debuggers.hg: tools/libxc/xc_domain

debuggers.hg

annotate tools/libxc/xc_domain_restore.c @ 20983:a948403c8f99

Remus: increase failover timeout from 500ms to 1s

500ms is aggressive enough to trigger split-brain under fairly
ordinary workloads, particularly for HVM. The long-term fix is to
integrate with a real HA monitor like linux HA.

Signed-off-by: Brendan Cully <brendan@cs.ubc.ca>

author	Keir Fraser <keir.fraser@citrix.com>
date	Fri Feb 12 09:23:10 2010 +0000 (2010-02-12)
parents	fbe8f32fa257
children	779c0ef9682c

rev	line source
mjw@1661	1 /******************************************************************************
Tim@14782	2 * xc_domain_restore.c
kaf24@9698	3 *
Tim@14782	4 * Restore the state of a guest session.
kaf24@9698	5 *
mjw@1661	6 * Copyright (c) 2003, K A Fraser.
Tim@14782	7 * Copyright (c) 2006, Intel Corporation
Tim@14782	8 * Copyright (c) 2007, XenSource Inc.
Tim@14782	9 *
Tim@14782	10 * This program is free software; you can redistribute it and/or modify it
Tim@14782	11 * under the terms and conditions of the GNU General Public License,
Tim@14782	12 * version 2, as published by the Free Software Foundation.
Tim@14782	13 *
Tim@14782	14 * This program is distributed in the hope it will be useful, but WITHOUT
Tim@14782	15 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
Tim@14782	16 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
Tim@14782	17 * more details.
Tim@14782	18 *
Tim@14782	19 * You should have received a copy of the GNU General Public License along with
Tim@14782	20 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
Tim@14782	21 * Place - Suite 330, Boston, MA 02111-1307 USA.
Tim@14782	22 *
mjw@1661	23 */
mjw@1661	24
cl349@6427	25 #include <stdlib.h>
cl349@6427	26 #include <unistd.h>
smh22@7740	27
cl349@6427	28 #include "xg_private.h"
smh22@7740	29 #include "xg_save_restore.h"
keir@14138	30 #include "xc_dom.h"
smh22@7740	31
Tim@14782	32 #include <xen/hvm/ioreq.h>
Tim@14782	33 #include <xen/hvm/params.h>
Tim@14782	34
keir@20587	35 struct restore_ctx {
keir@20587	36 unsigned long max_mfn; /* max mfn of the current host machine */
keir@20587	37 unsigned long hvirt_start; /* virtual starting address of the hypervisor */
keir@20587	38 unsigned int pt_levels; /* #levels of page tables used by the current guest */
keir@20587	39 unsigned long nr_pfns; /* number of 'in use' pfns in the guest (i.e. #P2M entries with a valid mfn) */
keir@20587	40 xen_pfn_t live_p2m; / Live mapping of the table mapping each PFN to its current MFN. */
keir@20587	41 xen_pfn_t p2m; / A table mapping each PFN to its new MFN. */
keir@20587	42 unsigned no_superpage_mem; /* If have enough continuous memory for super page allocation */
keir@20589	43 struct domain_info_context dinfo;
keir@20587	44 };
steven@14732	45
smh22@7740	46 /*
keir@19677	47 **
keir@19677	48 **
keir@19677	49 */
keir@19677	50 #define SUPERPAGE_PFN_SHIFT 9
keir@19677	51 #define SUPERPAGE_NR_PFNS (1UL << SUPERPAGE_PFN_SHIFT)
keir@19677	52
keir@20164	53 /*
keir@20164	54 * Setting bit 31 force to allocate super page even not all pfns come out,
keir@20164	55 * bit 30 indicate that not is in a super page tracking.
keir@20164	56 */
keir@20164	57 #define FORCE_SP_SHIFT 31
keir@20164	58 #define FORCE_SP_MASK (1UL << FORCE_SP_SHIFT)
keir@19677	59
keir@20164	60 #define INVALID_SUPER_PAGE ((1UL << 30) + 1)
keir@20164	61 #define SUPER_PAGE_START(pfn) (((pfn) & (SUPERPAGE_NR_PFNS-1)) == 0 )
keir@20164	62 #define SUPER_PAGE_TRACKING(pfn) ( (pfn) != INVALID_SUPER_PAGE )
keir@20164	63 #define SUPER_PAGE_DONE(pfn) ( SUPER_PAGE_START(pfn) )
keir@19677	64
keir@20591	65 static int super_page_populated(struct restore_ctx *ctx, unsigned long pfn)
keir@20164	66 {
keir@20164	67 int i;
keir@20164	68 pfn &= ~(SUPERPAGE_NR_PFNS - 1);
keir@20164	69 for ( i = pfn; i < pfn + SUPERPAGE_NR_PFNS; i++ )
keir@20164	70 {
keir@20587	71 if ( ctx->p2m[i] != INVALID_P2M_ENTRY )
keir@20164	72 return 1;
keir@20164	73 }
keir@20164	74 return 0;
keir@20164	75 }
keir@19677	76
keir@20164	77 /*
keir@20164	78 * Break a 2M page and move contents of [extent start, next_pfn-1] to
keir@20164	79 * some new allocated 4K pages
keir@20164	80 */
keir@20164	81 static int break_super_page(int xc_handle,
keir@20164	82 uint32_t dom,
keir@20591	83 struct restore_ctx *ctx,
keir@20164	84 xen_pfn_t next_pfn)
keir@20164	85 {
keir@20164	86 xen_pfn_t *page_array, start_pfn, mfn;
keir@20164	87 uint8_t ram_base, save_buf;
keir@20164	88 unsigned long i;
keir@20164	89 int tot_pfns, rc = 0;
keir@20164	90
keir@20164	91 tot_pfns = (next_pfn & (SUPERPAGE_NR_PFNS - 1));
keir@20164	92
keir@20164	93 start_pfn = next_pfn & ~(SUPERPAGE_NR_PFNS - 1);
keir@20164	94 for ( i = start_pfn; i < start_pfn + SUPERPAGE_NR_PFNS; i++ )
keir@20164	95 {
keir@20164	96 /* check the 2M page are populated */
keir@20587	97 if ( ctx->p2m[i] == INVALID_P2M_ENTRY ) {
keir@20164	98 DPRINTF("Previous super page was populated wrongly!\n");
keir@19677	99 return 1;
keir@19677	100 }
keir@20164	101 }
keir@20164	102
keir@20164	103 page_array = (xen_pfn_t)malloc(tot_pfns sizeof(xen_pfn_t));
keir@20164	104 save_buf = (uint8_t)malloc(tot_pfns PAGE_SIZE);
keir@20164	105
keir@20164	106 if ( !page_array \|\| !save_buf )
keir@20164	107 {
keir@20164	108 ERROR("alloc page_array failed\n");
keir@20164	109 errno = ENOMEM;
keir@20164	110 rc = 1;
keir@20164	111 goto out;
keir@20164	112 }
keir@20164	113
keir@20164	114 /* save previous super page contents */
keir@20164	115 for ( i = 0; i < tot_pfns; i++ )
keir@20164	116 {
keir@20164	117 /* only support HVM, as the mfn of the 2M page is missing */
keir@20164	118 page_array[i] = start_pfn + i;
keir@20164	119 }
keir@20164	120
keir@20837	121 ram_base = xc_map_foreign_pages(xc_handle, dom, PROT_READ,
keir@20164	122 page_array, tot_pfns);
keir@20164	123
keir@20164	124 if ( ram_base == NULL )
keir@20164	125 {
keir@20164	126 ERROR("map batch failed\n");
keir@20164	127 rc = 1;
keir@20164	128 goto out;
keir@20164	129 }
keir@20164	130
keir@20164	131 memcpy(save_buf, ram_base, tot_pfns * PAGE_SIZE);
keir@20164	132 munmap(ram_base, tot_pfns * PAGE_SIZE);
keir@20164	133
keir@20164	134 /* free the super page */
keir@20164	135 if ( xc_domain_memory_decrease_reservation(xc_handle, dom, 1,
keir@20164	136 SUPERPAGE_PFN_SHIFT, &start_pfn) != 0 )
keir@20164	137 {
keir@20164	138 ERROR("free 2M page failure @ 0x%ld.\n", next_pfn);
keir@20164	139 rc = 1;
keir@20164	140 goto out;
keir@20164	141 }
keir@20164	142
keir@20164	143 start_pfn = next_pfn & ~(SUPERPAGE_NR_PFNS - 1);
keir@20164	144 for ( i = start_pfn; i < start_pfn + SUPERPAGE_NR_PFNS; i++ )
keir@20164	145 {
keir@20587	146 ctx->p2m[i] = INVALID_P2M_ENTRY;
keir@20164	147 }
keir@20164	148
keir@20164	149 for ( i = start_pfn; i < start_pfn + tot_pfns; i++ )
keir@20164	150 {
keir@20164	151 mfn = i;
keir@20164	152 if (xc_domain_memory_populate_physmap(xc_handle, dom, 1, 0,
keir@20164	153 0, &mfn) != 0)
keir@19677	154 {
keir@20164	155 ERROR("Failed to allocate physical memory.!\n");
keir@20164	156 errno = ENOMEM;
keir@20164	157 rc = 1;
keir@20164	158 goto out;
keir@20164	159 }
keir@20587	160 ctx->p2m[i] = mfn;
keir@20164	161 }
keir@20164	162
keir@20164	163 /* restore contents */
keir@20164	164 for ( i = 0; i < tot_pfns; i++ )
keir@20164	165 {
keir@20164	166 page_array[i] = start_pfn + i;
keir@20164	167 }
keir@20164	168
keir@20837	169 ram_base = xc_map_foreign_pages(xc_handle, dom, PROT_WRITE,
keir@20164	170 page_array, tot_pfns);
keir@20164	171 if ( ram_base == NULL )
keir@20164	172 {
keir@20164	173 ERROR("map batch failed\n");
keir@20164	174 rc = 1;
keir@20164	175 goto out;
keir@20164	176 }
keir@20164	177
keir@20164	178 memcpy(ram_base, save_buf, tot_pfns * PAGE_SIZE);
keir@20164	179 munmap(ram_base, tot_pfns * PAGE_SIZE);
keir@20164	180
keir@20164	181 out:
keir@20164	182 free(page_array);
keir@20164	183 free(save_buf);
keir@20164	184 return rc;
keir@20164	185 }
keir@20164	186
keir@20164	187
keir@20164	188 /*
keir@20164	189 * According to pfn list allocate pages: one 2M page or series of 4K pages.
keir@20164	190 * Also optimistically allocate a 2M page even when not all pages in the 2M
keir@20164	191 * extent come out, and fix it up in next batch:
keir@20164	192 * If new pages fit the missing one in the 2M extent, do nothing; Else take
keir@20164	193 * place of the original 2M page by some 4K pages.
