xen_intr.c revision 1.21 1 /* $NetBSD: xen_intr.c,v 1.21 2020/04/06 19:26:00 jdolecek Exp $ */
2
3 /*-
4 * Copyright (c) 1998, 2001 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Charles M. Hannum, and by Jason R. Thorpe.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32 #include <sys/cdefs.h>
33 __KERNEL_RCSID(0, "$NetBSD: xen_intr.c,v 1.21 2020/04/06 19:26:00 jdolecek Exp $");
34
35 #include "opt_multiprocessor.h"
36
37 #include <sys/param.h>
38 #include <sys/kernel.h>
39 #include <sys/kmem.h>
40 #include <sys/cpu.h>
41 #include <sys/device.h>
42
43 #include <xen/evtchn.h>
44 #include <xen/xenfunc.h>
45
46 #include <uvm/uvm.h>
47
48 #include <machine/cpu.h>
49 #include <machine/intr.h>
50
51 #include "acpica.h"
52 #include "ioapic.h"
53 #include "lapic.h"
54 #include "pci.h"
55
56 #if NACPICA > 0
57 #include <dev/acpi/acpivar.h>
58 #endif
59
60 #if NIOAPIC > 0 || NACPICA > 0
61 #include <machine/i82093var.h>
62 #endif
63
64 #if NLAPIC > 0
65 #include <machine/i82489var.h>
66 #endif
67
68 #if NPCI > 0
69 #include <dev/pci/ppbreg.h>
70 #endif
71
72 #if defined(MULTIPROCESSOR)
73 static const char *xen_ipi_names[XEN_NIPIS] = XEN_IPI_NAMES;
74 #endif
75
76 /*
77 * Restore a value to cpl (unmasking interrupts). If any unmasked
78 * interrupts are pending, call Xspllower() to process them.
79 */
80 void xen_spllower(int nlevel);
81
82 void
83 xen_spllower(int nlevel)
84 {
85 struct cpu_info *ci = curcpu();
86 uint32_t xmask;
87 u_long psl;
88
89 if (ci->ci_ilevel <= nlevel)
90 return;
91
92 __insn_barrier();
93
94 xmask = XUNMASK(ci, nlevel);
95 psl = xen_read_psl();
96 x86_disable_intr();
97 if (ci->ci_xpending & xmask) {
98 KASSERT(psl == 0);
99 Xspllower(nlevel);
100 /* Xspllower does enable_intr() */
101 } else {
102 ci->ci_ilevel = nlevel;
103 xen_write_psl(psl);
104 }
105 }
106
107
108 #if !defined(XENPVHVM)
109 void
110 x86_disable_intr(void)
111 {
112 curcpu()->ci_vcpu->evtchn_upcall_mask = 1;
113 x86_lfence();
114 }
115
116 void
117 x86_enable_intr(void)
118 {
119 volatile struct vcpu_info *_vci = curcpu()->ci_vcpu;
120 __insn_barrier();
121 _vci->evtchn_upcall_mask = 0;
122 x86_lfence(); /* unmask then check (avoid races) */
123 if (__predict_false(_vci->evtchn_upcall_pending))
124 hypervisor_force_callback();
125 }
126
127 #endif /* !XENPVHVM */
128
129 u_long
130 xen_read_psl(void)
131 {
132
133 return (curcpu()->ci_vcpu->evtchn_upcall_mask);
134 }
135
136 void
137 xen_write_psl(u_long psl)
138 {
139 struct cpu_info *ci = curcpu();
140
141 ci->ci_vcpu->evtchn_upcall_mask = psl;
142 xen_rmb();
143 if (ci->ci_vcpu->evtchn_upcall_pending && psl == 0) {
144 hypervisor_force_callback();
145 }
146 }
147
148 void *
149 xen_intr_establish(int legacy_irq, struct pic *pic, int pin,
150 int type, int level, int (*handler)(void *), void *arg,
151 bool known_mpsafe)
152 {
153
154 return xen_intr_establish_xname(legacy_irq, pic, pin, type, level,
155 handler, arg, known_mpsafe, "XEN");
156 }
157
158 void *
159 xen_intr_establish_xname(int legacy_irq, struct pic *pic, int pin,
160 int type, int level, int (*handler)(void *), void *arg,
161 bool known_mpsafe, const char *xname)
162 {
163 const char *intrstr;
164 char intrstr_buf[INTRIDBUF];
165
166 if (pic->pic_type == PIC_XEN) {
167 struct intrhand *rih;
168
169 intrstr = intr_create_intrid(legacy_irq, pic, pin, intrstr_buf,
170 sizeof(intrstr_buf));
171
172 event_set_handler(pin, handler, arg, level, intrstr, xname,
173 known_mpsafe);
174
175 rih = kmem_zalloc(sizeof(*rih), cold ? KM_NOSLEEP : KM_SLEEP);
176 if (rih == NULL) {
177 printf("%s: can't allocate handler info\n", __func__);
178 return NULL;
179 }
180
181 /*
182 * XXX:
183 * This is just a copy for API conformance.
