xref: /netbsd-src/sys/arch/xen/x86/hypervisor_machdep.c (revision a5847cc334d9a7029f6352b847e9e8d71a0f9e0c)
1 /*	$NetBSD: hypervisor_machdep.c,v 1.16 2011/09/20 00:12:24 jym Exp $	*/
2 
3 /*
4  *
5  * Copyright (c) 2004 Christian Limpach.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /******************************************************************************
30  * hypervisor.c
31  *
32  * Communication to/from hypervisor.
33  *
34  * Copyright (c) 2002-2004, K A Fraser
35  *
36  * Permission is hereby granted, free of charge, to any person obtaining a copy
37  * of this software and associated documentation files (the "Software"), to
38  * deal in the Software without restriction, including without limitation the
39  * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
40  * sell copies of the Software, and to permit persons to whom the Software is
41  * furnished to do so, subject to the following conditions:
42  *
43  * The above copyright notice and this permission notice shall be included in
44  * all copies or substantial portions of the Software.
45  *
46  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
47  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
48  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
49  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
50  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
51  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
52  * DEALINGS IN THE SOFTWARE.
53  */
54 
55 
56 #include <sys/cdefs.h>
57 __KERNEL_RCSID(0, "$NetBSD: hypervisor_machdep.c,v 1.16 2011/09/20 00:12:24 jym Exp $");
58 
59 #include <sys/param.h>
60 #include <sys/systm.h>
61 #include <sys/kmem.h>
62 
63 #include <uvm/uvm_extern.h>
64 
65 #include <machine/vmparam.h>
66 #include <machine/pmap.h>
67 
68 #include <xen/xen.h>
69 #include <xen/hypervisor.h>
70 #include <xen/evtchn.h>
71 #include <xen/xenpmap.h>
72 
73 #include "opt_xen.h"
74 
75 /*
76  * arch-dependent p2m frame lists list (L3 and L2)
77  * used by Xen for save/restore mappings
78  */
79 static unsigned long * l3_p2m_page;
80 static unsigned long * l2_p2m_page;
81 static int l2_p2m_page_size; /* size of L2 page, in pages */
82 
83 static void build_p2m_frame_list_list(void);
84 static void update_p2m_frame_list_list(void);
85 
86 // #define PORT_DEBUG 4
87 // #define EARLY_DEBUG_EVENT
88 
89 /* callback function type */
90 typedef void (*iterate_func_t)(struct cpu_info *, unsigned int,
91 			       unsigned int, unsigned int, void *);
92 
93 static inline void
94 evt_iterate_bits(struct cpu_info *ci, volatile unsigned long *pendingl1,
95 		 volatile unsigned long *pendingl2,
96 		 volatile unsigned long *mask,
97 		 iterate_func_t iterate_pending, void *iterate_args)
98 {
99 
100 	KASSERT(pendingl1 != NULL);
101 	KASSERT(pendingl2 != NULL);
102 
103 	unsigned long l1, l2;
104 	unsigned int l1i, l2i, port;
105 
106 	l1 = xen_atomic_xchg(pendingl1, 0);
107 	while ((l1i = xen_ffs(l1)) != 0) {
108 		l1i--;
109 		l1 &= ~(1UL << l1i);
110 
111 		l2 = pendingl2[l1i] & (mask != NULL ? ~mask[l1i] : -1UL);
112 
113 		if (mask != NULL) xen_atomic_setbits_l(&mask[l1i], l2);
114 		xen_atomic_clearbits_l(&pendingl2[l1i], l2);
115 
116 		while ((l2i = xen_ffs(l2)) != 0) {
117 			l2i--;
118 			l2 &= ~(1UL << l2i);
119 
120 			port = (l1i << LONG_SHIFT) + l2i;
121 
122 			iterate_pending(ci, port, l1i, l2i, iterate_args);
123 		}
124 	}
125 }
126 
127 /*
128  * Set per-cpu "pending" information for outstanding events that
129  * cannot be processed now.