keir@20164	194 */
keir@20164	195 static int allocate_mfn_list(int xc_handle,
keir@20164	196 uint32_t dom,
keir@20591	197 struct restore_ctx *ctx,
keir@20164	198 unsigned long nr_extents,
keir@20164	199 xen_pfn_t *batch_buf,
keir@20164	200 xen_pfn_t *next_pfn,
keir@20164	201 int superpages)
keir@20164	202 {
keir@20164	203 unsigned int i;
keir@20164	204 unsigned long mfn, pfn, sp_pfn;
keir@20164	205
keir@20164	206 /Check if force super page, then clear it /
keir@20164	207 unsigned force_super_page = !!(*next_pfn & FORCE_SP_MASK);
keir@20164	208 *next_pfn &= ~FORCE_SP_MASK;
keir@20164	209
keir@20164	210 sp_pfn = *next_pfn;
keir@20164	211
keir@20164	212 if ( !superpages \|\|
keir@20587	213 ctx->no_superpage_mem \|\|
keir@20164	214 !SUPER_PAGE_TRACKING(sp_pfn) )
keir@20164	215 goto normal_page;
keir@20164	216
keir@20164	217 if ( !batch_buf )
keir@20164	218 {
keir@20164	219 /* Break previous 2M page, if 512 pages split across a batch boundary */
keir@20164	220 if ( SUPER_PAGE_TRACKING(sp_pfn) &&
keir@20164	221 !SUPER_PAGE_DONE(sp_pfn))
keir@20164	222 {
keir@20164	223 /* break previously allocated super page*/
keir@20591	224 if ( break_super_page(xc_handle, dom, ctx, sp_pfn) != 0 )
keir@20164	225 {
keir@20164	226 ERROR("Break previous super page fail!\n");
keir@20164	227 return 1;
keir@20164	228 }
keir@20164	229 }
keir@20164	230
keir@20164	231 /* follwing pages fit the order in 2M extent */
keir@20164	232 return 0;
keir@20164	233 }
keir@20164	234
keir@20164	235 /*
keir@20164	236 * We try to allocate a 2M page only when:
keir@20164	237 * user require this(superpages),
keir@20164	238 * AND have enough memory,
keir@20164	239 * AND is in the tracking,
keir@20164	240 * AND tracked all pages in 2M extent, OR partial 2M extent for speculation
keir@20164	241 * AND any page in 2M extent are not populated
keir@20164	242 */
keir@20164	243 if ( !SUPER_PAGE_DONE(sp_pfn) && !force_super_page )
keir@20164	244 goto normal_page;
keir@20164	245
keir@20164	246 pfn = batch_buf[0] & ~XEN_DOMCTL_PFINFO_LTAB_MASK;
keir@20591	247 if ( super_page_populated(ctx, pfn) )
keir@20164	248 goto normal_page;
keir@20164	249
keir@20164	250 pfn &= ~(SUPERPAGE_NR_PFNS - 1);
keir@20164	251 mfn = pfn;
keir@20164	252
keir@20164	253 if ( xc_domain_memory_populate_physmap(xc_handle, dom, 1,
keir@20164	254 SUPERPAGE_PFN_SHIFT, 0, &mfn) == 0)
keir@20164	255 {
keir@20164	256 for ( i = pfn; i < pfn + SUPERPAGE_NR_PFNS; i++, mfn++ )
keir@20164	257 {
keir@20587	258 ctx->p2m[i] = mfn;
keir@20164	259 }
keir@20164	260 return 0;
keir@20164	261 }
keir@20164	262 DPRINTF("No 2M page available for pfn 0x%lx, fall back to 4K page.\n",
keir@20164	263 pfn);
keir@20587	264 ctx->no_superpage_mem = 1;
keir@20164	265
keir@20164	266 normal_page:
keir@20164	267 if ( !batch_buf )
keir@20164	268 return 0;
keir@20164	269
keir@20164	270 /* End the tracking, if want a 2M page but end by 4K pages, */
keir@20164	271 *next_pfn = INVALID_SUPER_PAGE;
keir@20164	272
keir@20164	273 for ( i = 0; i < nr_extents; i++ )
keir@20164	274 {
keir@20164	275 unsigned long pagetype = batch_buf[i] & XEN_DOMCTL_PFINFO_LTAB_MASK;
keir@20164	276 if ( pagetype == XEN_DOMCTL_PFINFO_XTAB )
keir@20164	277 continue;
keir@20164	278
keir@20164	279 pfn = mfn = batch_buf[i] & ~XEN_DOMCTL_PFINFO_LTAB_MASK;
keir@20587	280 if ( ctx->p2m[pfn] == INVALID_P2M_ENTRY )
keir@20164	281 {
keir@20164	282 if (xc_domain_memory_populate_physmap(xc_handle, dom, 1, 0,
keir@20164	283 0, &mfn) != 0)
keir@20164	284 {
keir@20164	285 ERROR("Failed to allocate physical memory.! pfn=0x%lx, mfn=0x%lx.\n",
keir@20164	286 pfn, mfn);
keir@20164	287 errno = ENOMEM;
keir@20164	288 return 1;
keir@20164	289 }
keir@20587	290 ctx->p2m[pfn] = mfn;
keir@19677	291 }
keir@19677	292 }
keir@20164	293
keir@19677	294 return 0;
keir@19677	295 }
keir@19677	296
keir@19677	297 static int allocate_physmem(int xc_handle, uint32_t dom,
keir@20591	298 struct restore_ctx *ctx,
keir@19677	299 unsigned long *region_pfn_type, int region_size,
keir@19677	300 unsigned int hvm, xen_pfn_t *region_mfn, int superpages)
keir@19677	301 {
keir@20164	302 int i;
keir@19677	303 unsigned long pfn;
keir@19677	304 unsigned long pagetype;
keir@19677	305
keir@20164	306 /* Next expected pfn in order to track a possible 2M page */
keir@20164	307 static unsigned long required_pfn = INVALID_SUPER_PAGE;
keir@20164	308
keir@20164	309 /* Buffer of pfn list for 2M page, or series of 4K pages */
keir@20164	310 xen_pfn_t *batch_buf;
keir@20164	311 unsigned int batch_buf_len;
keir@20589	312 struct domain_info_context *dinfo = &ctx->dinfo;
keir@20164	313
keir@20164	314 if ( !superpages )
keir@20164	315 {
keir@20164	316 batch_buf = &region_pfn_type[0];
keir@20164	317 batch_buf_len = region_size;
keir@20164	318 goto alloc_page;
keir@20164	319 }
keir@20164	320
keir@20164	321 batch_buf = NULL;
keir@20164	322 batch_buf_len = 0;
keir@20164	323 /* This loop tracks the possible 2M page */
keir@20164	324 for (i = 0; i < region_size; i++)
keir@20164	325 {
keir@20164	326 pfn = region_pfn_type[i] & ~XEN_DOMCTL_PFINFO_LTAB_MASK;
keir@20164	327 pagetype = region_pfn_type[i] & XEN_DOMCTL_PFINFO_LTAB_MASK;
keir@20164	328
keir@20164	329 if (pagetype == XEN_DOMCTL_PFINFO_XTAB)
keir@20164	330 {
keir@20164	331 /* Do not start collecting pfns until get a valid pfn */
keir@20164	332 if ( batch_buf_len != 0 )
keir@20164	333 batch_buf_len++;
keir@20164	334 continue;
keir@20164	335 }
keir@20164	336
keir@20164	337 if ( SUPER_PAGE_START(pfn) )
keir@20164	338 {
keir@20164	339 /* Start of a 2M extent, populate previsous buf */
keir@20591	340 if ( allocate_mfn_list(xc_handle, dom, ctx,
keir@20164	341 batch_buf_len, batch_buf,
keir@20164	342 &required_pfn, superpages) != 0 )
keir@20164	343 {
keir@20164	344 errno = ENOMEM;
keir@20164	345 return 1;
keir@20164	346 }
keir@20164	347
keir@20164	348 /* start new tracking for 2M page */
keir@20164	349 batch_buf = &region_pfn_type[i];
keir@20164	350 batch_buf_len = 1;
keir@20164	351 required_pfn = pfn + 1;
keir@20164	352 }
keir@20164	353 else if ( pfn == required_pfn )
keir@20164	354 {
keir@20164	355 /* this page fit the 2M extent in order */
keir@20164	356 batch_buf_len++;
keir@20164	357 required_pfn++;
keir@20164	358 }
keir@20164	359 else if ( SUPER_PAGE_TRACKING(required_pfn) )
keir@20164	360 {
keir@20164	361 /* break of a 2M extent, populate previous buf */
keir@20591	362 if ( allocate_mfn_list(xc_handle, dom, ctx,
keir@20164	363 batch_buf_len, batch_buf,
keir@20164	364 &required_pfn, superpages) != 0 )
keir@20164	365 {
keir@20164	366 errno = ENOMEM;
keir@20164	367 return 1;
keir@20164	368 }
keir@20164	369 /* start new tracking for a series of 4K pages */
keir@20164	370 batch_buf = &region_pfn_type[i];
keir@20164	371 batch_buf_len = 1;
keir@20164	372 required_pfn = INVALID_SUPER_PAGE;
keir@20164	373 }
keir@20164	374 else
keir@20164	375 {
keir@20164	376 /* this page is 4K */
keir@20164	377 if ( !batch_buf )
keir@20164	378 batch_buf = &region_pfn_type[i];
keir@20164	379 batch_buf_len++;
keir@20164	380 }
keir@20164	381 }
keir@20164	382
keir@20164	383 /*
keir@20164	384 * populate rest batch_buf in the end.
keir@20164	385 * In a speculative way, we allocate a 2M page even when not see all the
keir@20164	386 * pages in order(set bit 31). If not require super page support,
keir@20164	387 * we can skip the tracking loop and come here directly.
keir@20164	388 * Speculative allocation can't be used for PV guest, as we have no mfn to
keir@20164	389 * map previous 2M mem range if need break it.
keir@20164	390 */
keir@20164	391 if ( SUPER_PAGE_TRACKING(required_pfn) &&
keir@20164	392 !SUPER_PAGE_DONE(required_pfn) )
keir@20164	393 {
keir@20164	394 if (hvm)
keir@20164	395 required_pfn \|= FORCE_SP_MASK;
keir@20164	396 else
keir@20164	397 required_pfn = INVALID_SUPER_PAGE;
keir@20164	398 }
keir@20164	399
keir@20164	400 alloc_page:
keir@20164	401 if ( batch_buf )
keir@20164	402 {
keir@20591	403 if ( allocate_mfn_list(xc_handle, dom, ctx,
keir@20164	404 batch_buf_len, batch_buf,
keir@20164	405 &required_pfn,
keir@20164	406 superpages) != 0 )
keir@20164	407 {
keir@20164	408 errno = ENOMEM;
keir@20164	409 return 1;
keir@20164	410 }
keir@20164	411 }
keir@20164	412
keir@19677	413 for (i = 0; i < region_size; i++)
keir@19677	414 {
keir@19677	415 pfn = region_pfn_type[i] & ~XEN_DOMCTL_PFINFO_LTAB_MASK;
keir@19677	416 pagetype = region_pfn_type[i] & XEN_DOMCTL_PFINFO_LTAB_MASK;
keir@19677	417
keir@20585	418 if ( pfn > dinfo->p2m_size )
keir@19677	419 {
keir@19677	420 ERROR("pfn out of range");
keir@19677	421 return 1;
keir@19677	422 }
keir@19677	423 if (pagetype == XEN_DOMCTL_PFINFO_XTAB)
keir@19677	424 {
keir@19677	425 region_mfn[i] = ~0UL;
keir@19677	426 }
keir@19677	427 else
keir@19677	428 {
keir@20587	429 if (ctx->p2m[pfn] == INVALID_P2M_ENTRY)
keir@19677	430 {
keir@20164	431 DPRINTF("Warning: pfn 0x%lx are not allocated!\n", pfn);
keir@20164	432 /XXX:allocate this page?/
keir@19677	433 }
keir@19677	434
keir@19677	435 /* setup region_mfn[] for batch map.
keir@19677	436 * For HVM guests, this interface takes PFNs, not MFNs */
keir@20587	437 region_mfn[i] = hvm ? pfn : ctx->p2m[pfn];
keir@19677	438 }
keir@19677	439 }
keir@19677	440 return 0;
keir@19677	441 }
keir@19677	442
keir@19677	443
keir@20453	444 /* set when a consistent image is available */
keir@20453	445 static int completed = 0;
keir@20453	446
keir@20983	447 #define HEARTBEAT_MS 1000
keir@20453	448
keir@20453	449 #ifndef __MINIOS__
keir@20453	450 static ssize_t read_exact_timed(int fd, void* buf, size_t size)
keir@20453	451 {
keir@20453	452 size_t offset = 0;
keir@20453	453 ssize_t len;
keir@20453	454 struct timeval tv;
keir@20453	455 fd_set rfds;
keir@20453	456
keir@20453	457 while ( offset < size )
keir@20453	458 {
keir@20453	459 if ( completed ) {
keir@20453	460 /* expect a heartbeat every HEARBEAT_MS ms maximum */
keir@20983	461 tv.tv_sec = HEARTBEAT_MS / 1000;
keir@20983	462 tv.tv_usec = (HEARTBEAT_MS % 1000) * 1000;
keir@20453	463
keir@20453	464 FD_ZERO(&rfds);
keir@20453	465 FD_SET(fd, &rfds);
keir@20453	466 len = select(fd + 1, &rfds, NULL, NULL, &tv);
keir@20453	467 if ( !FD_ISSET(fd, &rfds) ) {
keir@20453	468 fprintf(stderr, "read_exact_timed failed (select returned %zd)\n", len);
keir@20453	469 return -1;
keir@20453	470 }
keir@20453	471 }
keir@20453	472
keir@20453	473 len = read(fd, buf + offset, size - offset);
keir@20453	474 if ( (len == -1) && ((errno == EINTR) \|\| (errno == EAGAIN)) )
keir@20453	475 continue;
keir@20453	476 if ( len <= 0 )
keir@20453	477 return -1;
keir@20453	478 offset += len;
keir@20453	479 }
keir@20453	480
keir@20453	481 return 0;
keir@20453	482 }
keir@20453	483
keir@20453	484 #define read_exact read_exact_timed
keir@20453	485
keir@20453	486 #else
keir@20453	487 #define read_exact_timed read_exact
keir@20453	488 #endif
keir@19677	489 /*
kaf24@9698	490 ** In the state file (or during transfer), all page-table pages are
kaf24@9698	491 ** converted into a 'canonical' form where references to actual mfns
kaf24@9698	492 ** are replaced with references to the corresponding pfns.
kaf24@9698	493 ** This function inverts that operation, replacing the pfn values with
kaf24@9698	494 ** the (now known) appropriate mfn values.
