linux/arch/x86/xen/mmu_pv.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2
   3/*
   4 * Xen mmu operations
   5 *
   6 * This file contains the various mmu fetch and update operations.
   7 * The most important job they must perform is the mapping between the
   8 * domain's pfn and the overall machine mfns.
   9 *
  10 * Xen allows guests to directly update the pagetable, in a controlled
  11 * fashion.  In other words, the guest modifies the same pagetable
  12 * that the CPU actually uses, which eliminates the overhead of having
  13 * a separate shadow pagetable.
  14 *
  15 * In order to allow this, it falls on the guest domain to map its
  16 * notion of a "physical" pfn - which is just a domain-local linear
  17 * address - into a real "machine address" which the CPU's MMU can
  18 * use.
  19 *
  20 * A pgd_t/pmd_t/pte_t will typically contain an mfn, and so can be
  21 * inserted directly into the pagetable.  When creating a new
  22 * pte/pmd/pgd, it converts the passed pfn into an mfn.  Conversely,
  23 * when reading the content back with __(pgd|pmd|pte)_val, it converts
  24 * the mfn back into a pfn.
  25 *
  26 * The other constraint is that all pages which make up a pagetable
  27 * must be mapped read-only in the guest.  This prevents uncontrolled
  28 * guest updates to the pagetable.  Xen strictly enforces this, and
  29 * will disallow any pagetable update which will end up mapping a
  30 * pagetable page RW, and will disallow using any writable page as a
  31 * pagetable.
  32 *
  33 * Naively, when loading %cr3 with the base of a new pagetable, Xen
  34 * would need to validate the whole pagetable before going on.
  35 * Naturally, this is quite slow.  The solution is to "pin" a
  36 * pagetable, which enforces all the constraints on the pagetable even
  37 * when it is not actively in use.  This menas that Xen can be assured
  38 * that it is still valid when you do load it into %cr3, and doesn't
  39 * need to revalidate it.
  40 *
  41 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
  42 */
  43#include <linux/sched/mm.h>
  44#include <linux/highmem.h>
  45#include <linux/debugfs.h>
  46#include <linux/bug.h>
  47#include <linux/vmalloc.h>
  48#include <linux/export.h>
  49#include <linux/init.h>
  50#include <linux/gfp.h>
  51#include <linux/memblock.h>
  52#include <linux/seq_file.h>
  53#include <linux/crash_dump.h>
  54#include <linux/pgtable.h>
  55#ifdef CONFIG_KEXEC_CORE
  56#include <linux/kexec.h>
  57#endif
  58
  59#include <trace/events/xen.h>
  60
  61#include <asm/tlbflush.h>
  62#include <asm/fixmap.h>
  63#include <asm/mmu_context.h>
  64#include <asm/setup.h>
  65#include <asm/paravirt.h>
  66#include <asm/e820/api.h>
  67#include <asm/linkage.h>
  68#include <asm/page.h>
  69#include <asm/init.h>
  70#include <asm/memtype.h>
  71#include <asm/smp.h>
  72#include <asm/tlb.h>
  73
  74#include <asm/xen/hypercall.h>
  75#include <asm/xen/hypervisor.h>
  76
  77#include <xen/xen.h>
  78#include <xen/page.h>
  79#include <xen/interface/xen.h>
  80#include <xen/interface/hvm/hvm_op.h>
  81#include <xen/interface/version.h>
  82#include <xen/interface/memory.h>
  83#include <xen/hvc-console.h>
  84
  85#include "multicalls.h"
  86#include "mmu.h"
  87#include "debugfs.h"
  88
  89/* l3 pud for userspace vsyscall mapping */
  90static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
  91
  92/*
  93 * Protects atomic reservation decrease/increase against concurrent increases.
  94 * Also protects non-atomic updates of current_pages and balloon lists.
  95 */
  96static DEFINE_SPINLOCK(xen_reservation_lock);
  97
  98/*
  99 * Note about cr3 (pagetable base) values:
 100 *
 101 * xen_cr3 contains the current logical cr3 value; it contains the
 102 * last set cr3.  This may not be the current effective cr3, because
 103 * its update may be being lazily deferred.  However, a vcpu looking
 104 * at its own cr3 can use this value knowing that it everything will
 105 * be self-consistent.
 106 *
 107 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
 108 * hypercall to set the vcpu cr3 is complete (so it may be a little
 109 * out of date, but it will never be set early).  If one vcpu is
 110 * looking at another vcpu's cr3 value, it should use this variable.
 111 */
 112DEFINE_PER_CPU(unsigned long, xen_cr3);  /* cr3 stored as physaddr */
 113DEFINE_PER_CPU(unsigned long, xen_current_cr3);  /* actual vcpu cr3 */
 114
 115static phys_addr_t xen_pt_base, xen_pt_size __initdata;
 116
 117static DEFINE_STATIC_KEY_FALSE(xen_struct_pages_ready);
 118
 119/*
 120 * Just beyond the highest usermode address.  STACK_TOP_MAX has a
 121 * redzone above it, so round it up to a PGD boundary.
 122 */
 123#define USER_LIMIT      ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
 124
 125void make_lowmem_page_readonly(void *vaddr)
 126{
 127        pte_t *pte, ptev;
 128        unsigned long address = (unsigned long)vaddr;
 129        unsigned int level;
 130
 131        pte = lookup_address(address, &level);
 132        if (pte == NULL)
 133                return;         /* vaddr missing */
 134
 135        ptev = pte_wrprotect(*pte);
 136
 137        if (HYPERVISOR_update_va_mapping(address, ptev, 0))
 138                BUG();
 139}
 140
 141void make_lowmem_page_readwrite(void *vaddr)
 142{
 143        pte_t *pte, ptev;
 144        unsigned long address = (unsigned long)vaddr;
 145        unsigned int level;
 146
 147        pte = lookup_address(address, &level);
 148        if (pte == NULL)
 149                return;         /* vaddr missing */
 150
 151        ptev = pte_mkwrite(*pte);
 152
 153        if (HYPERVISOR_update_va_mapping(address, ptev, 0))
 154                BUG();
 155}
 156
 157
 158/*
 159 * During early boot all page table pages are pinned, but we do not have struct
 160 * pages, so return true until struct pages are ready.
