
// SPDX-License-Identifier: GPL-2.0-only |
H |
2 |
/* |
|
3 |
* Local APIC related interfaces to support IOAPIC, MSI, etc. |
|
4 |
* |
|
5 |
* Copyright (C) 1997, 1998, 1999, 2000, 2009 Ingo Molnar, Hajnalka Szabo |
|
6 |
* Moved from arch/x86/kernel/apic/io_apic.c. |
|
7 |
* Jiang Liu <jiang.liu@linux.intel.com> |
|
8 |
* Enable support of hierarchical irqdomains |
|
9 |
*/ |
|
10 |
#include <linux/interrupt.h> |
|
11 |
#include <linux/irq.h> |
|
12 |
#include <linux/seq_file.h> |
|
13 |
#include <linux/init.h> |
|
14 |
#include <linux/compiler.h> |
|
15 |
#include <linux/slab.h> |
|
16 |
#include <asm/irqdomain.h> |
|
17 |
#include <asm/hw_irq.h> |
|
18 |
#include <asm/traps.h> |
|
19 |
#include <asm/apic.h> |
|
20 |
#include <asm/i8259.h> |
|
21 |
#include <asm/desc.h> |
|
22 |
#include <asm/irq_remapping.h> |
|
23 |
|
|
24 |
#include <asm/trace/irq_vectors.h> |
|
25 |
|
|
26 |
struct apic_chip_data { |
|
27 |
struct irq_cfg hw_irq_cfg; |
|
28 |
unsigned int vector; |
|
29 |
unsigned int prev_vector; |
|
30 |
unsigned int cpu; |
|
31 |
unsigned int prev_cpu; |
|
32 |
unsigned int irq; |
|
33 |
struct hlist_node clist; |
|
34 |
unsigned int move_in_progress : 1, |
|
35 |
is_managed : 1, |
|
36 |
can_reserve : 1, |
|
37 |
has_reserved : 1; |
|
38 |
}; |
|
39 |
|
|
40 |
struct irq_domain *x86_vector_domain; |
|
41 |
EXPORT_SYMBOL_GPL(x86_vector_domain); |
2f529f
|
42 |
static DEFINE_HARD_SPINLOCK(vector_lock); |
a07526
|
43 |
static cpumask_var_t vector_searchmask; |
H |
44 |
static struct irq_chip lapic_controller; |
|
45 |
static struct irq_matrix *vector_matrix; |
|
46 |
#ifdef CONFIG_SMP |
|
47 |
static DEFINE_PER_CPU(struct hlist_head, cleanup_list); |
|
48 |
#endif |
|
49 |
|
|
50 |
void lock_vector_lock(void) |
|
51 |
{ |
|
52 |
/* Used to the online set of cpus does not change |
|
53 |
* during assign_irq_vector. |
|
54 |
*/ |
|
55 |
raw_spin_lock(&vector_lock); |
|
56 |
} |
|
57 |
|
|
58 |
void unlock_vector_lock(void) |
|
59 |
{ |
|
60 |
raw_spin_unlock(&vector_lock); |
|
61 |
} |
|
62 |
|
|
63 |
void init_irq_alloc_info(struct irq_alloc_info *info, |
|
64 |
const struct cpumask *mask) |
|
65 |
{ |
|
66 |
memset(info, 0, sizeof(*info)); |
|
67 |
info->mask = mask; |
|
68 |
} |
|
69 |
|
|
70 |
void copy_irq_alloc_info(struct irq_alloc_info *dst, struct irq_alloc_info *src) |
|
71 |
{ |
|
72 |
if (src) |
|
73 |
*dst = *src; |
|
74 |
else |
|
75 |
memset(dst, 0, sizeof(*dst)); |
|
76 |
} |
|
77 |
|
|
78 |
static struct apic_chip_data *apic_chip_data(struct irq_data *irqd) |
|
79 |
{ |
|
80 |
if (!irqd) |
|
81 |
return NULL; |
|
82 |
|
|
83 |
while (irqd->parent_data) |
|
84 |
irqd = irqd->parent_data; |
|
85 |
|
|
86 |
return irqd->chip_data; |
|
87 |
} |
|
88 |
|
|
89 |
struct irq_cfg *irqd_cfg(struct irq_data *irqd) |
|
90 |
{ |
|
91 |
struct apic_chip_data *apicd = apic_chip_data(irqd); |
|
92 |
|
|
93 |
return apicd ? &apicd->hw_irq_cfg : NULL; |
|
94 |
} |
|
95 |
EXPORT_SYMBOL_GPL(irqd_cfg); |
|
96 |
|
|
97 |
struct irq_cfg *irq_cfg(unsigned int irq) |
|
98 |
{ |
|
99 |
return irqd_cfg(irq_get_irq_data(irq)); |
|
100 |
} |
|
101 |
|
|
102 |
static struct apic_chip_data *alloc_apic_chip_data(int node) |
|
103 |
{ |
|
104 |
struct apic_chip_data *apicd; |
|
105 |
|
|
106 |
apicd = kzalloc_node(sizeof(*apicd), GFP_KERNEL, node); |
|
107 |
if (apicd) |
|
108 |
INIT_HLIST_NODE(&apicd->clist); |
|
109 |
return apicd; |
|
110 |
} |
|
111 |
|
|
112 |
static void free_apic_chip_data(struct apic_chip_data *apicd) |
|
113 |
{ |
|
114 |
kfree(apicd); |
|
115 |
} |
|
116 |
|
|
117 |
static void apic_update_irq_cfg(struct irq_data *irqd, unsigned int vector, |
|
118 |
unsigned int cpu) |
|
119 |
{ |
|
120 |
struct apic_chip_data *apicd = apic_chip_data(irqd); |
|
121 |
|
|
122 |
lockdep_assert_held(&vector_lock); |
|
123 |
|
|
124 |
apicd->hw_irq_cfg.vector = vector; |
|
125 |
apicd->hw_irq_cfg.dest_apicid = apic->calc_dest_apicid(cpu); |
|
126 |
irq_data_update_effective_affinity(irqd, cpumask_of(cpu)); |
|
127 |
trace_vector_config(irqd->irq, vector, cpu, |
|
128 |
apicd->hw_irq_cfg.dest_apicid); |
|
129 |
} |
|
130 |
|
|
131 |
static void apic_update_vector(struct irq_data *irqd, unsigned int newvec, |
|
132 |
unsigned int newcpu) |
|
133 |
{ |
|
134 |
struct apic_chip_data *apicd = apic_chip_data(irqd); |
|
135 |
struct irq_desc *desc = irq_data_to_desc(irqd); |
|
136 |
bool managed = irqd_affinity_is_managed(irqd); |
|
137 |
|
|
138 |
lockdep_assert_held(&vector_lock); |
|
139 |
|
|
140 |
trace_vector_update(irqd->irq, newvec, newcpu, apicd->vector, |
|
141 |
apicd->cpu); |
|
142 |
|
|
143 |
/* |
|
144 |
* If there is no vector associated or if the associated vector is |
|
145 |
* the shutdown vector, which is associated to make PCI/MSI |
|
146 |
* shutdown mode work, then there is nothing to release. Clear out |
|
147 |
* prev_vector for this and the offlined target case. |
|
148 |
*/ |
|
149 |
apicd->prev_vector = 0; |
|
150 |
if (!apicd->vector || apicd->vector == MANAGED_IRQ_SHUTDOWN_VECTOR) |
|
151 |
goto setnew; |
|
152 |
/* |
|
153 |
* If the target CPU of the previous vector is online, then mark |
|
154 |
* the vector as move in progress and store it for cleanup when the |
|
155 |
* first interrupt on the new vector arrives. If the target CPU is |
|
156 |
* offline then the regular release mechanism via the cleanup |
|
157 |
* vector is not possible and the vector can be immediately freed |
|
158 |
* in the underlying matrix allocator. |
|
159 |
*/ |
|
160 |
if (cpu_online(apicd->cpu)) { |
|
161 |
apicd->move_in_progress = true; |
|
162 |
apicd->prev_vector = apicd->vector; |
|
163 |
apicd->prev_cpu = apicd->cpu; |
|
164 |
WARN_ON_ONCE(apicd->cpu == newcpu); |
|
165 |
} else { |
|
166 |
irq_matrix_free(vector_matrix, apicd->cpu, apicd->vector, |
|
167 |
managed); |
|
168 |
} |
|
169 |
|
|
170 |
setnew: |
|
171 |
apicd->vector = newvec; |
|
172 |
apicd->cpu = newcpu; |
|
173 |
BUG_ON(!IS_ERR_OR_NULL(per_cpu(vector_irq, newcpu)[newvec])); |
|
174 |