smh22@7740	495 */
keir@20591	496 static int uncanonicalize_pagetable(int xc_handle, uint32_t dom, struct restore_ctx *ctx,
keir@20837	497 void *page, int superpages)
kaf24@9698	498 {
kaf24@9698	499 int i, pte_last;
kaf24@9698	500 unsigned long pfn;
kaf24@9698	501 uint64_t pte;
keir@20589	502 struct domain_info_context *dinfo = &ctx->dinfo;
smh22@7740	503
keir@20587	504 pte_last = PAGE_SIZE / ((ctx->pt_levels == 2)? 4 : 8);
smh22@7740	505
keir@14809	506 for ( i = 0; i < pte_last; i++ )
keir@14809	507 {
keir@20587	508 if ( ctx->pt_levels == 2 )
kaf24@9698	509 pte = ((uint32_t *)page)[i];
kaf24@9698	510 else
kaf24@9698	511 pte = ((uint64_t *)page)[i];
steven@13424	512
steven@13424	513 /* XXX SMH: below needs fixing for PROT_NONE etc */
keir@14809	514 if ( !(pte & _PAGE_PRESENT) )
steven@13424	515 continue;
steven@13424	516
kfraser@14005	517 pfn = (pte >> PAGE_SHIFT) & MFN_MASK_X86;
kfraser@14819	518
keir@19677	519 /* Allocate mfn if necessary */
keir@20587	520 if ( ctx->p2m[pfn] == INVALID_P2M_ENTRY )
keir@19677	521 {
keir@20164	522 unsigned long force_pfn = superpages ? FORCE_SP_MASK : pfn;
keir@20591	523 if (allocate_mfn_list(xc_handle, dom, ctx,
keir@20164	524 1, &pfn, &force_pfn, superpages) != 0)
keir@19677	525 return 0;
keir@19677	526 }
kfraser@14005	527 pte &= ~MADDR_MASK_X86;
keir@20587	528 pte \|= (uint64_t)ctx->p2m[pfn] << PAGE_SHIFT;
smh22@7886	529
keir@20587	530 if ( ctx->pt_levels == 2 )
steven@13424	531 ((uint32_t *)page)[i] = (uint32_t)pte;
steven@13424	532 else
steven@13424	533 ((uint64_t *)page)[i] = (uint64_t)pte;
smh22@7740	534 }
kaf24@9698	535
kaf24@9698	536 return 1;
smh22@7740	537 }
smh22@7740	538
steven@14732	539
Tim@14782	540 /* Load the p2m frame list, plus potential extended info chunk */
keir@20591	541 static xen_pfn_t load_p2m_frame_list(struct restore_ctx ctx,
keir@16257	542 int io_fd, int pae_extended_cr3, int ext_vcpucontext)
Tim@14782	543 {
Tim@14782	544 xen_pfn_t *p2m_frame_list;
keir@17918	545 vcpu_guest_context_any_t ctxt;
Tim@15955	546 xen_pfn_t p2m_fl_zero;
keir@20589	547 struct domain_info_context *dinfo = &ctx->dinfo;
Tim@14782	548
Tim@14782	549 /* Read first entry of P2M list, or extended-info signature (~0UL). */
keir@16408	550 if ( read_exact(io_fd, &p2m_fl_zero, sizeof(long)) )
keir@14809	551 {
keir@14809	552 ERROR("read extended-info signature failed");
keir@14809	553 return NULL;
keir@14809	554 }
Tim@14782	555
Tim@15955	556 if ( p2m_fl_zero == ~0UL )
keir@14809	557 {
Tim@14782	558 uint32_t tot_bytes;
Tim@14782	559
Tim@14782	560 /* Next 4 bytes: total size of following extended info. */
keir@16408	561 if ( read_exact(io_fd, &tot_bytes, sizeof(tot_bytes)) )
keir@14809	562 {
Tim@14782	563 ERROR("read extended-info size failed");
Tim@14782	564 return NULL;
Tim@14782	565 }
Tim@14782	566
keir@14809	567 while ( tot_bytes )
keir@14809	568 {
Tim@14782	569 uint32_t chunk_bytes;
Tim@14782	570 char chunk_sig[4];
Tim@14782	571
Tim@14782	572 /* 4-character chunk signature + 4-byte remaining chunk size. */
keir@16408	573 if ( read_exact(io_fd, chunk_sig, sizeof(chunk_sig)) \|\|
keir@16408	574 read_exact(io_fd, &chunk_bytes, sizeof(chunk_bytes)) \|\|
keir@16257	575 (tot_bytes < (chunk_bytes + 8)) )
keir@14809	576 {
Tim@14782	577 ERROR("read extended-info chunk signature failed");
Tim@14782	578 return NULL;
Tim@14782	579 }
Tim@14782	580 tot_bytes -= 8;
keir@14809	581
Tim@14782	582 /* VCPU context structure? */
keir@14809	583 if ( !strncmp(chunk_sig, "vcpu", 4) )
keir@14809	584 {
Tim@15955	585 /* Pick a guest word-size and PT depth from the ctxt size */
Tim@15955	586 if ( chunk_bytes == sizeof (ctxt.x32) )
Tim@15955	587 {
keir@20585	588 dinfo->guest_width = 4;
keir@20587	589 if ( ctx->pt_levels > 2 )
keir@20587	590 ctx->pt_levels = 3;
Tim@15955	591 }
Tim@15955	592 else if ( chunk_bytes == sizeof (ctxt.x64) )
Tim@15955	593 {
keir@20585	594 dinfo->guest_width = 8;
keir@20587	595 ctx->pt_levels = 4;
Tim@15955	596 }
Tim@15955	597 else
Tim@15955	598 {
Tim@15955	599 ERROR("bad extended-info context size %d", chunk_bytes);
Tim@15955	600 return NULL;
Tim@15955	601 }
Tim@15955	602
keir@16408	603 if ( read_exact(io_fd, &ctxt, chunk_bytes) )
keir@14809	604 {
Tim@14782	605 ERROR("read extended-info vcpu context failed");
Tim@14782	606 return NULL;
Tim@14782	607 }
Tim@15955	608 tot_bytes -= chunk_bytes;
Tim@15955	609 chunk_bytes = 0;
Tim@15955	610
Tim@15955	611 if ( GET_FIELD(&ctxt, vm_assist)
Tim@15955	612 & (1UL << VMASST_TYPE_pae_extended_cr3) )
Tim@14782	613 *pae_extended_cr3 = 1;
Tim@14782	614 }
keir@16257	615 else if ( !strncmp(chunk_sig, "extv", 4) )
keir@16257	616 {
keir@16257	617 *ext_vcpucontext = 1;
keir@16257	618 }
Tim@14782	619
Tim@14782	620 /* Any remaining bytes of this chunk: read and discard. */
keir@14809	621 while ( chunk_bytes )
keir@14809	622 {
Tim@15955	623 unsigned long sz = MIN(chunk_bytes, sizeof(xen_pfn_t));
keir@16408	624 if ( read_exact(io_fd, &p2m_fl_zero, sz) )
keir@14809	625 {
Tim@14782	626 ERROR("read-and-discard extended-info chunk bytes failed");
Tim@14782	627 return NULL;
Tim@14782	628 }
Tim@14782	629 chunk_bytes -= sz;
Tim@14782	630 tot_bytes -= sz;
Tim@14782	631 }
Tim@14782	632 }
keir@14809	633
Tim@14782	634 /* Now read the real first entry of P2M list. */
keir@16408	635 if ( read_exact(io_fd, &p2m_fl_zero, sizeof(xen_pfn_t)) )
keir@14809	636 {
Tim@14782	637 ERROR("read first entry of p2m_frame_list failed");
Tim@14782	638 return NULL;
Tim@14782	639 }
Tim@14782	640 }
keir@14809	641
Tim@15955	642 /* Now that we know the guest's word-size, can safely allocate
Tim@15955	643 * the p2m frame list */
keir@17074	644 if ( (p2m_frame_list = malloc(P2M_TOOLS_FL_SIZE)) == NULL )
Tim@15955	645 {
Tim@15955	646 ERROR("Couldn't allocate p2m_frame_list array");
Tim@15955	647 return NULL;
Tim@15955	648 }
Tim@15955	649
Tim@15955	650 /* First entry has already been read. */
Tim@15955	651 p2m_frame_list[0] = p2m_fl_zero;
keir@16408	652 if ( read_exact(io_fd, &p2m_frame_list[1],
keir@16408	653 (P2M_FL_ENTRIES - 1) * sizeof(xen_pfn_t)) )
keir@14809	654 {
keir@14809	655 ERROR("read p2m_frame_list failed");
keir@14809	656 return NULL;
Tim@14782	657 }
Tim@14782	658
Tim@14782	659 return p2m_frame_list;
Tim@14782	660 }
Tim@14782	661
keir@20452	662 typedef struct {
keir@20457	663 int ishvm;
keir@20457	664 union {
keir@20457	665 struct tailbuf_pv {
keir@20457	666 unsigned int pfncount;
keir@20457	667 unsigned long* pfntab;
keir@20457	668 unsigned int vcpucount;
keir@20457	669 unsigned char* vcpubuf;
keir@20457	670 unsigned char shared_info_page[PAGE_SIZE];
keir@20457	671 } pv;
keir@20457	672 struct tailbuf_hvm {
keir@20457	673 uint64_t magicpfns[3];
keir@20457	674 uint32_t hvmbufsize, reclen;
keir@20457	675 uint8_t* hvmbuf;
keir@20457	676 struct {
keir@20457	677 uint32_t magic;
keir@20457	678 uint32_t version;
keir@20457	679 uint64_t len;
keir@20457	680 } qemuhdr;
keir@20457	681 uint32_t qemubufsize;
keir@20457	682 uint8_t* qemubuf;
keir@20457	683 } hvm;
keir@20457	684 } u;
keir@20452	685 } tailbuf_t;
keir@20452	686
keir@20457	687 /* read stream until EOF, growing buffer as necssary */
keir@20457	688 static int compat_buffer_qemu(int fd, struct tailbuf_hvm *buf)
keir@20457	689 {
keir@20457	690 uint8_t qbuf, tmp;
keir@20457	691 int blen = 0, dlen = 0;
keir@20457	692 int rc;
keir@20457	693
keir@20457	694 /* currently save records tend to be about 7K */
keir@20457	695 blen = 8192;
keir@20457	696 if ( !(qbuf = malloc(blen)) ) {
keir@20457	697 ERROR("Error allocating QEMU buffer");
keir@20457	698 return -1;
keir@20457	699 }
keir@20457	700
keir@20457	701 while( (rc = read(fd, qbuf+dlen, blen-dlen)) > 0 ) {
keir@20457	702 DPRINTF("Read %d bytes of QEMU data\n", rc);
keir@20457	703 dlen += rc;
keir@20457	704
keir@20457	705 if (dlen == blen) {
keir@20457	706 DPRINTF("%d-byte QEMU buffer full, reallocating...\n", dlen);
keir@20457	707 blen += 4096;
keir@20457	708 tmp = realloc(qbuf, blen);
keir@20457	709 if ( !tmp ) {
keir@20457	710 ERROR("Error growing QEMU buffer to %d bytes", blen);
keir@20457	711 free(qbuf);
keir@20457	712 return -1;
keir@20457	713 }
keir@20457	714 qbuf = tmp;
keir@20457	715 }
keir@20457	716 }
keir@20457	717
keir@20457	718 if ( rc < 0 ) {
keir@20457	719 ERROR("Error reading QEMU data");
keir@20457	720 free(qbuf);
keir@20457	721 return -1;
keir@20457	722 }
keir@20457	723
keir@20457	724 if ( memcmp(qbuf, "QEVM", 4) ) {
keir@20457	725 ERROR("Invalid QEMU magic: 0x%08x", (unsigned long)qbuf);
keir@20457	726 free(qbuf);
keir@20457	727 return -1;
keir@20457	728 }
keir@20457	729
keir@20457	730 buf->qemubuf = qbuf;
keir@20457	731 buf->qemubufsize = dlen;
keir@20457	732
keir@20457	733 return 0;
keir@20457	734 }
keir@20457	735
keir@20457	736 static int buffer_qemu(int fd, struct tailbuf_hvm *buf)
keir@20457	737 {
keir@20457	738 uint32_t qlen;
keir@20457	739 uint8_t *tmp;
keir@20457	740
keir@20457	741 if ( read_exact(fd, &qlen, sizeof(qlen)) ) {
keir@20457	742 ERROR("Error reading QEMU header length");
keir@20457	743 return -1;
keir@20457	744 }
keir@20457	745
keir@20457	746 if ( qlen > buf->qemubufsize ) {
keir@20457	747 if ( buf->qemubuf) {
keir@20457	748 tmp = realloc(buf->qemubuf, qlen);
keir@20457	749 if ( tmp )
keir@20457	750 buf->qemubuf = tmp;
keir@20457	751 else {
keir@20457	752 ERROR("Error reallocating QEMU state buffer");
keir@20457	753 return -1;
keir@20457	754 }
keir@20457	755 } else {
keir@20457	756 buf->qemubuf = malloc(qlen);
keir@20457	757 if ( !buf->qemubuf ) {
keir@20457	758 ERROR("Error allocating QEMU state buffer");
keir@20457	759 return -1;
keir@20457	760 }
keir@20457	761 }
keir@20457	762 }
keir@20457	763 buf->qemubufsize = qlen;
keir@20457	764
keir@20457	765 if ( read_exact(fd, buf->qemubuf, buf->qemubufsize) ) {
keir@20457	766 ERROR("Error reading QEMU state");
keir@20457	767 return -1;
keir@20457	768 }
keir@20457	769
keir@20457	770 return 0;
keir@20457	771 }
keir@20457	772
keir@20457	773 static int dump_qemu(uint32_t dom, struct tailbuf_hvm *buf)
keir@20457	774 {
keir@20457	775 int saved_errno;
keir@20457	776 char path[256];
keir@20457	777 FILE *fp;
keir@20457	778
keir@20457	779 sprintf(path, "/var/lib/xen/qemu-save.%u", dom);
keir@20457	780 fp = fopen(path, "wb");
keir@20457	781 if ( !fp )
keir@20457	782 return -1;
keir@20457	783
keir@20457	784 DPRINTF("Writing %d bytes of QEMU data\n", buf->qemubufsize);
keir@20457	785 if ( fwrite(buf->qemubuf, 1, buf->qemubufsize, fp) != buf->qemubufsize) {
keir@20457	786 saved_errno = errno;
keir@20457	787 fclose(fp);
keir@20457	788 errno = saved_errno;
keir@20457	789 return -1;
keir@20457	790 }
keir@20457	791
keir@20457	792 fclose(fp);
keir@20457	793
keir@20457	794 return 0;
keir@20457	795 }
keir@20457	796
keir@20591	797 static int buffer_tail_hvm(struct restore_ctx ctx, struct tailbuf_hvm buf, int fd,
keir@20457	798 unsigned int max_vcpu_id, uint64_t vcpumap,
keir@20457	799 int ext_vcpucontext)
keir@20457	800 {
keir@20457	801 uint8_t *tmp;
keir@20457	802 unsigned char qemusig[21];
keir@20457	803
keir@20457	804 if ( read_exact(fd, buf->magicpfns, sizeof(buf->magicpfns)) ) {
keir@20457	805 ERROR("Error reading magic PFNs");
keir@20457	806 return -1;
keir@20457	807 }
keir@20457	808
keir@20457	809 if ( read_exact(fd, &buf->reclen, sizeof(buf->reclen)) ) {
keir@20457	810 ERROR("Error reading HVM params size");
keir@20457	811 return -1;
keir@20457	812 }
keir@20457	813
keir@20457	814 if ( buf->reclen > buf->hvmbufsize ) {
keir@20457	815 if ( buf->hvmbuf) {
keir@20457	816 tmp = realloc(buf->hvmbuf, buf->reclen);
keir@20457	817 if ( tmp ) {
keir@20457	818 buf->hvmbuf = tmp;
keir@20457	819 buf->hvmbufsize = buf->reclen;
keir@20457	820 } else {
keir@20457	821 ERROR("Error reallocating HVM param buffer");
keir@20457	822 return -1;
keir@20457	823 }
keir@20457	824 } else {
keir@20457	825 buf->hvmbuf = malloc(buf->reclen);
keir@20457	826 if ( !buf->hvmbuf ) {
keir@20457	827 ERROR("Error allocating HVM param buffer");
keir@20457	828 return -1;
keir@20457	829 }
keir@20457	830 buf->hvmbufsize = buf->reclen;
keir@20457	831 }
keir@20457	832 }
keir@20457	833
keir@20457	834 if ( read_exact(fd, buf->hvmbuf, buf->reclen) ) {
keir@20457	835 ERROR("Error reading HVM params");
keir@20457	836 return -1;
keir@20457	837 }
keir@20457	838
keir@20457	839 if ( read_exact(fd, qemusig, sizeof(qemusig)) ) {
keir@20457	840 ERROR("Error reading QEMU signature");
keir@20457	841 return -1;
keir@20457	842 }
keir@20457	843
keir@20457	844 /* The normal live-migration QEMU record has no length information.