 161 */
 162static bool xen_page_pinned(void *ptr)
 163{
 164        if (static_branch_likely(&xen_struct_pages_ready)) {
 165                struct page *page = virt_to_page(ptr);
 166
 167                return PagePinned(page);
 168        }
 169        return true;
 170}
 171
 172static void xen_extend_mmu_update(const struct mmu_update *update)
 173{
 174        struct multicall_space mcs;
 175        struct mmu_update *u;
 176
 177        mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
 178
 179        if (mcs.mc != NULL) {
 180                mcs.mc->args[1]++;
 181        } else {
 182                mcs = __xen_mc_entry(sizeof(*u));
 183                MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
 184        }
 185
 186        u = mcs.args;
 187        *u = *update;
 188}
 189
 190static void xen_extend_mmuext_op(const struct mmuext_op *op)
 191{
 192        struct multicall_space mcs;
 193        struct mmuext_op *u;
 194
 195        mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u));
 196
 197        if (mcs.mc != NULL) {
 198                mcs.mc->args[1]++;
 199        } else {
 200                mcs = __xen_mc_entry(sizeof(*u));
 201                MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
 202        }
 203
 204        u = mcs.args;
 205        *u = *op;
 206}
 207
 208static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
 209{
 210        struct mmu_update u;
 211
 212        preempt_disable();
 213
 214        xen_mc_batch();
 215
 216        /* ptr may be ioremapped for 64-bit pagetable setup */
 217        u.ptr = arbitrary_virt_to_machine(ptr).maddr;
 218        u.val = pmd_val_ma(val);
 219        xen_extend_mmu_update(&u);
 220
 221        xen_mc_issue(PARAVIRT_LAZY_MMU);
 222
 223        preempt_enable();
 224}
 225
 226static void xen_set_pmd(pmd_t *ptr, pmd_t val)
 227{
 228        trace_xen_mmu_set_pmd(ptr, val);
 229
 230        /* If page is not pinned, we can just update the entry
 231           directly */
 232        if (!xen_page_pinned(ptr)) {
 233                *ptr = val;
 234                return;
 235        }
 236
 237        xen_set_pmd_hyper(ptr, val);
 238}
 239
 240/*
 241 * Associate a virtual page frame with a given physical page frame
 242 * and protection flags for that frame.
 243 */
 244void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
 245{
 246        set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
 247}
 248
 249static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
 250{
 251        struct mmu_update u;
 252
 253        if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU)
 254                return false;
 255
 256        xen_mc_batch();
 257
 258        u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
 259        u.val = pte_val_ma(pteval);
 260        xen_extend_mmu_update(&u);
 261
 262        xen_mc_issue(PARAVIRT_LAZY_MMU);
 263
 264        return true;
 265}
 266
 267static inline void __xen_set_pte(pte_t *ptep, pte_t pteval)
 268{
 269        if (!xen_batched_set_pte(ptep, pteval)) {
 270                /*
 271                 * Could call native_set_pte() here and trap and
 272                 * emulate the PTE write, but a hypercall is much cheaper.
 273                 */
 274                struct mmu_update u;
 275
 276                u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
 277                u.val = pte_val_ma(pteval);
 278                HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF);
 279        }
 280}
 281
 282static void xen_set_pte(pte_t *ptep, pte_t pteval)
 283{
 284        trace_xen_mmu_set_pte(ptep, pteval);
 285        __xen_set_pte(ptep, pteval);
 286}
 287
 288pte_t xen_ptep_modify_prot_start(struct vm_area_struct *vma,
 289                                 unsigned long addr, pte_t *ptep)
 290{
 291        /* Just return the pte as-is.  We preserve the bits on commit */
 292        trace_xen_mmu_ptep_modify_prot_start(vma->vm_mm, addr, ptep, *ptep);
 293        return *ptep;
 294}
 295
 296void xen_ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr,
 297                                 pte_t *ptep, pte_t pte)
 298{
 299        struct mmu_update u;
 300
 301        trace_xen_mmu_ptep_modify_prot_commit(vma->vm_mm, addr, ptep, pte);
 302        xen_mc_batch();
 303
 304        u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
 305        u.val = pte_val_ma(pte);
 306        xen_extend_mmu_update(&u);
 307
 308        xen_mc_issue(PARAVIRT_LAZY_MMU);
 309}
 310
 311/* Assume pteval_t is equivalent to all the other *val_t types. */
 312static pteval_t pte_mfn_to_pfn(pteval_t val)
 313{
 314        if (val & _PAGE_PRESENT) {
 315                unsigned long mfn = (val & XEN_PTE_MFN_MASK) >> PAGE_SHIFT;
 316                unsigned long pfn = mfn_to_pfn(mfn);
 317
 318                pteval_t flags = val & PTE_FLAGS_MASK;
 319                if (unlikely(pfn == ~0))
 320                        val = flags & ~_PAGE_PRESENT;
 321                else
 322                        val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
 323        }
 324
 325        return val;
 326}
 327
 328static pteval_t pte_pfn_to_mfn(pteval_t val)
 329{
 330        if (val & _PAGE_PRESENT) {
 331                unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
 332                pteval_t flags = val & PTE_FLAGS_MASK;
 333                unsigned long mfn;
 334
 335                mfn = __pfn_to_mfn(pfn);
 336
 337                /*
 338                 * If there's no mfn for the pfn, then just create an
 339                 * empty non-present pte.  Unfortunately this loses
 340                 * information about the original pfn, so
 341                 * pte_mfn_to_pfn is asymmetric.
