Home | History | Annotate | Line # | Download | only in x86
hypervisor_machdep.c revision 1.27.12.1
      1  1.27.12.1  martin /*	$NetBSD: hypervisor_machdep.c,v 1.27.12.1 2014/09/22 11:15:57 martin Exp $	*/
      2        1.2  bouyer 
      3        1.2  bouyer /*
      4        1.2  bouyer  *
      5        1.2  bouyer  * Copyright (c) 2004 Christian Limpach.
      6        1.2  bouyer  * All rights reserved.
      7        1.2  bouyer  *
      8        1.2  bouyer  * Redistribution and use in source and binary forms, with or without
      9        1.2  bouyer  * modification, are permitted provided that the following conditions
     10        1.2  bouyer  * are met:
     11        1.2  bouyer  * 1. Redistributions of source code must retain the above copyright
     12        1.2  bouyer  *    notice, this list of conditions and the following disclaimer.
     13        1.2  bouyer  * 2. Redistributions in binary form must reproduce the above copyright
     14        1.2  bouyer  *    notice, this list of conditions and the following disclaimer in the
     15        1.2  bouyer  *    documentation and/or other materials provided with the distribution.
     16        1.2  bouyer  *
     17        1.2  bouyer  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
     18        1.2  bouyer  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     19        1.2  bouyer  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     20        1.2  bouyer  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
     21        1.2  bouyer  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     22        1.2  bouyer  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     23        1.2  bouyer  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     24        1.2  bouyer  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     25        1.2  bouyer  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     26        1.2  bouyer  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     27        1.2  bouyer  */
     28        1.2  bouyer 
     29        1.2  bouyer /******************************************************************************
     30        1.2  bouyer  * hypervisor.c
     31        1.2  bouyer  *
     32        1.2  bouyer  * Communication to/from hypervisor.
     33        1.2  bouyer  *
     34        1.2  bouyer  * Copyright (c) 2002-2004, K A Fraser
     35        1.2  bouyer  *
     36        1.2  bouyer  * Permission is hereby granted, free of charge, to any person obtaining a copy
     37        1.2  bouyer  * of this software and associated documentation files (the "Software"), to
     38        1.2  bouyer  * deal in the Software without restriction, including without limitation the
     39        1.2  bouyer  * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
     40        1.2  bouyer  * sell copies of the Software, and to permit persons to whom the Software is
     41        1.2  bouyer  * furnished to do so, subject to the following conditions:
     42        1.2  bouyer  *
     43        1.2  bouyer  * The above copyright notice and this permission notice shall be included in
     44        1.2  bouyer  * all copies or substantial portions of the Software.
     45        1.2  bouyer  *
     46        1.2  bouyer  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
     47        1.2  bouyer  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
     48        1.2  bouyer  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
     49        1.2  bouyer  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
     50        1.2  bouyer  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
     51        1.2  bouyer  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
     52        1.2  bouyer  * DEALINGS IN THE SOFTWARE.