keir@20457	845 * Short of reimplementing the QEMU parser, we're forced to just read
keir@20457	846 * until EOF. Remus gets around this by sending a different signature
keir@20457	847 * which includes a length prefix */
keir@20457	848 if ( !memcmp(qemusig, "QemuDeviceModelRecord", sizeof(qemusig)) )
keir@20457	849 return compat_buffer_qemu(fd, buf);
keir@20457	850 else if ( !memcmp(qemusig, "RemusDeviceModelState", sizeof(qemusig)) )
keir@20457	851 return buffer_qemu(fd, buf);
keir@20457	852
keir@20457	853 qemusig[20] = '\0';
keir@20457	854 ERROR("Invalid QEMU signature: %s", qemusig);
keir@20457	855 return -1;
keir@20457	856 }
keir@20457	857
keir@20591	858 static int buffer_tail_pv(struct restore_ctx ctx, struct tailbuf_pv buf, int fd,
keir@20457	859 unsigned int max_vcpu_id, uint64_t vcpumap,
keir@20457	860 int ext_vcpucontext)
keir@20452	861 {
keir@20452	862 unsigned int i;
keir@20452	863 size_t pfnlen, vcpulen;
keir@20589	864 struct domain_info_context *dinfo = &ctx->dinfo;
keir@20452	865
keir@20452	866 /* TODO: handle changing pfntab and vcpu counts */
keir@20452	867 /* PFN tab */
keir@20452	868 if ( read_exact(fd, &buf->pfncount, sizeof(buf->pfncount)) \|\|
keir@20452	869 (buf->pfncount > (1U << 28)) ) /* up to 1TB of address space */
keir@20452	870 {
keir@20452	871 ERROR("Error when reading pfn count");
keir@20452	872 return -1;
keir@20452	873 }
keir@20452	874 pfnlen = sizeof(unsigned long) * buf->pfncount;
keir@20452	875 if ( !(buf->pfntab) ) {
keir@20452	876 if ( !(buf->pfntab = malloc(pfnlen)) ) {
keir@20452	877 ERROR("Error allocating PFN tail buffer");
keir@20452	878 return -1;
keir@20452	879 }
keir@20452	880 }
keir@20452	881 // DPRINTF("Reading PFN tab: %d bytes\n", pfnlen);
keir@20452	882 if ( read_exact(fd, buf->pfntab, pfnlen) ) {
keir@20452	883 ERROR("Error when reading pfntab");
keir@20452	884 goto free_pfntab;
keir@20452	885 }
keir@20452	886
keir@20452	887 /* VCPU contexts */
keir@20452	888 buf->vcpucount = 0;
keir@20452	889 for (i = 0; i <= max_vcpu_id; i++) {
keir@20452	890 // DPRINTF("vcpumap: %llx, cpu: %d, bit: %llu\n", vcpumap, i, (vcpumap % (1ULL << i)));
keir@20452	891 if ( (!(vcpumap & (1ULL << i))) )
keir@20452	892 continue;
keir@20452	893 buf->vcpucount++;
keir@20452	894 }
keir@20452	895 // DPRINTF("VCPU count: %d\n", buf->vcpucount);
keir@20585	896 vcpulen = ((dinfo->guest_width == 8) ? sizeof(vcpu_guest_context_x86_64_t)
keir@20452	897 : sizeof(vcpu_guest_context_x86_32_t)) * buf->vcpucount;
keir@20452	898 if ( ext_vcpucontext )
keir@20452	899 vcpulen += 128 * buf->vcpucount;
keir@20452	900
keir@20452	901 if ( !(buf->vcpubuf) ) {
keir@20452	902 if ( !(buf->vcpubuf = malloc(vcpulen)) ) {
keir@20452	903 ERROR("Error allocating VCPU ctxt tail buffer");
keir@20452	904 goto free_pfntab;
keir@20452	905 }
keir@20452	906 }
keir@20452	907 // DPRINTF("Reading VCPUS: %d bytes\n", vcpulen);
keir@20452	908 if ( read_exact(fd, buf->vcpubuf, vcpulen) ) {
keir@20452	909 ERROR("Error when reading ctxt");
keir@20452	910 goto free_vcpus;
keir@20452	911 }
keir@20452	912
keir@20452	913 /* load shared_info_page */
keir@20452	914 // DPRINTF("Reading shared info: %lu bytes\n", PAGE_SIZE);
keir@20452	915 if ( read_exact(fd, buf->shared_info_page, PAGE_SIZE) ) {
keir@20452	916 ERROR("Error when reading shared info page");
keir@20452	917 goto free_vcpus;
keir@20452	918 }
keir@20452	919
keir@20452	920 return 0;
keir@20452	921
keir@20452	922 free_vcpus:
keir@20452	923 if (buf->vcpubuf) {
keir@20452	924 free (buf->vcpubuf);
keir@20452	925 buf->vcpubuf = NULL;
keir@20452	926 }
keir@20452	927 free_pfntab:
keir@20452	928 if (buf->pfntab) {
keir@20452	929 free (buf->pfntab);
keir@20452	930 buf->pfntab = NULL;
keir@20452	931 }
keir@20452	932
keir@20452	933 return -1;
keir@20452	934 }
keir@20452	935
keir@20591	936 static int buffer_tail(struct restore_ctx ctx, tailbuf_t buf, int fd, unsigned int max_vcpu_id,
keir@20457	937 uint64_t vcpumap, int ext_vcpucontext)
keir@20457	938 {
keir@20457	939 if ( buf->ishvm )
keir@20591	940 return buffer_tail_hvm(ctx, &buf->u.hvm, fd, max_vcpu_id, vcpumap,
keir@20457	941 ext_vcpucontext);
keir@20457	942 else
keir@20591	943 return buffer_tail_pv(ctx, &buf->u.pv, fd, max_vcpu_id, vcpumap,
keir@20457	944 ext_vcpucontext);
keir@20457	945 }
keir@20457	946
keir@20457	947 static void tailbuf_free_hvm(struct tailbuf_hvm *buf)
keir@20452	948 {
keir@20457	949 if ( buf->hvmbuf ) {
keir@20457	950 free(buf->hvmbuf);
keir@20457	951 buf->hvmbuf = NULL;
keir@20457	952 }
keir@20457	953 if ( buf->qemubuf ) {
keir@20457	954 free(buf->qemubuf);
keir@20457	955 buf->qemubuf = NULL;
keir@20457	956 }
keir@20457	957 }
keir@20457	958
keir@20457	959 static void tailbuf_free_pv(struct tailbuf_pv *buf)
keir@20457	960 {
keir@20457	961 if ( buf->vcpubuf ) {
keir@20452	962 free(buf->vcpubuf);
keir@20452	963 buf->vcpubuf = NULL;
keir@20452	964 }
keir@20457	965 if ( buf->pfntab ) {
keir@20452	966 free(buf->pfntab);
keir@20452	967 buf->pfntab = NULL;
keir@20452	968 }
keir@20452	969 }
keir@20452	970
keir@20457	971 static void tailbuf_free(tailbuf_t *buf)
keir@20457	972 {
keir@20457	973 if ( buf->ishvm )
keir@20457	974 tailbuf_free_hvm(&buf->u.hvm);
keir@20457	975 else
keir@20457	976 tailbuf_free_pv(&buf->u.pv);
keir@20457	977 }
keir@20457	978
keir@20452	979 typedef struct {
keir@20452	980 void* pages;
keir@20452	981 /* pages is of length nr_physpages, pfn_types is of length nr_pages */
keir@20452	982 unsigned int nr_physpages, nr_pages;
keir@20452	983
keir@20452	984 /* Types of the pfns in the current region */
keir@20452	985 unsigned long* pfn_types;
keir@20452	986
keir@20452	987 int verify;
keir@20452	988
keir@20452	989 int new_ctxt_format;
keir@20452	990 int max_vcpu_id;
keir@20452	991 uint64_t vcpumap;
keir@20452	992 uint64_t identpt;
keir@20452	993 uint64_t vm86_tss;
keir@20452	994 } pagebuf_t;
keir@20452	995
keir@20452	996 static int pagebuf_init(pagebuf_t* buf)
keir@20452	997 {
keir@20452	998 memset(buf, 0, sizeof(*buf));
keir@20452	999 return 0;
keir@20452	1000 }
keir@20452	1001
keir@20452	1002 static void pagebuf_free(pagebuf_t* buf)
keir@20452	1003 {
keir@20452	1004 if (buf->pages) {
keir@20452	1005 free(buf->pages);
keir@20452	1006 buf->pages = NULL;
keir@20452	1007 }
keir@20452	1008 if(buf->pfn_types) {
keir@20452	1009 free(buf->pfn_types);
keir@20452	1010 buf->pfn_types = NULL;
keir@20452	1011 }
keir@20452	1012 }
keir@20452	1013
keir@20452	1014 static int pagebuf_get_one(pagebuf_t* buf, int fd, int xch, uint32_t dom)
keir@20452	1015 {
keir@20452	1016 int count, countpages, oldcount, i;
keir@20452	1017 void* ptmp;
keir@20452	1018
keir@20452	1019 if ( read_exact(fd, &count, sizeof(count)) )
keir@20452	1020 {
keir@20452	1021 ERROR("Error when reading batch size");
keir@20452	1022 return -1;
keir@20452	1023 }
keir@20452	1024
keir@20452	1025 // DPRINTF("reading batch of %d pages\n", count);
keir@20452	1026
keir@20452	1027 if (!count) {
keir@20452	1028 // DPRINTF("Last batch read\n");
keir@20452	1029 return 0;
keir@20452	1030 } else if (count == -1) {
keir@20452	1031 DPRINTF("Entering page verify mode\n");
keir@20452	1032 buf->verify = 1;
keir@20452	1033 return pagebuf_get_one(buf, fd, xch, dom);
keir@20452	1034 } else if (count == -2) {
keir@20452	1035 buf->new_ctxt_format = 1;
keir@20452	1036 if ( read_exact(fd, &buf->max_vcpu_id, sizeof(buf->max_vcpu_id)) \|\|
keir@20452	1037 buf->max_vcpu_id >= 64 \|\| read_exact(fd, &buf->vcpumap,
keir@20452	1038 sizeof(uint64_t)) ) {
keir@20452	1039 ERROR("Error when reading max_vcpu_id");
keir@20452	1040 return -1;
keir@20452	1041 }
keir@20452	1042 // DPRINTF("Max VCPU ID: %d, vcpumap: %llx\n", buf->max_vcpu_id, buf->vcpumap);
keir@20452	1043 return pagebuf_get_one(buf, fd, xch, dom);
keir@20452	1044 } else if (count == -3) {
keir@20452	1045 /* Skip padding 4 bytes then read the EPT identity PT location. */
keir@20452	1046 if ( read_exact(fd, &buf->identpt, sizeof(uint32_t)) \|\|
keir@20452	1047 read_exact(fd, &buf->identpt, sizeof(uint64_t)) )
keir@20452	1048 {
keir@20452	1049 ERROR("error read the address of the EPT identity map");
keir@20452	1050 return -1;
keir@20452	1051 }
keir@20452	1052 // DPRINTF("EPT identity map address: %llx\n", buf->identpt);
keir@20452	1053 return pagebuf_get_one(buf, fd, xch, dom);
keir@20452	1054 } else if ( count == -4 ) {
keir@20452	1055 /* Skip padding 4 bytes then read the vm86 TSS location. */
keir@20452	1056 if ( read_exact(fd, &buf->vm86_tss, sizeof(uint32_t)) \|\|
keir@20452	1057 read_exact(fd, &buf->vm86_tss, sizeof(uint64_t)) )
keir@20452	1058 {
keir@20452	1059 ERROR("error read the address of the vm86 TSS");
keir@20452	1060 return -1;
keir@20452	1061 }
keir@20452	1062 // DPRINTF("VM86 TSS location: %llx\n", buf->vm86_tss);
keir@20452	1063 return pagebuf_get_one(buf, fd, xch, dom);
keir@20452	1064 } else if ( count == -5 ) {
keir@20452	1065 DPRINTF("xc_domain_restore start tmem\n");
keir@20452	1066 if ( xc_tmem_restore(xch, dom, fd) ) {
keir@20452	1067 ERROR("error reading/restoring tmem");
keir@20452	1068 return -1;
keir@20452	1069 }
keir@20452	1070 return pagebuf_get_one(buf, fd, xch, dom);
keir@20452	1071 }
keir@20452	1072 else if ( count == -6 ) {
keir@20452	1073 if ( xc_tmem_restore_extra(xch, dom, fd) ) {
keir@20452	1074 ERROR("error reading/restoring tmem extra");
keir@20452	1075 return -1;
keir@20452	1076 }
keir@20452	1077 return pagebuf_get_one(buf, fd, xch, dom);
keir@20539	1078 } else if ( count == -7 ) {
keir@20539	1079 uint32_t tsc_mode, khz, incarn;
keir@20539	1080 uint64_t nsec;
keir@20539	1081 if ( read_exact(fd, &tsc_mode, sizeof(uint32_t)) \|\|
keir@20539	1082 read_exact(fd, &nsec, sizeof(uint64_t)) \|\|
keir@20539	1083 read_exact(fd, &khz, sizeof(uint32_t)) \|\|
keir@20539	1084 read_exact(fd, &incarn, sizeof(uint32_t)) \|\|
keir@20539	1085 xc_domain_set_tsc_info(xch, dom, tsc_mode, nsec, khz, incarn) ) {
keir@20539	1086 ERROR("error reading/restoring tsc info");
keir@20539	1087 return -1;
keir@20539	1088 }
keir@20539	1089 return pagebuf_get_one(buf, fd, xch, dom);
keir@20452	1090 } else if ( (count > MAX_BATCH_SIZE) \|\| (count < 0) ) {
keir@20452	1091 ERROR("Max batch size exceeded (%d). Giving up.", count);
