xref: /netbsd-src/sys/arch/xen/x86/hypervisor_machdep.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: hypervisor_machdep.c,v 1.3 2007/12/12 22:16:32 bouyer 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 1.3 2007/12/12 22:16:32 bouyer Exp $");
63 
64 #include <sys/param.h>
65 #include <sys/systm.h>
66 
67 #include <xen/xen.h>
68 #include <xen/hypervisor.h>
69 #include <xen/evtchn.h>
70 #include <machine/atomic.h>
71 
72 #include "opt_xen.h"
73 
74 // #define PORT_DEBUG 4
75 // #define EARLY_DEBUG_EVENT
76 
77 int stipending(void);
78 int
79 stipending()
80 {
81 	uint32_t l1;
82 	unsigned long l2;
83 	unsigned int l1i, l2i, port;
84 	volatile shared_info_t *s = HYPERVISOR_shared_info;
85 	struct cpu_info *ci;
86 	int ret;
87 
88 	ret = 0;
89 	ci = curcpu();
90 
91 #if 0
92 	if (HYPERVISOR_shared_info->events)
93 		printf("stipending events %08lx mask %08lx ilevel %d\n",
94 		    HYPERVISOR_shared_info->events,
95 		    HYPERVISOR_shared_info->events_mask, ci->ci_ilevel);
96 #endif
97 
98 #ifdef EARLY_DEBUG_EVENT
99 	if (xen_atomic_test_bit(&s->evtchn_pending[0], debug_port)) {
100 		xen_debug_handler(NULL);
101 		xen_atomic_clear_bit(&s->evtchn_pending[0], debug_port);
102 	}
103 #endif
104 
105 	/*
106 	 * we're only called after STIC, so we know that we'll have to
107 	 * STI at the end
108 	 */
109 	while (s->vcpu_info[0].evtchn_upcall_pending) {
110 		cli();
111 		s->vcpu_info[0].evtchn_upcall_pending = 0;
112 		/* NB. No need for a barrier here -- XCHG is a barrier
113 		 * on x86. */
114 #ifdef XEN3
115 		l1 = xen_atomic_xchg(&s->vcpu_info[0].evtchn_pending_sel, 0);
116 #else
117 		l1 = xen_atomic_xchg(&s->evtchn_pending_sel, 0);
118 #endif
119 		while ((l1i = ffs(l1)) != 0) {
120 			l1i--;
121 			l1 &= ~(1 << l1i);
122 
123 			l2 = s->evtchn_pending[l1i] & ~s->evtchn_mask[l1i];
124 			/*
125 			 * mask and clear event. More efficient than calling
126 			 * hypervisor_mask/clear_event for each event.
127 			 */
128 			xen_atomic_setbits_l(&s->evtchn_mask[l1i], l2);
129 			xen_atomic_clearbits_l(&s->evtchn_pending[l1i], l2);
130 			while ((l2i = ffs(l2)) != 0) {
131 				l2i--;
132 				l2 &= ~(1 << l2i);
133 
134 				port = (l1i << 5) + l2i;
135 				if (evtsource[port]) {
136 					hypervisor_set_ipending(
137 					    evtsource[port]->ev_imask,
138 					    l1i, l2i);
139 					evtsource[port]->ev_evcnt.ev_count++;
140 					if (ret == 0 && ci->ci_ilevel <
141 					    evtsource[port]->ev_maxlevel)
142 						ret = 1;
143 				}
144 #ifdef DOM0OPS
145 				else
146 					xenevt_event(port);
147 #endif
148 			}
149 		}
150 		sti();
151 	}
152 
153 #if 0
154 	if (ci->ci_ipending & 0x1)
155 		printf("stipending events %08lx mask %08lx ilevel %d ipending %08x\n",
156 		    HYPERVISOR_shared_info->events,
157 		    HYPERVISOR_shared_info->events_mask, ci->ci_ilevel,
158 		    ci->ci_ipending);
159 #endif
160 
161 	return (ret);
162 }
163 
164 void
165 do_hypervisor_callback(struct intrframe *regs)
166 {
167 	uint32_t l1;
168 	unsigned long l2;
169 	unsigned int l1i, l2i, port;
170 	volatile shared_info_t *s = HYPERVISOR_shared_info;
171 	struct cpu_info *ci;
172 	int level;
173 
174 	ci = curcpu();
175 	level = ci->ci_ilevel;
176 
177 	// DDD printf("do_hypervisor_callback\n");
178 
179 #ifdef EARLY_DEBUG_EVENT
180 	if (xen_atomic_test_bit(&s->evtchn_pending[0], debug_port)) {
181 		xen_debug_handler(NULL);
182 		xen_atomic_clear_bit(&s->evtchn_pending[0], debug_port);
183 	}
184 #endif
185 
186 	while (s->vcpu_info[0].evtchn_upcall_pending) {
187 		s->vcpu_info[0].evtchn_upcall_pending = 0;
188 		/* NB. No need for a barrier here -- XCHG is a barrier
189 		 * on x86. */
190 #ifdef XEN3
191 		l1 = xen_atomic_xchg(&s->vcpu_info[0].evtchn_pending_sel, 0);
192 #else
193 		l1 = xen_atomic_xchg(&s->evtchn_pending_sel, 0);
194 #endif
195 		while ((l1i = ffs(l1)) != 0) {
196 			l1i--;
197 			l1 &= ~(1 << l1i);
198 
199 			l2 = s->evtchn_pending[l1i] & ~s->evtchn_mask[l1i];
200 			/*
201 			 * mask and clear the pending events.