     53        1.2  bouyer  */
     54        1.2  bouyer 
     55        1.2  bouyer 
     56        1.2  bouyer #include <sys/cdefs.h>
     57  1.27.12.1  martin __KERNEL_RCSID(0, "$NetBSD: hypervisor_machdep.c,v 1.27.12.1 2014/09/22 11:15:57 martin Exp $");
     58        1.2  bouyer 
     59        1.2  bouyer #include <sys/param.h>
     60        1.2  bouyer #include <sys/systm.h>
     61       1.10  bouyer #include <sys/kmem.h>
     62       1.10  bouyer 
     63       1.10  bouyer #include <uvm/uvm_extern.h>
     64       1.10  bouyer 
     65       1.10  bouyer #include <machine/vmparam.h>
     66       1.10  bouyer #include <machine/pmap.h>
     67        1.2  bouyer 
     68        1.2  bouyer #include <xen/xen.h>
     69        1.2  bouyer #include <xen/hypervisor.h>
     70        1.2  bouyer #include <xen/evtchn.h>
     71       1.10  bouyer #include <xen/xenpmap.h>
     72        1.2  bouyer 
     73        1.2  bouyer #include "opt_xen.h"
     74        1.2  bouyer 
     75       1.10  bouyer /*
     76       1.10  bouyer  * arch-dependent p2m frame lists list (L3 and L2)
     77       1.10  bouyer  * used by Xen for save/restore mappings
     78       1.10  bouyer  */
     79       1.10  bouyer static unsigned long * l3_p2m_page;
     80       1.10  bouyer static unsigned long * l2_p2m_page;
     81       1.10  bouyer static int l2_p2m_page_size; /* size of L2 page, in pages */
     82       1.10  bouyer 
     83       1.10  bouyer static void build_p2m_frame_list_list(void);
     84       1.10  bouyer static void update_p2m_frame_list_list(void);
     85       1.10  bouyer 
     86        1.2  bouyer // #define PORT_DEBUG 4
     87        1.2  bouyer // #define EARLY_DEBUG_EVENT
     88        1.2  bouyer 
     89       1.15  cherry /* callback function type */
     90       1.27  bouyer typedef void (*iterate_func_t)(unsigned int, unsigned int,
     91       1.27  bouyer 			       unsigned int, void *);
     92       1.15  cherry 
     93       1.15  cherry static inline void
     94       1.27  bouyer evt_iterate_bits(volatile unsigned long *pendingl1,
     95       1.15  cherry 		 volatile unsigned long *pendingl2,
     96       1.15  cherry 		 volatile unsigned long *mask,
     97       1.15  cherry 		 iterate_func_t iterate_pending, void *iterate_args)
     98       1.15  cherry {
     99       1.15  cherry 
    100       1.15  cherry 	KASSERT(pendingl1 != NULL);
    101       1.15  cherry 	KASSERT(pendingl2 != NULL);
    102       1.15  cherry 
    103       1.15  cherry 	unsigned long l1, l2;
    104       1.15  cherry 	unsigned int l1i, l2i, port;
    105       1.15  cherry 
    106       1.15  cherry 	l1 = xen_atomic_xchg(pendingl1, 0);
    107       1.15  cherry 	while ((l1i = xen_ffs(l1)) != 0) {
    108       1.15  cherry 		l1i--;
    109       1.15  cherry 		l1 &= ~(1UL << l1i);
    110       1.15  cherry 
    111       1.15  cherry 		l2 = pendingl2[l1i] & (mask != NULL ? ~mask[l1i] : -1UL);
    112       1.27  bouyer 		l2 &= curcpu()->ci_evtmask[l1i];
    113       1.15  cherry 
    114       1.15  cherry 		if (mask != NULL) xen_atomic_setbits_l(&mask[l1i], l2);
    115       1.15  cherry 		xen_atomic_clearbits_l(&pendingl2[l1i], l2);
    116       1.15  cherry 
    117       1.15  cherry 		while ((l2i = xen_ffs(l2)) != 0) {
    118       1.15  cherry 			l2i--;
    119       1.15  cherry 			l2 &= ~(1UL << l2i);
    120       1.15  cherry 
    121       1.15  cherry 			port = (l1i << LONG_SHIFT) + l2i;
    122       1.15  cherry 
    123       1.27  bouyer 			iterate_pending(port, l1i, l2i, iterate_args);
    124       1.15  cherry 		}
    125       1.15  cherry 	}
    126       1.15  cherry }
    127       1.15  cherry 
    128       1.15  cherry /*
    129       1.15  cherry  * Set per-cpu "pending" information for outstanding events that
    130       1.15  cherry  * cannot be processed now.