keir@20452	1092 return -1;
keir@20452	1093 }
keir@20452	1094
keir@20452	1095 oldcount = buf->nr_pages;
keir@20452	1096 buf->nr_pages += count;
keir@20452	1097 if (!buf->pfn_types) {
keir@20452	1098 if (!(buf->pfn_types = malloc(buf->nr_pages * sizeof(*(buf->pfn_types))))) {
keir@20452	1099 ERROR("Could not allocate PFN type buffer");
keir@20452	1100 return -1;
keir@20452	1101 }
keir@20452	1102 } else {
keir@20452	1103 if (!(ptmp = realloc(buf->pfn_types, buf->nr_pages * sizeof(*(buf->pfn_types))))) {
keir@20452	1104 ERROR("Could not reallocate PFN type buffer");
keir@20452	1105 return -1;
keir@20452	1106 }
keir@20452	1107 buf->pfn_types = ptmp;
keir@20452	1108 }
keir@20452	1109 if ( read_exact(fd, buf->pfn_types + oldcount, count * sizeof(*(buf->pfn_types)))) {
keir@20452	1110 ERROR("Error when reading region pfn types");
keir@20452	1111 return -1;
keir@20452	1112 }
keir@20452	1113
keir@20452	1114 countpages = count;
keir@20452	1115 for (i = oldcount; i < buf->nr_pages; ++i)
keir@20452	1116 if ((buf->pfn_types[i] & XEN_DOMCTL_PFINFO_LTAB_MASK) == XEN_DOMCTL_PFINFO_XTAB)
keir@20452	1117 --countpages;
keir@20452	1118
keir@20452	1119 if (!countpages)
keir@20452	1120 return count;
keir@20452	1121
keir@20452	1122 oldcount = buf->nr_physpages;
keir@20452	1123 buf->nr_physpages += countpages;
keir@20452	1124 if (!buf->pages) {
keir@20452	1125 if (!(buf->pages = malloc(buf->nr_physpages * PAGE_SIZE))) {
keir@20452	1126 ERROR("Could not allocate page buffer");
keir@20452	1127 return -1;
keir@20452	1128 }
keir@20452	1129 } else {
keir@20452	1130 if (!(ptmp = realloc(buf->pages, buf->nr_physpages * PAGE_SIZE))) {
keir@20452	1131 ERROR("Could not reallocate page buffer");
keir@20452	1132 return -1;
keir@20452	1133 }
keir@20452	1134 buf->pages = ptmp;
keir@20452	1135 }
keir@20452	1136 if ( read_exact(fd, buf->pages + oldcount * PAGE_SIZE, countpages * PAGE_SIZE) ) {
keir@20452	1137 ERROR("Error when reading pages");
keir@20452	1138 return -1;
keir@20452	1139 }
keir@20452	1140
keir@20452	1141 return count;
keir@20452	1142 }
keir@20452	1143
keir@20452	1144 static int pagebuf_get(pagebuf_t* buf, int fd, int xch, uint32_t dom)
keir@20452	1145 {
keir@20452	1146 int rc;
keir@20452	1147
keir@20452	1148 buf->nr_physpages = buf->nr_pages = 0;
keir@20452	1149
keir@20452	1150 do {
keir@20452	1151 rc = pagebuf_get_one(buf, fd, xch, dom);
keir@20452	1152 } while (rc > 0);
keir@20452	1153
keir@20452	1154 if (rc < 0)
keir@20452	1155 pagebuf_free(buf);
keir@20452	1156
keir@20452	1157 return rc;
keir@20452	1158 }
keir@20452	1159
keir@20591	1160 static int apply_batch(int xc_handle, uint32_t dom, struct restore_ctx *ctx,
keir@20591	1161 xen_pfn_t* region_mfn, unsigned long* pfn_type, int pae_extended_cr3,
keir@20452	1162 unsigned int hvm, struct xc_mmu* mmu,
keir@20452	1163 pagebuf_t* pagebuf, int curbatch, int superpages)
keir@20452	1164 {
keir@20452	1165 int i, j, curpage;
keir@20452	1166 /* used by debug verify code */
keir@20452	1167 unsigned long buf[PAGE_SIZE/sizeof(unsigned long)];
keir@20452	1168 /* Our mapping of the current region (batch) */
keir@20452	1169 char *region_base;
keir@20452	1170 /* A temporary mapping, and a copy, of one frame of guest memory. */
keir@20452	1171 unsigned long *page = NULL;
keir@20452	1172 int nraces = 0;
keir@20589	1173 struct domain_info_context *dinfo = &ctx->dinfo;
keir@20877	1174 int* pfn_err = NULL;
keir@20877	1175 int rc = -1;
keir@20452	1176
keir@20452	1177 unsigned long mfn, pfn, pagetype;
keir@20452	1178
keir@20452	1179 j = pagebuf->nr_pages - curbatch;
keir@20452	1180 if (j > MAX_BATCH_SIZE)
keir@20452	1181 j = MAX_BATCH_SIZE;
keir@20452	1182
keir@20591	1183 if (allocate_physmem(xc_handle, dom, ctx, &pagebuf->pfn_types[curbatch],
keir@20452	1184 j, hvm, region_mfn, superpages) != 0)
keir@20452	1185 {
keir@20452	1186 ERROR("allocate_physmem() failed\n");
keir@20452	1187 return -1;
keir@20452	1188 }
keir@20452	1189
keir@20452	1190 /* Map relevant mfns */
keir@20877	1191 pfn_err = calloc(j, sizeof(*pfn_err));
keir@20877	1192 region_base = xc_map_foreign_bulk(
keir@20877	1193 xc_handle, dom, PROT_WRITE, region_mfn, pfn_err, j);
keir@20452	1194
keir@20452	1195 if ( region_base == NULL )
keir@20452	1196 {
keir@20452	1197 ERROR("map batch failed");
keir@20877	1198 free(pfn_err);
keir@20452	1199 return -1;
keir@20452	1200 }
keir@20452	1201
keir@20452	1202 for ( i = 0, curpage = -1; i < j; i++ )
keir@20452	1203 {
keir@20452	1204 pfn = pagebuf->pfn_types[i + curbatch] & ~XEN_DOMCTL_PFINFO_LTAB_MASK;
keir@20452	1205 pagetype = pagebuf->pfn_types[i + curbatch] & XEN_DOMCTL_PFINFO_LTAB_MASK;
keir@20452	1206
keir@20452	1207 if ( pagetype == XEN_DOMCTL_PFINFO_XTAB )
keir@20452	1208 /* a bogus/unmapped page: skip it */
keir@20452	1209 continue;
keir@20452	1210
keir@20877	1211 if (pfn_err[i])
keir@20877	1212 {
keir@20877	1213 ERROR("unexpected PFN mapping failure");
keir@20877	1214 goto err_mapped;
keir@20877	1215 }
keir@20877	1216
keir@20452	1217 ++curpage;
keir@20452	1218
keir@20585	1219 if ( pfn > dinfo->p2m_size )
keir@20452	1220 {
keir@20452	1221 ERROR("pfn out of range");
keir@20877	1222 goto err_mapped;
keir@20452	1223 }
keir@20452	1224
keir@20452	1225 pfn_type[pfn] = pagetype;
keir@20452	1226
keir@20587	1227 mfn = ctx->p2m[pfn];
keir@20452	1228
keir@20452	1229 /* In verify mode, we use a copy; otherwise we work in place */
keir@20452	1230 page = pagebuf->verify ? (void )buf : (region_base + iPAGE_SIZE);
keir@20452	1231
keir@20452	1232 memcpy(page, pagebuf->pages + (curpage + curbatch) * PAGE_SIZE, PAGE_SIZE);
keir@20452	1233
keir@20452	1234 pagetype &= XEN_DOMCTL_PFINFO_LTABTYPE_MASK;
keir@20452	1235
keir@20452	1236 if ( (pagetype >= XEN_DOMCTL_PFINFO_L1TAB) &&
keir@20452	1237 (pagetype <= XEN_DOMCTL_PFINFO_L4TAB) )
keir@20452	1238 {
keir@20452	1239 /*
keir@20452	1240 ** A page table page - need to 'uncanonicalize' it, i.e.
keir@20452	1241 ** replace all the references to pfns with the corresponding
keir@20452	1242 ** mfns for the new domain.
keir@20452	1243 **
keir@20452	1244 ** On PAE we need to ensure that PGDs are in MFNs < 4G, and
keir@20452	1245 ** so we may need to update the p2m after the main loop.
keir@20452	1246 ** Hence we defer canonicalization of L1s until then.
keir@20452	1247 */
keir@20587	1248 if ((ctx->pt_levels != 3) \|\|
keir@20452	1249 pae_extended_cr3 \|\|
keir@20452	1250 (pagetype != XEN_DOMCTL_PFINFO_L1TAB)) {
keir@20452	1251
keir@20591	1252 if (!uncanonicalize_pagetable(xc_handle, dom, ctx,
keir@20837	1253 page, superpages)) {
keir@20452	1254 /*
keir@20452	1255 ** Failing to uncanonicalize a page table can be ok
keir@20452	1256 ** under live migration since the pages type may have
keir@20452	1257 ** changed by now (and we'll get an update later).
keir@20452	1258 */
keir@20452	1259 DPRINTF("PT L%ld race on pfn=%08lx mfn=%08lx\n",
keir@20452	1260 pagetype >> 28, pfn, mfn);
keir@20452	1261 nraces++;
keir@20452	1262 continue;
keir@20452	1263 }
keir@20452	1264 }
keir@20452	1265 }
keir@20452	1266 else if ( pagetype != XEN_DOMCTL_PFINFO_NOTAB )
keir@20452	1267 {
keir@20452	1268 ERROR("Bogus page type %lx page table is out of range: "
keir@20585	1269 "i=%d p2m_size=%lu", pagetype, i, dinfo->p2m_size);
keir@20877	1270 goto err_mapped;
keir@20452	1271 }
keir@20452	1272
keir@20452	1273 if ( pagebuf->verify )
keir@20452	1274 {
keir@20452	1275 int res = memcmp(buf, (region_base + i*PAGE_SIZE), PAGE_SIZE);
keir@20452	1276 if ( res )
keir@20452	1277 {
keir@20452	1278 int v;
keir@20452	1279
keir@20452	1280 DPRINTF("************** pfn=%lx type=%lx gotcs=%08lx "
keir@20452	1281 "actualcs=%08lx\n", pfn, pagebuf->pfn_types[pfn],
keir@20452	1282 csum_page(region_base + (i + curbatch)*PAGE_SIZE),
keir@20452	1283 csum_page(buf));
keir@20452	1284
keir@20452	1285 for ( v = 0; v < 4; v++ )
keir@20452	1286 {
keir@20452	1287 unsigned long p = (unsigned long )
keir@20452	1288 (region_base + i*PAGE_SIZE);
keir@20452	1289 if ( buf[v] != p[v] )
keir@20452	1290 DPRINTF(" %d: %08lx %08lx\n", v, buf[v], p[v]);
keir@20452	1291 }
keir@20452	1292 }
keir@20452	1293 }
keir@20452	1294
keir@20452	1295 if ( !hvm &&
keir@20452	1296 xc_add_mmu_update(xc_handle, mmu,
keir@20452	1297 (((unsigned long long)mfn) << PAGE_SHIFT)
keir@20452	1298 \| MMU_MACHPHYS_UPDATE, pfn) )
keir@20452	1299 {
keir@20452	1300 ERROR("failed machpys update mfn=%lx pfn=%lx", mfn, pfn);
keir@20877	1301 goto err_mapped;
keir@20452	1302 }
keir@20452	1303 } /* end of 'batch' for loop */
keir@20452	1304
keir@20877	1305 rc = nraces;
keir@20452	1306
keir@20877	1307 err_mapped:
keir@20877	1308 munmap(region_base, j*PAGE_SIZE);
keir@20877	1309 free(pfn_err);
keir@20877	1310
keir@20877	1311 return rc;
keir@20452	1312 }
keir@20452	1313
Tim@14782	1314 int xc_domain_restore(int xc_handle, int io_fd, uint32_t dom,
Tim@14782	1315 unsigned int store_evtchn, unsigned long *store_mfn,
Tim@14782	1316 unsigned int console_evtchn, unsigned long *console_mfn,
keir@19677	1317 unsigned int hvm, unsigned int pae, int superpages)
mjw@1661	1318 {
kfraser@11295	1319 DECLARE_DOMCTL;
keir@16257	1320 int rc = 1, frc, i, j, n, m, pae_extended_cr3 = 0, ext_vcpucontext = 0;
kaf24@9698	1321 unsigned long mfn, pfn;
mjw@1661	1322 unsigned int prev_pc, this_pc;
kaf24@9698	1323 int nraces = 0;
mjw@1661	1324
mjw@1661	1325 /* The new domain's shared-info frame number. */
mjw@1661	1326 unsigned long shared_info_frame;
cl349@2964	1327 unsigned char shared_info_page[PAGE_SIZE]; /* saved contents from file */
keir@17918	1328 shared_info_any_t *old_shared_info =
keir@17918	1329 (shared_info_any_t *)shared_info_page;
keir@17918	1330 shared_info_any_t *new_shared_info;
kaf24@9698	1331
mjw@1661	1332 /* A copy of the CPU context of the guest. */
keir@17918	1333 vcpu_guest_context_any_t ctxt;
mjw@1661	1334
smh22@7740	1335 /* A table containing the type of each PFN (/not/ MFN!). */
mjw@1661	1336 unsigned long *pfn_type = NULL;
mjw@1661	1337
mjw@1661	1338 /* A table of MFNs to map in the current region */
kaf24@10314	1339 xen_pfn_t *region_mfn = NULL;
mjw@1661	1340
mjw@1661	1341 /* A copy of the pfn-to-mfn table frame list. */
kaf24@10314	1342 xen_pfn_t *p2m_frame_list = NULL;
Tim@14782	1343