184 * The real ih is lost in the innards of
185 * event_set_handler(); where the details of
186 * biglock_wrapper etc are taken care of.
187 * All that goes away when we nuke event_set_handler()
188 * et. al. and unify with x86/intr.c
189 */
190 rih->ih_pin = pin; /* port */
191 rih->ih_fun = rih->ih_realfun = handler;
192 rih->ih_arg = rih->ih_realarg = arg;
193 rih->pic_type = pic->pic_type;
194 return rih;
195 } /* Else we assume pintr */
196
197 #if (NPCI > 0 || NISA > 0) && defined(XENPV) /* XXX: support PVHVM pirq */
198 struct pintrhand *pih;
199 int gsi;
200 int vector, evtchn;
201
202 KASSERTMSG(legacy_irq == -1 || (0 <= legacy_irq && legacy_irq < NUM_XEN_IRQS),
203 "bad legacy IRQ value: %d", legacy_irq);
204 KASSERTMSG(!(legacy_irq == -1 && pic == &i8259_pic),
205 "non-legacy IRQon i8259 ");
206
207 gsi = xen_pic_to_gsi(pic, pin);
208
209 intrstr = intr_create_intrid(gsi, pic, pin, intrstr_buf,
210 sizeof(intrstr_buf));
211
212 vector = xen_vec_alloc(gsi);
213
214 if (irq2port[gsi] == 0) {
215 extern struct cpu_info phycpu_info_primary; /* XXX */
216 struct cpu_info *ci = &phycpu_info_primary;
217
218 pic->pic_addroute(pic, ci, pin, vector, type);
219
220 evtchn = bind_pirq_to_evtch(gsi);
221 KASSERT(evtchn > 0);
222 KASSERT(evtchn < NR_EVENT_CHANNELS);
223 irq2port[gsi] = evtchn + 1;
224 xen_atomic_set_bit(&ci->ci_evtmask[0], evtchn);
225 } else {
226 /*
227 * Shared interrupt - we can't rebind.
228 * The port is shared instead.
229 */
230 evtchn = irq2port[gsi] - 1;
231 }
232
233 pih = pirq_establish(gsi, evtchn, handler, arg, level,
234 intrstr, xname, known_mpsafe);
235 pih->pic_type = pic->pic_type;
236 return pih;
237 #endif /* NPCI > 0 || NISA > 0 */
238
239 /* FALLTHROUGH */
240 return NULL;
241 }
242
243 /*
244 * Mask an interrupt source.
245 */
246 void
247 xen_intr_mask(struct intrhand *ih)
248 {
249 /* XXX */
250 panic("xen_intr_mask: not yet implemented.");
251 }
252
253 /*
254 * Unmask an interrupt source.
255 */
256 void
257 xen_intr_unmask(struct intrhand *ih)
258 {
259 /* XXX */
260 panic("xen_intr_unmask: not yet implemented.");
261 }
262
263 /*
264 * Deregister an interrupt handler.
265 */
266 void
267 xen_intr_disestablish(struct intrhand *ih)
268 {
269
270 if (ih->pic_type == PIC_XEN) {
271 event_remove_handler(ih->ih_pin, ih->ih_realfun,
272 ih->ih_realarg);
273 kmem_free(ih, sizeof(*ih));
274 return;
275 }
276 #if defined(DOM0OPS)
277 /*
278 * Cache state, to prevent a use after free situation with
279 * ih.
280 */
281
282 struct pintrhand *pih = (struct pintrhand *)ih;
283
284 int pirq = pih->pirq;
285 int port = pih->evtch;
286 KASSERT(irq2port[pirq] != 0);
287
288 pirq_disestablish(pih);
289
290 if (evtsource[port] == NULL) {
291 /*
292 * Last handler was removed by
293 * event_remove_handler().
294 *
295 * We can safely unbind the pirq now.