 342                 */
 343                if (unlikely(mfn == INVALID_P2M_ENTRY)) {
 344                        mfn = 0;
 345                        flags = 0;
 346                } else
 347                        mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT);
 348                val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
 349        }
 350
 351        return val;
 352}
 353
 354__visible pteval_t xen_pte_val(pte_t pte)
 355{
 356        pteval_t pteval = pte.pte;
 357
 358        return pte_mfn_to_pfn(pteval);
 359}
 360PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
 361
 362__visible pgdval_t xen_pgd_val(pgd_t pgd)
 363{
 364        return pte_mfn_to_pfn(pgd.pgd);
 365}
 366PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
 367
 368__visible pte_t xen_make_pte(pteval_t pte)
 369{
 370        pte = pte_pfn_to_mfn(pte);
 371
 372        return native_make_pte(pte);
 373}
 374PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
 375
 376__visible pgd_t xen_make_pgd(pgdval_t pgd)
 377{
 378        pgd = pte_pfn_to_mfn(pgd);
 379        return native_make_pgd(pgd);
 380}
 381PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
 382
 383__visible pmdval_t xen_pmd_val(pmd_t pmd)
 384{
 385        return pte_mfn_to_pfn(pmd.pmd);
 386}
 387PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
 388
 389static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
 390{
 391        struct mmu_update u;
 392
 393        preempt_disable();
 394
 395        xen_mc_batch();
 396
 397        /* ptr may be ioremapped for 64-bit pagetable setup */
 398        u.ptr = arbitrary_virt_to_machine(ptr).maddr;
 399        u.val = pud_val_ma(val);
 400        xen_extend_mmu_update(&u);
 401
 402        xen_mc_issue(PARAVIRT_LAZY_MMU);
 403
 404        preempt_enable();
 405}
 406
 407static void xen_set_pud(pud_t *ptr, pud_t val)
 408{
 409        trace_xen_mmu_set_pud(ptr, val);
 410
 411        /* If page is not pinned, we can just update the entry
 412           directly */
 413        if (!xen_page_pinned(ptr)) {
 414                *ptr = val;
 415                return;
 416        }
 417
 418        xen_set_pud_hyper(ptr, val);
 419}
 420
 421__visible pmd_t xen_make_pmd(pmdval_t pmd)
 422{
 423        pmd = pte_pfn_to_mfn(pmd);
 424        return native_make_pmd(pmd);
 425}
 426PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
 427
 428__visible pudval_t xen_pud_val(pud_t pud)
 429{
 430        return pte_mfn_to_pfn(pud.pud);
 431}
 432PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
 433
 434__visible pud_t xen_make_pud(pudval_t pud)
 435{
 436        pud = pte_pfn_to_mfn(pud);
 437
 438        return native_make_pud(pud);
 439}
 440PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
 441
 442static pgd_t *xen_get_user_pgd(pgd_t *pgd)
 443{
 444        pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
 445        unsigned offset = pgd - pgd_page;
 446        pgd_t *user_ptr = NULL;
 447
 448        if (offset < pgd_index(USER_LIMIT)) {
 449                struct page *page = virt_to_page(pgd_page);
 450                user_ptr = (pgd_t *)page->private;
 451                if (user_ptr)
 452                        user_ptr += offset;
 453        }
 454
 455        return user_ptr;
 456}
 457
 458static void __xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
 459{
 460        struct mmu_update u;
 461
 462        u.ptr = virt_to_machine(ptr).maddr;
 463        u.val = p4d_val_ma(val);
 464        xen_extend_mmu_update(&u);
 465}
 466
 467/*
 468 * Raw hypercall-based set_p4d, intended for in early boot before
 469 * there's a page structure.  This implies:
 470 *  1. The only existing pagetable is the kernel's
 471 *  2. It is always pinned
 472 *  3. It has no user pagetable attached to it
 473 */
 474static void __init xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
 475{
 476        preempt_disable();
 477
 478        xen_mc_batch();
 479
 480        __xen_set_p4d_hyper(ptr, val);
 481
 482        xen_mc_issue(PARAVIRT_LAZY_MMU);
 483
 484        preempt_enable();
 485}
 486
 487static void xen_set_p4d(p4d_t *ptr, p4d_t val)
 488{
 489        pgd_t *user_ptr = xen_get_user_pgd((pgd_t *)ptr);
 490        pgd_t pgd_val;
 491
 492        trace_xen_mmu_set_p4d(ptr, (p4d_t *)user_ptr, val);
 493
 494        /* If page is not pinned, we can just update the entry
 495           directly */
 496        if (!xen_page_pinned(ptr)) {
 497                *ptr = val;
 498                if (user_ptr) {
 499                        WARN_ON(xen_page_pinned(user_ptr));
 500                        pgd_val.pgd = p4d_val_ma(val);
 501                        *user_ptr = pgd_val;
 502                }
 503                return;
 504        }
 505
 506        /* If it's pinned, then we can at least batch the kernel and
 507           user updates together. */
 508        xen_mc_batch();
 509
 510        __xen_set_p4d_hyper(ptr, val);
 511        if (user_ptr)
 512                __xen_set_p4d_hyper((p4d_t *)user_ptr, val);
 513
 514        xen_mc_issue(PARAVIRT_LAZY_MMU);
 515}
 516
 517#if CONFIG_PGTABLE_LEVELS >= 5
 518__visible p4dval_t xen_p4d_val(p4d_t p4d)
 519{
 520        return pte_mfn_to_pfn(p4d.p4d);
 521}
 522PV_CALLEE_SAVE_REGS_THUNK(xen_p4d_val);
 523
 524__visible p4d_t xen_make_p4d(p4dval_t p4d)
 525{
 526        p4d = pte_pfn_to_mfn(p4d);
 527
 528        return native_make_p4d(p4d);
 529}
 530PV_CALLEE_SAVE_REGS_THUNK(xen_make_p4d);
 531#endif  /* CONFIG_PGTABLE_LEVELS >= 5 */
 532
 533static void xen_pmd_walk(struct mm_struct *mm, pmd_t *pmd,
 534                         void (*func)(struct mm_struct *mm, struct page *,
 535                                      enum pt_level),
 536                         bool last, unsigned long limit)
 537{
 538        int i, nr;
 539
 540        nr = last ? pmd_index(limit) + 1 : PTRS_PER_PMD;
 541        for (i = 0; i < nr; i++) {
 542                if (!pmd_none(pmd[i]))
 543                        (*func)(mm, pmd_page(pmd[i]), PT_PTE);
 544        }
 545}
 546
 547static void xen_pud_walk(struct mm_struct *mm, pud_t *pud,
 548                         void (*func)(struct mm_struct *mm, struct page *,
 549                                      enum pt_level),
 550                         bool last, unsigned long limit)
 551{
 552        int i, nr;
 553
 554        nr = last ? pud_index(limit) + 1 : PTRS_PER_PUD;
 555        for (i = 0; i < nr; i++) {
 556                pmd_t *pmd;
 557
 558                if (pud_none(pud[i]))
 559                        continue;
 560
 561                pmd = pmd_offset(&pud[i], 0);
 562                if (PTRS_PER_PMD > 1)
 563                        (*func)(mm, virt_to_page(pmd), PT_PMD);
 564                xen_pmd_walk(mm, pmd, func, last && i == nr - 1, limit);
 565        }
 566}
 567
 568static void xen_p4d_walk(struct mm_struct *mm, p4d_t *p4d,
 569                         void (*func)(struct mm_struct *mm, struct page *,
 570                                      enum pt_level),
 571                         bool last, unsigned long limit)
 572{
 573        pud_t *pud;
 574
 575
 576        if (p4d_none(*p4d))
 577                return;
 578
 579        pud = pud_offset(p4d, 0);
 580        if (PTRS_PER_PUD > 1)
 581                (*func)(mm, virt_to_page(pud), PT_PUD);
 582        xen_pud_walk(mm, pud, func, last, limit);
 583}
 584
 585/*
 586 * (Yet another) pagetable walker.  This one is intended for pinning a
 587 * pagetable.  This means that it walks a pagetable and calls the
 588 * callback function on each page it finds making up the page table,
 589 * at every level.  It walks the entire pagetable, but it only bothers
 590 * pinning pte pages which are below limit.  In the normal case this
 591 * will be STACK_TOP_MAX, but at boot we need to pin up to
 592 * FIXADDR_TOP.