cl349@6659	1344 /* A temporary mapping of the guest's start_info page. */
keir@17918	1345 start_info_any_t *start_info;
mjw@1661	1346
steven@13424	1347 /* Our mapping of the current region (batch) */
cl349@5014	1348 char *region_base;
mjw@1661	1349
kfraser@14808	1350 struct xc_mmu *mmu = NULL;
mjw@1661	1351
kaf24@4457	1352 struct mmuext_op pin[MAX_PIN_BATCH];
kaf24@9698	1353 unsigned int nr_pins;
kaf24@4457	1354
kfraser@14236	1355 uint64_t vcpumap = 1ULL;
kfraser@14236	1356 unsigned int max_vcpu_id = 0;
kfraser@14388	1357 int new_ctxt_format = 0;
smh22@7740	1358
keir@20452	1359 pagebuf_t pagebuf;
keir@20452	1360 tailbuf_t tailbuf, tmptail;
keir@20452	1361 void* vcpup;
keir@20452	1362
keir@20591	1363 static struct restore_ctx _ctx = {
keir@20591	1364 .live_p2m = NULL,
keir@20591	1365 .p2m = NULL,
keir@20591	1366 .no_superpage_mem = 0,
keir@20591	1367 };
keir@20591	1368 static struct restore_ctx *ctx = &_ctx;
keir@20589	1369 struct domain_info_context *dinfo = &ctx->dinfo;
keir@20589	1370
keir@20452	1371 pagebuf_init(&pagebuf);
keir@20452	1372 memset(&tailbuf, 0, sizeof(tailbuf));
keir@20457	1373 tailbuf.ishvm = hvm;
keir@20452	1374
steven@14732	1375 /* For info only */
keir@20587	1376 ctx->nr_pfns = 0;
steven@14732	1377
keir@20164	1378 /* Always try to allocate 2M pages for HVM */
keir@20164	1379 if ( hvm )
keir@20164	1380 superpages = 1;
keir@20164	1381
keir@20585	1382 if ( read_exact(io_fd, &dinfo->p2m_size, sizeof(unsigned long)) )
kfraser@14754	1383 {
kfraser@14754	1384 ERROR("read: p2m_size");
kfraser@14754	1385 goto out;
kfraser@14754	1386 }
keir@20585	1387 DPRINTF("xc_domain_restore start: p2m_size = %lx\n", dinfo->p2m_size);
smh22@7740	1388
Tim@15955	1389 if ( !get_platform_info(xc_handle, dom,
keir@20587	1390 &ctx->max_mfn, &ctx->hvirt_start, &ctx->pt_levels, &dinfo->guest_width) )
Tim@14782	1391 {
Tim@15955	1392 ERROR("Unable to get platform info.");
Tim@15955	1393 return 1;
Tim@15955	1394 }
Tim@15955	1395
Tim@15955	1396 /* The current word size of the guest isn't very interesting; for now
Tim@15955	1397 * assume the guest will be the same as we are. We'll fix that later
Tim@15955	1398 * if we discover otherwise. */
keir@20585	1399 dinfo->guest_width = sizeof(unsigned long);
keir@20587	1400 ctx->pt_levels = (dinfo->guest_width == 8) ? 4 : (ctx->pt_levels == 2) ? 2 : 3;
Tim@15955	1401
Tim@15955	1402 if ( !hvm )
Tim@15955	1403 {
Tim@15955	1404 /* Load the p2m frame list, plus potential extended info chunk */
keir@20591	1405 p2m_frame_list = load_p2m_frame_list(ctx,
keir@16257	1406 io_fd, &pae_extended_cr3, &ext_vcpucontext);
Tim@15955	1407 if ( !p2m_frame_list )
Tim@15955	1408 goto out;
Tim@15955	1409
Tim@15955	1410 /* Now that we know the word size, tell Xen about it */
Tim@14782	1411 memset(&domctl, 0, sizeof(domctl));
Tim@14782	1412 domctl.domain = dom;
Tim@14782	1413 domctl.cmd = XEN_DOMCTL_set_address_size;
keir@20585	1414 domctl.u.address_size.size = dinfo->guest_width * 8;
keir@16257	1415 frc = do_domctl(xc_handle, &domctl);
keir@16257	1416 if ( frc != 0 )
keir@14809	1417 {
Tim@14782	1418 ERROR("Unable to set guest address size.");
Tim@14782	1419 goto out;
Tim@14782	1420 }
Christian@14235	1421 }
smh22@7740	1422
cl349@5091	1423 /* We want zeroed memory so use calloc rather than malloc. */
keir@20587	1424 ctx->p2m = calloc(dinfo->p2m_size, sizeof(xen_pfn_t));
keir@20585	1425 pfn_type = calloc(dinfo->p2m_size, sizeof(unsigned long));
keir@17462	1426
keir@20887	1427 region_mfn = xc_memalign(PAGE_SIZE, ROUNDUP(
keir@17462	1428 MAX_BATCH_SIZE * sizeof(xen_pfn_t), PAGE_SHIFT));
mjw@1661	1429
keir@20587	1430 if ( (ctx->p2m == NULL) \|\| (pfn_type == NULL) \|\|
keir@19677	1431 (region_mfn == NULL) )
keir@14809	1432 {
kfraser@11814	1433 ERROR("memory alloc failed");
mjw@1661	1434 errno = ENOMEM;
mjw@1661	1435 goto out;
mjw@1661	1436 }
kaf24@9698	1437
keir@17462	1438 memset(region_mfn, 0,
keir@17462	1439 ROUNDUP(MAX_BATCH_SIZE * sizeof(xen_pfn_t), PAGE_SHIFT));
keir@17462	1440
keir@14809	1441 if ( lock_pages(region_mfn, sizeof(xen_pfn_t) * MAX_BATCH_SIZE) )
keir@14809	1442 {
kfraser@11895	1443 ERROR("Could not lock region_mfn");
iap10@2202	1444 goto out;
iap10@2202	1445 }
mjw@1661	1446
mjw@1661	1447 /* Get the domain's shared-info frame. */
kfraser@11295	1448 domctl.cmd = XEN_DOMCTL_getdomaininfo;
kfraser@11295	1449 domctl.domain = (domid_t)dom;
keir@14809	1450 if ( xc_domctl(xc_handle, &domctl) < 0 )
keir@14809	1451 {
kfraser@11814	1452 ERROR("Could not get information on new domain");
mjw@1661	1453 goto out;
mjw@1661	1454 }
kfraser@11295	1455 shared_info_frame = domctl.u.getdomaininfo.shared_info_frame;
mjw@1661	1456
steven@13424	1457 /* Mark all PFNs as invalid; we allocate on demand */
keir@20585	1458 for ( pfn = 0; pfn < dinfo->p2m_size; pfn++ )
keir@20587	1459 ctx->p2m[pfn] = INVALID_P2M_ENTRY;
smh22@7886	1460
kfraser@14808	1461 mmu = xc_alloc_mmu_updates(xc_handle, dom);
kfraser@14808	1462 if ( mmu == NULL )
kfraser@14808	1463 {
kfraser@11814	1464 ERROR("Could not initialise for MMU updates");
mjw@1661	1465 goto out;
mjw@1661	1466 }
mjw@1661	1467
smh22@7740	1468 DPRINTF("Reloading memory pages: 0%%\n");
mjw@1661	1469
mjw@1661	1470 /*
mjw@1661	1471 * Now simply read each saved frame into its new machine frame.
mjw@1661	1472 * We uncanonicalise page tables as we go.
mjw@1661	1473 */
mjw@1661	1474 prev_pc = 0;
mjw@1661	1475
keir@14142	1476 n = m = 0;
keir@20452	1477 loadpages:
keir@14809	1478 for ( ; ; )
keir@14809	1479 {
keir@20452	1480 int j, curbatch;
mjw@1661	1481
keir@20585	1482 this_pc = (n * 100) / dinfo->p2m_size;
kaf24@1683	1483 if ( (this_pc - prev_pc) >= 5 )
kaf24@1683	1484 {
cl349@5091	1485 PPRINTF("\b\b\b\b%3d%%", this_pc);
mjw@1661	1486 prev_pc = this_pc;
mjw@1661	1487 }
mjw@1661	1488
keir@20452	1489 if ( !completed ) {
keir@20452	1490 pagebuf.nr_physpages = pagebuf.nr_pages = 0;
keir@20452	1491 if ( pagebuf_get_one(&pagebuf, io_fd, xc_handle, dom) < 0 ) {
keir@20452	1492 ERROR("Error when reading batch\n");
keir@20452	1493 goto out;
keir@20452	1494 }
mjw@1661	1495 }
keir@20452	1496 j = pagebuf.nr_pages;
mjw@1661	1497
cl349@5091	1498 PPRINTF("batch %d\n",j);
kaf24@9698	1499
keir@20452	1500 if ( j == 0 ) {
keir@20452	1501 /* catch vcpu updates */
keir@20452	1502 if (pagebuf.new_ctxt_format) {
keir@20452	1503 vcpumap = pagebuf.vcpumap;
keir@20452	1504 max_vcpu_id = pagebuf.max_vcpu_id;
keir@18929	1505 }
keir@20452	1506 /* should this be deferred? does it change? */
keir@20452	1507 if ( pagebuf.identpt )
keir@20452	1508 xc_set_hvm_param(xc_handle, dom, HVM_PARAM_IDENT_PT, pagebuf.identpt);
keir@20452	1509 if ( pagebuf.vm86_tss )
keir@20452	1510 xc_set_hvm_param(xc_handle, dom, HVM_PARAM_VM86_TSS, pagebuf.vm86_tss);
kaf24@1683	1511 break; /* our work here is done */
mjw@1661	1512 }
mjw@1661	1513
keir@20452	1514 /* break pagebuf into batches */
keir@20452	1515 curbatch = 0;
keir@20452	1516 while ( curbatch < j ) {
keir@20452	1517 int brc;
kaf24@9698	1518
keir@20591	1519 brc = apply_batch(xc_handle, dom, ctx, region_mfn, pfn_type,
keir@20452	1520 pae_extended_cr3, hvm, mmu, &pagebuf, curbatch, superpages);
keir@20452	1521 if ( brc < 0 )
mjw@1661	1522 goto out;
mjw@1661	1523
keir@20452	1524 nraces += brc;
smh22@7740	1525
keir@20452	1526 curbatch += MAX_BATCH_SIZE;
keir@20452	1527 }
mjw@1661	1528
keir@20452	1529 pagebuf.nr_physpages = pagebuf.nr_pages = 0;
mjw@1661	1530
keir@20452	1531 n += j; /* crude stats */
keir@14142	1532
keir@14142	1533 /*
keir@14142	1534 * Discard cache for portion of file read so far up to last
keir@14142	1535 * page boundary every 16MB or so.
keir@14142	1536 */
keir@14142	1537 m += j;
keir@14142	1538 if ( m > MAX_PAGECACHE_USAGE )
keir@14142	1539 {
keir@14142	1540 discard_file_cache(io_fd, 0 /* no flush */);
keir@14142	1541 m = 0;
keir@14142	1542 }
mjw@1661	1543 }
mjw@1661	1544
kfraser@10383	1545 /*
kfraser@10383	1546 * Ensure we flush all machphys updates before potential PAE-specific
kfraser@10383	1547 * reallocations below.
kfraser@10383	1548 */
keir@14809	1549 if ( !hvm && xc_flush_mmu_updates(xc_handle, mmu) )
keir@14809	1550 {
kfraser@14808	1551 ERROR("Error doing flush_mmu_updates()");
kfraser@10383	1552 goto out;
kfraser@10383	1553 }
kfraser@10383	1554
keir@20452	1555 // DPRINTF("Received all pages (%d races)\n", nraces);
mjw@1661	1556
keir@20452	1557 if ( !completed ) {
keir@20453	1558 int flags = 0;
keir@20453	1559
keir@20591	1560 if ( buffer_tail(ctx, &tailbuf, io_fd, max_vcpu_id, vcpumap,
keir@20452	1561 ext_vcpucontext) < 0 ) {
keir@20452	1562 ERROR ("error buffering image tail");
keir@20452	1563 goto out;
keir@20452	1564 }
keir@20452	1565 completed = 1;
keir@20453	1566 /* shift into nonblocking mode for the remainder */
keir@20453	1567 if ( (flags = fcntl(io_fd, F_GETFL,0)) < 0 )
keir@20453	1568 flags = 0;
keir@20453	1569 fcntl(io_fd, F_SETFL, flags \| O_NONBLOCK);
keir@20452	1570 }
keir@20452	1571
keir@20452	1572 // DPRINTF("Buffered checkpoint\n");
keir@20452	1573
keir@20452	1574 if ( pagebuf_get(&pagebuf, io_fd, xc_handle, dom) ) {
keir@20452	1575 ERROR("error when buffering batch, finishing\n");
keir@20452	1576 goto finish;
keir@20452	1577 }
keir@20452	1578 memset(&tmptail, 0, sizeof(tmptail));
keir@20457	1579 tmptail.ishvm = hvm;
keir@20591	1580 if ( buffer_tail(ctx, &tmptail, io_fd, max_vcpu_id, vcpumap,
keir@20452	1581 ext_vcpucontext) < 0 ) {
keir@20452	1582 ERROR ("error buffering image tail, finishing");
keir@20452	1583 goto finish;
keir@20452	1584 }
keir@20452	1585 tailbuf_free(&tailbuf);
keir@20452	1586 memcpy(&tailbuf, &tmptail, sizeof(tailbuf));
keir@20452	1587
keir@20452	1588 goto loadpages;
keir@20452	1589
keir@20452	1590 finish:
keir@20457	1591 if ( hvm )
keir@20457	1592 goto finish_hvm;
keir@20452	1593
keir@20587	1594 if ( (ctx->pt_levels == 3) && !pae_extended_cr3 )
keir@14809	1595 {
kaf24@10304	1596 /*
kaf24@10304	1597 ** XXX SMH on PAE we need to ensure PGDs are in MFNs < 4G. This
kaf24@10304	1598 ** is a little awkward and involves (a) finding all such PGDs and
kaf24@10304	1599 ** replacing them with 'lowmem' versions; (b) upating the p2m[]
kaf24@10304	1600 ** with the new info; and (c) canonicalizing all the L1s using the
kaf24@10304	1601 ** (potentially updated) p2m[].