per_cpu(vector_irq, newcpu)[newvec] = desc; |
|
175 |
} |
|
176 |
|
|
177 |
static void vector_assign_managed_shutdown(struct irq_data *irqd) |
|
178 |
{ |
|
179 |
unsigned int cpu = cpumask_first(cpu_online_mask); |
|
180 |
|
|
181 |
apic_update_irq_cfg(irqd, MANAGED_IRQ_SHUTDOWN_VECTOR, cpu); |
|
182 |
} |
|
183 |
|
|
184 |
static int reserve_managed_vector(struct irq_data *irqd) |
|
185 |
{ |
|
186 |
const struct cpumask *affmsk = irq_data_get_affinity_mask(irqd); |
|
187 |
struct apic_chip_data *apicd = apic_chip_data(irqd); |
|
188 |
unsigned long flags; |
|
189 |
int ret; |
|
190 |
|
|
191 |
raw_spin_lock_irqsave(&vector_lock, flags); |
|
192 |
apicd->is_managed = true; |
|
193 |
ret = irq_matrix_reserve_managed(vector_matrix, affmsk); |
|
194 |
raw_spin_unlock_irqrestore(&vector_lock, flags); |
|
195 |
trace_vector_reserve_managed(irqd->irq, ret); |
|
196 |
return ret; |
|
197 |
} |
|
198 |
|
|
199 |
static void reserve_irq_vector_locked(struct irq_data *irqd) |
|
200 |
{ |
|
201 |
struct apic_chip_data *apicd = apic_chip_data(irqd); |
|
202 |
|
|
203 |
irq_matrix_reserve(vector_matrix); |
|
204 |
apicd->can_reserve = true; |
|
205 |
apicd->has_reserved = true; |
|
206 |
irqd_set_can_reserve(irqd); |
|
207 |
trace_vector_reserve(irqd->irq, 0); |
|
208 |
vector_assign_managed_shutdown(irqd); |
|
209 |
} |
|
210 |
|
|
211 |
static int reserve_irq_vector(struct irq_data *irqd) |
|
212 |
{ |
|
213 |
unsigned long flags; |
|
214 |
|
|
215 |
raw_spin_lock_irqsave(&vector_lock, flags); |
|
216 |
reserve_irq_vector_locked(irqd); |
|
217 |
raw_spin_unlock_irqrestore(&vector_lock, flags); |
|
218 |
return 0; |
|
219 |
} |
|
220 |
|
|
221 |
static int |
|
222 |
assign_vector_locked(struct irq_data *irqd, const struct cpumask *dest) |
|
223 |
{ |
|
224 |
struct apic_chip_data *apicd = apic_chip_data(irqd); |
|
225 |
bool resvd = apicd->has_reserved; |
|
226 |
unsigned int cpu = apicd->cpu; |
|
227 |
int vector = apicd->vector; |
|
228 |
|
|
229 |
lockdep_assert_held(&vector_lock); |
|
230 |
|
|
231 |
/* |
|
232 |
* If the current target CPU is online and in the new requested |
|
233 |
* affinity mask, there is no point in moving the interrupt from |
|
234 |
* one CPU to another. |
|
235 |
*/ |
|
236 |
if (vector && cpu_online(cpu) && cpumask_test_cpu(cpu, dest)) |
|
237 |
return 0; |
|
238 |
|
|
239 |
/* |
|
240 |
* Careful here. @apicd might either have move_in_progress set or |
|
241 |
* be enqueued for cleanup. Assigning a new vector would either |
|
242 |
* leave a stale vector on some CPU around or in case of a pending |
|
243 |
* cleanup corrupt the hlist. |
|
244 |
*/ |
|
245 |
if (apicd->move_in_progress || !hlist_unhashed(&apicd->clist)) |
|
246 |
return -EBUSY; |
|
247 |
|
|
248 |
vector = irq_matrix_alloc(vector_matrix, dest, resvd, &cpu); |
|
249 |
trace_vector_alloc(irqd->irq, vector, resvd, vector); |
|
250 |
if (vector < 0) |
|
251 |
return vector; |
|
252 |
apic_update_vector(irqd, vector, cpu); |
|
253 |
apic_update_irq_cfg(irqd, vector, cpu); |
|
254 |
|
|
255 |
return 0; |
|
256 |
} |
|
257 |
|
|
258 |
static int assign_irq_vector(struct irq_data *irqd, const struct cpumask *dest) |
|
259 |
{ |
|
260 |
unsigned long flags; |
|
261 |
int ret; |
|
262 |
|
|
263 |
raw_spin_lock_irqsave(&vector_lock, flags); |
|
264 |
cpumask_and(vector_searchmask, dest, cpu_online_mask); |
|
265 |
ret = assign_vector_locked(irqd, vector_searchmask); |
|
266 |
raw_spin_unlock_irqrestore(&vector_lock, flags); |
|
267 |
return ret; |
|
268 |
} |
|
269 |
|
|
270 |
static int assign_irq_vector_any_locked(struct irq_data *irqd) |
|
271 |
{ |
|
272 |
/* Get the affinity mask - either irq_default_affinity or (user) set */ |
|
273 |
const struct cpumask *affmsk = irq_data_get_affinity_mask(irqd); |
|
274 |
int node = irq_data_get_node(irqd); |
|
275 |
|
|
276 |
if (node != NUMA_NO_NODE) { |
|
277 |
/* Try the intersection of @affmsk and node mask */ |
|
278 |
cpumask_and(vector_searchmask, cpumask_of_node(node), affmsk); |
|
279 |
if (!assign_vector_locked(irqd, vector_searchmask)) |
|
280 |
return 0; |
|
281 |
} |
|
282 |
|
|
283 |
/* Try the full affinity mask */ |
|
284 |
cpumask_and(vector_searchmask, affmsk, cpu_online_mask); |
|
285 |
if (!assign_vector_locked(irqd, vector_searchmask)) |
|
286 |
return 0; |
|
287 |
|
|
288 |
if (node != NUMA_NO_NODE) { |
|
289 |
/* Try the node mask */ |
|
290 |
if (!assign_vector_locked(irqd, cpumask_of_node(node))) |
|
291 |
return 0; |
|
292 |
} |
|
293 |
|
|
294 |
/* Try the full online mask */ |
|
295 |
return assign_vector_locked(irqd, cpu_online_mask); |
|
296 |
} |
|
297 |
|
|
298 |
static int |
|
299 |
assign_irq_vector_policy(struct irq_data *irqd, struct irq_alloc_info *info) |
|
300 |
{ |
|
301 |
if (irqd_affinity_is_managed(irqd)) |
|
302 |
return reserve_managed_vector(irqd); |
|
303 |
if (info->mask) |
|
304 |
return assign_irq_vector(irqd, info->mask); |
|
305 |
/* |
|
306 |
* Make only a global reservation with no guarantee. A real vector |
|
307 |
* is associated at activation time. |
|
308 |
*/ |
|
309 |
return reserve_irq_vector(irqd); |
|
310 |
} |
|
311 |
|
|
312 |
static int |
|
313 |
assign_managed_vector(struct irq_data *irqd, const struct cpumask *dest) |
|
314 |
{ |
|
315 |
const struct cpumask *affmsk = irq_data_get_affinity_mask(irqd); |
|
316 |
struct apic_chip_data *apicd = apic_chip_data(irqd); |
|
317 |
int vector, cpu; |
|
318 |
|
|
319 |
cpumask_and(vector_searchmask, dest, affmsk); |
|
320 |
|
|
321 |
/* set_affinity might call here for nothing */ |
|
322 |
if (apicd->vector && cpumask_test_cpu(apicd->cpu, vector_searchmask)) |
|
323 |
return 0; |
|
324 |
vector = irq_matrix_alloc_managed(vector_matrix, vector_searchmask, |
|
325 |
&cpu); |
|
326 |
trace_vector_alloc_managed(irqd->irq, vector, vector); |
|
327 |
if (vector < 0) |
|
328 |
return vector; |
|
329 |
apic_update_vector(irqd, vector, cpu); |
|
330 |
apic_update_irq_cfg(irqd, vector, cpu); |
|
331 |
return 0; |
|
332 |
} |
|
333 |
|
|
334 |