 593 *
 594 * We must skip the Xen hole in the middle of the address space, just after
 595 * the big x86-64 virtual hole.
 596 */
 597static void __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
 598                           void (*func)(struct mm_struct *mm, struct page *,
 599                                        enum pt_level),
 600                           unsigned long limit)
 601{
 602        int i, nr;
 603        unsigned hole_low = 0, hole_high = 0;
 604
 605        /* The limit is the last byte to be touched */
 606        limit--;
 607        BUG_ON(limit >= FIXADDR_TOP);
 608
 609        /*
 610         * 64-bit has a great big hole in the middle of the address
 611         * space, which contains the Xen mappings.
 612         */
 613        hole_low = pgd_index(GUARD_HOLE_BASE_ADDR);
 614        hole_high = pgd_index(GUARD_HOLE_END_ADDR);
 615
 616        nr = pgd_index(limit) + 1;
 617        for (i = 0; i < nr; i++) {
 618                p4d_t *p4d;
 619
 620                if (i >= hole_low && i < hole_high)
 621                        continue;
 622
 623                if (pgd_none(pgd[i]))
 624                        continue;
 625
 626                p4d = p4d_offset(&pgd[i], 0);
 627                xen_p4d_walk(mm, p4d, func, i == nr - 1, limit);
 628        }
 629
 630        /* Do the top level last, so that the callbacks can use it as
 631           a cue to do final things like tlb flushes. */
 632        (*func)(mm, virt_to_page(pgd), PT_PGD);
 633}
 634
 635static void xen_pgd_walk(struct mm_struct *mm,
 636                         void (*func)(struct mm_struct *mm, struct page *,
 637                                      enum pt_level),
 638                         unsigned long limit)
 639{
 640        __xen_pgd_walk(mm, mm->pgd, func, limit);
 641}
 642
 643/* If we're using split pte locks, then take the page's lock and
 644   return a pointer to it.  Otherwise return NULL. */
 645static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
 646{
 647        spinlock_t *ptl = NULL;
 648
 649#if USE_SPLIT_PTE_PTLOCKS
 650        ptl = ptlock_ptr(page);
 651        spin_lock_nest_lock(ptl, &mm->page_table_lock);
 652#endif
 653
 654        return ptl;
 655}
 656
 657static void xen_pte_unlock(void *v)
 658{
 659        spinlock_t *ptl = v;
 660        spin_unlock(ptl);
 661}
 662
 663static void xen_do_pin(unsigned level, unsigned long pfn)
 664{
 665        struct mmuext_op op;
 666
 667        op.cmd = level;
 668        op.arg1.mfn = pfn_to_mfn(pfn);
 669
 670        xen_extend_mmuext_op(&op);
 671}
 672
 673static void xen_pin_page(struct mm_struct *mm, struct page *page,
 674                         enum pt_level level)
 675{
 676        unsigned pgfl = TestSetPagePinned(page);
 677
 678        if (!pgfl) {
 679                void *pt = lowmem_page_address(page);
 680                unsigned long pfn = page_to_pfn(page);
 681                struct multicall_space mcs = __xen_mc_entry(0);
 682                spinlock_t *ptl;
 683
 684                /*
 685                 * We need to hold the pagetable lock between the time
 686                 * we make the pagetable RO and when we actually pin
 687                 * it.  If we don't, then other users may come in and
 688                 * attempt to update the pagetable by writing it,
 689                 * which will fail because the memory is RO but not
 690                 * pinned, so Xen won't do the trap'n'emulate.
 691                 *
 692                 * If we're using split pte locks, we can't hold the
 693                 * entire pagetable's worth of locks during the
 694                 * traverse, because we may wrap the preempt count (8
 695                 * bits).  The solution is to mark RO and pin each PTE
 696                 * page while holding the lock.  This means the number
 697                 * of locks we end up holding is never more than a
 698                 * batch size (~32 entries, at present).
 699                 *
 700                 * If we're not using split pte locks, we needn't pin
 701                 * the PTE pages independently, because we're
 702                 * protected by the overall pagetable lock.
 703                 */
 704                ptl = NULL;
 705                if (level == PT_PTE)
 706                        ptl = xen_pte_lock(page, mm);
 707
 708                MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
 709                                        pfn_pte(pfn, PAGE_KERNEL_RO),
 710                                        level == PT_PGD ? UVMF_TLB_FLUSH : 0);
 711
 712                if (ptl) {
 713                        xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
 714
 715                        /* Queue a deferred unlock for when this batch
 716                           is completed. */
 717                        xen_mc_callback(xen_pte_unlock, ptl);
 718                }
 719        }
 720}
 721
 722/* This is called just after a mm has been created, but it has not
 723   been used yet.  We need to make sure that its pagetable is all
 724   read-only, and can be pinned. */
 725static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
 726{
 727        pgd_t *user_pgd = xen_get_user_pgd(pgd);
 728
 729        trace_xen_mmu_pgd_pin(mm, pgd);
 730
 731        xen_mc_batch();
 732
 733        __xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT);
 734
 735        xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
 736
 737        if (user_pgd) {
 738                xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
 739                xen_do_pin(MMUEXT_PIN_L4_TABLE,
 740                           PFN_DOWN(__pa(user_pgd)));
 741        }
 742
 743        xen_mc_issue(0);
 744}
 745
 746static void xen_pgd_pin(struct mm_struct *mm)
 747{
 748        __xen_pgd_pin(mm, mm->pgd);
 749}
 750
 751/*
 752 * On save, we need to pin all pagetables to make sure they get their
 753 * mfns turned into pfns.  Search the list for any unpinned pgds and pin
 754 * them (unpinned pgds are not currently in use, probably because the
 755 * process is under construction or destruction).