130  */
131 
132 static inline void
133 evt_set_pending(struct cpu_info *ci, unsigned int port, unsigned int l1i,
134 		unsigned int l2i, void *args)
135 {
136 
137 	KASSERT(args != NULL);
138 	KASSERT(ci != NULL);
139 
140 	int *ret = args;
141 
142 	if (evtsource[port]) {
143 		hypervisor_set_ipending(ci, evtsource[port]->ev_imask,
144 		    l1i, l2i);
145 		evtsource[port]->ev_evcnt.ev_count++;
146 		if (*ret == 0 && ci->ci_ilevel <
147 		    evtsource[port]->ev_maxlevel)
148 			*ret = 1;
149 	}
150 #ifdef DOM0OPS
151 	else  {
152 		/* set pending event */
153 		xenevt_setipending(l1i, l2i);
154 	}
155 #endif
156 }
157 
158 int stipending(void);
159 int
160 stipending(void)
161 {
162 	volatile shared_info_t *s = HYPERVISOR_shared_info;
163 	struct cpu_info *ci;
164 	volatile struct vcpu_info *vci;
165 	int ret;
166 
167 	ret = 0;
168 	ci = curcpu();
169 	vci = ci->ci_vcpu;
170 
171 #if 0
172 	if (HYPERVISOR_shared_info->events)
173 		printf("stipending events %08lx mask %08lx ilevel %d\n",
174 		    HYPERVISOR_shared_info->events,
175 		    HYPERVISOR_shared_info->events_mask, ci->ci_ilevel);
176 #endif
177 
178 #ifdef EARLY_DEBUG_EVENT
179 	if (xen_atomic_test_bit(&s->evtchn_pending[0], debug_port)) {
180 		xen_debug_handler(NULL);
181 		xen_atomic_clear_bit(&s->evtchn_pending[0], debug_port);
182 	}
183 #endif
184 
185 	/*
186 	 * we're only called after STIC, so we know that we'll have to
187 	 * STI at the end
188 	 */
189 
190 	while (vci->evtchn_upcall_pending) {
191 		cli();
192 
193 		vci->evtchn_upcall_pending = 0;
194 
195 		evt_iterate_bits(ci, &vci->evtchn_pending_sel,
196 		    s->evtchn_pending, s->evtchn_mask,
197 		    evt_set_pending, &ret);
198 
199 		sti();
200 	}
201 
202 #if 0
203 	if (ci->ci_ipending & 0x1)
204 		printf("stipending events %08lx mask %08lx ilevel %d ipending %08x\n",
205 		    HYPERVISOR_shared_info->events,
206 		    HYPERVISOR_shared_info->events_mask, ci->ci_ilevel,
207 		    ci->ci_ipending);
208 #endif
209 
210 	return (ret);
211 }
212 
213 /* Iterate through pending events and call the event handler */
214 
215 static inline void
216 evt_do_hypervisor_callback(struct cpu_info *ci, unsigned int port,
217 			   unsigned int l1i, unsigned int l2i, void *args)
218 {
219 	KASSERT(args != NULL);
220 	KASSERT(ci == curcpu());
221 
222 	struct intrframe *regs = args;
223 
224 #ifdef PORT_DEBUG
225 	if (port == PORT_DEBUG)
226 		printf("do_hypervisor_callback event %d\n", port);
227 #endif
228 	if (evtsource[port])
229 		call_evtchn_do_event(port, regs);
230 #ifdef DOM0OPS
231 	else  {
232 		if (ci->ci_ilevel < IPL_HIGH) {
233 			/* fast path */
234 			int oipl = ci->ci_ilevel;
235 			ci->ci_ilevel = IPL_HIGH;
236 			call_xenevt_event(port);
237 			ci->ci_ilevel = oipl;
238 		} else {
239 			/* set pending event */
240 			xenevt_setipending(l1i, l2i);
241 		}
242 	}
243 #endif
244 }
245 
246 void
247 do_hypervisor_callback(struct intrframe *regs)
