hypervisor_machdep.c revision 1.2.4.2 1 1.2.4.2 ad /* $NetBSD: hypervisor_machdep.c,v 1.2.4.2 2007/12/03 19:04:40 ad Exp $ */
2 1.2.4.2 ad
3 1.2.4.2 ad /*
4 1.2.4.2 ad *
5 1.2.4.2 ad * Copyright (c) 2004 Christian Limpach.
6 1.2.4.2 ad * All rights reserved.
7 1.2.4.2 ad *
8 1.2.4.2 ad * Redistribution and use in source and binary forms, with or without
9 1.2.4.2 ad * modification, are permitted provided that the following conditions
10 1.2.4.2 ad * are met:
11 1.2.4.2 ad * 1. Redistributions of source code must retain the above copyright
12 1.2.4.2 ad * notice, this list of conditions and the following disclaimer.
13 1.2.4.2 ad * 2. Redistributions in binary form must reproduce the above copyright
14 1.2.4.2 ad * notice, this list of conditions and the following disclaimer in the
15 1.2.4.2 ad * documentation and/or other materials provided with the distribution.
16 1.2.4.2 ad * 3. All advertising materials mentioning features or use of this software
17 1.2.4.2 ad * must display the following acknowledgement:
18 1.2.4.2 ad * This product includes software developed by Christian Limpach.
19 1.2.4.2 ad * 4. The name of the author may not be used to endorse or promote products
20 1.2.4.2 ad * derived from this software without specific prior written permission.
21 1.2.4.2 ad *
22 1.2.4.2 ad * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23 1.2.4.2 ad * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 1.2.4.2 ad * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 1.2.4.2 ad * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26 1.2.4.2 ad * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27 1.2.4.2 ad * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 1.2.4.2 ad * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 1.2.4.2 ad * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 1.2.4.2 ad * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31 1.2.4.2 ad * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 1.2.4.2 ad */
33 1.2.4.2 ad
34 1.2.4.2 ad /******************************************************************************
35 1.2.4.2 ad * hypervisor.c
36 1.2.4.2 ad *
37 1.2.4.2 ad * Communication to/from hypervisor.
38 1.2.4.2 ad *
39 1.2.4.2 ad * Copyright (c) 2002-2004, K A Fraser
40 1.2.4.2 ad *
41 1.2.4.2 ad * Permission is hereby granted, free of charge, to any person obtaining a copy
42 1.2.4.2 ad * of this software and associated documentation files (the "Software"), to
43 1.2.4.2 ad * deal in the Software without restriction, including without limitation the
44 1.2.4.2 ad * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
45 1.2.4.2 ad * sell copies of the Software, and to permit persons to whom the Software is
46 1.2.4.2 ad * furnished to do so, subject to the following conditions:
47 1.2.4.2 ad *
48 1.2.4.2 ad * The above copyright notice and this permission notice shall be included in
49 1.2.4.2 ad * all copies or substantial portions of the Software.
50 1.2.4.2 ad *
51 1.2.4.2 ad * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
52 1.2.4.2 ad * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
53 1.2.4.2 ad * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
54 1.2.4.2 ad * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
55 1.2.4.2 ad * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
56 1.2.4.2 ad * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
57 1.2.4.2 ad * DEALINGS IN THE SOFTWARE.
58 1.2.4.2 ad */
59 1.2.4.2 ad
60 1.2.4.2 ad
61 1.2.4.2 ad #include <sys/cdefs.h>
62 1.2.4.2 ad __KERNEL_RCSID(0, "$NetBSD: hypervisor_machdep.c,v 1.2.4.2 2007/12/03 19:04:40 ad Exp $");
63 1.2.4.2 ad