    131       1.15  cherry  */
    132       1.15  cherry 
    133       1.15  cherry static inline void
    134       1.27  bouyer evt_set_pending(unsigned int port, unsigned int l1i,
    135       1.15  cherry 		unsigned int l2i, void *args)
    136       1.15  cherry {
    137       1.15  cherry 
    138       1.15  cherry 	KASSERT(args != NULL);
    139       1.15  cherry 
    140       1.15  cherry 	int *ret = args;
    141       1.15  cherry 
    142       1.15  cherry 	if (evtsource[port]) {
    143       1.27  bouyer 		hypervisor_set_ipending(evtsource[port]->ev_imask, l1i, l2i);
    144       1.15  cherry 		evtsource[port]->ev_evcnt.ev_count++;
    145       1.27  bouyer 		if (*ret == 0 && curcpu()->ci_ilevel <
    146       1.15  cherry 		    evtsource[port]->ev_maxlevel)
    147       1.15  cherry 			*ret = 1;
    148       1.15  cherry 	}
    149       1.15  cherry #ifdef DOM0OPS
    150       1.15  cherry 	else  {
    151       1.15  cherry 		/* set pending event */
    152       1.15  cherry 		xenevt_setipending(l1i, l2i);
    153       1.15  cherry 	}
    154       1.15  cherry #endif
    155       1.15  cherry }
    156       1.15  cherry 
    157        1.2  bouyer int stipending(void);
    158        1.2  bouyer int
    159        1.7  cegger stipending(void)
    160        1.2  bouyer {
    161        1.2  bouyer 	volatile shared_info_t *s = HYPERVISOR_shared_info;
    162        1.2  bouyer 	struct cpu_info *ci;
    163        1.8  cegger 	volatile struct vcpu_info *vci;
    164        1.2  bouyer 	int ret;
    165        1.2  bouyer 
    166        1.2  bouyer 	ret = 0;
    167        1.2  bouyer 	ci = curcpu();
    168        1.8  cegger 	vci = ci->ci_vcpu;
    169        1.2  bouyer 
    170        1.2  bouyer #if 0
    171        1.2  bouyer 	if (HYPERVISOR_shared_info->events)
    172        1.2  bouyer 		printf("stipending events %08lx mask %08lx ilevel %d\n",
    173        1.2  bouyer 		    HYPERVISOR_shared_info->events,
    174        1.2  bouyer 		    HYPERVISOR_shared_info->events_mask, ci->ci_ilevel);
    175        1.2  bouyer #endif
    176        1.2  bouyer 
    177        1.2  bouyer #ifdef EARLY_DEBUG_EVENT
    178        1.2  bouyer 	if (xen_atomic_test_bit(&s->evtchn_pending[0], debug_port)) {
    179        1.2  bouyer 		xen_debug_handler(NULL);
    180        1.2  bouyer 		xen_atomic_clear_bit(&s->evtchn_pending[0], debug_port);
    181        1.2  bouyer 	}
    182        1.2  bouyer #endif
    183        1.2  bouyer 
    184        1.2  bouyer 	/*
    185        1.2  bouyer 	 * we're only called after STIC, so we know that we'll have to
    186        1.2  bouyer 	 * STI at the end
    187        1.2  bouyer 	 */
    188       1.15  cherry 
    189        1.8  cegger 	while (vci->evtchn_upcall_pending) {
    190       1.25  bouyer 		cli();
    191       1.15  cherry 
    192        1.8  cegger 		vci->evtchn_upcall_pending = 0;
    193       1.15  cherry 
    194       1.27  bouyer 		evt_iterate_bits(&vci->evtchn_pending_sel,
    195       1.15  cherry 		    s->evtchn_pending, s->evtchn_mask,
    196       1.15  cherry 		    evt_set_pending, &ret);
    197       1.15  cherry 
    198       1.25  bouyer 		sti();
    199        1.2  bouyer 	}
    200        1.2  bouyer 
    201        1.2  bouyer #if 0
    202        1.2  bouyer 	if (ci->ci_ipending & 0x1)
