1 /* $NetBSD: hypervisor_machdep.c,v 1.11 2008/10/21 15:46:32 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 1.11 2008/10/21 15:46:32 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 #ifdef XEN3 81 /* 82 * arch-dependent p2m frame lists list (L3 and L2) 83 * used by Xen for save/restore mappings 84 */ 85 static unsigned long * l3_p2m_page; 86 static unsigned long * l2_p2m_page; 87 static int l2_p2m_page_size; /* size of L2 page, in pages */ 88 89 static void build_p2m_frame_list_list(void); 90 static void update_p2m_frame_list_list(void); 91 #endif /* XEN3 */ 92 93 // #define PORT_DEBUG 4 94 // #define EARLY_DEBUG_EVENT 95 96 int stipending(void); 97 int 98 stipending(void) 99 { 100 unsigned long l1; 101 unsigned long l2; 102 unsigned int l1i, l2i, port; 103 volatile shared_info_t *s = HYPERVISOR_shared_info; 104 struct cpu_info *ci; 105 volatile struct vcpu_info *vci; 106 int ret; 107 108 ret = 0; 109 ci = curcpu(); 110 vci = ci->ci_vcpu; 111 112 #if 0 113 if (HYPERVISOR_shared_info->events) 114 printf("stipending events %08lx mask %08lx ilevel %d\n", 115 HYPERVISOR_shared_info->events, 116 HYPERVISOR_shared_info->events_mask, ci->ci_ilevel); 117 #endif 118 119 #ifdef EARLY_DEBUG_EVENT 120 if (xen_atomic_test_bit(&s->evtchn_pending[0], debug_port)) { 121 xen_debug_handler(NULL); 122 xen_atomic_clear_bit(&s->evtchn_pending[0], debug_port); 123 } 124 #endif 125 126 /* 127 * we're only called after STIC, so we know that we'll have to 128 * STI at the end 129 */ 130 while (vci->evtchn_upcall_pending) { 131 cli(); 132 vci->evtchn_upcall_pending = 0; 133 /* NB. No need for a barrier here -- XCHG is a barrier 134 * on x86. */ 135 #ifdef XEN3 136 l1 = xen_atomic_xchg(&vci->evtchn_pending_sel, 0); 137 #else 138 l1 = xen_atomic_xchg(&s->evtchn_pending_sel, 0); 139 #endif 140 while ((l1i = xen_ffs(l1)) != 0) { 141 l1i--; 142 l1 &= ~(1UL << l1i); 143 144 l2 = s->evtchn_pending[l1i] & ~s->evtchn_mask[l1i]; 145 /* 146 * mask and clear event. More efficient than calling 147 * hypervisor_mask/clear_event for each event. 148 */ 149 xen_atomic_setbits_l(&s->evtchn_mask[l1i], l2); 150 xen_atomic_clearbits_l(&s->evtchn_pending[l1i], l2); 151 while ((l2i = xen_ffs(l2)) != 0) { 152 l2i--; 153 l2 &= ~(1UL << l2i); 154 155 port = (l1i << LONG_SHIFT) + l2i; 156 if (evtsource[port]) { 157 hypervisor_set_ipending( 158 evtsource[port]->ev_imask, 159 l1i, l2i); 160 evtsource[port]->ev_evcnt.ev_count++; 161 if (ret == 0 && ci->ci_ilevel < 162 evtsource[port]->ev_maxlevel) 163 ret = 1; 164 } 165 #ifdef DOM0OPS 166 else { 167 /* set pending event */ 168 xenevt_setipending(l1i, l2i); 169 } 170 #endif 171 } 172 } 173 sti(); 174 } 175 176 #if 0 177 if (ci->ci_ipending & 0x1) 178 printf("stipending events %08lx mask %08lx ilevel %d ipending %08x\n", 179 HYPERVISOR_shared_info->events, 180 HYPERVISOR_shared_info->events_mask, ci->ci_ilevel, 181 ci->ci_ipending); 182 #endif 183 184 return (ret); 185 } 186 187 void 188 do_hypervisor_callback(struct intrframe *regs) 189 { 190 unsigned long l1; 191 unsigned long l2; 192 unsigned int l1i, l2i, port; 193 volatile shared_info_t *s = HYPERVISOR_shared_info; 194 struct cpu_info *ci; 195 volatile struct vcpu_info *vci; 196 int level; 197 198 ci = curcpu(); 199 vci = ci->ci_vcpu; 200 level = ci->ci_ilevel; 201 202 // DDD printf("do_hypervisor_callback\n"); 203 204 #ifdef EARLY_DEBUG_EVENT 205 if (xen_atomic_test_bit(&s->evtchn_pending[0], debug_port)) { 206 xen_debug_handler(NULL); 207 xen_atomic_clear_bit(&s->evtchn_pending[0], debug_port); 208 } 209 #endif 210 211 while (vci->evtchn_upcall_pending) { 212 vci->evtchn_upcall_pending = 0; 213 /* NB. No need for a barrier here -- XCHG is a barrier 214 * on x86. */ 215 #ifdef XEN3 216 l1 = xen_atomic_xchg(&vci->evtchn_pending_sel, 0); 217 #else 218 l1 = xen_atomic_xchg(&s->evtchn_pending_sel, 0); 219 #endif 220 while ((l1i = xen_ffs(l1)) != 0) { 221 l1i--; 222 l1 &= ~(1UL << l1i); 223 224 l2 = s->evtchn_pending[l1i] & ~s->evtchn_mask[l1i]; 225 /* 226 * mask and clear the pending events. 