64 1.2.4.2 ad #include <sys/param.h>
65 1.2.4.2 ad #include <sys/systm.h>
66 1.2.4.2 ad
67 1.2.4.2 ad #include <xen/xen.h>
68 1.2.4.2 ad #include <xen/hypervisor.h>
69 1.2.4.2 ad #include <xen/evtchn.h>
70 1.2.4.2 ad #include <machine/atomic.h>
71 1.2.4.2 ad
72 1.2.4.2 ad #include "opt_xen.h"
73 1.2.4.2 ad
74 1.2.4.2 ad // #define PORT_DEBUG 4
75 1.2.4.2 ad // #define EARLY_DEBUG_EVENT
76 1.2.4.2 ad
77 1.2.4.2 ad #ifdef EARLY_DEBUG_EVENT
78 1.2.4.2 ad extern int debug_port;
79 1.2.4.2 ad extern int xen_debug_handler(void *);
80 1.2.4.2 ad #endif
81 1.2.4.2 ad
82 1.2.4.2 ad int stipending(void);
83 1.2.4.2 ad int
84 1.2.4.2 ad stipending()
85 1.2.4.2 ad {
86 1.2.4.2 ad uint32_t l1;
87 1.2.4.2 ad unsigned long l2;
88 1.2.4.2 ad unsigned int l1i, l2i, port;
89 1.2.4.2 ad volatile shared_info_t *s = HYPERVISOR_shared_info;
90 1.2.4.2 ad struct cpu_info *ci;
91 1.2.4.2 ad int ret;
92 1.2.4.2 ad
93 1.2.4.2 ad ret = 0;
94 1.2.4.2 ad ci = curcpu();
95 1.2.4.2 ad
96 1.2.4.2 ad #if 0
97 1.2.4.2 ad if (HYPERVISOR_shared_info->events)
98 1.2.4.2 ad printf("stipending events %08lx mask %08lx ilevel %d\n",
99 1.2.4.2 ad HYPERVISOR_shared_info->events,
100 1.2.4.2 ad HYPERVISOR_shared_info->events_mask, ci->ci_ilevel);
101 1.2.4.2 ad #endif
102 1.2.4.2 ad
103 1.2.4.2 ad #ifdef EARLY_DEBUG_EVENT
104 1.2.4.2 ad if (xen_atomic_test_bit(&s->evtchn_pending[0], debug_port)) {
105 1.2.4.2 ad xen_debug_handler(NULL);
106 1.2.4.2 ad xen_atomic_clear_bit(&s->evtchn_pending[0], debug_port);
107 1.2.4.2 ad }
108 1.2.4.2 ad #endif
109 1.2.4.2 ad
110 1.2.4.2 ad /*
111 1.2.4.2 ad * we're only called after STIC, so we know that we'll have to
112 1.2.4.2 ad * STI at the end
113 1.2.4.2 ad */
114 1.2.4.2 ad while (s->vcpu_info[0].evtchn_upcall_pending) {
115 1.2.4.2 ad cli();
116 1.2.4.2 ad s->vcpu_info[0].evtchn_upcall_pending = 0;
117 1.2.4.2 ad /* NB. No need for a barrier here -- XCHG is a barrier
118 1.2.4.2 ad * on x86. */
119 1.2.4.2 ad #ifdef XEN3
120 1.2.4.2 ad l1 = xen_atomic_xchg(&s->vcpu_info[0].evtchn_pending_sel, 0);
121 1.2.4.2 ad #else
122 1.2.4.2 ad l1 = xen_atomic_xchg(&s->evtchn_pending_sel, 0);
123 1.2.4.2 ad #endif
124 1.2.4.2 ad while ((l1i = ffs(l1)) != 0) {
125 1.2.4.2 ad l1i--;
126 1.2.4.2 ad l1 &= ~(1 << l1i);
127 1.2.4.2 ad
128 1.2.4.2 ad l2 = s->evtchn_pending[l1i] & ~s->evtchn_mask[l1i];
129 1.2.4.2 ad /*
130 1.2.4.2 ad * mask and clear event. More efficient than calling
131 1.2.4.2 ad * hypervisor_mask/clear_event for each event.
132 1.2.4.2 ad */
133 1.2.4.2 ad xen_atomic_setbits_l(&s->evtchn_mask[l1i], l2);
134 1.2.4.2 ad xen_atomic_clearbits_l(&s->evtchn_pending[l1i], l2);
135 1.2.4.2 ad while ((l2i = ffs(l2)) != 0) {
136 1.2.4.2 ad l2i--;
137 1.2.4.2 ad l2 &= ~(1 << l2i);
138 1.2.4.2 ad
139 1.2.4.2 ad port = (l1i << 5) + l2i;
140 1.2.4.2 ad if (evtsource[port]) {
141 1.2.4.2 ad hypervisor_set_ipending(
142 1.2.4.2 ad evtsource[port]->ev_imask,
143 1.2.4.2 ad l1i, l2i);
144 1.2.4.2 ad evtsource[port]->ev_evcnt.ev_count++;
145 1.2.4.2 ad if (ret == 0 && ci->ci_ilevel <
146 1.2.4.2 ad evtsource[port]->ev_maxlevel)
147 1.2.4.2 ad ret = 1;
148 1.2.4.2 ad }
149 1.2.4.2 ad #ifdef DOM0OPS
150 1.2.4.2 ad else
151 1.2.4.2 ad xenevt_event(port);
152 1.2.4.2 ad #endif
153 1.2.4.2 ad }
154 1.2.4.2 ad }
155 1.2.4.2 ad sti();
156 1.2.4.2 ad }
157 1.2.4.2 ad
158 1.2.4.2 ad #if 0
159 1.2.4.2 ad if (ci->ci_ipending & 0x1)
160 1.2.4.2 ad printf("stipending events %08lx mask %08lx ilevel %d ipending %08x\n",
161 1.2.4.2 ad HYPERVISOR_shared_info->events,
162 1.2.4.2 ad HYPERVISOR_shared_info->events_mask, ci->ci_ilevel,
163 1.2.4.2 ad ci->ci_ipending);
164 1.2.4.2 ad #endif
