xref: /netbsd-src/sys/arch/xen/x86/hypervisor_machdep.c (revision 10ad5ffa714ce1a679dcc9dd8159648df2d67b5a)
1 /*	$NetBSD: hypervisor_machdep.c,v 1.12 2009/07/29 12:02:08 cegger 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  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *      This product includes software developed by Christian Limpach.
19  * 4. The name of the author may not be used to endorse or promote products
20  *    derived from this software without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 /******************************************************************************
35  * hypervisor.c
36  *
37  * Communication to/from hypervisor.
38  *
39  * Copyright (c) 2002-2004, K A Fraser
40  *
41  * Permission is hereby granted, free of charge, to any person obtaining a copy
42  * of this software and associated documentation files (the "Software"), to
43  * deal in the Software without restriction, including without limitation the
44  * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
45  * sell copies of the Software, and to permit persons to whom the Software is
46  * furnished to do so, subject to the following conditions:
47  *
48  * The above copyright notice and this permission notice shall be included in
49  * all copies or substantial portions of the Software.
50  *
51  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
52  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
53  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
54  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
55  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
56  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
57  * DEALINGS IN THE SOFTWARE.
58  */
59 
60 
61 #include <sys/cdefs.h>
62 __KERNEL_RCSID(0, "$NetBSD: hypervisor_machdep.c,v 1.12 2009/07/29 12:02:08 cegger Exp $");
63 
64 #include <sys/param.h>
65 #include <sys/systm.h>
66 #include <sys/kmem.h>
67 
68 #include <uvm/uvm_extern.h>
69 
70 #include <machine/vmparam.h>
71 #include <machine/pmap.h>
72 
73 #include <xen/xen.h>
74 #include <xen/hypervisor.h>
75 #include <xen/evtchn.h>
76 #include <xen/xenpmap.h>
77 
78 #include "opt_xen.h"
79 
80 /*
81  * arch-dependent p2m frame lists list (L3 and L2)
82  * used by Xen for save/restore mappings
83  */
84 static unsigned long * l3_p2m_page;
85 static unsigned long * l2_p2m_page;
86 static int l2_p2m_page_size; /* size of L2 page, in pages */
87 
88 static void build_p2m_frame_list_list(void);
89 static void update_p2m_frame_list_list(void);
90 
91 // #define PORT_DEBUG 4
92 // #define EARLY_DEBUG_EVENT
93 
94 int stipending(void);
95 int
96 stipending(void)
97 {
98 	unsigned long l1;
99 	unsigned long l2;
100 	unsigned int l1i, l2i, port;
101 	volatile shared_info_t *s = HYPERVISOR_shared_info;
102 	struct cpu_info *ci;
103 	volatile struct vcpu_info *vci;
104 	int ret;
105 
106 	ret = 0;
107 	ci = curcpu();
108 	vci = ci->ci_vcpu;
109 
110 #if 0
111 	if (HYPERVISOR_shared_info->events)
112 		printf("stipending events %08lx mask %08lx ilevel %d\n",
113 		    HYPERVISOR_shared_info->events,
114 		    HYPERVISOR_shared_info->events_mask, ci->ci_ilevel);
115 #endif
116 
117 #ifdef EARLY_DEBUG_EVENT
118 	if (xen_atomic_test_bit(&s->evtchn_pending[0], debug_port)) {
119 		xen_debug_handler(NULL);
120 		xen_atomic_clear_bit(&s->evtchn_pending[0], debug_port);
121 	}
122 #endif
123 
124 	/*
125 	 * we're only called after STIC, so we know that we'll have to
126 	 * STI at the end
127 	 */
128 	while (vci->evtchn_upcall_pending) {
129 		cli();
130 		vci->evtchn_upcall_pending = 0;
131 		/* NB. No need for a barrier here -- XCHG is a barrier
132 		 * on x86. */
133 		l1 = xen_atomic_xchg(&vci->evtchn_pending_sel, 0);
134 		while ((l1i = xen_ffs(l1)) != 0) {
135 			l1i--;
136 			l1 &= ~(1UL << l1i);
137 
138 			l2 = s->evtchn_pending[l1i] & ~s->evtchn_mask[l1i];
139 			/*
140 			 * mask and clear event. More efficient than calling
141 			 * hypervisor_mask/clear_event for each event.