 756 *
 757 * Expected to be called in stop_machine() ("equivalent to taking
 758 * every spinlock in the system"), so the locking doesn't really
 759 * matter all that much.
 760 */
 761void xen_mm_pin_all(void)
 762{
 763        struct page *page;
 764
 765        spin_lock(&pgd_lock);
 766
 767        list_for_each_entry(page, &pgd_list, lru) {
 768                if (!PagePinned(page)) {
 769                        __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
 770                        SetPageSavePinned(page);
 771                }
 772        }
 773
 774        spin_unlock(&pgd_lock);
 775}
 776
 777static void __init xen_mark_pinned(struct mm_struct *mm, struct page *page,
 778                                   enum pt_level level)
 779{
 780        SetPagePinned(page);
 781}
 782
 783/*
 784 * The init_mm pagetable is really pinned as soon as its created, but
 785 * that's before we have page structures to store the bits.  So do all
 786 * the book-keeping now once struct pages for allocated pages are
 787 * initialized. This happens only after memblock_free_all() is called.
 788 */
 789static void __init xen_after_bootmem(void)
 790{
 791        static_branch_enable(&xen_struct_pages_ready);
 792        SetPagePinned(virt_to_page(level3_user_vsyscall));
 793        xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
 794}
 795
 796static void xen_unpin_page(struct mm_struct *mm, struct page *page,
 797                           enum pt_level level)
 798{
 799        unsigned pgfl = TestClearPagePinned(page);
 800
 801        if (pgfl) {
 802                void *pt = lowmem_page_address(page);
 803                unsigned long pfn = page_to_pfn(page);
 804                spinlock_t *ptl = NULL;
 805                struct multicall_space mcs;
 806
 807                /*
 808                 * Do the converse to pin_page.  If we're using split
 809                 * pte locks, we must be holding the lock for while
 810                 * the pte page is unpinned but still RO to prevent
 811                 * concurrent updates from seeing it in this
 812                 * partially-pinned state.
 813                 */
 814                if (level == PT_PTE) {
 815                        ptl = xen_pte_lock(page, mm);
 816
 817                        if (ptl)
 818                                xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
 819                }
 820
 821                mcs = __xen_mc_entry(0);
 822
 823                MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
 824                                        pfn_pte(pfn, PAGE_KERNEL),
 825                                        level == PT_PGD ? UVMF_TLB_FLUSH : 0);
 826
 827                if (ptl) {
 828                        /* unlock when batch completed */
 829                        xen_mc_callback(xen_pte_unlock, ptl);
 830                }
 831        }
 832}
 833
 834/* Release a pagetables pages back as normal RW */
 835static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
 836{
 837        pgd_t *user_pgd = xen_get_user_pgd(pgd);
 838
 839        trace_xen_mmu_pgd_unpin(mm, pgd);
 840
 841        xen_mc_batch();
 842
 843        xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
 844
 845        if (user_pgd) {
 846                xen_do_pin(MMUEXT_UNPIN_TABLE,
 847                           PFN_DOWN(__pa(user_pgd)));
 848                xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
 849        }
 850
 851        __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
 852
 853        xen_mc_issue(0);
 854}
 855
 856static void xen_pgd_unpin(struct mm_struct *mm)
 857{
 858        __xen_pgd_unpin(mm, mm->pgd);
 859}
 860
 861/*
 862 * On resume, undo any pinning done at save, so that the rest of the
 863 * kernel doesn't see any unexpected pinned pagetables.
 864 */
 865void xen_mm_unpin_all(void)
 866{
 867        struct page *page;
 868
 869        spin_lock(&pgd_lock);
 870
 871        list_for_each_entry(page, &pgd_list, lru) {
 872                if (PageSavePinned(page)) {
 873                        BUG_ON(!PagePinned(page));
 874                        __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
 875                        ClearPageSavePinned(page);
 876                }
 877        }
 878
 879        spin_unlock(&pgd_lock);
 880}
 881
 882static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
 883{
 884        spin_lock(&next->page_table_lock);
 885        xen_pgd_pin(next);
 886        spin_unlock(&next->page_table_lock);
 887}
 888
 889static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
 890{
 891        spin_lock(&mm->page_table_lock);
 892        xen_pgd_pin(mm);
 893        spin_unlock(&mm->page_table_lock);
 894}
 895
 896static void drop_mm_ref_this_cpu(void *info)
 897{
 898        struct mm_struct *mm = info;
 899
 900        if (this_cpu_read(cpu_tlbstate.loaded_mm) == mm)
 901                leave_mm(smp_processor_id());
 902
 903        /*
 904         * If this cpu still has a stale cr3 reference, then make sure
 905         * it has been flushed.
 906         */
 907        if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd))
 908                xen_mc_flush();
 909}
 910
 911#ifdef CONFIG_SMP
 912/*
 913 * Another cpu may still have their %cr3 pointing at the pagetable, so
 914 * we need to repoint it somewhere else before we can unpin it.
 915 */
 916static void xen_drop_mm_ref(struct mm_struct *mm)
 917{
 918        cpumask_var_t mask;
 919        unsigned cpu;
 920
 921        drop_mm_ref_this_cpu(mm);
 922
 923        /* Get the "official" set of cpus referring to our pagetable. */
 924        if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
 925                for_each_online_cpu(cpu) {
 926                        if (per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
 927                                continue;
 928                        smp_call_function_single(cpu, drop_mm_ref_this_cpu, mm, 1);
 929                }
 930                return;
 931        }
 932
 933        /*
 934         * It's possible that a vcpu may have a stale reference to our
 935         * cr3, because its in lazy mode, and it hasn't yet flushed
 936         * its set of pending hypercalls yet.  In this case, we can
 937         * look at its actual current cr3 value, and force it to flush
 938         * if needed.
 939         */
 940        cpumask_clear(mask);
 941        for_each_online_cpu(cpu) {
 942                if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
 943                        cpumask_set_cpu(cpu, mask);
 944        }
 945
 946        smp_call_function_many(mask, drop_mm_ref_this_cpu, mm, 1);
 947        free_cpumask_var(mask);
 948}
 949#else
 950static void xen_drop_mm_ref(struct mm_struct *mm)
 951{
 952        drop_mm_ref_this_cpu(mm);
 953}
 954#endif
 955
 956/*
 957 * While a process runs, Xen pins its pagetables, which means that the
 958 * hypervisor forces it to be read-only, and it controls all updates
 959 * to it.  This means that all pagetable updates have to go via the
 960 * hypervisor, which is moderately expensive.