    203        1.2  bouyer 		printf("stipending events %08lx mask %08lx ilevel %d ipending %08x\n",
    204        1.2  bouyer 		    HYPERVISOR_shared_info->events,
    205        1.2  bouyer 		    HYPERVISOR_shared_info->events_mask, ci->ci_ilevel,
    206        1.2  bouyer 		    ci->ci_ipending);
    207        1.2  bouyer #endif
    208        1.2  bouyer 
    209        1.2  bouyer 	return (ret);
    210        1.2  bouyer }
    211        1.2  bouyer 
    212       1.15  cherry /* Iterate through pending events and call the event handler */
    213       1.15  cherry 
    214       1.15  cherry static inline void
    215       1.27  bouyer evt_do_hypervisor_callback(unsigned int port, unsigned int l1i,
    216       1.27  bouyer 			   unsigned int l2i, void *args)
    217       1.15  cherry {
    218       1.15  cherry 	KASSERT(args != NULL);
    219       1.15  cherry 
    220       1.27  bouyer 	struct cpu_info *ci = curcpu();
    221       1.15  cherry 	struct intrframe *regs = args;
    222       1.15  cherry 
    223       1.15  cherry #ifdef PORT_DEBUG
    224       1.15  cherry 	if (port == PORT_DEBUG)
    225       1.15  cherry 		printf("do_hypervisor_callback event %d\n", port);
    226       1.15  cherry #endif
    227       1.22  cherry 	if (evtsource[port]) {
    228       1.22  cherry 		ci->ci_idepth++;
    229       1.22  cherry 		evtchn_do_event(port, regs);
    230       1.22  cherry 		ci->ci_idepth--;
    231       1.22  cherry 	}
    232       1.15  cherry #ifdef DOM0OPS
    233       1.15  cherry 	else  {
    234       1.15  cherry 		if (ci->ci_ilevel < IPL_HIGH) {
    235       1.15  cherry 			/* fast path */
    236       1.15  cherry 			int oipl = ci->ci_ilevel;
    237       1.15  cherry 			ci->ci_ilevel = IPL_HIGH;
    238       1.22  cherry 			ci->ci_idepth++;
    239       1.22  cherry 			xenevt_event(port);
    240       1.22  cherry 			ci->ci_idepth--;
    241       1.15  cherry 			ci->ci_ilevel = oipl;
    242       1.15  cherry 		} else {
    243       1.15  cherry 			/* set pending event */
    244       1.15  cherry 			xenevt_setipending(l1i, l2i);
    245       1.15  cherry 		}
    246       1.15  cherry 	}
    247       1.15  cherry #endif
    248       1.15  cherry }
    249       1.15  cherry 
    250        1.2  bouyer void
    251        1.2  bouyer do_hypervisor_callback(struct intrframe *regs)
    252        1.2  bouyer {
    253        1.2  bouyer 	volatile shared_info_t *s = HYPERVISOR_shared_info;
    254        1.2  bouyer 	struct cpu_info *ci;
    255        1.8  cegger 	volatile struct vcpu_info *vci;
    256  1.27.12.1  martin 	int level __diagused;
    257        1.2  bouyer 
    258        1.2  bouyer 	ci = curcpu();
    259        1.8  cegger 	vci = ci->ci_vcpu;
    260        1.2  bouyer 	level = ci->ci_ilevel;
    261        1.2  bouyer 
    262        1.2  bouyer 	// DDD printf("do_hypervisor_callback\n");
    263        1.2  bouyer 
    264        1.2  bouyer #ifdef EARLY_DEBUG_EVENT
    265        1.2  bouyer 	if (xen_atomic_test_bit(&s->evtchn_pending[0], debug_port)) {
    266        1.2  bouyer 		xen_debug_handler(NULL);
    267        1.2  bouyer 		xen_atomic_clear_bit(&s->evtchn_pending[0], debug_port);
    268        1.2  bouyer 	}
    269        1.2  bouyer #endif
    270        1.2  bouyer 
    271        1.8  cegger 	while (vci->evtchn_upcall_pending) {