static void clear_irq_vector(struct irq_data *irqd) |
|
335 |
{ |
|
336 |
struct apic_chip_data *apicd = apic_chip_data(irqd); |
|
337 |
bool managed = irqd_affinity_is_managed(irqd); |
|
338 |
unsigned int vector = apicd->vector; |
|
339 |
|
|
340 |
lockdep_assert_held(&vector_lock); |
|
341 |
|
|
342 |
if (!vector) |
|
343 |
return; |
|
344 |
|
|
345 |
trace_vector_clear(irqd->irq, vector, apicd->cpu, apicd->prev_vector, |
|
346 |
apicd->prev_cpu); |
|
347 |
|
|
348 |
per_cpu(vector_irq, apicd->cpu)[vector] = VECTOR_SHUTDOWN; |
|
349 |
irq_matrix_free(vector_matrix, apicd->cpu, vector, managed); |
|
350 |
apicd->vector = 0; |
|
351 |
|
|
352 |
/* Clean up move in progress */ |
|
353 |
vector = apicd->prev_vector; |
|
354 |
if (!vector) |
|
355 |
return; |
|
356 |
|
|
357 |
per_cpu(vector_irq, apicd->prev_cpu)[vector] = VECTOR_SHUTDOWN; |
|
358 |
irq_matrix_free(vector_matrix, apicd->prev_cpu, vector, managed); |
|
359 |
apicd->prev_vector = 0; |
|
360 |
apicd->move_in_progress = 0; |
|
361 |
hlist_del_init(&apicd->clist); |
|
362 |
} |
|
363 |
|
|
364 |
static void x86_vector_deactivate(struct irq_domain *dom, struct irq_data *irqd) |
|
365 |
{ |
|
366 |
struct apic_chip_data *apicd = apic_chip_data(irqd); |
|
367 |
unsigned long flags; |
|
368 |
|
|
369 |
trace_vector_deactivate(irqd->irq, apicd->is_managed, |
|
370 |
apicd->can_reserve, false); |
|
371 |
|
|
372 |
/* Regular fixed assigned interrupt */ |
|
373 |
if (!apicd->is_managed && !apicd->can_reserve) |
|
374 |
return; |
|
375 |
/* If the interrupt has a global reservation, nothing to do */ |
|
376 |
if (apicd->has_reserved) |
|
377 |
return; |
|
378 |
|
|
379 |
raw_spin_lock_irqsave(&vector_lock, flags); |
|
380 |
clear_irq_vector(irqd); |
|
381 |
if (apicd->can_reserve) |
|
382 |
reserve_irq_vector_locked(irqd); |
|
383 |
else |
|
384 |
vector_assign_managed_shutdown(irqd); |
|
385 |
raw_spin_unlock_irqrestore(&vector_lock, flags); |
|
386 |
} |
|
387 |
|
|
388 |
static int activate_reserved(struct irq_data *irqd) |
|
389 |
{ |
|
390 |
struct apic_chip_data *apicd = apic_chip_data(irqd); |
|
391 |
int ret; |
|
392 |
|
|
393 |
ret = assign_irq_vector_any_locked(irqd); |
|
394 |
if (!ret) { |
|
395 |
apicd->has_reserved = false; |
|
396 |
/* |
|
397 |
* Core might have disabled reservation mode after |
|
398 |
* allocating the irq descriptor. Ideally this should |
|
399 |
* happen before allocation time, but that would require |
|
400 |
* completely convoluted ways of transporting that |
|
401 |
* information. |
|
402 |
*/ |
|
403 |
if (!irqd_can_reserve(irqd)) |
|
404 |
apicd->can_reserve = false; |
|
405 |
} |
|
406 |
|
|
407 |
/* |
|
408 |
* Check to ensure that the effective affinity mask is a subset |
|
409 |
* the user supplied affinity mask, and warn the user if it is not |
|
410 |
*/ |
|
411 |
if (!cpumask_subset(irq_data_get_effective_affinity_mask(irqd), |
|
412 |
irq_data_get_affinity_mask(irqd))) { |
|
413 |
pr_warn("irq %u: Affinity broken due to vector space exhaustion.\n", |
|
414 |
irqd->irq); |
|
415 |
} |
|
416 |
|
|
417 |
return ret; |
|
418 |
} |
|
419 |
|
|
420 |
static int activate_managed(struct irq_data *irqd) |
|
421 |
{ |
|
422 |
const struct cpumask *dest = irq_data_get_affinity_mask(irqd); |
|
423 |
int ret; |
|
424 |
|
|
425 |
cpumask_and(vector_searchmask, dest, cpu_online_mask); |
|
426 |
if (WARN_ON_ONCE(cpumask_empty(vector_searchmask))) { |
|
427 |
/* Something in the core code broke! Survive gracefully */ |
|
428 |
pr_err("Managed startup for irq %u, but no CPU\n", irqd->irq); |
|
429 |
return -EINVAL; |
|
430 |
} |
|
431 |
|
|
432 |
ret = assign_managed_vector(irqd, vector_searchmask); |
|
433 |
/* |
|
434 |
* This should not happen. The vector reservation got buggered. Handle |
|
435 |
* it gracefully. |
|
436 |
*/ |
|
437 |
if (WARN_ON_ONCE(ret < 0)) { |
|
438 |
pr_err("Managed startup irq %u, no vector available\n", |
|
439 |
irqd->irq); |
|
440 |
} |
|
441 |
return ret; |
|
442 |
} |
|
443 |
|
|
444 |
static int x86_vector_activate(struct irq_domain *dom, struct irq_data *irqd, |
|
445 |
bool reserve) |
|
446 |
{ |
|
447 |
struct apic_chip_data *apicd = apic_chip_data(irqd); |
|
448 |
unsigned long flags; |
|
449 |
int ret = 0; |
|
450 |
|
|
451 |
trace_vector_activate(irqd->irq, apicd->is_managed, |
|
452 |
apicd->can_reserve, reserve); |
|
453 |
|
|
454 |
raw_spin_lock_irqsave(&vector_lock, flags); |
|
455 |
if (!apicd->can_reserve && !apicd->is_managed) |
|
456 |
assign_irq_vector_any_locked(irqd); |
|
457 |
else if (reserve || irqd_is_managed_and_shutdown(irqd)) |
|
458 |
vector_assign_managed_shutdown(irqd); |
|
459 |
else if (apicd->is_managed) |
|
460 |
ret = activate_managed(irqd); |
|
461 |
else if (apicd->has_reserved) |
|
462 |
ret = activate_reserved(irqd); |
|
463 |
raw_spin_unlock_irqrestore(&vector_lock, flags); |
|
464 |
return ret; |
|
465 |
} |
|
466 |
|
|
467 |
static void vector_free_reserved_and_managed(struct irq_data *irqd) |
|
468 |
{ |
|
469 |
const struct cpumask *dest = irq_data_get_affinity_mask(irqd); |
|
470 |
struct apic_chip_data *apicd = apic_chip_data(irqd); |
|
471 |
|
|
472 |
trace_vector_teardown(irqd->irq, apicd->is_managed, |
|
473 |
apicd->has_reserved); |
|
474 |
|
|
475 |
if (apicd->has_reserved) |
|
476 |
irq_matrix_remove_reserved(vector_matrix); |
|
477 |
if (apicd->is_managed) |
|
478 |
irq_matrix_remove_managed(vector_matrix, dest); |
|
479 |
} |
|
480 |
|
|
481 |
static void x86_vector_free_irqs(struct irq_domain *domain, |
|
482 |
unsigned int virq, unsigned int nr_irqs) |
|
483 |
{ |
|
484 |
struct apic_chip_data *apicd; |
|
485 |
struct irq_data *irqd; |
|
486 |
unsigned long flags; |
|
487 |
int i; |
|
488 |
|
|
489 |
for (i = 0; i < nr_irqs; i++) { |
|
490 |
irqd = irq_domain_get_irq_data(x86_vector_domain, virq + i); |
|
491 |
if (irqd && irqd->chip_data) { |
|
492 |
raw_spin_lock_irqsave(&vector_lock, flags); |
|
493 |
clear_irq_vector(irqd); |
|
494 |
vector_free_reserved_and_managed(irqd); |
|
495 |