296 */
297
298 port = unbind_pirq_from_evtch(pirq);
299 KASSERT(port == pih->evtch);
300 irq2port[pirq] = 0;
301 }
302 #endif
303 return;
304 }
305
306 /* MI interface for kern_cpu.c */
307 void xen_cpu_intr_redistribute(void);
308
309 void
310 xen_cpu_intr_redistribute(void)
311 {
312 KASSERT(mutex_owned(&cpu_lock));
313 KASSERT(mp_online);
314
315 return;
316 }
317
318 /* MD - called by x86/cpu.c */
319 #if defined(INTRSTACKSIZE)
320 static inline bool
321 redzone_const_or_false(bool x)
322 {
323 #ifdef DIAGNOSTIC
324 return x;
325 #else
326 return false;
327 #endif /* !DIAGNOSTIC */
328 }
329
330 static inline int
331 redzone_const_or_zero(int x)
332 {
333 return redzone_const_or_false(true) ? x : 0;
334 }
335 #endif
336
337 void xen_cpu_intr_init(struct cpu_info *);
338 void
339 xen_cpu_intr_init(struct cpu_info *ci)
340 {
341 int i; /* XXX: duplicate */
342
343 ci->ci_xunmask[0] = 0xfffffffe;
344 for (i = 1; i < NIPL; i++)
345 ci->ci_xunmask[i] = ci->ci_xunmask[i - 1] & ~(1 << i);
346
347 #if defined(INTRSTACKSIZE)
348 vaddr_t istack;
349
350 /*
351 * If the red zone is activated, protect both the top and
352 * the bottom of the stack with an unmapped page.
353 */
354 istack = uvm_km_alloc(kernel_map,
355 INTRSTACKSIZE + redzone_const_or_zero(2 * PAGE_SIZE), 0,
356 UVM_KMF_WIRED|UVM_KMF_ZERO);
357 if (redzone_const_or_false(true)) {
358 pmap_kremove(istack, PAGE_SIZE);
359 pmap_kremove(istack + INTRSTACKSIZE + PAGE_SIZE, PAGE_SIZE);
360 pmap_update(pmap_kernel());
361 }
362
363 /*
364 * 33 used to be 1. Arbitrarily reserve 32 more register_t's
365 * of space for ddb(4) to examine some subroutine arguments
366 * and to hunt for the next stack frame.
367 */
368 ci->ci_intrstack = (char *)istack + redzone_const_or_zero(PAGE_SIZE) +
369 INTRSTACKSIZE - 33 * sizeof(register_t);
370 #endif
371
372 #ifdef MULTIPROCESSOR
373 for (i = 0; i < XEN_NIPIS; i++)
374 evcnt_attach_dynamic(&ci->ci_ipi_events[i], EVCNT_TYPE_MISC,
375 NULL, device_xname(ci->ci_dev), xen_ipi_names[i]);
376 #endif
377
378 ci->ci_idepth = -1;
379 }
380
381 /*
382 * Everything below from here is duplicated from x86/intr.c
383 * When intr.c and xen_intr.c are unified, these will need to be
384 * merged.
385 */
386
387 u_int xen_cpu_intr_count(struct cpu_info *ci);
388
389 u_int
390 xen_cpu_intr_count(struct cpu_info *ci)
391 {
392
393 KASSERT(ci->ci_nintrhand >= 0);
394
395 return ci->ci_nintrhand;
396 }
397
398 static const char *
399 xen_intr_string(int port, char *buf, size_t len, struct pic *pic)
400 {
401 KASSERT(pic->pic_type == PIC_XEN);
402
403 KASSERT(port >= 0);
404 KASSERT(port < NR_EVENT_CHANNELS);
405
406 snprintf(buf, len, "%s channel %d", pic->pic_name, port);
407
408 return buf;
409 }
410
411 static const char *
412 legacy_intr_string(int ih, char *buf, size_t len, struct pic *pic)
413 {
414 int legacy_irq;
415
416 KASSERT(pic->pic_type == PIC_I8259);
417 #if NLAPIC > 0
418 KASSERT(APIC_IRQ_ISLEGACY(ih));
419
420 legacy_irq = APIC_IRQ_LEGACY_IRQ(ih);
421 #else
422 legacy_irq = ih;
423 #endif
424 KASSERT(legacy_irq >= 0 && legacy_irq < 16);
425
426 snprintf(buf, len, "%s pin %d", pic->pic_name, legacy_irq);
427
428 return buf;
429 }
430
431 const char * xintr_string(intr_handle_t ih, char *buf, size_t len);
432
433 const char *
434 xintr_string(intr_handle_t ih, char *buf, size_t len)
435 {
436 #if NIOAPIC > 0
437 struct ioapic_softc *pic;
438 #endif
439
440 if (ih == 0)
441 panic("%s: bogus handle 0x%" PRIx64, __func__, ih);
442
443 #if NIOAPIC > 0
444 if (ih & APIC_INT_VIA_APIC) {
445 pic = ioapic_find(APIC_IRQ_APIC(ih));
446 if (pic != NULL) {
447 snprintf(buf, len, "%s pin %d",
448 device_xname(pic->sc_dev), APIC_IRQ_PIN(ih));
449 } else {
450 snprintf(buf, len,
451 "apic %d int %d (irq %d)",
452 APIC_IRQ_APIC(ih),
453 APIC_IRQ_PIN(ih),
454 APIC_IRQ_LEGACY_IRQ(ih));
455 }
456 } else
457 snprintf(buf, len, "irq %d", APIC_IRQ_LEGACY_IRQ(ih));
458
459 #elif NLAPIC > 0
460 snprintf(buf, len, "irq %d", APIC_IRQ_LEGACY_IRQ(ih));
461 #else
462 snprintf(buf, len, "irq %d", (int) ih);
463 #endif
464 return buf;
465
466 }
467
468 /*
469 * Create an interrupt id such as "ioapic0 pin 9". This interrupt id is used
470 * by MI code and intrctl(8).