    272        1.8  cegger 		vci->evtchn_upcall_pending = 0;
    273        1.2  bouyer 
    274       1.27  bouyer 		evt_iterate_bits(&vci->evtchn_pending_sel,
    275       1.15  cherry 		    s->evtchn_pending, s->evtchn_mask,
    276       1.15  cherry 		    evt_do_hypervisor_callback, regs);
    277        1.2  bouyer 	}
    278        1.2  bouyer 
    279        1.2  bouyer #ifdef DIAGNOSTIC
    280        1.2  bouyer 	if (level != ci->ci_ilevel)
    281        1.2  bouyer 		printf("hypervisor done %08x level %d/%d ipending %08x\n",
    282        1.8  cegger 		    (uint)vci->evtchn_pending_sel,
    283        1.2  bouyer 		    level, ci->ci_ilevel, ci->ci_ipending);
    284        1.2  bouyer #endif
    285        1.2  bouyer }
    286        1.2  bouyer 
    287        1.2  bouyer void
    288       1.17  cherry hypervisor_send_event(struct cpu_info *ci, unsigned int ev)
    289       1.17  cherry {
    290       1.17  cherry 	KASSERT(ci != NULL);
    291       1.17  cherry 
    292       1.17  cherry 	volatile shared_info_t *s = HYPERVISOR_shared_info;
    293       1.17  cherry 	volatile struct vcpu_info *vci = ci->ci_vcpu;
    294       1.17  cherry 
    295       1.17  cherry #ifdef PORT_DEBUG
    296       1.17  cherry 	if (ev == PORT_DEBUG)
    297       1.17  cherry 		printf("hypervisor_send_event %d\n", ev);
    298       1.17  cherry #endif
    299       1.17  cherry 
    300       1.17  cherry 	xen_atomic_set_bit(&s->evtchn_pending[0], ev);
    301       1.17  cherry 
    302       1.21  cherry 	if (__predict_false(ci == curcpu())) {
    303       1.20  cherry 		xen_atomic_set_bit(&vci->evtchn_pending_sel,
    304       1.20  cherry 		    ev >> LONG_SHIFT);
    305       1.20  cherry 		xen_atomic_set_bit(&vci->evtchn_upcall_pending, 0);
    306       1.20  cherry 	}
    307       1.17  cherry 
    308       1.17  cherry 	xen_atomic_clear_bit(&s->evtchn_mask[0], ev);
    309       1.17  cherry 
    310       1.17  cherry 	if (__predict_true(ci == curcpu())) {
    311       1.17  cherry 		hypervisor_force_callback();
    312       1.17  cherry 	} else {
    313       1.18  bouyer 		if (__predict_false(xen_send_ipi(ci, XEN_IPI_HVCB))) {
    314       1.18  bouyer 			panic("xen_send_ipi(cpu%d, XEN_IPI_HVCB) failed\n",
    315       1.18  bouyer 			    (int) ci->ci_cpuid);
    316       1.17  cherry 		}
    317       1.17  cherry 	}
    318       1.17  cherry }
    319       1.17  cherry 
    320       1.17  cherry void
    321        1.2  bouyer hypervisor_unmask_event(unsigned int ev)
    322        1.2  bouyer {
    323        1.2  bouyer 	volatile shared_info_t *s = HYPERVISOR_shared_info;
    324       1.18  bouyer 	CPU_INFO_ITERATOR cii;
    325       1.18  bouyer 	struct cpu_info *ci;
    326       1.18  bouyer 	volatile struct vcpu_info *vci;
    327        1.8  cegger 
    328        1.2  bouyer #ifdef PORT_DEBUG
    329        1.2  bouyer 	if (ev == PORT_DEBUG)
    330        1.2  bouyer 		printf("hypervisor_unmask_event %d\n", ev);
    331        1.2  bouyer #endif
    332        1.2  bouyer 
    333        1.2  bouyer 	xen_atomic_clear_bit(&s->evtchn_mask[0], ev);
    334        1.2  bouyer 	/*
    335        1.2  bouyer 	 * The following is basically the equivalent of
    336        1.2  bouyer 	 * 'hw_resend_irq'. Just like a real IO-APIC we 'lose the
    337        1.2  bouyer 	 * interrupt edge' if the channel is masked.