apicd = irqd->chip_data; |
|
496 |
irq_domain_reset_irq_data(irqd); |
|
497 |
raw_spin_unlock_irqrestore(&vector_lock, flags); |
|
498 |
free_apic_chip_data(apicd); |
|
499 |
} |
|
500 |
} |
|
501 |
} |
|
502 |
|
|
503 |
static bool vector_configure_legacy(unsigned int virq, struct irq_data *irqd, |
|
504 |
struct apic_chip_data *apicd) |
|
505 |
{ |
|
506 |
unsigned long flags; |
|
507 |
bool realloc = false; |
|
508 |
|
|
509 |
apicd->vector = ISA_IRQ_VECTOR(virq); |
|
510 |
apicd->cpu = 0; |
|
511 |
|
|
512 |
raw_spin_lock_irqsave(&vector_lock, flags); |
|
513 |
/* |
|
514 |
* If the interrupt is activated, then it must stay at this vector |
|
515 |
* position. That's usually the timer interrupt (0). |
|
516 |
*/ |
|
517 |
if (irqd_is_activated(irqd)) { |
|
518 |
trace_vector_setup(virq, true, 0); |
|
519 |
apic_update_irq_cfg(irqd, apicd->vector, apicd->cpu); |
|
520 |
} else { |
|
521 |
/* Release the vector */ |
|
522 |
apicd->can_reserve = true; |
|
523 |
irqd_set_can_reserve(irqd); |
|
524 |
clear_irq_vector(irqd); |
|
525 |
realloc = true; |
|
526 |
} |
|
527 |
raw_spin_unlock_irqrestore(&vector_lock, flags); |
|
528 |
return realloc; |
|
529 |
} |
|
530 |
|
|
531 |
static int x86_vector_alloc_irqs(struct irq_domain *domain, unsigned int virq, |
|
532 |
unsigned int nr_irqs, void *arg) |
|
533 |
{ |
|
534 |
struct irq_alloc_info *info = arg; |
|
535 |
struct apic_chip_data *apicd; |
|
536 |
struct irq_data *irqd; |
|
537 |
int i, err, node; |
|
538 |
|
|
539 |
if (disable_apic) |
|
540 |
return -ENXIO; |
|
541 |
|
|
542 |
/* Currently vector allocator can't guarantee contiguous allocations */ |
|
543 |
if ((info->flags & X86_IRQ_ALLOC_CONTIGUOUS_VECTORS) && nr_irqs > 1) |
|
544 |
return -ENOSYS; |
|
545 |
|
|
546 |
for (i = 0; i < nr_irqs; i++) { |
|
547 |
irqd = irq_domain_get_irq_data(domain, virq + i); |
|
548 |
BUG_ON(!irqd); |
|
549 |
node = irq_data_get_node(irqd); |
|
550 |
WARN_ON_ONCE(irqd->chip_data); |
|
551 |
apicd = alloc_apic_chip_data(node); |
|
552 |
if (!apicd) { |
|
553 |
err = -ENOMEM; |
|
554 |
goto error; |
|
555 |
} |
|
556 |
|
|
557 |
apicd->irq = virq + i; |
|
558 |
irqd->chip = &lapic_controller; |
|
559 |
irqd->chip_data = apicd; |
|
560 |
irqd->hwirq = virq + i; |
|
561 |
irqd_set_single_target(irqd); |
|
562 |
/* |
|
563 |
* Prevent that any of these interrupts is invoked in |
|
564 |
* non interrupt context via e.g. generic_handle_irq() |
|
565 |
* as that can corrupt the affinity move state. |
|
566 |
*/ |
|
567 |
irqd_set_handle_enforce_irqctx(irqd); |
|
568 |
|
|
569 |
/* Don't invoke affinity setter on deactivated interrupts */ |
|
570 |
irqd_set_affinity_on_activate(irqd); |
|
571 |
|
|
572 |
/* |
|
573 |
* Legacy vectors are already assigned when the IOAPIC |
|
574 |
* takes them over. They stay on the same vector. This is |
|
575 |
* required for check_timer() to work correctly as it might |
|
576 |
* switch back to legacy mode. Only update the hardware |
|
577 |
* config. |
|
578 |
*/ |
|
579 |
if (info->flags & X86_IRQ_ALLOC_LEGACY) { |
|
580 |
if (!vector_configure_legacy(virq + i, irqd, apicd)) |
|
581 |
continue; |
|
582 |
} |
|
583 |
|
|
584 |
err = assign_irq_vector_policy(irqd, info); |
|
585 |
trace_vector_setup(virq + i, false, err); |
|
586 |
if (err) { |
|
587 |
irqd->chip_data = NULL; |
|
588 |
free_apic_chip_data(apicd); |
|
589 |
goto error; |
|
590 |
} |
|
591 |
} |
|
592 |
|
|
593 |
return 0; |
|
594 |
|
|
595 |
error: |
|
596 |
x86_vector_free_irqs(domain, virq, i); |
|
597 |
return err; |
|
598 |
} |
|
599 |
|
|
600 |
#ifdef CONFIG_GENERIC_IRQ_DEBUGFS |
|
601 |
static void x86_vector_debug_show(struct seq_file *m, struct irq_domain *d, |
|
602 |
struct irq_data *irqd, int ind) |
|
603 |
{ |
|
604 |
struct apic_chip_data apicd; |
|
605 |
unsigned long flags; |
|
606 |
int irq; |
|
607 |
|
|
608 |
if (!irqd) { |
|
609 |
irq_matrix_debug_show(m, vector_matrix, ind); |
|
610 |
return; |
|
611 |
} |
|
612 |
|
|
613 |
irq = irqd->irq; |
|
614 |
if (irq < nr_legacy_irqs() && !test_bit(irq, &io_apic_irqs)) { |
|
615 |
seq_printf(m, "%*sVector: %5d\n", ind, "", ISA_IRQ_VECTOR(irq)); |
|
616 |
seq_printf(m, "%*sTarget: Legacy PIC all CPUs\n", ind, ""); |
|
617 |
return; |
|
618 |
} |
|
619 |
|
|
620 |
if (!irqd->chip_data) { |
|
621 |
seq_printf(m, "%*sVector: Not assigned\n", ind, ""); |
|
622 |
return; |
|
623 |
} |
|
624 |
|
|
625 |
raw_spin_lock_irqsave(&vector_lock, flags); |
|
626 |
memcpy(&apicd, irqd->chip_data, sizeof(apicd)); |
|
627 |
raw_spin_unlock_irqrestore(&vector_lock, flags); |
|
628 |
|
|
629 |
seq_printf(m, "%*sVector: %5u\n", ind, "", apicd.vector); |
|
630 |
seq_printf(m, "%*sTarget: %5u\n", ind, "", apicd.cpu); |
|
631 |
if (apicd.prev_vector) { |
|
632 |
seq_printf(m, "%*sPrevious vector: %5u\n", ind, "", apicd.prev_vector); |
|
633 |
seq_printf(m, "%*sPrevious target: %5u\n", ind, "", apicd.prev_cpu); |
|
634 |
} |
|
635 |
seq_printf(m, "%*smove_in_progress: %u\n", ind, "", apicd.move_in_progress ? 1 : 0); |
|
636 |
seq_printf(m, "%*sis_managed: %u\n", ind, "", apicd.is_managed ? 1 : 0); |
|
637 |
seq_printf(m, "%*scan_reserve: %u\n", ind, "", apicd.can_reserve ? 1 : 0); |
|
638 |
seq_printf(m, "%*shas_reserved: %u\n", ind, "", apicd.has_reserved ? 1 : 0); |
|
639 |
seq_printf(m, "%*scleanup_pending: %u\n", ind, "", !hlist_unhashed(&apicd.clist)); |
|
640 |
} |
|
641 |
#endif |
|
642 |
|
|
643 |
static const struct irq_domain_ops x86_vector_domain_ops = { |
|
644 |
.alloc = x86_vector_alloc_irqs, |
|
645 |
.free = x86_vector_free_irqs, |
|
646 |
.activate = x86_vector_activate, |
|
647 |
.deactivate = x86_vector_deactivate, |
|
648 |
#ifdef CONFIG_GENERIC_IRQ_DEBUGFS |
|
649 |
.debug_show = x86_vector_debug_show, |
|
650 |
#endif |
|
651 |
}; |
|
652 |
|
|
653 |
int __init arch_probe_nr_irqs(void) |
|
654 |
{ |
|
655 |
int nr; |
|
656 |
|
|
657 |
if (nr_irqs > (NR_VECTORS * nr_cpu_ids)) |
|
658 |
nr_irqs = NR_VECTORS * nr_cpu_ids; |
|
659 |
|
|
660 |
nr = (gsi_top + nr_legacy_irqs()) + 8 * nr_cpu_ids; |