248 {
249 	volatile shared_info_t *s = HYPERVISOR_shared_info;
250 	struct cpu_info *ci;
251 	volatile struct vcpu_info *vci;
252 	int level;
253 
254 	ci = curcpu();
255 	vci = ci->ci_vcpu;
256 	level = ci->ci_ilevel;
257 
258 	// DDD printf("do_hypervisor_callback\n");
259 
260 #ifdef EARLY_DEBUG_EVENT
261 	if (xen_atomic_test_bit(&s->evtchn_pending[0], debug_port)) {
262 		xen_debug_handler(NULL);
263 		xen_atomic_clear_bit(&s->evtchn_pending[0], debug_port);
264 	}
265 #endif
266 
267 	while (vci->evtchn_upcall_pending) {
268 		vci->evtchn_upcall_pending = 0;
269 
270 		evt_iterate_bits(ci, &vci->evtchn_pending_sel,
271 		    s->evtchn_pending, s->evtchn_mask,
272 		    evt_do_hypervisor_callback, regs);
273 	}
274 
275 #ifdef DIAGNOSTIC
276 	if (level != ci->ci_ilevel)
277 		printf("hypervisor done %08x level %d/%d ipending %08x\n",
278 		    (uint)vci->evtchn_pending_sel,
279 		    level, ci->ci_ilevel, ci->ci_ipending);
280 #endif
281 }
282 
283 void
284 hypervisor_unmask_event(unsigned int ev)
285 {
286 	volatile shared_info_t *s = HYPERVISOR_shared_info;
287 	volatile struct vcpu_info *vci = curcpu()->ci_vcpu;
288 
289 #ifdef PORT_DEBUG
290 	if (ev == PORT_DEBUG)
291 		printf("hypervisor_unmask_event %d\n", ev);
292 #endif
293 
294 	xen_atomic_clear_bit(&s->evtchn_mask[0], ev);
295 	/*
296 	 * The following is basically the equivalent of
297 	 * 'hw_resend_irq'. Just like a real IO-APIC we 'lose the
298 	 * interrupt edge' if the channel is masked.
299 	 */
300 	if (xen_atomic_test_bit(&s->evtchn_pending[0], ev) &&
301 	    !xen_atomic_test_and_set_bit(&vci->evtchn_pending_sel, ev>>LONG_SHIFT)) {
302 		xen_atomic_set_bit(&vci->evtchn_upcall_pending, 0);
303 		if (!vci->evtchn_upcall_mask)
304 			hypervisor_force_callback();
305 	}
306 }
307 
308 void
309 hypervisor_mask_event(unsigned int ev)
310 {
311 	volatile shared_info_t *s = HYPERVISOR_shared_info;
312 #ifdef PORT_DEBUG
313 	if (ev == PORT_DEBUG)
314 		printf("hypervisor_mask_event %d\n", ev);
315 #endif
316 
317 	xen_atomic_set_bit(&s->evtchn_mask[0], ev);
318 }
319 
320 void
321 hypervisor_clear_event(unsigned int ev)
322 {
323 	volatile shared_info_t *s = HYPERVISOR_shared_info;
324 #ifdef PORT_DEBUG
325 	if (ev == PORT_DEBUG)
326 		printf("hypervisor_clear_event %d\n", ev);
327 #endif
328 
329 	xen_atomic_clear_bit(&s->evtchn_pending[0], ev);
330 }
331 
332 static inline void
333 evt_enable_event(struct cpu_info *ci, unsigned int port,
334 		 unsigned int l1i, unsigned int l2i, void *args)
335 {
336 	KASSERT(ci != NULL);
337 	KASSERT(args == NULL);
338 	hypervisor_enable_event(port);
339 }
340 
341 void
342 hypervisor_enable_ipl(unsigned int ipl)
343 {
344 	struct cpu_info *ci = curcpu();
345 
346 	/*
347 	 * enable all events for ipl. As we only set an event in ipl_evt_mask
348 	 * for its lowest IPL, and pending IPLs are processed high to low,
349 	 * we know that all callback for this event have been processed.