    338        1.2  bouyer 	 */
    339       1.18  bouyer 	if (!xen_atomic_test_bit(&s->evtchn_pending[0], ev))
    340       1.18  bouyer 		return;
    341       1.18  bouyer 
    342       1.18  bouyer 	for (CPU_INFO_FOREACH(cii, ci)) {
    343       1.18  bouyer 		if (!xen_atomic_test_bit(&ci->ci_evtmask[0], ev))
    344       1.18  bouyer 			continue;
    345       1.18  bouyer 		vci = ci->ci_vcpu;
    346       1.19  cherry 		if (__predict_true(ci == curcpu())) {
    347       1.19  cherry 			if (!xen_atomic_test_and_set_bit(&vci->evtchn_pending_sel,
    348       1.19  cherry 				ev>>LONG_SHIFT))
    349       1.19  cherry 				xen_atomic_set_bit(&vci->evtchn_upcall_pending, 0);
    350       1.19  cherry 		}
    351       1.18  bouyer 		if (!vci->evtchn_upcall_mask) {
    352       1.18  bouyer 			if (__predict_true(ci == curcpu())) {
    353       1.18  bouyer 				hypervisor_force_callback();
    354       1.18  bouyer 			} else {
    355       1.18  bouyer 				if (__predict_false(
    356       1.18  bouyer 				    xen_send_ipi(ci, XEN_IPI_HVCB))) {
    357       1.18  bouyer 					panic("xen_send_ipi(cpu%d, "
    358       1.18  bouyer 					    "XEN_IPI_HVCB) failed\n",
    359       1.18  bouyer 					    (int) ci->ci_cpuid);
    360       1.18  bouyer 				}
    361       1.18  bouyer 			}
    362       1.18  bouyer 		}
    363        1.2  bouyer 	}
    364        1.2  bouyer }
    365        1.2  bouyer 
    366        1.2  bouyer void
    367        1.2  bouyer hypervisor_mask_event(unsigned int ev)
    368        1.2  bouyer {
    369        1.2  bouyer 	volatile shared_info_t *s = HYPERVISOR_shared_info;
    370        1.2  bouyer #ifdef PORT_DEBUG
    371        1.2  bouyer 	if (ev == PORT_DEBUG)
    372        1.2  bouyer 		printf("hypervisor_mask_event %d\n", ev);
    373        1.2  bouyer #endif
    374        1.2  bouyer 
    375        1.2  bouyer 	xen_atomic_set_bit(&s->evtchn_mask[0], ev);
    376        1.2  bouyer }
    377        1.2  bouyer 
    378        1.2  bouyer void
    379        1.2  bouyer hypervisor_clear_event(unsigned int ev)
    380        1.2  bouyer {
    381        1.2  bouyer 	volatile shared_info_t *s = HYPERVISOR_shared_info;
    382        1.2  bouyer #ifdef PORT_DEBUG
    383        1.2  bouyer 	if (ev == PORT_DEBUG)
    384        1.2  bouyer 		printf("hypervisor_clear_event %d\n", ev);
    385        1.2  bouyer #endif
    386        1.2  bouyer 
    387        1.2  bouyer 	xen_atomic_clear_bit(&s->evtchn_pending[0], ev);
    388        1.2  bouyer }
    389        1.2  bouyer 
    390       1.15  cherry static inline void
    391       1.27  bouyer evt_enable_event(unsigned int port, unsigned int l1i,
    392       1.27  bouyer 		 unsigned int l2i, void *args)
    393       1.15  cherry {
    394       1.15  cherry 	KASSERT(args == NULL);
    395       1.15  cherry 	hypervisor_enable_event(port);
    396       1.15  cherry }
    397       1.15  cherry 
    398        1.2  bouyer void
    399        1.2  bouyer hypervisor_enable_ipl(unsigned int ipl)
    400        1.2  bouyer {
    401        1.2  bouyer 	struct cpu_info *ci = curcpu();
    402        1.2  bouyer 
    403        1.2  bouyer 	/*
    404        1.2  bouyer 	 * enable all events for ipl. As we only set an event in ipl_evt_mask
    405        1.2  bouyer 	 * for its lowest IPL, and pending IPLs are processed high to low,
    406        1.2  bouyer 	 * we know that all callback for this event have been processed.