 961 *
 962 * Since we're pulling the pagetable down, we switch to use init_mm,
 963 * unpin old process pagetable and mark it all read-write, which
 964 * allows further operations on it to be simple memory accesses.
 965 *
 966 * The only subtle point is that another CPU may be still using the
 967 * pagetable because of lazy tlb flushing.  This means we need need to
 968 * switch all CPUs off this pagetable before we can unpin it.
 969 */
 970static void xen_exit_mmap(struct mm_struct *mm)
 971{
 972        get_cpu();              /* make sure we don't move around */
 973        xen_drop_mm_ref(mm);
 974        put_cpu();
 975
 976        spin_lock(&mm->page_table_lock);
 977
 978        /* pgd may not be pinned in the error exit path of execve */
 979        if (xen_page_pinned(mm->pgd))
 980                xen_pgd_unpin(mm);
 981
 982        spin_unlock(&mm->page_table_lock);
 983}
 984
 985static void xen_post_allocator_init(void);
 986
 987static void __init pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
 988{
 989        struct mmuext_op op;
 990
 991        op.cmd = cmd;
 992        op.arg1.mfn = pfn_to_mfn(pfn);
 993        if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
 994                BUG();
 995}
 996
 997static void __init xen_cleanhighmap(unsigned long vaddr,
 998                                    unsigned long vaddr_end)
 999{
1000        unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
1001        pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr);
1002
1003        /* NOTE: The loop is more greedy than the cleanup_highmap variant.
1004         * We include the PMD passed in on _both_ boundaries. */
1005        for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PTRS_PER_PMD));
1006                        pmd++, vaddr += PMD_SIZE) {
1007                if (pmd_none(*pmd))
1008                        continue;
1009                if (vaddr < (unsigned long) _text || vaddr > kernel_end)
1010                        set_pmd(pmd, __pmd(0));
1011        }
1012        /* In case we did something silly, we should crash in this function
1013         * instead of somewhere later and be confusing. */
1014        xen_mc_flush();
1015}
1016
1017/*
1018 * Make a page range writeable and free it.
1019 */
1020static void __init xen_free_ro_pages(unsigned long paddr, unsigned long size)
1021{
1022        void *vaddr = __va(paddr);
1023        void *vaddr_end = vaddr + size;
1024
1025        for (; vaddr < vaddr_end; vaddr += PAGE_SIZE)
1026                make_lowmem_page_readwrite(vaddr);
1027
1028        memblock_free(paddr, size);
1029}
1030
1031static void __init xen_cleanmfnmap_free_pgtbl(void *pgtbl, bool unpin)
1032{
1033        unsigned long pa = __pa(pgtbl) & PHYSICAL_PAGE_MASK;
1034
1035        if (unpin)
1036                pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa));
1037        ClearPagePinned(virt_to_page(__va(pa)));
1038        xen_free_ro_pages(pa, PAGE_SIZE);
1039}
1040
1041static void __init xen_cleanmfnmap_pmd(pmd_t *pmd, bool unpin)
1042{
1043        unsigned long pa;
1044        pte_t *pte_tbl;
1045        int i;
1046
1047        if (pmd_large(*pmd)) {
1048                pa = pmd_val(*pmd) & PHYSICAL_PAGE_MASK;
1049                xen_free_ro_pages(pa, PMD_SIZE);
1050                return;
1051        }
1052
1053        pte_tbl = pte_offset_kernel(pmd, 0);
1054        for (i = 0; i < PTRS_PER_PTE; i++) {
1055                if (pte_none(pte_tbl[i]))
1056                        continue;
1057                pa = pte_pfn(pte_tbl[i]) << PAGE_SHIFT;
1058                xen_free_ro_pages(pa, PAGE_SIZE);
1059        }
1060        set_pmd(pmd, __pmd(0));
1061        xen_cleanmfnmap_free_pgtbl(pte_tbl, unpin);
1062}
1063
1064static void __init xen_cleanmfnmap_pud(pud_t *pud, bool unpin)
1065{
1066        unsigned long pa;
1067        pmd_t *pmd_tbl;
1068        int i;
1069
1070        if (pud_large(*pud)) {
1071                pa = pud_val(*pud) & PHYSICAL_PAGE_MASK;
1072                xen_free_ro_pages(pa, PUD_SIZE);
1073                return;
1074        }
1075
1076        pmd_tbl = pmd_offset(pud, 0);
1077        for (i = 0; i < PTRS_PER_PMD; i++) {
1078                if (pmd_none(pmd_tbl[i]))
1079                        continue;
1080                xen_cleanmfnmap_pmd(pmd_tbl + i, unpin);
1081        }
1082        set_pud(pud, __pud(0));
1083        xen_cleanmfnmap_free_pgtbl(pmd_tbl, unpin);
1084}
1085
1086static void __init xen_cleanmfnmap_p4d(p4d_t *p4d, bool unpin)
1087{
1088        unsigned long pa;
1089        pud_t *pud_tbl;
1090        int i;
1091
1092        if (p4d_large(*p4d)) {
1093                pa = p4d_val(*p4d) & PHYSICAL_PAGE_MASK;
1094                xen_free_ro_pages(pa, P4D_SIZE);
1095                return;
1096        }
1097
1098        pud_tbl = pud_offset(p4d, 0);
1099        for (i = 0; i < PTRS_PER_PUD; i++) {
1100                if (pud_none(pud_tbl[i]))
1101                        continue;
1102                xen_cleanmfnmap_pud(pud_tbl + i, unpin);
1103        }
1104        set_p4d(p4d, __p4d(0));
1105        xen_cleanmfnmap_free_pgtbl(pud_tbl, unpin);
1106}
1107
1108/*
1109 * Since it is well isolated we can (and since it is perhaps large we should)
1110 * also free the page tables mapping the initial P->M table.