165 1.2.4.2 ad
166 1.2.4.2 ad return (ret);
167 1.2.4.2 ad }
168 1.2.4.2 ad
169 1.2.4.2 ad void
170 1.2.4.2 ad do_hypervisor_callback(struct intrframe *regs)
171 1.2.4.2 ad {
172 1.2.4.2 ad uint32_t l1;
173 1.2.4.2 ad unsigned long l2;
174 1.2.4.2 ad unsigned int l1i, l2i, port;
175 1.2.4.2 ad volatile shared_info_t *s = HYPERVISOR_shared_info;
176 1.2.4.2 ad struct cpu_info *ci;
177 1.2.4.2 ad int level;
178 1.2.4.2 ad
179 1.2.4.2 ad ci = curcpu();
180 1.2.4.2 ad level = ci->ci_ilevel;
181 1.2.4.2 ad
182 1.2.4.2 ad // DDD printf("do_hypervisor_callback\n");
183 1.2.4.2 ad
184 1.2.4.2 ad #ifdef EARLY_DEBUG_EVENT
185 1.2.4.2 ad if (xen_atomic_test_bit(&s->evtchn_pending[0], debug_port)) {
186 1.2.4.2 ad xen_debug_handler(NULL);
187 1.2.4.2 ad xen_atomic_clear_bit(&s->evtchn_pending[0], debug_port);
188 1.2.4.2 ad }
189 1.2.4.2 ad #endif
190 1.2.4.2 ad
191 1.2.4.2 ad while (s->vcpu_info[0].evtchn_upcall_pending) {
192 1.2.4.2 ad s->vcpu_info[0].evtchn_upcall_pending = 0;
193 1.2.4.2 ad /* NB. No need for a barrier here -- XCHG is a barrier
194 1.2.4.2 ad * on x86. */
195 1.2.4.2 ad #ifdef XEN3
196 1.2.4.2 ad l1 = xen_atomic_xchg(&s->vcpu_info[0].evtchn_pending_sel, 0);
197 1.2.4.2 ad #else
198 1.2.4.2 ad l1 = xen_atomic_xchg(&s->evtchn_pending_sel, 0);
199 1.2.4.2 ad #endif
200 1.2.4.2 ad while ((l1i = ffs(l1)) != 0) {
201 1.2.4.2 ad l1i--;
202 1.2.4.2 ad l1 &= ~(1 << l1i);
203 1.2.4.2 ad
204 1.2.4.2 ad l2 = s->evtchn_pending[l1i] & ~s->evtchn_mask[l1i];
205 1.2.4.2 ad /*
206 1.2.4.2 ad * mask and clear the pending events.
207 1.2.4.2 ad * Doing it here for all event that will be processed
208 1.2.4.2 ad * avoids a race with stipending (which can be called
209 1.2.4.2 ad * though evtchn_do_event->splx) that could cause an event to
210 1.2.4.2 ad * be both processed and marked pending.
211 1.2.4.2 ad */
212 1.2.4.2 ad xen_atomic_setbits_l(&s->evtchn_mask[l1i], l2);
213 1.2.4.2 ad xen_atomic_clearbits_l(&s->evtchn_pending[l1i], l2);
214 1.2.4.2 ad
215 1.2.4.2 ad while ((l2i = ffs(l2)) != 0) {
216 1.2.4.2 ad l2i--;
217 1.2.4.2 ad l2 &= ~(1 << l2i);
218 1.2.4.2 ad
219 1.2.4.2 ad port = (l1i << 5) + l2i;
220 1.2.4.2 ad #ifdef PORT_DEBUG
221 1.2.4.2 ad if (port == PORT_DEBUG)
222 1.2.4.2 ad printf("do_hypervisor_callback event %d\n", port);
223 1.2.4.2 ad #endif
224 1.2.4.2 ad if (evtsource[port])
225 1.2.4.2 ad call_evtchn_do_event(port, regs);
226 1.2.4.2 ad #ifdef DOM0OPS
227 1.2.4.2 ad else
228 1.2.4.2 ad xenevt_event(port);
229 1.2.4.2 ad #endif
230 1.2.4.2 ad }
231 1.2.4.2 ad }
232 1.2.4.2 ad }
233 1.2.4.2 ad
234 1.2.4.2 ad #ifdef DIAGNOSTIC
235 1.2.4.2 ad if (level != ci->ci_ilevel)
236 1.2.4.2 ad printf("hypervisor done %08x level %d/%d ipending %08x\n",
237 1.2.4.2 ad #ifdef XEN3
238 1.2.4.2 ad (uint)HYPERVISOR_shared_info->vcpu_info[0].evtchn_pending_sel,
239 1.2.4.2 ad #else
240 1.2.4.2 ad (uint)HYPERVISOR_shared_info->evtchn_pending_sel,
241 1.2.4.2 ad #endif
242 1.2.4.2 ad level, ci->ci_ilevel, ci->ci_ipending);
243 1.2.4.2 ad #endif
244 1.2.4.2 ad }
245 1.2.4.2 ad
246 1.2.4.2 ad void
247 1.2.4.2 ad hypervisor_unmask_event(unsigned int ev)
248 1.2.4.2 ad {
249 1.2.4.2 ad volatile shared_info_t *s = HYPERVISOR_shared_info;
250 1.2.4.2 ad #ifdef PORT_DEBUG
251 1.2.4.2 ad if (ev == PORT_DEBUG)
252 1.2.4.2 ad printf("hypervisor_unmask_event %d\n", ev);
253 1.2.4.2 ad #endif
254 1.2.4.2 ad
255 1.2.4.2 ad xen_atomic_clear_bit(&s->evtchn_mask[0], ev);
256 1.2.4.2 ad /*
257 1.2.4.2 ad * The following is basically the equivalent of
258 1.2.4.2 ad * 'hw_resend_irq'. Just like a real IO-APIC we 'lose the
259 1.2.4.2 ad * interrupt edge' if the channel is masked.