kaf24@10304	1602 **
kaf24@10304	1603 ** This is relatively slow (and currently involves two passes through
kaf24@10304	1604 ** the pfn_type[] array), but at least seems to be correct. May wish
kaf24@10304	1605 ** to consider more complex approaches to optimize this later.
kaf24@10304	1606 */
kaf24@10304	1607
kaf24@10304	1608 int j, k;
steven@13424	1609
kaf24@10304	1610 /* First pass: find all L3TABs current in > 4G mfns and get new mfns */
keir@20585	1611 for ( i = 0; i < dinfo->p2m_size; i++ )
kfraser@11295	1612 {
kfraser@11295	1613 if ( ((pfn_type[i] & XEN_DOMCTL_PFINFO_LTABTYPE_MASK) ==
kfraser@11295	1614 XEN_DOMCTL_PFINFO_L3TAB) &&
keir@20587	1615 (ctx->p2m[i] > 0xfffffUL) )
kfraser@11295	1616 {
kaf24@10304	1617 unsigned long new_mfn;
kaf24@10304	1618 uint64_t l3ptes[4];
kaf24@10304	1619 uint64_t *l3tab;
kaf24@10304	1620
kaf24@10304	1621 l3tab = (uint64_t *)
kaf24@10304	1622 xc_map_foreign_range(xc_handle, dom, PAGE_SIZE,
keir@20587	1623 PROT_READ, ctx->p2m[i]);
kaf24@10304	1624
keir@14809	1625 for ( j = 0; j < 4; j++ )
kaf24@10304	1626 l3ptes[j] = l3tab[j];
kaf24@10304	1627
kaf24@10304	1628 munmap(l3tab, PAGE_SIZE);
kaf24@10304	1629
keir@20587	1630 new_mfn = xc_make_page_below_4G(xc_handle, dom, ctx->p2m[i]);
keir@14809	1631 if ( !new_mfn )
keir@14809	1632 {
kfraser@11814	1633 ERROR("Couldn't get a page below 4GB :-(");
kaf24@10304	1634 goto out;
kaf24@10304	1635 }
kaf24@10304	1636
keir@20587	1637 ctx->p2m[i] = new_mfn;
keir@14809	1638 if ( xc_add_mmu_update(xc_handle, mmu,
keir@14809	1639 (((unsigned long long)new_mfn)
keir@14809	1640 << PAGE_SHIFT) \|
keir@14809	1641 MMU_MACHPHYS_UPDATE, i) )
keir@14809	1642 {
kfraser@11814	1643 ERROR("Couldn't m2p on PAE root pgdir");
kaf24@10304	1644 goto out;
kaf24@10304	1645 }
kaf24@10304	1646
kaf24@10304	1647 l3tab = (uint64_t *)
kaf24@10304	1648 xc_map_foreign_range(xc_handle, dom, PAGE_SIZE,
keir@20587	1649 PROT_READ \| PROT_WRITE, ctx->p2m[i]);
kaf24@10304	1650
keir@14809	1651 for ( j = 0; j < 4; j++ )
kaf24@10304	1652 l3tab[j] = l3ptes[j];
kaf24@10304	1653
kaf24@10304	1654 munmap(l3tab, PAGE_SIZE);
kaf24@10304	1655 }
kaf24@10304	1656 }
kaf24@10304	1657
kaf24@10304	1658 /* Second pass: find all L1TABs and uncanonicalize them */
kaf24@10304	1659 j = 0;
kaf24@10304	1660
keir@20585	1661 for ( i = 0; i < dinfo->p2m_size; i++ )
kfraser@11295	1662 {
kfraser@11295	1663 if ( ((pfn_type[i] & XEN_DOMCTL_PFINFO_LTABTYPE_MASK) ==
kfraser@11295	1664 XEN_DOMCTL_PFINFO_L1TAB) )
kfraser@11295	1665 {
keir@20587	1666 region_mfn[j] = ctx->p2m[i];
kaf24@10304	1667 j++;
kaf24@10304	1668 }
kaf24@10304	1669
keir@20585	1670 if ( (i == (dinfo->p2m_size-1)) \|\| (j == MAX_BATCH_SIZE) )
keir@14809	1671 {
keir@20837	1672 region_base = xc_map_foreign_pages(
keir@14809	1673 xc_handle, dom, PROT_READ \| PROT_WRITE, region_mfn, j);
keir@14809	1674 if ( region_base == NULL )
keir@14809	1675 {
kfraser@11814	1676 ERROR("map batch failed");
kaf24@10304	1677 goto out;
kaf24@10304	1678 }
kaf24@10304	1679
keir@14809	1680 for ( k = 0; k < j; k++ )
keir@14809	1681 {
keir@14809	1682 if ( !uncanonicalize_pagetable(
keir@20837	1683 xc_handle, dom, ctx,
keir@19677	1684 region_base + k*PAGE_SIZE, superpages) )
keir@14809	1685 {
kfraser@11814	1686 ERROR("failed uncanonicalize pt!");
kaf24@10304	1687 goto out;
kaf24@10304	1688 }
kaf24@10304	1689 }
kaf24@10304	1690
kaf24@10304	1691 munmap(region_base, j*PAGE_SIZE);
kaf24@10304	1692 j = 0;
kaf24@10304	1693 }
kaf24@10304	1694 }
kaf24@10304	1695
keir@14809	1696 if ( xc_flush_mmu_updates(xc_handle, mmu) )
keir@14809	1697 {
kfraser@14808	1698 ERROR("Error doing xc_flush_mmu_updates()");
kfraser@10383	1699 goto out;
kfraser@10383	1700 }
kaf24@10304	1701 }
kaf24@10304	1702
mjw@1661	1703 /*
mjw@1661	1704 * Pin page tables. Do this after writing to them as otherwise Xen
mjw@1661	1705 * will barf when doing the type-checking.
mjw@1661	1706 */
kaf24@9698	1707 nr_pins = 0;
keir@20585	1708 for ( i = 0; i < dinfo->p2m_size; i++ )
kfraser@11295	1709 {
kfraser@11295	1710 if ( (pfn_type[i] & XEN_DOMCTL_PFINFO_LPINTAB) == 0 )
kaf24@4457	1711 continue;
smh22@7886	1712
kfraser@11295	1713 switch ( pfn_type[i] & XEN_DOMCTL_PFINFO_LTABTYPE_MASK )
kfraser@11295	1714 {
kfraser@11295	1715 case XEN_DOMCTL_PFINFO_L1TAB:
kaf24@4457	1716 pin[nr_pins].cmd = MMUEXT_PIN_L1_TABLE;
kaf24@9698	1717 break;
kaf24@9698	1718
kfraser@11295	1719 case XEN_DOMCTL_PFINFO_L2TAB:
kaf24@4457	1720 pin[nr_pins].cmd = MMUEXT_PIN_L2_TABLE;
kaf24@9698	1721 break;
kaf24@9698	1722
kfraser@11295	1723 case XEN_DOMCTL_PFINFO_L3TAB:
smh22@7740	1724 pin[nr_pins].cmd = MMUEXT_PIN_L3_TABLE;
kaf24@9698	1725 break;
smh22@7740	1726
kfraser@11295	1727 case XEN_DOMCTL_PFINFO_L4TAB:
smh22@7740	1728 pin[nr_pins].cmd = MMUEXT_PIN_L4_TABLE;
kaf24@9698	1729 break;
kaf24@9698	1730
kaf24@9698	1731 default:
kaf24@9698	1732 continue;
smh22@7740	1733 }
smh22@7740	1734
keir@20587	1735 pin[nr_pins].arg1.mfn = ctx->p2m[i];
kaf24@9698	1736 nr_pins++;
smh22@7886	1737
kaf24@10513	1738 /* Batch full? Then flush. */
keir@14809	1739 if ( nr_pins == MAX_PIN_BATCH )
keir@14809	1740 {
keir@14809	1741 if ( xc_mmuext_op(xc_handle, pin, nr_pins, dom) < 0 )
keir@14809	1742 {
kfraser@11814	1743 ERROR("Failed to pin batch of %d page tables", nr_pins);
kaf24@10513	1744 goto out;
kaf24@10513	1745 }
kaf24@10513	1746 nr_pins = 0;
kaf24@10513	1747 }
kaf24@10513	1748 }
kaf24@10513	1749
kaf24@10513	1750 /* Flush final partial batch. */
keir@14809	1751 if ( (nr_pins != 0) && (xc_mmuext_op(xc_handle, pin, nr_pins, dom) < 0) )
keir@14809	1752 {
kfraser@11814	1753 ERROR("Failed to pin batch of %d page tables", nr_pins);
kaf24@10513	1754 goto out;
iap10@2507	1755 }
mjw@1661	1756
cl349@5091	1757 DPRINTF("\b\b\b\b100%%\n");
keir@20587	1758 DPRINTF("Memory reloaded (%ld pages)\n", ctx->nr_pfns);
mjw@1661	1759
iap10@2291	1760 /* Get the list of PFNs that are not in the psuedo-phys map */
iap10@2291	1761 {
keir@20452	1762 int nr_frees = 0;
kaf24@2624	1763
keir@20457	1764 for ( i = 0; i < tailbuf.u.pv.pfncount; i++ )
keir@14809	1765 {
keir@20457	1766 unsigned long pfn = tailbuf.u.pv.pfntab[i];
smh22@7740	1767
keir@20587	1768 if ( ctx->p2m[pfn] != INVALID_P2M_ENTRY )
keir@14809	1769 {
keir@20452	1770 /* pfn is not in physmap now, but was at some point during
steven@13424	1771 the save/migration process - need to free it */
keir@20587	1772 tailbuf.u.pv.pfntab[nr_frees++] = ctx->p2m[pfn];
keir@20587	1773 ctx->p2m[pfn] = INVALID_P2M_ENTRY; /* not in pseudo-physical map */
steven@13424	1774 }
kaf24@6775	1775 }
kaf24@9698	1776
keir@14809	1777 if ( nr_frees > 0 )
keir@14809	1778 {
kaf24@6506	1779 struct xen_memory_reservation reservation = {
steven@13424	1780 .nr_extents = nr_frees,
kaf24@6506	1781 .extent_order = 0,
kaf24@6510	1782 .domid = dom
kaf24@6506	1783 };
keir@20457	1784 set_xen_guest_handle(reservation.extent_start, tailbuf.u.pv.pfntab);
smh22@7740	1785
keir@16257	1786 if ( (frc = xc_memory_op(xc_handle, XENMEM_decrease_reservation,
keir@16257	1787 &reservation)) != nr_frees )
keir@14809	1788 {
keir@16257	1789 ERROR("Could not decrease reservation : %d", frc);
kaf24@6775	1790 goto out;
keir@14809	1791 }
keir@14809	1792 else
keir@20457	1793 DPRINTF("Decreased reservation by %d pages\n", tailbuf.u.pv.pfncount);
kaf24@9698	1794 }
iap10@2291	1795 }
iap10@2291	1796
keir@17462	1797 if ( lock_pages(&ctxt, sizeof(ctxt)) )
keir@17462	1798 {
keir@17462	1799 ERROR("Unable to lock ctxt");
keir@17462	1800 return 1;
keir@17462	1801 }
keir@17462	1802
keir@20457	1803 vcpup = tailbuf.u.pv.vcpubuf;
keir@14809	1804 for ( i = 0; i <= max_vcpu_id; i++ )
keir@14809	1805 {
keir@14809	1806 if ( !(vcpumap & (1ULL << i)) )
kfraser@14236	1807 continue;
kfraser@14236	1808
keir@20585	1809 memcpy(&ctxt, vcpup, ((dinfo->guest_width == 8) ? sizeof(ctxt.x64)
keir@20452	1810 : sizeof(ctxt.x32)));
keir@20585	1811 vcpup += (dinfo->guest_width == 8) ? sizeof(ctxt.x64) : sizeof(ctxt.x32);
keir@20452	1812
keir@20452	1813 DPRINTF("read VCPU %d\n", i);
kfraser@14236	1814
kfraser@14388	1815 if ( !new_ctxt_format )
Tim@15955	1816 SET_FIELD(&ctxt, flags, GET_FIELD(&ctxt, flags) \| VGCF_online);
kfraser@14388	1817
keir@14809	1818 if ( i == 0 )
keir@14809	1819 {
kfraser@14236	1820 /*
kfraser@14236	1821 * Uncanonicalise the suspend-record frame number and poke
kfraser@14236	1822 * resume record.