471 */
472 const char * xen_intr_create_intrid(int legacy_irq, struct pic *pic,
473 int pin, char *buf, size_t len);
474
475 const char *
476 xen_intr_create_intrid(int legacy_irq, struct pic *pic, int pin, char *buf, size_t len)
477 {
478 int ih = 0;
479
480 #if NPCI > 0
481 #if defined(__HAVE_PCI_MSI_MSIX)
482 if ((pic->pic_type == PIC_MSI) || (pic->pic_type == PIC_MSIX)) {
483 uint64_t pih;
484 int dev, vec;
485
486 dev = msipic_get_devid(pic);
487 vec = pin;
488 pih = __SHIFTIN((uint64_t)dev, MSI_INT_DEV_MASK)
489 | __SHIFTIN((uint64_t)vec, MSI_INT_VEC_MASK)
490 | APIC_INT_VIA_MSI;
491 if (pic->pic_type == PIC_MSI)
492 MSI_INT_MAKE_MSI(pih);
493 else if (pic->pic_type == PIC_MSIX)
494 MSI_INT_MAKE_MSIX(pih);
495
496 return x86_pci_msi_string(NULL, pih, buf, len);
497 }
498 #endif /* __HAVE_PCI_MSI_MSIX */
499 #endif
500
501 if (pic->pic_type == PIC_XEN) {
502 ih = pin; /* Port == pin */
503 return xen_intr_string(pin, buf, len, pic);
504 }
505
506 /*
507 * If the device is pci, "legacy_irq" is alway -1. Least 8 bit of "ih"
508 * is only used in intr_string() to show the irq number.
509 * If the device is "legacy"(such as floppy), it should not use
510 * intr_string().
511 */
512 if (pic->pic_type == PIC_I8259) {
513 ih = legacy_irq;
514 return legacy_intr_string(ih, buf, len, pic);
515 }
516
517 #if NIOAPIC > 0 || NACPICA > 0
518 ih = ((pic->pic_apicid << APIC_INT_APIC_SHIFT) & APIC_INT_APIC_MASK)
519 | ((pin << APIC_INT_PIN_SHIFT) & APIC_INT_PIN_MASK);
520 if (pic->pic_type == PIC_IOAPIC) {
521 ih |= APIC_INT_VIA_APIC;
522 }
523 ih |= pin;
524 return intr_string(ih, buf, len);
525 #endif
526
527 return NULL; /* No pic found! */
528 }
529
530 #if !defined(XENPVHVM)
531 __strong_alias(spllower, xen_spllower);
532 __strong_alias(x86_read_psl, xen_read_psl);
533 __strong_alias(x86_write_psl, xen_write_psl);
534
535 __strong_alias(intr_string, xintr_string);
536 __strong_alias(intr_create_intrid, xen_intr_create_intrid);
537 __strong_alias(intr_establish, xen_intr_establish);
538 __strong_alias(intr_establish_xname, xen_intr_establish_xname);
539 __strong_alias(intr_mask, xen_intr_mask);
540 __strong_alias(intr_unmask, xen_intr_unmask);
541 __strong_alias(intr_disestablish, xen_intr_disestablish);
542 __strong_alias(cpu_intr_redistribute, xen_cpu_intr_redistribute);
543 __strong_alias(cpu_intr_count, xen_cpu_intr_count);
544 __strong_alias(cpu_intr_init, xen_cpu_intr_init);
545 #endif /* !XENPVHVM */
546