260 1.2.4.2 ad */
261 1.2.4.2 ad if (xen_atomic_test_bit(&s->evtchn_pending[0], ev) &&
262 1.2.4.2 ad #ifdef XEN3
263 1.2.4.2 ad !xen_atomic_test_and_set_bit(&s->vcpu_info[0].evtchn_pending_sel, ev>>5)) {
264 1.2.4.2 ad #else
265 1.2.4.2 ad !xen_atomic_test_and_set_bit(&s->evtchn_pending_sel, ev>>5)) {
266 1.2.4.2 ad #endif
267 1.2.4.2 ad xen_atomic_set_bit(&s->vcpu_info[0].evtchn_upcall_pending, 0);
268 1.2.4.2 ad if (!s->vcpu_info[0].evtchn_upcall_mask)
269 1.2.4.2 ad hypervisor_force_callback();
270 1.2.4.2 ad }
271 1.2.4.2 ad }
272 1.2.4.2 ad
273 1.2.4.2 ad void
274 1.2.4.2 ad hypervisor_mask_event(unsigned int ev)
275 1.2.4.2 ad {
276 1.2.4.2 ad volatile shared_info_t *s = HYPERVISOR_shared_info;
277 1.2.4.2 ad #ifdef PORT_DEBUG
278 1.2.4.2 ad if (ev == PORT_DEBUG)
279 1.2.4.2 ad printf("hypervisor_mask_event %d\n", ev);
280 1.2.4.2 ad #endif
281 1.2.4.2 ad
282 1.2.4.2 ad xen_atomic_set_bit(&s->evtchn_mask[0], ev);
283 1.2.4.2 ad }
284 1.2.4.2 ad
285 1.2.4.2 ad void
286 1.2.4.2 ad hypervisor_clear_event(unsigned int ev)
287 1.2.4.2 ad {
288 1.2.4.2 ad volatile shared_info_t *s = HYPERVISOR_shared_info;
289 1.2.4.2 ad #ifdef PORT_DEBUG
290 1.2.4.2 ad if (ev == PORT_DEBUG)
291 1.2.4.2 ad printf("hypervisor_clear_event %d\n", ev);
292 1.2.4.2 ad #endif
293 1.2.4.2 ad
294 1.2.4.2 ad xen_atomic_clear_bit(&s->evtchn_pending[0], ev);
295 1.2.4.2 ad }
296 1.2.4.2 ad
297 1.2.4.2 ad void
298 1.2.4.2 ad hypervisor_enable_ipl(unsigned int ipl)
299 1.2.4.2 ad {
300 1.2.4.2 ad u_int32_t l1, l2;
301 1.2.4.2 ad int l1i, l2i;
302 1.2.4.2 ad struct cpu_info *ci = curcpu();
303 1.2.4.2 ad
304 1.2.4.2 ad /*
305 1.2.4.2 ad * enable all events for ipl. As we only set an event in ipl_evt_mask
306 1.2.4.2 ad * for its lowest IPL, and pending IPLs are processed high to low,
307 1.2.4.2 ad * we know that all callback for this event have been processed.
308 1.2.4.2 ad */
309 1.2.4.2 ad
310 1.2.4.2 ad l1 = ci->ci_isources[ipl]->ipl_evt_mask1;
311 1.2.4.2 ad ci->ci_isources[ipl]->ipl_evt_mask1 = 0;
312 1.2.4.2 ad while ((l1i = ffs(l1)) != 0) {
313 1.2.4.2 ad l1i--;
314 1.2.4.2 ad l1 &= ~(1 << l1i);
315 1.2.4.2 ad l2 = ci->ci_isources[ipl]->ipl_evt_mask2[l1i];
316 1.2.4.2 ad ci->ci_isources[ipl]->ipl_evt_mask2[l1i] = 0;
317 1.2.4.2 ad while ((l2i = ffs(l2)) != 0) {
318 1.2.4.2 ad int evtch;
319 1.2.4.2 ad
320 1.2.4.2 ad l2i--;
321 1.2.4.2 ad l2 &= ~(1 << l2i);
322 1.2.4.2 ad
323 1.2.4.2 ad evtch = (l1i << 5) + l2i;
324 1.2.4.2 ad hypervisor_enable_event(evtch);
325 1.2.4.2 ad }
326 1.2.4.2 ad }
327 1.2.4.2 ad }
328 1.2.4.2 ad
329 1.2.4.2 ad void
330 1.2.4.2 ad hypervisor_set_ipending(u_int32_t iplmask, int l1, int l2)
331 1.2.4.2 ad {
332 1.2.4.2 ad int ipl;
333 1.2.4.2 ad struct cpu_info *ci = curcpu();
334 1.2.4.2 ad
335 1.2.4.2 ad /* set pending bit for the appropriate IPLs */
336 1.2.4.2 ad ci->ci_ipending |= iplmask;
337 1.2.4.2 ad
338 1.2.4.2 ad /*
339 1.2.4.2 ad * And set event pending bit for the lowest IPL. As IPL are handled
340 1.2.4.2 ad * from high to low, this ensure that all callbacks will have been
341 1.2.4.2 ad * called when we ack the event
342 1.2.4.2 ad */
343 1.2.4.2 ad ipl = ffs(iplmask);
344 1.2.4.2 ad KASSERT(ipl > 0);
345 1.2.4.2 ad ipl--;
346 1.2.4.2 ad ci->ci_isources[ipl]->ipl_evt_mask1 |= 1 << l1;
347 1.2.4.2 ad ci->ci_isources[ipl]->ipl_evt_mask2[l1] |= 1 << l2;
348 1.2.4.2 ad }
349 1.2.4.2 ad /* $NetBSD: hypervisor_machdep.c,v 1.2.4.2 2007/12/03 19:04:40 ad Exp $ */
350 1.2.4.2 ad
351 1.2.4.2 ad /*
352 1.2.4.2 ad *
353 1.2.4.2 ad * Copyright (c) 2004 Christian Limpach.