|
661 |
#if defined(CONFIG_PCI_MSI) |
|
662 |
/* |
|
663 |
* for MSI and HT dyn irq |
|
664 |
*/ |
|
665 |
if (gsi_top <= NR_IRQS_LEGACY) |
|
666 |
nr += 8 * nr_cpu_ids; |
|
667 |
else |
|
668 |
nr += gsi_top * 16; |
|
669 |
#endif |
|
670 |
if (nr < nr_irqs) |
|
671 |
nr_irqs = nr; |
|
672 |
|
|
673 |
/* |
|
674 |
* We don't know if PIC is present at this point so we need to do |
|
675 |
* probe() to get the right number of legacy IRQs. |
|
676 |
*/ |
|
677 |
return legacy_pic->probe(); |
|
678 |
} |
|
679 |
|
|
680 |
void lapic_assign_legacy_vector(unsigned int irq, bool replace) |
|
681 |
{ |
|
682 |
/* |
|
683 |
* Use assign system here so it wont get accounted as allocated |
|
684 |
* and moveable in the cpu hotplug check and it prevents managed |
|
685 |
* irq reservation from touching it. |
|
686 |
*/ |
|
687 |
irq_matrix_assign_system(vector_matrix, ISA_IRQ_VECTOR(irq), replace); |
|
688 |
} |
|
689 |
|
|
690 |
void __init lapic_update_legacy_vectors(void) |
|
691 |
{ |
|
692 |
unsigned int i; |
|
693 |
|
|
694 |
if (IS_ENABLED(CONFIG_X86_IO_APIC) && nr_ioapics > 0) |
|
695 |
return; |
|
696 |
|
|
697 |
/* |
|
698 |
* If the IO/APIC is disabled via config, kernel command line or |
|
699 |
* lack of enumeration then all legacy interrupts are routed |
|
700 |
* through the PIC. Make sure that they are marked as legacy |
|
701 |
* vectors. PIC_CASCADE_IRQ has already been marked in |
|
702 |
* lapic_assign_system_vectors(). |
|
703 |
*/ |
|
704 |
for (i = 0; i < nr_legacy_irqs(); i++) { |
|
705 |
if (i != PIC_CASCADE_IR) |
|
706 |
lapic_assign_legacy_vector(i, true); |
|
707 |
} |
|
708 |
} |
|
709 |
|
|
710 |
void __init lapic_assign_system_vectors(void) |
|
711 |
{ |
|
712 |
unsigned int i, vector = 0; |
|
713 |
|
|
714 |
for_each_set_bit_from(vector, system_vectors, NR_VECTORS) |
|
715 |
irq_matrix_assign_system(vector_matrix, vector, false); |
|
716 |
|
|
717 |
if (nr_legacy_irqs() > 1) |
|
718 |
lapic_assign_legacy_vector(PIC_CASCADE_IR, false); |
|
719 |
|
|
720 |
/* System vectors are reserved, online it */ |
|
721 |
irq_matrix_online(vector_matrix); |
|
722 |
|
|
723 |
/* Mark the preallocated legacy interrupts */ |
|
724 |
for (i = 0; i < nr_legacy_irqs(); i++) { |
|
725 |
if (i != PIC_CASCADE_IR) |
|
726 |
irq_matrix_assign(vector_matrix, ISA_IRQ_VECTOR(i)); |
|
727 |
} |
|
728 |
} |
|
729 |
|
|
730 |
int __init arch_early_irq_init(void) |
|
731 |
{ |
|
732 |
struct fwnode_handle *fn; |
|
733 |
|
|
734 |
fn = irq_domain_alloc_named_fwnode("VECTOR"); |
|
735 |
BUG_ON(!fn); |
|
736 |
x86_vector_domain = irq_domain_create_tree(fn, &x86_vector_domain_ops, |
|
737 |
NULL); |
|
738 |
BUG_ON(x86_vector_domain == NULL); |
|
739 |
irq_set_default_host(x86_vector_domain); |
|
740 |
|
|
741 |
BUG_ON(!alloc_cpumask_var(&vector_searchmask, GFP_KERNEL)); |
|
742 |
|
|
743 |
/* |
|
744 |
* Allocate the vector matrix allocator data structure and limit the |
|
745 |
* search area. |
|
746 |
*/ |
|
747 |
vector_matrix = irq_alloc_matrix(NR_VECTORS, FIRST_EXTERNAL_VECTOR, |
|
748 |
FIRST_SYSTEM_VECTOR); |
|
749 |
BUG_ON(!vector_matrix); |
|
750 |
|
|
751 |
return arch_early_ioapic_init(); |
|
752 |
} |
|
753 |
|
|
754 |
#ifdef CONFIG_SMP |
|
755 |
|
|
756 |
static struct irq_desc *__setup_vector_irq(int vector) |
|
757 |
{ |
|
758 |
int isairq = vector - ISA_IRQ_VECTOR(0); |
|
759 |
|
2f529f
|
760 |
/* Copy the cleanup vector if irqs are pipelined. */ |
H |
761 |
if (IS_ENABLED(CONFIG_IRQ_PIPELINE) && |
|
762 |
vector == IRQ_MOVE_CLEANUP_VECTOR) |
|
763 |
return irq_to_desc(IRQ_MOVE_CLEANUP_VECTOR); /* 1:1 mapping */ |
a07526
|
764 |
/* Check whether the irq is in the legacy space */ |
H |
765 |
if (isairq < 0 || isairq >= nr_legacy_irqs()) |
|
766 |
return VECTOR_UNUSED; |
|
767 |
/* Check whether the irq is handled by the IOAPIC */ |
|
768 |
if (test_bit(isairq, &io_apic_irqs)) |
|
769 |
return VECTOR_UNUSED; |
|
770 |
return irq_to_desc(isairq); |
|
771 |
} |
|
772 |
|
|
773 |
/* Online the local APIC infrastructure and initialize the vectors */ |
|
774 |
void lapic_online(void) |
|
775 |
{ |
|
776 |
unsigned int vector; |
|
777 |
|
|
778 |
lockdep_assert_held(&vector_lock); |
|
779 |
|
|
780 |
/* Online the vector matrix array for this CPU */ |
|
781 |
irq_matrix_online(vector_matrix); |
|
782 |
|
|
783 |
/* |
|
784 |
* The interrupt affinity logic never targets interrupts to offline |
|
785 |
* CPUs. The exception are the legacy PIC interrupts. In general |
|
786 |
* they are only targeted to CPU0, but depending on the platform |
|
787 |
* they can be distributed to any online CPU in hardware. The |
|
788 |
* kernel has no influence on that. So all active legacy vectors |
|
789 |
* must be installed on all CPUs. All non legacy interrupts can be |
|
790 |
* cleared. |
|
791 |
*/ |
|
792 |
for (vector = 0; vector < NR_VECTORS; vector++) |
|
793 |
this_cpu_write(vector_irq[vector], __setup_vector_irq(vector)); |
|
794 |
} |
|
795 |
|
|
796 |
void lapic_offline(void) |
|
797 |
{ |
2f529f
|
798 |
unsigned long flags; |
H |
799 |
|
|
800 |
raw_spin_lock_irqsave(&vector_lock, flags); |
a07526
|
801 |
irq_matrix_offline(vector_matrix); |
2f529f
|
802 |
raw_spin_unlock_irqrestore(&vector_lock, flags); |
a07526
|
803 |
} |
H |
804 |
|
|
805 |
static int apic_set_affinity(struct irq_data *irqd, |
|
806 |
const struct cpumask *dest, bool force) |
|
807 |
{ |
|
808 |
int err; |
2f529f
|
809 |
|
H |
810 |
WARN_ON_ONCE(irqs_pipelined() && !hard_irqs_disabled()); |
a07526
|
811 |
|
H |
812 |
if (WARN_ON_ONCE(!irqd_is_activated(irqd))) |
|
813 |
return -EIO; |
|
814 |
|
|
815 |
raw_spin_lock(&vector_lock); |
|
816 |
cpumask_and(vector_searchmask, dest, cpu_online_mask); |
|
817 |
if (irqd_affinity_is_managed(irqd)) |
|
818 |
err = assign_managed_vector(irqd, vector_searchmask); |
|
819 |
else |
|
820 |
err = assign_vector_locked(irqd, vector_searchmask); |
|
821 |
raw_spin_unlock(&vector_lock); |
|
822 |
return err ? err : IRQ_SET_MASK_OK; |
|
823 |
} |
|
824 |
|
|