202 			 * Doing it here for all event that will be processed
203 			 * avoids a race with stipending (which can be called
204 			 * though evtchn_do_event->splx) that could cause an event to
205 			 * be both processed and marked pending.
206 			 */
207 			xen_atomic_setbits_l(&s->evtchn_mask[l1i], l2);
208 			xen_atomic_clearbits_l(&s->evtchn_pending[l1i], l2);
209 
210 			while ((l2i = ffs(l2)) != 0) {
211 				l2i--;
212 				l2 &= ~(1 << l2i);
213 
214 				port = (l1i << 5) + l2i;
215 #ifdef PORT_DEBUG
216 				if (port == PORT_DEBUG)
217 					printf("do_hypervisor_callback event %d\n", port);
218 #endif
219 				if (evtsource[port])
220 					call_evtchn_do_event(port, regs);
221 #ifdef DOM0OPS
222 				else
223 					xenevt_event(port);
224 #endif
225 			}
226 		}
227 	}
228 
229 #ifdef DIAGNOSTIC
230 	if (level != ci->ci_ilevel)
231 		printf("hypervisor done %08x level %d/%d ipending %08x\n",
232 #ifdef XEN3
233 		    (uint)HYPERVISOR_shared_info->vcpu_info[0].evtchn_pending_sel,
234 #else
235 		    (uint)HYPERVISOR_shared_info->evtchn_pending_sel,
236 #endif
237 		    level, ci->ci_ilevel, ci->ci_ipending);
238 #endif
239 }
240 
241 void
242 hypervisor_unmask_event(unsigned int ev)
243 {
244 	volatile shared_info_t *s = HYPERVISOR_shared_info;
245 #ifdef PORT_DEBUG
246 	if (ev == PORT_DEBUG)
247 		printf("hypervisor_unmask_event %d\n", ev);
248 #endif
249 
250 	xen_atomic_clear_bit(&s->evtchn_mask[0], ev);
251 	/*
252 	 * The following is basically the equivalent of
253 	 * 'hw_resend_irq'. Just like a real IO-APIC we 'lose the
254 	 * interrupt edge' if the channel is masked.
255 	 */
256 	if (xen_atomic_test_bit(&s->evtchn_pending[0], ev) &&
257 #ifdef XEN3
258 	    !xen_atomic_test_and_set_bit(&s->vcpu_info[0].evtchn_pending_sel, ev>>5)) {
259 #else
260 	    !xen_atomic_test_and_set_bit(&s->evtchn_pending_sel, ev>>5)) {
261 #endif
262 		xen_atomic_set_bit(&s->vcpu_info[0].evtchn_upcall_pending, 0);
263 		if (!s->vcpu_info[0].evtchn_upcall_mask)
264 			hypervisor_force_callback();
265 	}
266 }
267 
268 void
269 hypervisor_mask_event(unsigned int ev)
270 {
271 	volatile shared_info_t *s = HYPERVISOR_shared_info;
272 #ifdef PORT_DEBUG
273 	if (ev == PORT_DEBUG)
274 		printf("hypervisor_mask_event %d\n", ev);
275 #endif
276 
277 	xen_atomic_set_bit(&s->evtchn_mask[0], ev);
278 }
279 
280 void
281 hypervisor_clear_event(unsigned int ev)
282 {
283 	volatile shared_info_t *s = HYPERVISOR_shared_info;
284 #ifdef PORT_DEBUG
285 	if (ev == PORT_DEBUG)
286 		printf("hypervisor_clear_event %d\n", ev);
287 #endif
288 
289 	xen_atomic_clear_bit(&s->evtchn_pending[0], ev);
290 }
291 
292 void
293 hypervisor_enable_ipl(unsigned int ipl)
294 {
295 	u_int32_t l1, l2;
296 	int l1i, l2i;
297 	struct cpu_info *ci = curcpu();
298 
299 	/*
300 	 * enable all events for ipl. As we only set an event in ipl_evt_mask
301 	 * for its lowest IPL, and pending IPLs are processed high to low,
302 	 * we know that all callback for this event have been processed.
303 	 */
304 
305 	l1 = ci->ci_isources[ipl]->ipl_evt_mask1;
306 	ci->ci_isources[ipl]->ipl_evt_mask1 = 0;
307 	while ((l1i = ffs(l1)) != 0) {
308 		l1i--;
309 		l1 &= ~(1 << l1i);
310 		l2 = ci->ci_isources[ipl]->ipl_evt_mask2[l1i];
311 		ci->ci_isources[ipl]->ipl_evt_mask2[l1i] = 0;
312 		while ((l2i = ffs(l2)) != 0) {
313 			int evtch;
314 
315 			l2i--;
316 			l2 &= ~(1 << l2i);
317 
318 			evtch = (l1i << 5) + l2i;
319 			hypervisor_enable_event(evtch);
320 		}
321 	}
322 }
323 
324 void
325 hypervisor_set_ipending(u_int32_t iplmask, int l1, int l2)
326 {
327 	int ipl;
328 	struct cpu_info *ci = curcpu();
329 
330 	/* set pending bit for the appropriate IPLs */
331 	ci->ci_ipending |= iplmask;
332 
333 	/*
334 	 * And set event pending bit for the lowest IPL. As IPL are handled
335 	 * from high to low, this ensure that all callbacks will have been
336 	 * called when we ack the event
337 	 */
338 	ipl = ffs(iplmask);
339 	KASSERT(ipl > 0);
340 	ipl--;
341 	ci->ci_isources[ipl]->ipl_evt_mask1 |= 1 << l1;
342 	ci->ci_isources[ipl]->ipl_evt_mask2[l1] |= 1 << l2;
343 }
344