354 1.2.4.2 ad * All rights reserved.
355 1.2.4.2 ad *
356 1.2.4.2 ad * Redistribution and use in source and binary forms, with or without
357 1.2.4.2 ad * modification, are permitted provided that the following conditions
358 1.2.4.2 ad * are met:
359 1.2.4.2 ad * 1. Redistributions of source code must retain the above copyright
360 1.2.4.2 ad * notice, this list of conditions and the following disclaimer.
361 1.2.4.2 ad * 2. Redistributions in binary form must reproduce the above copyright
362 1.2.4.2 ad * notice, this list of conditions and the following disclaimer in the
363 1.2.4.2 ad * documentation and/or other materials provided with the distribution.
364 1.2.4.2 ad * 3. All advertising materials mentioning features or use of this software
365 1.2.4.2 ad * must display the following acknowledgement:
366 1.2.4.2 ad * This product includes software developed by Christian Limpach.
367 1.2.4.2 ad * 4. The name of the author may not be used to endorse or promote products
368 1.2.4.2 ad * derived from this software without specific prior written permission.
369 1.2.4.2 ad *
370 1.2.4.2 ad * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
371 1.2.4.2 ad * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
372 1.2.4.2 ad * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
373 1.2.4.2 ad * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
374 1.2.4.2 ad * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
375 1.2.4.2 ad * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
376 1.2.4.2 ad * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
377 1.2.4.2 ad * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
378 1.2.4.2 ad * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
379 1.2.4.2 ad * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
380 1.2.4.2 ad */
381 1.2.4.2 ad
382 1.2.4.2 ad /******************************************************************************
383 1.2.4.2 ad * hypervisor.c
384 1.2.4.2 ad *
385 1.2.4.2 ad * Communication to/from hypervisor.
386 1.2.4.2 ad *
387 1.2.4.2 ad * Copyright (c) 2002-2004, K A Fraser
388 1.2.4.2 ad *
389 1.2.4.2 ad * Permission is hereby granted, free of charge, to any person obtaining a copy
390 1.2.4.2 ad * of this software and associated documentation files (the "Software"), to
391 1.2.4.2 ad * deal in the Software without restriction, including without limitation the
392 1.2.4.2 ad * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
393 1.2.4.2 ad * sell copies of the Software, and to permit persons to whom the Software is
394 1.2.4.2 ad * furnished to do so, subject to the following conditions:
395 1.2.4.2 ad *
396 1.2.4.2 ad * The above copyright notice and this permission notice shall be included in
397 1.2.4.2 ad * all copies or substantial portions of the Software.
398 1.2.4.2 ad *
399 1.2.4.2 ad * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
400 1.2.4.2 ad * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
401 1.2.4.2 ad * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
402 1.2.4.2 ad * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
403 1.2.4.2 ad * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
404 1.2.4.2 ad * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
405 1.2.4.2 ad * DEALINGS IN THE SOFTWARE.
406 1.2.4.2 ad */
407 1.2.4.2 ad
408 1.2.4.2 ad
409 1.2.4.2 ad #include <sys/cdefs.h>
410 1.2.4.2 ad __KERNEL_RCSID(0, "$NetBSD: hypervisor_machdep.c,v 1.2.4.2 2007/12/03 19:04:40 ad Exp $");
411 1.2.4.2 ad
412 1.2.4.2 ad #include <sys/param.h>
413 1.2.4.2 ad #include <sys/systm.h>
414 1.2.4.2 ad