825 |
#else |
|
826 |
# define apic_set_affinity NULL |
|
827 |
#endif |
|
828 |
|
|
829 |
static int apic_retrigger_irq(struct irq_data *irqd) |
|
830 |
{ |
|
831 |
struct apic_chip_data *apicd = apic_chip_data(irqd); |
|
832 |
unsigned long flags; |
|
833 |
|
|
834 |
raw_spin_lock_irqsave(&vector_lock, flags); |
|
835 |
apic->send_IPI(apicd->cpu, apicd->vector); |
|
836 |
raw_spin_unlock_irqrestore(&vector_lock, flags); |
|
837 |
|
|
838 |
return 1; |
|
839 |
} |
|
840 |
|
2f529f
|
841 |
#if defined(CONFIG_IRQ_PIPELINE) && \ |
H |
842 |
defined(CONFIG_GENERIC_PENDING_IRQ) |
|
843 |
|
|
844 |
static void apic_deferred_irq_move(struct irq_work *work) |
|
845 |
{ |
|
846 |
struct irq_data *irqd; |
|
847 |
struct irq_desc *desc; |
|
848 |
unsigned long flags; |
|
849 |
|
|
850 |
irqd = container_of(work, struct irq_data, move_work); |
|
851 |
desc = irq_data_to_desc(irqd); |
|
852 |
raw_spin_lock_irqsave(&desc->lock, flags); |
|
853 |
__irq_move_irq(irqd); |
|
854 |
raw_spin_unlock_irqrestore(&desc->lock, flags); |
|
855 |
} |
|
856 |
|
|
857 |
static inline void apic_move_irq(struct irq_data *irqd) |
|
858 |
{ |
|
859 |
if (irqd_is_setaffinity_pending(irqd) && |
|
860 |
!irqd_is_setaffinity_blocked(irqd)) { |
|
861 |
init_irq_work(&irqd->move_work, apic_deferred_irq_move); |
|
862 |
irq_work_queue(&irqd->move_work); |
|
863 |
} |
|
864 |
} |
|
865 |
|
|
866 |
#else |
|
867 |
|
|
868 |
static inline void apic_move_irq(struct irq_data *irqd) |
a07526
|
869 |
{ |
H |
870 |
irq_move_irq(irqd); |
2f529f
|
871 |
} |
H |
872 |
|
|
873 |
#endif |
|
874 |
|
|
875 |
void apic_ack_irq(struct irq_data *irqd) |
|
876 |
{ |
|
877 |
apic_move_irq(irqd); |
|
878 |
__ack_APIC_irq(); |
a07526
|
879 |
} |
H |
880 |
|
|
881 |
void apic_ack_edge(struct irq_data *irqd) |
|
882 |
{ |
|
883 |
irq_complete_move(irqd_cfg(irqd)); |
|
884 |
apic_ack_irq(irqd); |
|
885 |
} |
|
886 |
|
|
887 |
static struct irq_chip lapic_controller = { |
|
888 |
.name = "APIC", |
|
889 |
.irq_ack = apic_ack_edge, |
|
890 |
.irq_set_affinity = apic_set_affinity, |
|
891 |
.irq_compose_msi_msg = x86_vector_msi_compose_msg, |
|
892 |
.irq_retrigger = apic_retrigger_irq, |
|
893 |
}; |
|
894 |
|
|
895 |
#ifdef CONFIG_SMP |
|
896 |
|
|
897 |
static void free_moved_vector(struct apic_chip_data *apicd) |
|
898 |
{ |
|
899 |
unsigned int vector = apicd->prev_vector; |
|
900 |
unsigned int cpu = apicd->prev_cpu; |
|
901 |
bool managed = apicd->is_managed; |
|
902 |
|
|
903 |
/* |
|
904 |
* Managed interrupts are usually not migrated away |
|
905 |
* from an online CPU, but CPU isolation 'managed_irq' |
|
906 |
* can make that happen. |
|
907 |
* 1) Activation does not take the isolation into account |
|
908 |
* to keep the code simple |
|
909 |
* 2) Migration away from an isolated CPU can happen when |
|
910 |
* a non-isolated CPU which is in the calculated |
|
911 |
* affinity mask comes online. |
|
912 |
*/ |
|
913 |
trace_vector_free_moved(apicd->irq, cpu, vector, managed); |
|
914 |
irq_matrix_free(vector_matrix, cpu, vector, managed); |
|
915 |
per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED; |
|
916 |
hlist_del_init(&apicd->clist); |
|
917 |
apicd->prev_vector = 0; |
|
918 |
apicd->move_in_progress = 0; |
|
919 |
} |
|
920 |
|
2f529f
|
921 |
DEFINE_IDTENTRY_SYSVEC_PIPELINED(IRQ_MOVE_CLEANUP_VECTOR, |
H |
922 |
sysvec_irq_move_cleanup) |
a07526
|
923 |
{ |
H |
924 |
struct hlist_head *clhead = this_cpu_ptr(&cleanup_list); |
|
925 |
struct apic_chip_data *apicd; |
|
926 |
struct hlist_node *tmp; |
2f529f
|
927 |
unsigned long flags; |
a07526
|
928 |
|
H |
929 |
ack_APIC_irq(); |
|
930 |
/* Prevent vectors vanishing under us */ |
2f529f
|
931 |
raw_spin_lock_irqsave(&vector_lock, flags); |
a07526
|
932 |
|
H |
933 |
hlist_for_each_entry_safe(apicd, tmp, clhead, clist) { |
|
934 |
unsigned int irr, vector = apicd->prev_vector; |
|
935 |
|
|
936 |
/* |
|
937 |
* Paranoia: Check if the vector that needs to be cleaned |
|
938 |
* up is registered at the APICs IRR. If so, then this is |
|
939 |
* not the best time to clean it up. Clean it up in the |
|
940 |
* next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR |
|
941 |
* to this CPU. IRQ_MOVE_CLEANUP_VECTOR is the lowest |
|
942 |
* priority external vector, so on return from this |
|
943 |
* interrupt the device interrupt will happen first. |
|
944 |
*/ |
|
945 |
irr = apic_read(APIC_IRR + (vector / 32 * 0x10)); |
|
946 |
if (irr & (1U << (vector % 32))) { |
|
947 |
apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR); |
|
948 |
continue; |
|
949 |
} |
|
950 |
free_moved_vector(apicd); |
|
951 |
} |
|
952 |
|
2f529f
|
953 |
raw_spin_unlock_irqrestore(&vector_lock, flags); |
a07526
|
954 |
} |
H |
955 |
|
|
956 |
static void __send_cleanup_vector(struct apic_chip_data *apicd) |
|
957 |
{ |
2f529f
|
958 |
unsigned long flags; |
a07526
|
959 |
unsigned int cpu; |
H |
960 |
|
2f529f
|
961 |
raw_spin_lock_irqsave(&vector_lock, flags); |
a07526
|
962 |
apicd->move_in_progress = 0; |
H |
963 |
cpu = apicd->prev_cpu; |
|
964 |
if (cpu_online(cpu)) { |
|
965 |
hlist_add_head(&apicd->clist, per_cpu_ptr(&cleanup_list, cpu)); |
|
966 |
apic->send_IPI(cpu, IRQ_MOVE_CLEANUP_VECTOR); |
|
967 |
} else { |
|
968 |
apicd->prev_vector = 0; |
|
969 |
} |
2f529f
|
970 |
raw_spin_unlock_irqrestore(&vector_lock, flags); |
a07526
|
971 |
} |
H |
972 |
|
|
973 |
void send_cleanup_vector(struct irq_cfg *cfg) |
|
974 |
{ |
|
975 |
struct apic_chip_data *apicd; |
|
976 |
|
|
977 |
apicd = container_of(cfg, struct apic_chip_data, hw_irq_cfg); |
|
978 |
if (apicd->move_in_progress) |
|
979 |
__send_cleanup_vector(apicd); |
|
980 |
} |
|
981 |
|
|
982 |
void irq_complete_move(struct irq_cfg *cfg) |
|
983 |
{ |
|
984 |
struct apic_chip_data *apicd; |
|
985 |
|
|
986 |
apicd = container_of(cfg, struct apic_chip_data, hw_irq_cfg); |
|
987 |
if (likely(!apicd->move_in_progress)) |
|
988 |
return; |
|
989 |
|
|
990 |
/* |
|
991 |
* If the interrupt arrived on the new target CPU, cleanup the |