1111 */
1112static void __init xen_cleanmfnmap(unsigned long vaddr)
1113{
1114        pgd_t *pgd;
1115        p4d_t *p4d;
1116        bool unpin;
1117
1118        unpin = (vaddr == 2 * PGDIR_SIZE);
1119        vaddr &= PMD_MASK;
1120        pgd = pgd_offset_k(vaddr);
1121        p4d = p4d_offset(pgd, 0);
1122        if (!p4d_none(*p4d))
1123                xen_cleanmfnmap_p4d(p4d, unpin);
1124}
1125
1126static void __init xen_pagetable_p2m_free(void)
1127{
1128        unsigned long size;
1129        unsigned long addr;
1130
1131        size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1132
1133        /* No memory or already called. */
1134        if ((unsigned long)xen_p2m_addr == xen_start_info->mfn_list)
1135                return;
1136
1137        /* using __ka address and sticking INVALID_P2M_ENTRY! */
1138        memset((void *)xen_start_info->mfn_list, 0xff, size);
1139
1140        addr = xen_start_info->mfn_list;
1141        /*
1142         * We could be in __ka space.
1143         * We roundup to the PMD, which means that if anybody at this stage is
1144         * using the __ka address of xen_start_info or
1145         * xen_start_info->shared_info they are in going to crash. Fortunately
1146         * we have already revectored in xen_setup_kernel_pagetable.
1147         */
1148        size = roundup(size, PMD_SIZE);
1149
1150        if (addr >= __START_KERNEL_map) {
1151                xen_cleanhighmap(addr, addr + size);
1152                size = PAGE_ALIGN(xen_start_info->nr_pages *
1153                                  sizeof(unsigned long));
1154                memblock_free(__pa(addr), size);
1155        } else {
1156                xen_cleanmfnmap(addr);
1157        }
1158}
1159
1160static void __init xen_pagetable_cleanhighmap(void)
1161{
1162        unsigned long size;
1163        unsigned long addr;
1164
1165        /* At this stage, cleanup_highmap has already cleaned __ka space
1166         * from _brk_limit way up to the max_pfn_mapped (which is the end of
1167         * the ramdisk). We continue on, erasing PMD entries that point to page
1168         * tables - do note that they are accessible at this stage via __va.
1169         * As Xen is aligning the memory end to a 4MB boundary, for good
1170         * measure we also round up to PMD_SIZE * 2 - which means that if
1171         * anybody is using __ka address to the initial boot-stack - and try
1172         * to use it - they are going to crash. The xen_start_info has been
1173         * taken care of already in xen_setup_kernel_pagetable. */
1174        addr = xen_start_info->pt_base;
1175        size = xen_start_info->nr_pt_frames * PAGE_SIZE;
1176
1177        xen_cleanhighmap(addr, roundup(addr + size, PMD_SIZE * 2));
1178        xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base));
1179}
1180
1181static void __init xen_pagetable_p2m_setup(void)
1182{
1183        xen_vmalloc_p2m_tree();
1184
1185        xen_pagetable_p2m_free();
1186
1187        xen_pagetable_cleanhighmap();
1188
1189        /* And revector! Bye bye old array */
1190        xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
1191}
1192
1193static void __init xen_pagetable_init(void)
1194{
1195        paging_init();
1196        xen_post_allocator_init();
1197
1198        xen_pagetable_p2m_setup();
1199
1200        /* Allocate and initialize top and mid mfn levels for p2m structure */
1201        xen_build_mfn_list_list();
1202
1203        /* Remap memory freed due to conflicts with E820 map */
1204        xen_remap_memory();
1205        xen_setup_mfn_list_list();
1206}
1207static void xen_write_cr2(unsigned long cr2)
1208{
1209        this_cpu_read(xen_vcpu)->arch.cr2 = cr2;
1210}
1211
1212static noinline void xen_flush_tlb(void)
1213{
1214        struct mmuext_op *op;
1215        struct multicall_space mcs;
1216
1217        preempt_disable();
1218
1219        mcs = xen_mc_entry(sizeof(*op));
1220
1221        op = mcs.args;
1222        op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1223        MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1224
1225        xen_mc_issue(PARAVIRT_LAZY_MMU);
1226
1227        preempt_enable();
1228}
1229
1230static void xen_flush_tlb_one_user(unsigned long addr)
1231{
1232        struct mmuext_op *op;
1233        struct multicall_space mcs;
1234
1235        trace_xen_mmu_flush_tlb_one_user(addr);
1236
1237        preempt_disable();
1238
1239        mcs = xen_mc_entry(sizeof(*op));
1240        op = mcs.args;
1241        op->cmd = MMUEXT_INVLPG_LOCAL;
1242        op->arg1.linear_addr = addr & PAGE_MASK;
1243        MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1244
1245        xen_mc_issue(PARAVIRT_LAZY_MMU);
1246
1247        preempt_enable();
1248}
1249
1250static void xen_flush_tlb_multi(const struct cpumask *cpus,
1251                                const struct flush_tlb_info *info)
1252{
1253        struct {
1254                struct mmuext_op op;
1255                DECLARE_BITMAP(mask, NR_CPUS);
1256        } *args;
1257        struct multicall_space mcs;
1258        const size_t mc_entry_size = sizeof(args->op) +
1259                sizeof(args->mask[0]) * BITS_TO_LONGS(num_possible_cpus());
1260
1261        trace_xen_mmu_flush_tlb_multi(cpus, info->mm, info->start, info->end);
1262
1263        if (cpumask_empty(cpus))
1264                return;         /* nothing to do */
1265
1266        mcs = xen_mc_entry(mc_entry_size);
1267        args = mcs.args;
1268        args->op.arg2.vcpumask = to_cpumask(args->mask);
1269
1270        /* Remove any offline CPUs */
1271        cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1272
1273        args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1274        if (info->end != TLB_FLUSH_ALL &&
1275            (info->end - info->start) <= PAGE_SIZE) {
1276                args->op.cmd = MMUEXT_INVLPG_MULTI;
1277                args->op.arg1.linear_addr = info->start;
1278        }
1279
1280        MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1281
1282        xen_mc_issue(PARAVIRT_LAZY_MMU);
1283}
1284
1285static unsigned long xen_read_cr3(void)
1286{
1287        return this_cpu_read(xen_cr3);
1288}
1289
1290static void set_current_cr3(void *v)
1291{
1292        this_cpu_write(xen_current_cr3, (unsigned long)v);
1293}
1294
1295static void __xen_write_cr3(bool kernel, unsigned long cr3)
1296{
1297        struct mmuext_op op;
1298        unsigned long mfn;
1299
1300        trace_xen_mmu_write_cr3(kernel, cr3);
1301
1302        if (cr3)
1303                mfn = pfn_to_mfn(PFN_DOWN(cr3));
1304        else
1305                mfn = 0;
1306
1307        WARN_ON(mfn == 0 && kernel);
1308
1309        op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1310        op.arg1.mfn = mfn;
1311
1312        xen_extend_mmuext_op(&op);
1313
1314        if (kernel) {
1315                this_cpu_write(xen_cr3, cr3);
1316
1317                /* Update xen_current_cr3 once the batch has actually
1318                   been submitted. */
1319                xen_mc_callback(set_current_cr3, (void *)cr3);
1320        }
1321}
1322static void xen_write_cr3(unsigned long cr3)
1323{
1324        pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1325
1326        BUG_ON(preemptible());
1327
1328        xen_mc_batch();  /* disables interrupts */
1329
1330        /* Update while interrupts are disabled, so its atomic with
1331           respect to ipis */
1332        this_cpu_write(xen_cr3, cr3);
1333
1334        __xen_write_cr3(true, cr3);
1335
1336        if (user_pgd)
1337                __xen_write_cr3(false, __pa(user_pgd));
1338        else
1339                __xen_write_cr3(false, 0);
1340
1341        xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
1342}
1343
1344/*
1345 * At the start of the day - when Xen launches a guest, it has already
1346 * built pagetables for the guest. We diligently look over them
1347 * in xen_setup_kernel_pagetable and graft as appropriate them in the
1348 * init_top_pgt and its friends. Then when we are happy we load
1349 * the new init_top_pgt - and continue on.