415 1.2.4.2 ad #include <xen/xen.h>
416 1.2.4.2 ad #include <xen/hypervisor.h>
417 1.2.4.2 ad #include <xen/evtchn.h>
418 1.2.4.2 ad #include <machine/atomic.h>
419 1.2.4.2 ad
420 1.2.4.2 ad #include "opt_xen.h"
421 1.2.4.2 ad
422 1.2.4.2 ad // #define PORT_DEBUG 4
423 1.2.4.2 ad // #define EARLY_DEBUG_EVENT
424 1.2.4.2 ad
425 1.2.4.2 ad #ifdef EARLY_DEBUG_EVENT
426 1.2.4.2 ad extern int debug_port;
427 1.2.4.2 ad extern int xen_debug_handler(void *);
428 1.2.4.2 ad #endif
429 1.2.4.2 ad
430 1.2.4.2 ad int stipending(void);
431 1.2.4.2 ad int
432 1.2.4.2 ad stipending()
433 1.2.4.2 ad {
434 1.2.4.2 ad uint32_t l1;
435 1.2.4.2 ad unsigned long l2;
436 1.2.4.2 ad unsigned int l1i, l2i, port;
437 1.2.4.2 ad volatile shared_info_t *s = HYPERVISOR_shared_info;
438 1.2.4.2 ad struct cpu_info *ci;
439 1.2.4.2 ad int ret;
440 1.2.4.2 ad
441 1.2.4.2 ad ret = 0;
442 1.2.4.2 ad ci = curcpu();
443 1.2.4.2 ad
444 1.2.4.2 ad #if 0
445 1.2.4.2 ad if (HYPERVISOR_shared_info->events)
446 1.2.4.2 ad printf("stipending events %08lx mask %08lx ilevel %d\n",
447 1.2.4.2 ad HYPERVISOR_shared_info->events,
448 1.2.4.2 ad HYPERVISOR_shared_info->events_mask, ci->ci_ilevel);
449 1.2.4.2 ad #endif
450 1.2.4.2 ad
451 1.2.4.2 ad #ifdef EARLY_DEBUG_EVENT
452 1.2.4.2 ad if (xen_atomic_test_bit(&s->evtchn_pending[0], debug_port)) {
453 1.2.4.2 ad xen_debug_handler(NULL);
454 1.2.4.2 ad xen_atomic_clear_bit(&s->evtchn_pending[0], debug_port);
455 1.2.4.2 ad }
456 1.2.4.2 ad #endif
457 1.2.4.2 ad
458 1.2.4.2 ad /*
459 1.2.4.2 ad * we're only called after STIC, so we know that we'll have to
460 1.2.4.2 ad * STI at the end
461 1.2.4.2 ad */
462 1.2.4.2 ad while (s->vcpu_info[0].evtchn_upcall_pending) {
463 1.2.4.2 ad cli();
464 1.2.4.2 ad s->vcpu_info[0].evtchn_upcall_pending = 0;
465 1.2.4.2 ad /* NB. No need for a barrier here -- XCHG is a barrier
466 1.2.4.2 ad * on x86. */
467 1.2.4.2 ad #ifdef XEN3
468 1.2.4.2 ad l1 = xen_atomic_xchg(&s->vcpu_info[0].evtchn_pending_sel, 0);
469 1.2.4.2 ad #else
470 1.2.4.2 ad l1 = xen_atomic_xchg(&s->evtchn_pending_sel, 0);
471 1.2.4.2 ad #endif
472 1.2.4.2 ad while ((l1i = ffs(l1)) != 0) {
473 1.2.4.2 ad l1i--;
474 1.2.4.2 ad l1 &= ~(1 << l1i);
475 1.2.4.2 ad
476 1.2.4.2 ad l2 = s->evtchn_pending[l1i] & ~s->evtchn_mask[l1i];
477 1.2.4.2 ad /*
478 1.2.4.2 ad * mask and clear event. More efficient than calling
479 1.2.4.2 ad * hypervisor_mask/clear_event for each event.
480 1.2.4.2 ad */
481 1.2.4.2 ad xen_atomic_setbits_l(&s->evtchn_mask[l1i], l2);
482 1.2.4.2 ad xen_atomic_clearbits_l(&s->evtchn_pending[l1i], l2);
483 1.2.4.2 ad while ((l2i = ffs(l2)) != 0) {
484 1.2.4.2 ad l2i--;
485 1.2.4.2 ad l2 &= ~(1 << l2i);
486 1.2.4.2 ad
487 1.2.4.2 ad port = (l1i << 5) + l2i;
488 1.2.4.2 ad if (evtsource[port]) {
489 1.2.4.2 ad hypervisor_set_ipending(
490 1.2.4.2 ad evtsource[port]->ev_imask,
491 1.2.4.2 ad l1i, l2i);
492 1.2.4.2 ad evtsource[port]->ev_evcnt.ev_count++;
493 1.2.4.2 ad if (ret == 0 && ci->ci_ilevel <
494 1.2.4.2 ad evtsource[port]->ev_maxlevel)
495 1.2.4.2 ad ret = 1;
496 1.2.4.2 ad }
497 1.2.4.2 ad #ifdef DOM0OPS
498 1.2.4.2 ad else
499 1.2.4.2 ad xenevt_event(port);
500 1.2.4.2 ad #endif
501 1.2.4.2 ad }
502 1.2.4.2 ad }
503 1.2.4.2 ad sti();
504 1.2.4.2 ad }
505 1.2.4.2 ad
506 1.2.4.2 ad #if 0
507 1.2.4.2 ad if (ci->ci_ipending & 0x1)
508 1.2.4.2 ad printf("stipending events %08lx mask %08lx ilevel %d ipending %08x\n",
509 1.2.4.2 ad HYPERVISOR_shared_info->events,
510 1.2.4.2 ad HYPERVISOR_shared_info->events_mask, ci->ci_ilevel,
511 1.2.4.2 ad ci->ci_ipending);
512 1.2.4.2 ad #endif
513 1.2.4.2 ad
514 1.2.4.2 ad return (ret);