|
992 |
* vector on the old target CPU. A vector check is not required |
|
993 |
* because an interrupt can never move from one vector to another |
|
994 |
* on the same CPU. |
|
995 |
*/ |
|
996 |
if (apicd->cpu == smp_processor_id()) |
|
997 |
__send_cleanup_vector(apicd); |
|
998 |
} |
|
999 |
|
|
1000 |
/* |
|
1001 |
* Called from fixup_irqs() with @desc->lock held and interrupts disabled. |
|
1002 |
*/ |
|
1003 |
void irq_force_complete_move(struct irq_desc *desc) |
|
1004 |
{ |
|
1005 |
struct apic_chip_data *apicd; |
|
1006 |
struct irq_data *irqd; |
|
1007 |
unsigned int vector; |
2f529f
|
1008 |
|
H |
1009 |
WARN_ON_ONCE(irqs_pipelined() && !hard_irqs_disabled()); |
a07526
|
1010 |
|
H |
1011 |
/* |
|
1012 |
* The function is called for all descriptors regardless of which |
|
1013 |
* irqdomain they belong to. For example if an IRQ is provided by |
|
1014 |
* an irq_chip as part of a GPIO driver, the chip data for that |
|
1015 |
* descriptor is specific to the irq_chip in question. |
|
1016 |
* |
|
1017 |
* Check first that the chip_data is what we expect |
|
1018 |
* (apic_chip_data) before touching it any further. |
|
1019 |
*/ |
|
1020 |
irqd = irq_domain_get_irq_data(x86_vector_domain, |
|
1021 |
irq_desc_get_irq(desc)); |
|
1022 |
if (!irqd) |
|
1023 |
return; |
|
1024 |
|
|
1025 |
raw_spin_lock(&vector_lock); |
|
1026 |
apicd = apic_chip_data(irqd); |
|
1027 |
if (!apicd) |
|
1028 |
goto unlock; |
|
1029 |
|
|
1030 |
/* |
|
1031 |
* If prev_vector is empty, no action required. |
|
1032 |
*/ |
|
1033 |
vector = apicd->prev_vector; |
|
1034 |
if (!vector) |
|
1035 |
goto unlock; |
|
1036 |
|
|
1037 |
/* |
|
1038 |
* This is tricky. If the cleanup of the old vector has not been |
|
1039 |
* done yet, then the following setaffinity call will fail with |
|
1040 |
* -EBUSY. This can leave the interrupt in a stale state. |
|
1041 |
* |
|
1042 |
* All CPUs are stuck in stop machine with interrupts disabled so |
|
1043 |
* calling __irq_complete_move() would be completely pointless. |
|
1044 |
* |
|
1045 |
* 1) The interrupt is in move_in_progress state. That means that we |
|
1046 |
* have not seen an interrupt since the io_apic was reprogrammed to |
|
1047 |
* the new vector. |
|
1048 |
* |
|
1049 |
* 2) The interrupt has fired on the new vector, but the cleanup IPIs |
|
1050 |
* have not been processed yet. |
|
1051 |
*/ |
|
1052 |
if (apicd->move_in_progress) { |
|
1053 |
/* |
|
1054 |
* In theory there is a race: |
|
1055 |
* |
|
1056 |
* set_ioapic(new_vector) <-- Interrupt is raised before update |
|
1057 |
* is effective, i.e. it's raised on |
|
1058 |
* the old vector. |
|
1059 |
* |
|
1060 |
* So if the target cpu cannot handle that interrupt before |
|
1061 |
* the old vector is cleaned up, we get a spurious interrupt |
|
1062 |
* and in the worst case the ioapic irq line becomes stale. |
|
1063 |
* |
|
1064 |
* But in case of cpu hotplug this should be a non issue |
|
1065 |
* because if the affinity update happens right before all |
|
1066 |
* cpus rendevouz in stop machine, there is no way that the |
|
1067 |
* interrupt can be blocked on the target cpu because all cpus |
|
1068 |
* loops first with interrupts enabled in stop machine, so the |
|
1069 |
* old vector is not yet cleaned up when the interrupt fires. |
|
1070 |
* |
|
1071 |
* So the only way to run into this issue is if the delivery |
|
1072 |
* of the interrupt on the apic/system bus would be delayed |
|
1073 |
* beyond the point where the target cpu disables interrupts |
|
1074 |
* in stop machine. I doubt that it can happen, but at least |
|
1075 |
* there is a theroretical chance. Virtualization might be |
|
1076 |
* able to expose this, but AFAICT the IOAPIC emulation is not |
|
1077 |
* as stupid as the real hardware. |
|
1078 |
* |
|
1079 |
* Anyway, there is nothing we can do about that at this point |
|
1080 |
* w/o refactoring the whole fixup_irq() business completely. |
|
1081 |
* We print at least the irq number and the old vector number, |
|
1082 |
* so we have the necessary information when a problem in that |
|
1083 |
* area arises. |
|
1084 |
*/ |
|
1085 |
pr_warn("IRQ fixup: irq %d move in progress, old vector %d\n", |
|
1086 |
irqd->irq, vector); |
|
1087 |
} |
|
1088 |
free_moved_vector(apicd); |
|
1089 |
unlock: |
|
1090 |
raw_spin_unlock(&vector_lock); |
|
1091 |
} |
|
1092 |
|
|
1093 |
#ifdef CONFIG_HOTPLUG_CPU |
|
1094 |
/* |
|
1095 |
* Note, this is not accurate accounting, but at least good enough to |
|
1096 |
* prevent that the actual interrupt move will run out of vectors. |
|
1097 |
*/ |
|
1098 |
int lapic_can_unplug_cpu(void) |
|
1099 |
{ |
|
1100 |
unsigned int rsvd, avl, tomove, cpu = smp_processor_id(); |
2f529f
|
1101 |
unsigned long flags; |
a07526
|
1102 |
int ret = 0; |
H |
1103 |
|
2f529f
|
1104 |
raw_spin_lock_irqsave(&vector_lock, flags); |
a07526
|
1105 |
tomove = irq_matrix_allocated(vector_matrix); |
H |
1106 |
avl = irq_matrix_available(vector_matrix, true); |
|
1107 |
if (avl < tomove) { |
|
1108 |
pr_warn("CPU %u has %u vectors, %u available. Cannot disable CPU\n", |
|
1109 |
cpu, tomove, avl); |
|
1110 |
ret = -ENOSPC; |
|
1111 |
goto out; |
|
1112 |
} |
|
1113 |
rsvd = irq_matrix_reserved(vector_matrix); |
|
1114 |
if (avl < rsvd) { |
|
1115 |
pr_warn("Reserved vectors %u > available %u. IRQ request may fail\n", |
|
1116 |
rsvd, avl); |
|
1117 |
} |
|
1118 |
out: |
2f529f
|
1119 |
raw_spin_unlock_irqrestore(&vector_lock, flags); |
a07526
|
1120 |
return ret; |
H |
1121 |
} |
|
1122 |
#endif /* HOTPLUG_CPU */ |
|
1123 |
#endif /* SMP */ |
|
1124 |
|
|
1125 |
static void __init print_APIC_field(int base) |
|
1126 |
{ |
|
1127 |
int i; |
|
1128 |
|
|
1129 |
printk(KERN_DEBUG); |
|
1130 |
|
|
1131 |
for (i = 0; i < 8; i++) |
|
1132 |
pr_cont("%08x", apic_read(base + i*0x10)); |
|
1133 |
|
|
1134 |
pr_cont("\n"); |
|
1135 |
} |
|
1136 |
|
|
1137 |
static void __init print_local_APIC(void *dummy) |
|
1138 |
{ |
|
1139 |