227 * Doing it here for all event that will be processed 228 * avoids a race with stipending (which can be called 229 * though evtchn_do_event->splx) that could cause an 230 * event to be both processed and marked pending. 231 */ 232 xen_atomic_setbits_l(&s->evtchn_mask[l1i], l2); 233 xen_atomic_clearbits_l(&s->evtchn_pending[l1i], l2); 234 235 while ((l2i = xen_ffs(l2)) != 0) { 236 l2i--; 237 l2 &= ~(1UL << l2i); 238 239 port = (l1i << LONG_SHIFT) + l2i; 240 #ifdef PORT_DEBUG 241 if (port == PORT_DEBUG) 242 printf("do_hypervisor_callback event %d\n", port); 243 #endif 244 if (evtsource[port]) 245 call_evtchn_do_event(port, regs); 246 #ifdef DOM0OPS 247 else { 248 if (ci->ci_ilevel < IPL_HIGH) { 249 /* fast path */ 250 int oipl = ci->ci_ilevel; 251 ci->ci_ilevel = IPL_HIGH; 252 call_xenevt_event(port); 253 ci->ci_ilevel = oipl; 254 } else { 255 /* set pending event */ 256 xenevt_setipending(l1i, l2i); 257 } 258 } 259 #endif 260 } 261 } 262 } 263 264 #ifdef DIAGNOSTIC 265 if (level != ci->ci_ilevel) 266 printf("hypervisor done %08x level %d/%d ipending %08x\n", 267 #ifdef XEN3 268 (uint)vci->evtchn_pending_sel, 269 #else 270 (uint)HYPERVISOR_shared_info->evtchn_pending_sel, 271 #endif 272 level, ci->ci_ilevel, ci->ci_ipending); 273 #endif 274 } 275 276 void 277 hypervisor_unmask_event(unsigned int ev) 278 { 279 volatile shared_info_t *s = HYPERVISOR_shared_info; 280 volatile struct vcpu_info *vci = curcpu()->ci_vcpu; 281 282 #ifdef PORT_DEBUG 283 if (ev == PORT_DEBUG) 284 printf("hypervisor_unmask_event %d\n", ev); 285 #endif 286 287 xen_atomic_clear_bit(&s->evtchn_mask[0], ev); 288 /* 289 * The following is basically the equivalent of 290 * 'hw_resend_irq'. Just like a real IO-APIC we 'lose the 291 * interrupt edge' if the channel is masked. 292 */ 293 if (xen_atomic_test_bit(&s->evtchn_pending[0], ev) && 294 #ifdef XEN3 295 !xen_atomic_test_and_set_bit(&vci->evtchn_pending_sel, ev>>LONG_SHIFT)) { 296 #else 297 !xen_atomic_test_and_set_bit(&s->evtchn_pending_sel, ev>>LONG_SHIFT)) { 298 #endif 299 xen_atomic_set_bit(&vci->evtchn_upcall_pending, 0); 300 if (!vci->evtchn_upcall_mask) 301 hypervisor_force_callback(); 302 } 303 } 304 305 void 306 hypervisor_mask_event(unsigned int ev) 307 { 308 volatile shared_info_t *s = HYPERVISOR_shared_info; 309 #ifdef PORT_DEBUG 310 if (ev == PORT_DEBUG) 311 printf("hypervisor_mask_event %d\n", ev); 312 #endif 313 314 xen_atomic_set_bit(&s->evtchn_mask[0], ev); 315 } 316 317 void 318 hypervisor_clear_event(unsigned int ev) 319 { 320 volatile shared_info_t *s = HYPERVISOR_shared_info; 321 #ifdef PORT_DEBUG 322 if (ev == PORT_DEBUG) 323 printf("hypervisor_clear_event %d\n", ev); 324 #endif 325 326 xen_atomic_clear_bit(&s->evtchn_pending[0], ev); 327 } 328 329 void 330 hypervisor_enable_ipl(unsigned int ipl) 331 { 332 u_long l1, l2; 333 int l1i, l2i; 334 struct cpu_info *ci = curcpu(); 335 336 /* 337 * enable all events for ipl. As we only set an event in ipl_evt_mask 338 * for its lowest IPL, and pending IPLs are processed high to low, 339 * we know that all callback for this event have been processed. 