515 1.2.4.2 ad }
516 1.2.4.2 ad
517 1.2.4.2 ad void
518 1.2.4.2 ad do_hypervisor_callback(struct intrframe *regs)
519 1.2.4.2 ad {
520 1.2.4.2 ad uint32_t l1;
521 1.2.4.2 ad unsigned long l2;
522 1.2.4.2 ad unsigned int l1i, l2i, port;
523 1.2.4.2 ad volatile shared_info_t *s = HYPERVISOR_shared_info;
524 1.2.4.2 ad struct cpu_info *ci;
525 1.2.4.2 ad int level;
526 1.2.4.2 ad
527 1.2.4.2 ad ci = curcpu();
528 1.2.4.2 ad level = ci->ci_ilevel;
529 1.2.4.2 ad
530 1.2.4.2 ad // DDD printf("do_hypervisor_callback\n");
531 1.2.4.2 ad
532 1.2.4.2 ad #ifdef EARLY_DEBUG_EVENT
533 1.2.4.2 ad if (xen_atomic_test_bit(&s->evtchn_pending[0], debug_port)) {
534 1.2.4.2 ad xen_debug_handler(NULL);
535 1.2.4.2 ad xen_atomic_clear_bit(&s->evtchn_pending[0], debug_port);
536 1.2.4.2 ad }
537 1.2.4.2 ad #endif
538 1.2.4.2 ad
539 1.2.4.2 ad while (s->vcpu_info[0].evtchn_upcall_pending) {
540 1.2.4.2 ad s->vcpu_info[0].evtchn_upcall_pending = 0;
541 1.2.4.2 ad /* NB. No need for a barrier here -- XCHG is a barrier
542 1.2.4.2 ad * on x86. */
543 1.2.4.2 ad #ifdef XEN3
544 1.2.4.2 ad l1 = xen_atomic_xchg(&s->vcpu_info[0].evtchn_pending_sel, 0);
545 1.2.4.2 ad #else
546 1.2.4.2 ad l1 = xen_atomic_xchg(&s->evtchn_pending_sel, 0);
547 1.2.4.2 ad #endif
548 1.2.4.2 ad while ((l1i = ffs(l1)) != 0) {
549 1.2.4.2 ad l1i--;
550 1.2.4.2 ad l1 &= ~(1 << l1i);
551 1.2.4.2 ad
552 1.2.4.2 ad l2 = s->evtchn_pending[l1i] & ~s->evtchn_mask[l1i];
553 1.2.4.2 ad /*
554 1.2.4.2 ad * mask and clear the pending events.
555 1.2.4.2 ad * Doing it here for all event that will be processed
556 1.2.4.2 ad * avoids a race with stipending (which can be called
557 1.2.4.2 ad * though evtchn_do_event->splx) that could cause an event to
558 1.2.4.2 ad * be both processed and marked pending.
559 1.2.4.2 ad */
560 1.2.4.2 ad xen_atomic_setbits_l(&s->evtchn_mask[l1i], l2);
561 1.2.4.2 ad xen_atomic_clearbits_l(&s->evtchn_pending[l1i], l2);
562 1.2.4.2 ad
563 1.2.4.2 ad while ((l2i = ffs(l2)) != 0) {
564 1.2.4.2 ad l2i--;
565 1.2.4.2 ad l2 &= ~(1 << l2i);
566 1.2.4.2 ad
567 1.2.4.2 ad port = (l1i << 5) + l2i;
568 1.2.4.2 ad #ifdef PORT_DEBUG
569 1.2.4.2 ad if (port == PORT_DEBUG)
570 1.2.4.2 ad printf("do_hypervisor_callback event %d\n", port);
571 1.2.4.2 ad #endif
572 1.2.4.2 ad if (evtsource[port])
573 1.2.4.2 ad call_evtchn_do_event(port, regs);
574 1.2.4.2 ad #ifdef DOM0OPS
575 1.2.4.2 ad else
576 1.2.4.2 ad xenevt_event(port);
577 1.2.4.2 ad #endif
578 1.2.4.2 ad }
579 1.2.4.2 ad }
580 1.2.4.2 ad }
581 1.2.4.2 ad
582 1.2.4.2 ad #ifdef DIAGNOSTIC
583 1.2.4.2 ad if (level != ci->ci_ilevel)
584 1.2.4.2 ad printf("hypervisor done %08x level %d/%d ipending %08x\n",
585 1.2.4.2 ad #ifdef XEN3
586 1.2.4.2 ad (uint)HYPERVISOR_shared_info->vcpu_info[0].evtchn_pending_sel,
587 1.2.4.2 ad #else
588 1.2.4.2 ad (uint)HYPERVISOR_shared_info->evtchn_pending_sel,
589 1.2.4.2 ad #endif
590 1.2.4.2 ad level, ci->ci_ilevel, ci->ci_ipending);
591 1.2.4.2 ad #endif
592 1.2.4.2 ad }
593 1.2.4.2 ad
594 1.2.4.2 ad void
595 1.2.4.2 ad hypervisor_unmask_event(unsigned int ev)
596 1.2.4.2 ad {
597 1.2.4.2 ad volatile shared_info_t *s = HYPERVISOR_shared_info;
598 1.2.4.2 ad #ifdef PORT_DEBUG
599 1.2.4.2 ad if (ev == PORT_DEBUG)
600 1.2.4.2 ad printf("hypervisor_unmask_event %d\n", ev);
601 1.2.4.2 ad #endif
602 1.2.4.2 ad
603 1.2.4.2 ad xen_atomic_clear_bit(&s->evtchn_mask[0], ev);
604 1.2.4.2 ad /*
605 1.2.4.2 ad * The following is basically the equivalent of
606 1.2.4.2 ad * 'hw_resend_irq'. Just like a real IO-APIC we 'lose the
607 1.2.4.2 ad * interrupt edge' if the channel is masked.