unsigned int i, v, ver, maxlvt; |
|
1140 |
u64 icr; |
|
1141 |
|
|
1142 |
pr_debug("printing local APIC contents on CPU#%d/%d:\n", |
|
1143 |
smp_processor_id(), hard_smp_processor_id()); |
|
1144 |
v = apic_read(APIC_ID); |
|
1145 |
pr_info("... APIC ID: %08x (%01x)\n", v, read_apic_id()); |
|
1146 |
v = apic_read(APIC_LVR); |
|
1147 |
pr_info("... APIC VERSION: %08x\n", v); |
|
1148 |
ver = GET_APIC_VERSION(v); |
|
1149 |
maxlvt = lapic_get_maxlvt(); |
|
1150 |
|
|
1151 |
v = apic_read(APIC_TASKPRI); |
|
1152 |
pr_debug("... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK); |
|
1153 |
|
|
1154 |
/* !82489DX */ |
|
1155 |
if (APIC_INTEGRATED(ver)) { |
|
1156 |
if (!APIC_XAPIC(ver)) { |
|
1157 |
v = apic_read(APIC_ARBPRI); |
|
1158 |
pr_debug("... APIC ARBPRI: %08x (%02x)\n", |
|
1159 |
v, v & APIC_ARBPRI_MASK); |
|
1160 |
} |
|
1161 |
v = apic_read(APIC_PROCPRI); |
|
1162 |
pr_debug("... APIC PROCPRI: %08x\n", v); |
|
1163 |
} |
|
1164 |
|
|
1165 |
/* |
|
1166 |
* Remote read supported only in the 82489DX and local APIC for |
|
1167 |
* Pentium processors. |
|
1168 |
*/ |
|
1169 |
if (!APIC_INTEGRATED(ver) || maxlvt == 3) { |
|
1170 |
v = apic_read(APIC_RRR); |
|
1171 |
pr_debug("... APIC RRR: %08x\n", v); |
|
1172 |
} |
|
1173 |
|
|
1174 |
v = apic_read(APIC_LDR); |
|
1175 |
pr_debug("... APIC LDR: %08x\n", v); |
|
1176 |
if (!x2apic_enabled()) { |
|
1177 |
v = apic_read(APIC_DFR); |
|
1178 |
pr_debug("... APIC DFR: %08x\n", v); |
|
1179 |
} |
|
1180 |
v = apic_read(APIC_SPIV); |
|
1181 |
pr_debug("... APIC SPIV: %08x\n", v); |
|
1182 |
|
|
1183 |
pr_debug("... APIC ISR field:\n"); |
|
1184 |
print_APIC_field(APIC_ISR); |
|
1185 |
pr_debug("... APIC TMR field:\n"); |
|
1186 |
print_APIC_field(APIC_TMR); |
|
1187 |
pr_debug("... APIC IRR field:\n"); |
|
1188 |
print_APIC_field(APIC_IRR); |
|
1189 |
|
|
1190 |
/* !82489DX */ |
|
1191 |
if (APIC_INTEGRATED(ver)) { |
|
1192 |
/* Due to the Pentium erratum 3AP. */ |
|
1193 |
if (maxlvt > 3) |
|
1194 |
apic_write(APIC_ESR, 0); |
|
1195 |
|
|
1196 |
v = apic_read(APIC_ESR); |
|
1197 |
pr_debug("... APIC ESR: %08x\n", v); |
|
1198 |
} |
|
1199 |
|
|
1200 |
icr = apic_icr_read(); |
|
1201 |
pr_debug("... APIC ICR: %08x\n", (u32)icr); |
|
1202 |
pr_debug("... APIC ICR2: %08x\n", (u32)(icr >> 32)); |
|
1203 |
|
|
1204 |
v = apic_read(APIC_LVTT); |
|
1205 |
pr_debug("... APIC LVTT: %08x\n", v); |
|
1206 |
|
|
1207 |
if (maxlvt > 3) { |
|
1208 |
/* PC is LVT#4. */ |
|
1209 |
v = apic_read(APIC_LVTPC); |
|
1210 |
pr_debug("... APIC LVTPC: %08x\n", v); |
|
1211 |
} |
|
1212 |
v = apic_read(APIC_LVT0); |
|
1213 |
pr_debug("... APIC LVT0: %08x\n", v); |
|
1214 |
v = apic_read(APIC_LVT1); |
|
1215 |
pr_debug("... APIC LVT1: %08x\n", v); |
|
1216 |
|
|
1217 |
if (maxlvt > 2) { |
|
1218 |
/* ERR is LVT#3. */ |
|
1219 |
v = apic_read(APIC_LVTERR); |
|
1220 |
pr_debug("... APIC LVTERR: %08x\n", v); |
|
1221 |
} |
|
1222 |
|
|
1223 |
v = apic_read(APIC_TMICT); |
|
1224 |
pr_debug("... APIC TMICT: %08x\n", v); |
|
1225 |
v = apic_read(APIC_TMCCT); |
|
1226 |
pr_debug("... APIC TMCCT: %08x\n", v); |
|
1227 |
v = apic_read(APIC_TDCR); |
|
1228 |
pr_debug("... APIC TDCR: %08x\n", v); |
|
1229 |
|
|
1230 |
if (boot_cpu_has(X86_FEATURE_EXTAPIC)) { |
|
1231 |
v = apic_read(APIC_EFEAT); |
|
1232 |
maxlvt = (v >> 16) & 0xff; |
|
1233 |
pr_debug("... APIC EFEAT: %08x\n", v); |
|
1234 |
v = apic_read(APIC_ECTRL); |
|
1235 |
pr_debug("... APIC ECTRL: %08x\n", v); |
|
1236 |
for (i = 0; i < maxlvt; i++) { |
|
1237 |
v = apic_read(APIC_EILVTn(i)); |
|
1238 |
pr_debug("... APIC EILVT%d: %08x\n", i, v); |
|
1239 |
} |
|
1240 |
} |
|
1241 |
pr_cont("\n"); |
|
1242 |
} |
|
1243 |
|
|
1244 |
static void __init print_local_APICs(int maxcpu) |
|
1245 |
{ |
|
1246 |
int cpu; |
|
1247 |
|
|
1248 |
if (!maxcpu) |
|
1249 |
return; |
|
1250 |
|
|
1251 |
preempt_disable(); |
|
1252 |
for_each_online_cpu(cpu) { |
|
1253 |
if (cpu >= maxcpu) |
|
1254 |
break; |
|
1255 |
smp_call_function_single(cpu, print_local_APIC, NULL, 1); |
|
1256 |
} |
|
1257 |
preempt_enable(); |
|
1258 |
} |
|
1259 |
|
|
1260 |
static void __init print_PIC(void) |
|
1261 |
{ |
|
1262 |
unsigned int v; |
|
1263 |
unsigned long flags; |
|
1264 |
|
|
1265 |
if (!nr_legacy_irqs()) |
|
1266 |
return; |
|
1267 |
|
|
1268 |
pr_debug("\nprinting PIC contents\n"); |
|
1269 |
|
|
1270 |
raw_spin_lock_irqsave(&i8259A_lock, flags); |
|
1271 |
|
|
1272 |
v = inb(0xa1) << 8 | inb(0x21); |
|
1273 |
pr_debug("... PIC IMR: %04x\n", v); |
|
1274 |
|
|
1275 |
v = inb(0xa0) << 8 | inb(0x20); |
|
1276 |
pr_debug("... PIC IRR: %04x\n", v); |
|
1277 |
|
|
1278 |
outb(0x0b, 0xa0); |
|
1279 |
outb(0x0b, 0x20); |
|
1280 |
v = inb(0xa0) << 8 | inb(0x20); |
|
1281 |
outb(0x0a, 0xa0); |
|
1282 |
outb(0x0a, 0x20); |
|
1283 |
|
|
1284 |
raw_spin_unlock_irqrestore(&i8259A_lock, flags); |
|
1285 |
|
|
1286 |
pr_debug("... PIC ISR: %04x\n", v); |
|
1287 |
|
|
1288 |
v = inb(0x4d1) << 8 | inb(0x4d0); |
|
1289 |
pr_debug("... PIC ELCR: %04x\n", v); |
|
1290 |
} |
|
1291 |
|
|
1292 |
static int show_lapic __initdata = 1; |
|
1293 |
static __init int setup_show_lapic(char *arg) |
|
1294 |
{ |
|
1295 |
int num = -1; |
|
1296 |
|
|
1297 |
if (strcmp(arg, "all") == 0) { |
|
1298 |
show_lapic = CONFIG_NR_CPUS; |
|
1299 |
} else { |
|
1300 |
get_option(&arg, &num); |
|
1301 |
if (num >= 0) |
|
1302 |
show_lapic = num; |
|
1303 |
} |
|
1304 |
|
|
1305 |
return 1; |
|
1306 |
} |
|
1307 |
__setup("show_lapic=", setup_show_lapic); |
|
1308 |
|
|
1309 |
static int __init print_ICs(void) |
|
1310 |
{ |
|
1311 |
if (apic_verbosity == APIC_QUIET) |
|
1312 |
return 0; |
|
1313 |
|
|
1314 |
print_PIC(); |
|
1315 |
|
|
1316 |
/* don't print out if apic is not there */ |
|
1317 |
if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config()) |
|
1318 |
return 0; |
|
1319 |
|
|
1320 |
print_local_APICs(show_lapic); |
|
1321 |
print_IO_APICs(); |
|
1322 |
|
|
1323 |
return 0; |
|
1324 |
} |
|
1325 |
|
|
1326 |
late_initcall(print_ICs); |