    407        1.2  bouyer 	 */
    408        1.2  bouyer 
    409       1.27  bouyer 	evt_iterate_bits(&ci->ci_isources[ipl]->ipl_evt_mask1,
    410       1.15  cherry 	    ci->ci_isources[ipl]->ipl_evt_mask2, NULL,
    411       1.15  cherry 	    evt_enable_event, NULL);
    412        1.2  bouyer 
    413        1.2  bouyer }
    414        1.2  bouyer 
    415        1.2  bouyer void
    416       1.27  bouyer hypervisor_set_ipending(uint32_t iplmask, int l1, int l2)
    417        1.2  bouyer {
    418       1.27  bouyer 
    419       1.27  bouyer 	/* This function is not re-entrant */
    420       1.27  bouyer 	KASSERT(x86_read_psl() != 0);
    421       1.27  bouyer 
    422        1.2  bouyer 	int ipl;
    423       1.27  bouyer 	struct cpu_info *ci = curcpu();
    424        1.2  bouyer 
    425        1.2  bouyer 	/* set pending bit for the appropriate IPLs */
    426        1.2  bouyer 	ci->ci_ipending |= iplmask;
    427        1.2  bouyer 
    428        1.2  bouyer 	/*
    429        1.2  bouyer 	 * And set event pending bit for the lowest IPL. As IPL are handled
    430        1.2  bouyer 	 * from high to low, this ensure that all callbacks will have been
    431        1.2  bouyer 	 * called when we ack the event
    432        1.2  bouyer 	 */
    433        1.2  bouyer 	ipl = ffs(iplmask);
    434        1.2  bouyer 	KASSERT(ipl > 0);
    435        1.2  bouyer 	ipl--;
    436       1.15  cherry 	KASSERT(ipl < NIPL);
    437       1.15  cherry 	KASSERT(ci->ci_isources[ipl] != NULL);
    438        1.6  bouyer 	ci->ci_isources[ipl]->ipl_evt_mask1 |= 1UL << l1;
    439        1.6  bouyer 	ci->ci_isources[ipl]->ipl_evt_mask2[l1] |= 1UL << l2;
    440       1.25  bouyer 	if (__predict_false(ci != curcpu())) {
    441       1.25  bouyer 		if (xen_send_ipi(ci, XEN_IPI_HVCB)) {
    442       1.25  bouyer 			panic("hypervisor_set_ipending: "
    443       1.25  bouyer 			    "xen_send_ipi(cpu%d, XEN_IPI_HVCB) failed\n",
    444       1.25  bouyer 			    (int) ci->ci_cpuid);
    445       1.25  bouyer 		}
    446       1.25  bouyer 	}
    447        1.2  bouyer }
    448       1.10  bouyer 
    449       1.10  bouyer void
    450       1.12  cegger hypervisor_machdep_attach(void)
    451       1.12  cegger {
    452       1.10  bouyer  	/* dom0 does not require the arch-dependent P2M translation table */
    453       1.16     jym 	if (!xendomain_is_dom0()) {
    454       1.10  bouyer 		build_p2m_frame_list_list();
    455       1.16     jym 		sysctl_xen_suspend_setup();
    456       1.10  bouyer 	}
    457       1.10  bouyer }
    458       1.10  bouyer 
    459       1.16     jym void
    460       1.16     jym hypervisor_machdep_resume(void)
    461       1.16     jym {
    462       1.16     jym 	/* dom0 does not require the arch-dependent P2M translation table */
    463       1.16     jym 	if (!xendomain_is_dom0())
    464       1.16     jym 		update_p2m_frame_list_list();
    465       1.16     jym }
    466       1.16     jym 
    467       1.10  bouyer /*
    468       1.10  bouyer  * Generate the p2m_frame_list_list table,
    469       1.10  bouyer  * needed for guest save/restore
    470       1.10  bouyer  */
    471       1.10  bouyer static void
    472       1.12  cegger build_p2m_frame_list_list(void)
    473       1.12  cegger {
    474       1.10  bouyer         int fpp; /* number of page (frame) pointer per page */
    475       1.10  bouyer         unsigned long max_pfn;
    476       1.10  bouyer         /*
    477       1.10  bouyer          * The p2m list is composed of three levels of indirection,
    478       1.10  bouyer          * each layer containing MFNs pointing to lower level pages
    479       1.10  bouyer          * The indirection is used to convert a given PFN to its MFN
    480       1.10  bouyer          * Each N level page can point to @fpp (N-1) level pages
    481       1.10  bouyer          * For example, for x86 32bit, we have:
    482       1.10  bouyer          * - PAGE_SIZE: 4096 bytes
    483       1.10  bouyer          * - fpp: 1024 (one L3 page can address 1024 L2 pages)