350 	 */
351 
352 	evt_iterate_bits(ci, &ci->ci_isources[ipl]->ipl_evt_mask1,
353 	    ci->ci_isources[ipl]->ipl_evt_mask2, NULL,
354 	    evt_enable_event, NULL);
355 
356 }
357 
358 void
359 hypervisor_set_ipending(struct cpu_info *ci, uint32_t iplmask, int l1, int l2)
360 {
361 	int ipl;
362 
363 	/* set pending bit for the appropriate IPLs */
364 	ci->ci_ipending |= iplmask;
365 
366 	/*
367 	 * And set event pending bit for the lowest IPL. As IPL are handled
368 	 * from high to low, this ensure that all callbacks will have been
369 	 * called when we ack the event
370 	 */
371 	ipl = ffs(iplmask);
372 	KASSERT(ipl > 0);
373 	ipl--;
374 	KASSERT(ipl < NIPL);
375 	KASSERT(ci->ci_isources[ipl] != NULL);
376 	ci->ci_isources[ipl]->ipl_evt_mask1 |= 1UL << l1;
377 	ci->ci_isources[ipl]->ipl_evt_mask2[l1] |= 1UL << l2;
378 }
379 
380 void
381 hypervisor_machdep_attach(void)
382 {
383  	/* dom0 does not require the arch-dependent P2M translation table */
384 	if (!xendomain_is_dom0()) {
385 		build_p2m_frame_list_list();
386 		sysctl_xen_suspend_setup();
387 	}
388 }
389 
390 void
391 hypervisor_machdep_resume(void)
392 {
393 	/* dom0 does not require the arch-dependent P2M translation table */
394 	if (!xendomain_is_dom0())
395 		update_p2m_frame_list_list();
396 }
397 
398 /*
399  * Generate the p2m_frame_list_list table,
400  * needed for guest save/restore
401  */
402 static void
403 build_p2m_frame_list_list(void)
404 {
405         int fpp; /* number of page (frame) pointer per page */
406         unsigned long max_pfn;
407         /*
408          * The p2m list is composed of three levels of indirection,
409          * each layer containing MFNs pointing to lower level pages
410          * The indirection is used to convert a given PFN to its MFN
411          * Each N level page can point to @fpp (N-1) level pages
412          * For example, for x86 32bit, we have:
413          * - PAGE_SIZE: 4096 bytes
414          * - fpp: 1024 (one L3 page can address 1024 L2 pages)
415          * A L1 page contains the list of MFN we are looking for
416          */
417         max_pfn = xen_start_info.nr_pages;
418         fpp = PAGE_SIZE / sizeof(xen_pfn_t);
419 
420         /* we only need one L3 page */
421         l3_p2m_page = (vaddr_t *)uvm_km_alloc(kernel_map, PAGE_SIZE,
422 	    PAGE_SIZE, UVM_KMF_WIRED | UVM_KMF_NOWAIT);
423         if (l3_p2m_page == NULL)
424                 panic("could not allocate memory for l3_p2m_page");
425 
426         /*
427          * Determine how many L2 pages we need for the mapping
428          * Each L2 can map a total of @fpp L1 pages
429          */
430         l2_p2m_page_size = howmany(max_pfn, fpp);
431 
432         l2_p2m_page = (vaddr_t *)uvm_km_alloc(kernel_map,
433 	    l2_p2m_page_size * PAGE_SIZE,
434 	    PAGE_SIZE, UVM_KMF_WIRED | UVM_KMF_NOWAIT);
435         if (l2_p2m_page == NULL)
436                 panic("could not allocate memory for l2_p2m_page");
437 
438         /* We now have L3 and L2 pages ready, update L1 mapping */
439         update_p2m_frame_list_list();
440 
441 }
442 
443 /*
444  * Update the L1 p2m_frame_list_list mapping (during guest boot or resume)
445  */
446 static void
447 update_p2m_frame_list_list(void)
448 {
449         int i;
450         int fpp; /* number of page (frame) pointer per page */
451         unsigned long max_pfn;
452 
453         max_pfn = xen_start_info.nr_pages;
454         fpp = PAGE_SIZE / sizeof(xen_pfn_t);
455 
456         for (i = 0; i < l2_p2m_page_size; i++) {
457                 /*
458                  * Each time we start a new L2 page,
459                  * store its MFN in the L3 page
460                  */
461                 if ((i % fpp) == 0) {
462                         l3_p2m_page[i/fpp] = vtomfn(
463                                 (vaddr_t)&l2_p2m_page[i]);
464                 }
465                 /*
466                  * we use a shortcut
467                  * since @xpmap_phys_to_machine_mapping array
468                  * already contains PFN to MFN mapping, we just
469                  * set the l2_p2m_page MFN pointer to the MFN of the
470                  * according frame of @xpmap_phys_to_machine_mapping
471                  */
472                 l2_p2m_page[i] = vtomfn((vaddr_t)
473                         &xpmap_phys_to_machine_mapping[i*fpp]);
474         }
475 
476         HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
477                                         vtomfn((vaddr_t)l3_p2m_page);
478         HYPERVISOR_shared_info->arch.max_pfn = max_pfn;
479 
480 }
481