608 1.2.4.2 ad */
609 1.2.4.2 ad if (xen_atomic_test_bit(&s->evtchn_pending[0], ev) &&
610 1.2.4.2 ad #ifdef XEN3
611 1.2.4.2 ad !xen_atomic_test_and_set_bit(&s->vcpu_info[0].evtchn_pending_sel, ev>>5)) {
612 1.2.4.2 ad #else
613 1.2.4.2 ad !xen_atomic_test_and_set_bit(&s->evtchn_pending_sel, ev>>5)) {
614 1.2.4.2 ad #endif
615 1.2.4.2 ad xen_atomic_set_bit(&s->vcpu_info[0].evtchn_upcall_pending, 0);
616 1.2.4.2 ad if (!s->vcpu_info[0].evtchn_upcall_mask)
617 1.2.4.2 ad hypervisor_force_callback();
618 1.2.4.2 ad }
619 1.2.4.2 ad }
620 1.2.4.2 ad
621 1.2.4.2 ad void
622 1.2.4.2 ad hypervisor_mask_event(unsigned int ev)
623 1.2.4.2 ad {
624 1.2.4.2 ad volatile shared_info_t *s = HYPERVISOR_shared_info;
625 1.2.4.2 ad #ifdef PORT_DEBUG
626 1.2.4.2 ad if (ev == PORT_DEBUG)
627 1.2.4.2 ad printf("hypervisor_mask_event %d\n", ev);
628 1.2.4.2 ad #endif
629 1.2.4.2 ad
630 1.2.4.2 ad xen_atomic_set_bit(&s->evtchn_mask[0], ev);
631 1.2.4.2 ad }
632 1.2.4.2 ad
633 1.2.4.2 ad void
634 1.2.4.2 ad hypervisor_clear_event(unsigned int ev)
635 1.2.4.2 ad {
636 1.2.4.2 ad volatile shared_info_t *s = HYPERVISOR_shared_info;
637 1.2.4.2 ad #ifdef PORT_DEBUG
638 1.2.4.2 ad if (ev == PORT_DEBUG)
639 1.2.4.2 ad printf("hypervisor_clear_event %d\n", ev);
640 1.2.4.2 ad #endif
641 1.2.4.2 ad
642 1.2.4.2 ad xen_atomic_clear_bit(&s->evtchn_pending[0], ev);
643 1.2.4.2 ad }
644 1.2.4.2 ad
645 1.2.4.2 ad void
646 1.2.4.2 ad hypervisor_enable_ipl(unsigned int ipl)
647 1.2.4.2 ad {
648 1.2.4.2 ad u_int32_t l1, l2;
649 1.2.4.2 ad int l1i, l2i;
650 1.2.4.2 ad struct cpu_info *ci = curcpu();
651 1.2.4.2 ad
652 1.2.4.2 ad /*
653 1.2.4.2 ad * enable all events for ipl. As we only set an event in ipl_evt_mask
654 1.2.4.2 ad * for its lowest IPL, and pending IPLs are processed high to low,
655 1.2.4.2 ad * we know that all callback for this event have been processed.
656 1.2.4.2 ad */
657 1.2.4.2 ad
658 1.2.4.2 ad l1 = ci->ci_isources[ipl]->ipl_evt_mask1;
659 1.2.4.2 ad ci->ci_isources[ipl]->ipl_evt_mask1 = 0;
660 1.2.4.2 ad while ((l1i = ffs(l1)) != 0) {
661 1.2.4.2 ad l1i--;
662 1.2.4.2 ad l1 &= ~(1 << l1i);
663 1.2.4.2 ad l2 = ci->ci_isources[ipl]->ipl_evt_mask2[l1i];
664 1.2.4.2 ad ci->ci_isources[ipl]->ipl_evt_mask2[l1i] = 0;
665 1.2.4.2 ad while ((l2i = ffs(l2)) != 0) {
666 1.2.4.2 ad int evtch;
667 1.2.4.2 ad
668 1.2.4.2 ad l2i--;
669 1.2.4.2 ad l2 &= ~(1 << l2i);
670 1.2.4.2 ad
671 1.2.4.2 ad evtch = (l1i << 5) + l2i;
672 1.2.4.2 ad hypervisor_enable_event(evtch);
673 1.2.4.2 ad }
674 1.2.4.2 ad }
675 1.2.4.2 ad }
676 1.2.4.2 ad
677 1.2.4.2 ad void
678 1.2.4.2 ad hypervisor_set_ipending(u_int32_t iplmask, int l1, int l2)
679 1.2.4.2 ad {
680 1.2.4.2 ad int ipl;
681 1.2.4.2 ad struct cpu_info *ci = curcpu();
682 1.2.4.2 ad
683 1.2.4.2 ad /* set pending bit for the appropriate IPLs */
684 1.2.4.2 ad ci->ci_ipending |= iplmask;
685 1.2.4.2 ad
686 1.2.4.2 ad /*
687 1.2.4.2 ad * And set event pending bit for the lowest IPL. As IPL are handled
688 1.2.4.2 ad * from high to low, this ensure that all callbacks will have been
689 1.2.4.2 ad * called when we ack the event
690 1.2.4.2 ad */
691 1.2.4.2 ad ipl = ffs(iplmask);
692 1.2.4.2 ad KASSERT(ipl > 0);
693 1.2.4.2 ad ipl--;
694 1.2.4.2 ad ci->ci_isources[ipl]->ipl_evt_mask1 |= 1 << l1;
695 1.2.4.2 ad ci->ci_isources[ipl]->ipl_evt_mask2[l1] |= 1 << l2;
696 1.2.4.2 ad }
697