142 			 */
143 			xen_atomic_setbits_l(&s->evtchn_mask[l1i], l2);
144 			xen_atomic_clearbits_l(&s->evtchn_pending[l1i], l2);
145 			while ((l2i = xen_ffs(l2)) != 0) {
146 				l2i--;
147 				l2 &= ~(1UL << l2i);
148 
149 				port = (l1i << LONG_SHIFT) + l2i;
150 				if (evtsource[port]) {
151 					hypervisor_set_ipending(
152 					    evtsource[port]->ev_imask,
153 					    l1i, l2i);
154 					evtsource[port]->ev_evcnt.ev_count++;
155 					if (ret == 0 && ci->ci_ilevel <
156 					    evtsource[port]->ev_maxlevel)
157 						ret = 1;
158 				}
159 #ifdef DOM0OPS
160 				else  {
161 					/* set pending event */
162 					xenevt_setipending(l1i, l2i);
163 				}
164 #endif
165 			}
166 		}
167 		sti();
168 	}
169 
170 #if 0
171 	if (ci->ci_ipending & 0x1)
172 		printf("stipending events %08lx mask %08lx ilevel %d ipending %08x\n",
173 		    HYPERVISOR_shared_info->events,
174 		    HYPERVISOR_shared_info->events_mask, ci->ci_ilevel,
175 		    ci->ci_ipending);
176 #endif
177 
178 	return (ret);
179 }
180 
181 void
182 do_hypervisor_callback(struct intrframe *regs)
183 {
184 	unsigned long l1;
185 	unsigned long l2;
186 	unsigned int l1i, l2i, port;
187 	volatile shared_info_t *s = HYPERVISOR_shared_info;
188 	struct cpu_info *ci;
189 	volatile struct vcpu_info *vci;
190 	int level;
191 
192 	ci = curcpu();
193 	vci = ci->ci_vcpu;
194 	level = ci->ci_ilevel;
195 
196 	// DDD printf("do_hypervisor_callback\n");
197 
198 #ifdef EARLY_DEBUG_EVENT
199 	if (xen_atomic_test_bit(&s->evtchn_pending[0], debug_port)) {
200 		xen_debug_handler(NULL);
201 		xen_atomic_clear_bit(&s->evtchn_pending[0], debug_port);
202 	}
203 #endif
204 
205 	while (vci->evtchn_upcall_pending) {
206 		vci->evtchn_upcall_pending = 0;
207 		/* NB. No need for a barrier here -- XCHG is a barrier
208 		 * on x86. */
209 		l1 = xen_atomic_xchg(&vci->evtchn_pending_sel, 0);
210 		while ((l1i = xen_ffs(l1)) != 0) {
211 			l1i--;
212 			l1 &= ~(1UL << l1i);
213 
214 			l2 = s->evtchn_pending[l1i] & ~s->evtchn_mask[l1i];
215 			/*
216 			 * mask and clear the pending events.
217 			 * Doing it here for all event that will be processed
218 			 * avoids a race with stipending (which can be called
219 			 * though evtchn_do_event->splx) that could cause an
220 			 * event to be both processed and marked pending.