kfraser@14236	1823 */
Tim@15955	1824 pfn = GET_FIELD(&ctxt, user_regs.edx);
keir@20585	1825 if ( (pfn >= dinfo->p2m_size) \|\|
keir@14809	1826 (pfn_type[pfn] != XEN_DOMCTL_PFINFO_NOTAB) )
keir@14809	1827 {
kfraser@14236	1828 ERROR("Suspend record frame number is bad");
kfraser@14236	1829 goto out;
kfraser@14236	1830 }
keir@20587	1831 mfn = ctx->p2m[pfn];
Tim@15955	1832 SET_FIELD(&ctxt, user_regs.edx, mfn);
kfraser@14236	1833 start_info = xc_map_foreign_range(
kfraser@14236	1834 xc_handle, dom, PAGE_SIZE, PROT_READ \| PROT_WRITE, mfn);
keir@20585	1835 SET_FIELD(start_info, nr_pages, dinfo->p2m_size);
Tim@15955	1836 SET_FIELD(start_info, shared_info, shared_info_frame<<PAGE_SHIFT);
Tim@15955	1837 SET_FIELD(start_info, flags, 0);
keir@20587	1838 *store_mfn = ctx->p2m[GET_FIELD(start_info, store_mfn)];
Tim@15955	1839 SET_FIELD(start_info, store_mfn, *store_mfn);
Tim@15955	1840 SET_FIELD(start_info, store_evtchn, store_evtchn);
keir@20587	1841 *console_mfn = ctx->p2m[GET_FIELD(start_info, console.domU.mfn)];
Tim@15955	1842 SET_FIELD(start_info, console.domU.mfn, *console_mfn);
Tim@15955	1843 SET_FIELD(start_info, console.domU.evtchn, console_evtchn);
kfraser@14236	1844 munmap(start_info, PAGE_SIZE);
kfraser@14236	1845 }
kfraser@14236	1846 /* Uncanonicalise each GDT frame number. */
Tim@15955	1847 if ( GET_FIELD(&ctxt, gdt_ents) > 8192 )
keir@14809	1848 {
kfraser@14236	1849 ERROR("GDT entry count out of range");
kfraser@14236	1850 goto out;
kfraser@14236	1851 }
kfraser@14236	1852
Tim@15955	1853 for ( j = 0; (512*j) < GET_FIELD(&ctxt, gdt_ents); j++ )
keir@14809	1854 {
Tim@15955	1855 pfn = GET_FIELD(&ctxt, gdt_frames[j]);
keir@20585	1856 if ( (pfn >= dinfo->p2m_size) \|\|
keir@14809	1857 (pfn_type[pfn] != XEN_DOMCTL_PFINFO_NOTAB) )
keir@14809	1858 {
Tim@15955	1859 ERROR("GDT frame number %i (0x%lx) is bad",
Tim@15955	1860 j, (unsigned long)pfn);
kfraser@14236	1861 goto out;
kfraser@14236	1862 }
keir@20587	1863 SET_FIELD(&ctxt, gdt_frames[j], ctx->p2m[pfn]);
kfraser@14236	1864 }
kfraser@14236	1865 /* Uncanonicalise the page table base pointer. */
keir@17074	1866 pfn = UNFOLD_CR3(GET_FIELD(&ctxt, ctrlreg[3]));
kfraser@14236	1867
keir@20585	1868 if ( pfn >= dinfo->p2m_size )
keir@14809	1869 {
steven@14732	1870 ERROR("PT base is bad: pfn=%lu p2m_size=%lu type=%08lx",
keir@20585	1871 pfn, dinfo->p2m_size, pfn_type[pfn]);
kfraser@14236	1872 goto out;
kfraser@14236	1873 }
kfraser@14236	1874
kfraser@14236	1875 if ( (pfn_type[pfn] & XEN_DOMCTL_PFINFO_LTABTYPE_MASK) !=
keir@20587	1876 ((unsigned long)ctx->pt_levels<<XEN_DOMCTL_PFINFO_LTAB_SHIFT) )
keir@14809	1877 {
kfraser@14236	1878 ERROR("PT base is bad. pfn=%lu nr=%lu type=%08lx %08lx",
keir@20585	1879 pfn, dinfo->p2m_size, pfn_type[pfn],
keir@20587	1880 (unsigned long)ctx->pt_levels<<XEN_DOMCTL_PFINFO_LTAB_SHIFT);
kfraser@14236	1881 goto out;
kfraser@14236	1882 }
keir@20587	1883 SET_FIELD(&ctxt, ctrlreg[3], FOLD_CR3(ctx->p2m[pfn]));
kfraser@14236	1884
keir@14254	1885 /* Guest pagetable (x86/64) stored in otherwise-unused CR1. */
keir@20587	1886 if ( (ctx->pt_levels == 4) && (ctxt.x64.ctrlreg[1] & 1) )
keir@14254	1887 {
keir@17074	1888 pfn = UNFOLD_CR3(ctxt.x64.ctrlreg[1] & ~1);
keir@20585	1889 if ( pfn >= dinfo->p2m_size )
keir@14809	1890 {
Tim@15955	1891 ERROR("User PT base is bad: pfn=%lu p2m_size=%lu",
keir@20585	1892 pfn, dinfo->p2m_size);
keir@14254	1893 goto out;
keir@14254	1894 }
keir@14254	1895 if ( (pfn_type[pfn] & XEN_DOMCTL_PFINFO_LTABTYPE_MASK) !=
keir@20587	1896 ((unsigned long)ctx->pt_levels<<XEN_DOMCTL_PFINFO_LTAB_SHIFT) )
keir@14809	1897 {
keir@14254	1898 ERROR("User PT base is bad. pfn=%lu nr=%lu type=%08lx %08lx",
keir@20585	1899 pfn, dinfo->p2m_size, pfn_type[pfn],
keir@20587	1900 (unsigned long)ctx->pt_levels<<XEN_DOMCTL_PFINFO_LTAB_SHIFT);
keir@14254	1901 goto out;
keir@14254	1902 }
keir@20587	1903 ctxt.x64.ctrlreg[1] = FOLD_CR3(ctx->p2m[pfn]);
keir@14254	1904 }
kfraser@14236	1905 domctl.cmd = XEN_DOMCTL_setvcpucontext;
kfraser@14236	1906 domctl.domain = (domid_t)dom;
kfraser@14236	1907 domctl.u.vcpucontext.vcpu = i;
Tim@15955	1908 set_xen_guest_handle(domctl.u.vcpucontext.ctxt, &ctxt.c);
keir@16257	1909 frc = xc_domctl(xc_handle, &domctl);
keir@16257	1910 if ( frc != 0 )
keir@14809	1911 {
kfraser@14236	1912 ERROR("Couldn't build vcpu%d", i);
kfraser@14236	1913 goto out;
kfraser@14236	1914 }
keir@16257	1915
keir@16257	1916 if ( !ext_vcpucontext )
keir@16257	1917 continue;
keir@20452	1918 memcpy(&domctl.u.ext_vcpucontext, vcpup, 128);
keir@20452	1919 vcpup += 128;
keir@16257	1920 domctl.cmd = XEN_DOMCTL_set_ext_vcpucontext;
keir@16257	1921 domctl.domain = dom;
keir@16257	1922 frc = xc_domctl(xc_handle, &domctl);
keir@16257	1923 if ( frc != 0 )
keir@16257	1924 {
keir@16257	1925 ERROR("Couldn't set extended vcpu%d info\n", i);
keir@16257	1926 goto out;
keir@16257	1927 }
mjw@1661	1928 }
mjw@1661	1929
keir@20457	1930 memcpy(shared_info_page, tailbuf.u.pv.shared_info_page, PAGE_SIZE);
keir@20452	1931
keir@20452	1932 DPRINTF("Completed checkpoint load\n");
mjw@1661	1933
ian@15744	1934 /* Restore contents of shared-info page. No checking needed. */
ian@15744	1935 new_shared_info = xc_map_foreign_range(
ian@15744	1936 xc_handle, dom, PAGE_SIZE, PROT_WRITE, shared_info_frame);
ian@15744	1937
ian@15744	1938 /* restore saved vcpu_info and arch specific info */
Tim@15955	1939 MEMCPY_FIELD(new_shared_info, old_shared_info, vcpu_info);
Tim@15955	1940 MEMCPY_FIELD(new_shared_info, old_shared_info, arch);
mjw@1661	1941
ian@15744	1942 /* clear any pending events and the selector */
Tim@15955	1943 MEMSET_ARRAY_FIELD(new_shared_info, evtchn_pending, 0);
keir@19826	1944 for ( i = 0; i < XEN_LEGACY_MAX_VCPUS; i++ )
Tim@15955	1945 SET_FIELD(new_shared_info, vcpu_info[i].evtchn_pending_sel, 0);
ian@15744	1946
ian@15744	1947 /* mask event channels */
Tim@15955	1948 MEMSET_ARRAY_FIELD(new_shared_info, evtchn_mask, 0xff);
ian@15744	1949
ian@15744	1950 /* leave wallclock time. set by hypervisor */
ian@15744	1951 munmap(new_shared_info, PAGE_SIZE);
kaf24@9698	1952
mjw@1661	1953 /* Uncanonicalise the pfn-to-mfn table frame-number list. */
keir@14809	1954 for ( i = 0; i < P2M_FL_ENTRIES; i++ )
keir@14809	1955 {
smh22@7740	1956 pfn = p2m_frame_list[i];
keir@20585	1957 if ( (pfn >= dinfo->p2m_size) \|\| (pfn_type[pfn] != XEN_DOMCTL_PFINFO_NOTAB) )
keir@14809	1958 {
Tim@15955	1959 ERROR("PFN-to-MFN frame number %i (%#lx) is bad", i, pfn);
mjw@1661	1960 goto out;
mjw@1661	1961 }
keir@20587	1962 p2m_frame_list[i] = ctx->p2m[pfn];
mjw@1661	1963 }
kaf24@9698	1964
smh22@7740	1965 /* Copy the P2M we've constructed to the 'live' P2M */
keir@20837	1966 if ( !(ctx->live_p2m = xc_map_foreign_pages(xc_handle, dom, PROT_WRITE,
keir@14809	1967 p2m_frame_list, P2M_FL_ENTRIES)) )
keir@14809	1968 {
kfraser@11814	1969 ERROR("Couldn't map p2m table");
mjw@1661	1970 goto out;
mjw@1661	1971 }
mjw@1661	1972
Tim@15955	1973 /* If the domain we're restoring has a different word size to ours,
keir@18367	1974 * we need to adjust the live_p2m assignment appropriately */
keir@20585	1975 if ( dinfo->guest_width > sizeof (xen_pfn_t) )
keir@20585	1976 for ( i = dinfo->p2m_size - 1; i >= 0; i-- )
keir@20587	1977 ((int64_t *)ctx->live_p2m)[i] = (long)ctx->p2m[i];
keir@20585	1978 else if ( dinfo->guest_width < sizeof (xen_pfn_t) )
keir@20585	1979 for ( i = 0; i < dinfo->p2m_size; i++ )
keir@20587	1980 ((uint32_t *)ctx->live_p2m)[i] = ctx->p2m[i];
keir@18367	1981 else
keir@20587	1982 memcpy(ctx->live_p2m, ctx->p2m, dinfo->p2m_size * sizeof(xen_pfn_t));
keir@20587	1983 munmap(ctx->live_p2m, P2M_FL_ENTRIES * PAGE_SIZE);
mjw@1661	1984
cl349@5091	1985 DPRINTF("Domain ready to be built.\n");
Tim@14782	1986 rc = 0;
keir@20457	1987 goto out;
keir@20457	1988
keir@20457	1989 finish_hvm:
keir@20457	1990 /* Dump the QEMU state to a state file for QEMU to load */
keir@20457	1991 if ( dump_qemu(dom, &tailbuf.u.hvm) ) {
keir@20457	1992 ERROR("Error dumping QEMU state to file");
keir@20457	1993 goto out;
keir@20457	1994 }
keir@20457	1995
keir@20457	1996 /* These comms pages need to be zeroed at the start of day */
keir@20457	1997 if ( xc_clear_domain_page(xc_handle, dom, tailbuf.u.hvm.magicpfns[0]) \|\|
keir@20457	1998 xc_clear_domain_page(xc_handle, dom, tailbuf.u.hvm.magicpfns[1]) \|\|
keir@20457	1999 xc_clear_domain_page(xc_handle, dom, tailbuf.u.hvm.magicpfns[2]) )
keir@20457	2000 {
keir@20457	2001 ERROR("error zeroing magic pages");
keir@20457	2002 goto out;
keir@20457	2003 }
keir@20457	2004
keir@20457	2005 if ( (frc = xc_set_hvm_param(xc_handle, dom,
keir@20457	2006 HVM_PARAM_IOREQ_PFN, tailbuf.u.hvm.magicpfns[0]))
keir@20457	2007 \|\| (frc = xc_set_hvm_param(xc_handle, dom,
keir@20457	2008 HVM_PARAM_BUFIOREQ_PFN, tailbuf.u.hvm.magicpfns[1]))
keir@20457	2009 \|\| (frc = xc_set_hvm_param(xc_handle, dom,
keir@20457	2010 HVM_PARAM_STORE_PFN, tailbuf.u.hvm.magicpfns[2]))
keir@20457	2011 \|\| (frc = xc_set_hvm_param(xc_handle, dom,
keir@20457	2012 HVM_PARAM_PAE_ENABLED, pae))
keir@20457	2013 \|\| (frc = xc_set_hvm_param(xc_handle, dom,
keir@20457	2014 HVM_PARAM_STORE_EVTCHN,
keir@20457	2015 store_evtchn)) )
keir@20457	2016 {
keir@20457	2017 ERROR("error setting HVM params: %i", frc);
keir@20457	2018 goto out;
keir@20457	2019 }
keir@20457	2020 *store_mfn = tailbuf.u.hvm.magicpfns[2];
keir@20457	2021
keir@20457	2022 frc = xc_domain_hvm_setcontext(xc_handle, dom, tailbuf.u.hvm.hvmbuf,
keir@20457	2023 tailbuf.u.hvm.reclen);
keir@20457	2024 if ( frc )
keir@20457	2025 {
keir@20457	2026 ERROR("error setting the HVM context");
keir@20457	2027 goto out;
keir@20457	2028 }
keir@20457	2029
keir@20457	2030 /* HVM success! */
keir@20457	2031 rc = 0;
cl349@2791	2032
mjw@1661	2033 out:
kaf24@1683	2034 if ( (rc != 0) && (dom != 0) )
mjw@1661	2035 xc_domain_destroy(xc_handle, dom);
vh249@6159	2036 free(mmu);
keir@20587	2037 free(ctx->p2m);
vh249@6159	2038 free(pfn_type);
keir@20457	2039 tailbuf_free(&tailbuf);
mjw@1661	2040
keir@14142	2041 /* discard cache for save file */
keir@14142	2042 discard_file_cache(io_fd, 1 /flush/);
keir@14142	2043
cl349@5091	2044 DPRINTF("Restore exit with rc=%d\n", rc);
steven@13424	2045
mjw@1661	2046 return rc;
mjw@1661	2047 }
keir@19677	2048 /*
keir@19677	2049 * Local variables:
keir@19677	2050 * mode: C
keir@19677	2051 * c-set-style: "BSD"
keir@19677	2052 * c-basic-offset: 4
keir@19677	2053 * tab-width: 4
keir@19677	2054 * indent-tabs-mode: nil
keir@19677	2055 * End:
keir@19677	2056 */