    484       1.10  bouyer          * A L1 page contains the list of MFN we are looking for
    485       1.10  bouyer          */
    486       1.10  bouyer         max_pfn = xen_start_info.nr_pages;
    487       1.14     jym         fpp = PAGE_SIZE / sizeof(xen_pfn_t);
    488       1.10  bouyer 
    489       1.10  bouyer         /* we only need one L3 page */
    490       1.14     jym         l3_p2m_page = (vaddr_t *)uvm_km_alloc(kernel_map, PAGE_SIZE,
    491       1.14     jym 	    PAGE_SIZE, UVM_KMF_WIRED | UVM_KMF_NOWAIT);
    492       1.10  bouyer         if (l3_p2m_page == NULL)
    493       1.10  bouyer                 panic("could not allocate memory for l3_p2m_page");
    494       1.10  bouyer 
    495       1.10  bouyer         /*
    496       1.10  bouyer          * Determine how many L2 pages we need for the mapping
    497       1.10  bouyer          * Each L2 can map a total of @fpp L1 pages
    498       1.10  bouyer          */
    499       1.10  bouyer         l2_p2m_page_size = howmany(max_pfn, fpp);
    500       1.10  bouyer 
    501       1.14     jym         l2_p2m_page = (vaddr_t *)uvm_km_alloc(kernel_map,
    502       1.14     jym 	    l2_p2m_page_size * PAGE_SIZE,
    503       1.14     jym 	    PAGE_SIZE, UVM_KMF_WIRED | UVM_KMF_NOWAIT);
    504       1.10  bouyer         if (l2_p2m_page == NULL)
    505       1.10  bouyer                 panic("could not allocate memory for l2_p2m_page");
    506       1.10  bouyer 
    507       1.10  bouyer         /* We now have L3 and L2 pages ready, update L1 mapping */
    508       1.10  bouyer         update_p2m_frame_list_list();
    509       1.10  bouyer 
    510       1.10  bouyer }
    511       1.10  bouyer 
    512       1.10  bouyer /*
    513       1.10  bouyer  * Update the L1 p2m_frame_list_list mapping (during guest boot or resume)
    514       1.10  bouyer  */
    515       1.10  bouyer static void
    516       1.12  cegger update_p2m_frame_list_list(void)
    517       1.12  cegger {
    518       1.10  bouyer         int i;
    519       1.10  bouyer         int fpp; /* number of page (frame) pointer per page */
    520       1.10  bouyer         unsigned long max_pfn;
    521       1.10  bouyer 
    522       1.10  bouyer         max_pfn = xen_start_info.nr_pages;
    523       1.14     jym         fpp = PAGE_SIZE / sizeof(xen_pfn_t);
    524       1.10  bouyer 
    525       1.10  bouyer         for (i = 0; i < l2_p2m_page_size; i++) {
    526       1.10  bouyer                 /*
    527       1.10  bouyer                  * Each time we start a new L2 page,
    528       1.10  bouyer                  * store its MFN in the L3 page
    529       1.10  bouyer                  */
    530       1.10  bouyer                 if ((i % fpp) == 0) {
    531       1.10  bouyer                         l3_p2m_page[i/fpp] = vtomfn(
    532       1.10  bouyer                                 (vaddr_t)&l2_p2m_page[i]);
    533       1.10  bouyer                 }
    534       1.10  bouyer                 /*
    535       1.10  bouyer                  * we use a shortcut
    536       1.10  bouyer                  * since @xpmap_phys_to_machine_mapping array
    537       1.10  bouyer                  * already contains PFN to MFN mapping, we just
    538       1.10  bouyer                  * set the l2_p2m_page MFN pointer to the MFN of the
    539       1.10  bouyer                  * according frame of @xpmap_phys_to_machine_mapping
    540       1.10  bouyer                  */
    541       1.10  bouyer                 l2_p2m_page[i] = vtomfn((vaddr_t)
    542       1.10  bouyer                         &xpmap_phys_to_machine_mapping[i*fpp]);
    543       1.10  bouyer         }
    544       1.10  bouyer 
    545       1.10  bouyer         HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
    546       1.10  bouyer                                         vtomfn((vaddr_t)l3_p2m_page);
    547       1.10  bouyer         HYPERVISOR_shared_info->arch.max_pfn = max_pfn;
    548       1.10  bouyer 
    549       1.10  bouyer }
    550