221 			 */
222 			xen_atomic_setbits_l(&s->evtchn_mask[l1i], l2);
223 			xen_atomic_clearbits_l(&s->evtchn_pending[l1i], l2);
224 
225 			while ((l2i = xen_ffs(l2)) != 0) {
226 				l2i--;
227 				l2 &= ~(1UL << l2i);
228 
229 				port = (l1i << LONG_SHIFT) + l2i;
230 #ifdef PORT_DEBUG
231 				if (port == PORT_DEBUG)
232 					printf("do_hypervisor_callback event %d\n", port);
233 #endif
234 				if (evtsource[port])
235 					call_evtchn_do_event(port, regs);
236 #ifdef DOM0OPS
237 				else  {
238 					if (ci->ci_ilevel < IPL_HIGH) {
239 						/* fast path */
240 						int oipl = ci->ci_ilevel;
241 						ci->ci_ilevel = IPL_HIGH;
242 						call_xenevt_event(port);
243 						ci->ci_ilevel = oipl;
244 					} else {
245 						/* set pending event */
246 						xenevt_setipending(l1i, l2i);
247 					}
248 				}
249 #endif
250 			}
251 		}
252 	}
253 
254 #ifdef DIAGNOSTIC
255 	if (level != ci->ci_ilevel)
256 		printf("hypervisor done %08x level %d/%d ipending %08x\n",
257 		    (uint)vci->evtchn_pending_sel,
258 		    level, ci->ci_ilevel, ci->ci_ipending);
259 #endif
260 }
261 
262 void
263 hypervisor_unmask_event(unsigned int ev)
264 {
265 	volatile shared_info_t *s = HYPERVISOR_shared_info;
266 	volatile struct vcpu_info *vci = curcpu()->ci_vcpu;
267 
268 #ifdef PORT_DEBUG
269 	if (ev == PORT_DEBUG)
270 		printf("hypervisor_unmask_event %d\n", ev);
271 #endif
272 
273 	xen_atomic_clear_bit(&s->evtchn_mask[0], ev);
274 	/*
275 	 * The following is basically the equivalent of
276 	 * 'hw_resend_irq'. Just like a real IO-APIC we 'lose the
277 	 * interrupt edge' if the channel is masked.
278 	 */
279 	if (xen_atomic_test_bit(&s->evtchn_pending[0], ev) &&
280 	    !xen_atomic_test_and_set_bit(&vci->evtchn_pending_sel, ev>>LONG_SHIFT)) {
281 		xen_atomic_set_bit(&vci->evtchn_upcall_pending, 0);
282 		if (!vci->evtchn_upcall_mask)
283 			hypervisor_force_callback();
284 	}
285 }
286 
287 void
288 hypervisor_mask_event(unsigned int ev)
289 {
290 	volatile shared_info_t *s = HYPERVISOR_shared_info;
291 #ifdef PORT_DEBUG
292 	if (ev == PORT_DEBUG)
293 		printf("hypervisor_mask_event %d\n", ev);
294 #endif
295 
296 	xen_atomic_set_bit(&s->evtchn_mask[0], ev);
297 }
298 
299 void
300 hypervisor_clear_event(unsigned int ev)
301 {
302 	volatile shared_info_t *s = HYPERVISOR_shared_info;
303 #ifdef PORT_DEBUG
304 	if (ev == PORT_DEBUG)
305 		printf("hypervisor_clear_event %d\n", ev);
306 #endif
307 
308 	xen_atomic_clear_bit(&s->evtchn_pending[0], ev);
309 }
310 
311 void
312 hypervisor_enable_ipl(unsigned int ipl)
313 {
314 	u_long l1, l2;
315 	int l1i, l2i;
316 	struct cpu_info *ci = curcpu();
317 
318 	/*
319 	 * enable all events for ipl. As we only set an event in ipl_evt_mask
320 	 * for its lowest IPL, and pending IPLs are processed high to low,
321 	 * we know that all callback for this event have been processed.