340 */ 341 342 l1 = ci->ci_isources[ipl]->ipl_evt_mask1; 343 ci->ci_isources[ipl]->ipl_evt_mask1 = 0; 344 while ((l1i = xen_ffs(l1)) != 0) { 345 l1i--; 346 l1 &= ~(1UL << l1i); 347 l2 = ci->ci_isources[ipl]->ipl_evt_mask2[l1i]; 348 ci->ci_isources[ipl]->ipl_evt_mask2[l1i] = 0; 349 while ((l2i = xen_ffs(l2)) != 0) { 350 int evtch; 351 352 l2i--; 353 l2 &= ~(1UL << l2i); 354 355 evtch = (l1i << LONG_SHIFT) + l2i; 356 hypervisor_enable_event(evtch); 357 } 358 } 359 } 360 361 void 362 hypervisor_set_ipending(uint32_t iplmask, int l1, int l2) 363 { 364 int ipl; 365 struct cpu_info *ci = curcpu(); 366 367 /* set pending bit for the appropriate IPLs */ 368 ci->ci_ipending |= iplmask; 369 370 /* 371 * And set event pending bit for the lowest IPL. As IPL are handled 372 * from high to low, this ensure that all callbacks will have been 373 * called when we ack the event 374 */ 375 ipl = ffs(iplmask); 376 KASSERT(ipl > 0); 377 ipl--; 378 ci->ci_isources[ipl]->ipl_evt_mask1 |= 1UL << l1; 379 ci->ci_isources[ipl]->ipl_evt_mask2[l1] |= 1UL << l2; 380 } 381 382 void 383 hypervisor_machdep_attach(void) { 384 385 #ifdef XEN3 386 /* dom0 does not require the arch-dependent P2M translation table */ 387 if ( !xendomain_is_dom0() ) { 388 build_p2m_frame_list_list(); 389 } 390 #endif 391 392 } 393 394 #ifdef XEN3 395 /* 396 * Generate the p2m_frame_list_list table, 397 * needed for guest save/restore 398 */ 399 static void 400 build_p2m_frame_list_list(void) { 401 402 int fpp; /* number of page (frame) pointer per page */ 403 unsigned long max_pfn; 404 /* 405 * The p2m list is composed of three levels of indirection, 406 * each layer containing MFNs pointing to lower level pages 407 * The indirection is used to convert a given PFN to its MFN 408 * Each N level page can point to @fpp (N-1) level pages 409 * For example, for x86 32bit, we have: 410 * - PAGE_SIZE: 4096 bytes 411 * - fpp: 1024 (one L3 page can address 1024 L2 pages) 412 * A L1 page contains the list of MFN we are looking for 413 */ 414 max_pfn = xen_start_info.nr_pages; 415 fpp = PAGE_SIZE / sizeof(paddr_t); 416 417 /* we only need one L3 page */ 418 l3_p2m_page = kmem_alloc(PAGE_SIZE, KM_NOSLEEP); 419 if (l3_p2m_page == NULL) 420 panic("could not allocate memory for l3_p2m_page"); 421 422 /* 423 * Determine how many L2 pages we need for the mapping 424 * Each L2 can map a total of @fpp L1 pages 425 */ 426 l2_p2m_page_size = howmany(max_pfn, fpp); 427 428 l2_p2m_page = kmem_alloc(l2_p2m_page_size * PAGE_SIZE, KM_NOSLEEP); 429 if (l2_p2m_page == NULL) 430 panic("could not allocate memory for l2_p2m_page"); 431 432 /* We now have L3 and L2 pages ready, update L1 mapping */ 433 update_p2m_frame_list_list(); 434 435 } 436 437 /* 438 * Update the L1 p2m_frame_list_list mapping (during guest boot or resume) 439 */ 440 static void 441 update_p2m_frame_list_list(void) { 442 443 int i; 444 int fpp; /* number of page (frame) pointer per page */ 445 unsigned long max_pfn; 446 447 max_pfn = xen_start_info.nr_pages; 448 fpp = PAGE_SIZE / sizeof(paddr_t); 449 450 for (i = 0; i < l2_p2m_page_size; i++) { 451 /* 452 * Each time we start a new L2 page, 453 * store its MFN in the L3 page 454 */ 455 if ((i % fpp) == 0) { 456 l3_p2m_page[i/fpp] = vtomfn( 457 (vaddr_t)&l2_p2m_page[i]); 458 } 459 /* 460 * we use a shortcut 461 * since @xpmap_phys_to_machine_mapping array 462 * already contains PFN to MFN mapping, we just 463 * set the l2_p2m_page MFN pointer to the MFN of the 464 * according frame of @xpmap_phys_to_machine_mapping 465 */ 466 l2_p2m_page[i] = vtomfn((vaddr_t) 467 &xpmap_phys_to_machine_mapping[i*fpp]); 468 } 469 470 HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list = 471 vtomfn((vaddr_t)l3_p2m_page); 472 HYPERVISOR_shared_info->arch.max_pfn = max_pfn; 473 474 } 475 #endif /* XEN3 */ 476