322 	 */
323 
324 	l1 = ci->ci_isources[ipl]->ipl_evt_mask1;
325 	ci->ci_isources[ipl]->ipl_evt_mask1 = 0;
326 	while ((l1i = xen_ffs(l1)) != 0) {
327 		l1i--;
328 		l1 &= ~(1UL << l1i);
329 		l2 = ci->ci_isources[ipl]->ipl_evt_mask2[l1i];
330 		ci->ci_isources[ipl]->ipl_evt_mask2[l1i] = 0;
331 		while ((l2i = xen_ffs(l2)) != 0) {
332 			int evtch;
333 
334 			l2i--;
335 			l2 &= ~(1UL << l2i);
336 
337 			evtch = (l1i << LONG_SHIFT) + l2i;
338 			hypervisor_enable_event(evtch);
339 		}
340 	}
341 }
342 
343 void
344 hypervisor_set_ipending(uint32_t iplmask, int l1, int l2)
345 {
346 	int ipl;
347 	struct cpu_info *ci = curcpu();
348 
349 	/* set pending bit for the appropriate IPLs */
350 	ci->ci_ipending |= iplmask;
351 
352 	/*
353 	 * And set event pending bit for the lowest IPL. As IPL are handled
354 	 * from high to low, this ensure that all callbacks will have been
355 	 * called when we ack the event
356 	 */
357 	ipl = ffs(iplmask);
358 	KASSERT(ipl > 0);
359 	ipl--;
360 	ci->ci_isources[ipl]->ipl_evt_mask1 |= 1UL << l1;
361 	ci->ci_isources[ipl]->ipl_evt_mask2[l1] |= 1UL << l2;
362 }
363 
364 void
365 hypervisor_machdep_attach(void)
366 {
367  	/* dom0 does not require the arch-dependent P2M translation table */
368 	if ( !xendomain_is_dom0() ) {
369 		build_p2m_frame_list_list();
370 	}
371 }
372 
373 /*
374  * Generate the p2m_frame_list_list table,
375  * needed for guest save/restore
376  */
377 static void
378 build_p2m_frame_list_list(void)
379 {
380         int fpp; /* number of page (frame) pointer per page */
381         unsigned long max_pfn;
382         /*
383          * The p2m list is composed of three levels of indirection,
384          * each layer containing MFNs pointing to lower level pages
385          * The indirection is used to convert a given PFN to its MFN
386          * Each N level page can point to @fpp (N-1) level pages
387          * For example, for x86 32bit, we have:
388          * - PAGE_SIZE: 4096 bytes
389          * - fpp: 1024 (one L3 page can address 1024 L2 pages)
390          * A L1 page contains the list of MFN we are looking for
391          */
392         max_pfn = xen_start_info.nr_pages;
393         fpp = PAGE_SIZE / sizeof(paddr_t);
394 
395         /* we only need one L3 page */
396         l3_p2m_page = kmem_alloc(PAGE_SIZE, KM_NOSLEEP);
397         if (l3_p2m_page == NULL)
398                 panic("could not allocate memory for l3_p2m_page");
399 
400         /*
401          * Determine how many L2 pages we need for the mapping
402          * Each L2 can map a total of @fpp L1 pages
403          */
404         l2_p2m_page_size = howmany(max_pfn, fpp);
405 
406         l2_p2m_page = kmem_alloc(l2_p2m_page_size * PAGE_SIZE, KM_NOSLEEP);
407         if (l2_p2m_page == NULL)
408                 panic("could not allocate memory for l2_p2m_page");
409 
410         /* We now have L3 and L2 pages ready, update L1 mapping */
411         update_p2m_frame_list_list();
412 
413 }
414 
415 /*
416  * Update the L1 p2m_frame_list_list mapping (during guest boot or resume)
417  */
418 static void
419 update_p2m_frame_list_list(void)
420 {
421         int i;
422         int fpp; /* number of page (frame) pointer per page */
423         unsigned long max_pfn;
424 
425         max_pfn = xen_start_info.nr_pages;
426         fpp = PAGE_SIZE / sizeof(paddr_t);
427 
428         for (i = 0; i < l2_p2m_page_size; i++) {
429                 /*
430                  * Each time we start a new L2 page,
431                  * store its MFN in the L3 page
432                  */
433                 if ((i % fpp) == 0) {
434                         l3_p2m_page[i/fpp] = vtomfn(
435                                 (vaddr_t)&l2_p2m_page[i]);
436                 }
437                 /*
438                  * we use a shortcut
439                  * since @xpmap_phys_to_machine_mapping array
440                  * already contains PFN to MFN mapping, we just
441                  * set the l2_p2m_page MFN pointer to the MFN of the
442                  * according frame of @xpmap_phys_to_machine_mapping
443                  */
444                 l2_p2m_page[i] = vtomfn((vaddr_t)
445                         &xpmap_phys_to_machine_mapping[i*fpp]);
446         }
447 
448         HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
449                                         vtomfn((vaddr_t)l3_p2m_page);
450         HYPERVISOR_shared_info->arch.max_pfn = max_pfn;
451 
452 }
453