1350 *
1351 * The generic code starts (start_kernel) and 'init_mem_mapping' sets
1352 * up the rest of the pagetables. When it has completed it loads the cr3.
1353 * N.B. that baremetal would start at 'start_kernel' (and the early
1354 * #PF handler would create bootstrap pagetables) - so we are running
1355 * with the same assumptions as what to do when write_cr3 is executed
1356 * at this point.
1357 *
1358 * Since there are no user-page tables at all, we have two variants
1359 * of xen_write_cr3 - the early bootup (this one), and the late one
1360 * (xen_write_cr3). The reason we have to do that is that in 64-bit
1361 * the Linux kernel and user-space are both in ring 3 while the
1362 * hypervisor is in ring 0.
1363 */
1364static void __init xen_write_cr3_init(unsigned long cr3)
1365{
1366        BUG_ON(preemptible());
1367
1368        xen_mc_batch();  /* disables interrupts */
1369
1370        /* Update while interrupts are disabled, so its atomic with
1371           respect to ipis */
1372        this_cpu_write(xen_cr3, cr3);
1373
1374        __xen_write_cr3(true, cr3);
1375
1376        xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
1377}
1378
1379static int xen_pgd_alloc(struct mm_struct *mm)
1380{
1381        pgd_t *pgd = mm->pgd;
1382        struct page *page = virt_to_page(pgd);
1383        pgd_t *user_pgd;
1384        int ret = -ENOMEM;
1385
1386        BUG_ON(PagePinned(virt_to_page(pgd)));
1387        BUG_ON(page->private != 0);
1388
1389        user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1390        page->private = (unsigned long)user_pgd;
1391
1392        if (user_pgd != NULL) {
1393#ifdef CONFIG_X86_VSYSCALL_EMULATION
1394                user_pgd[pgd_index(VSYSCALL_ADDR)] =
1395                        __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1396#endif
1397                ret = 0;
1398        }
1399
1400        BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1401
1402        return ret;
1403}
1404
1405static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1406{
1407        pgd_t *user_pgd = xen_get_user_pgd(pgd);
1408
1409        if (user_pgd)
1410                free_page((unsigned long)user_pgd);
1411}
1412
1413/*
1414 * Init-time set_pte while constructing initial pagetables, which
1415 * doesn't allow RO page table pages to be remapped RW.
1416 *
1417 * If there is no MFN for this PFN then this page is initially
1418 * ballooned out so clear the PTE (as in decrease_reservation() in
1419 * drivers/xen/balloon.c).
1420 *
1421 * Many of these PTE updates are done on unpinned and writable pages
1422 * and doing a hypercall for these is unnecessary and expensive.  At
1423 * this point it is not possible to tell if a page is pinned or not,
1424 * so always write the PTE directly and rely on Xen trapping and
1425 * emulating any updates as necessary.
1426 */
1427__visible pte_t xen_make_pte_init(pteval_t pte)
1428{
1429        unsigned long pfn;
1430
1431        /*
1432         * Pages belonging to the initial p2m list mapped outside the default
1433         * address range must be mapped read-only. This region contains the
1434         * page tables for mapping the p2m list, too, and page tables MUST be
1435         * mapped read-only.
1436         */
1437        pfn = (pte & PTE_PFN_MASK) >> PAGE_SHIFT;
1438        if (xen_start_info->mfn_list < __START_KERNEL_map &&
1439            pfn >= xen_start_info->first_p2m_pfn &&
1440            pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames)
1441                pte &= ~_PAGE_RW;
1442
1443        pte = pte_pfn_to_mfn(pte);
1444        return native_make_pte(pte);
1445}
1446PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_init);
1447
1448static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1449{
1450        __xen_set_pte(ptep, pte);
1451}
1452
1453/* Early in boot, while setting up the initial pagetable, assume
1454   everything is pinned. */
1455static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1456{
1457#ifdef CONFIG_FLATMEM
1458        BUG_ON(mem_map);        /* should only be used early */
1459#endif
1460        make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1461        pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1462}
1463
1464/* Used for pmd and pud */
1465static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1466{
1467#ifdef CONFIG_FLATMEM
1468        BUG_ON(mem_map);        /* should only be used early */
1469#endif
1470        make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1471}
1472
1473/* Early release_pte assumes that all pts are pinned, since there's
1474   only init_mm and anything attached to that is pinned. */
1475static void __init xen_release_pte_init(unsigned long pfn)
1476{
1477        pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1478        make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1479}
1480
1481static void __init xen_release_pmd_init(unsigned long pfn)
1482{
1483        make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1484}
1485
1486static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1487{
1488        struct multicall_space mcs;
1489        struct mmuext_op *op;
1490
1491        mcs = __xen_mc_entry(sizeof(*op));
1492        op = mcs.args;
1493        op->cmd = cmd;
1494        op->arg1.mfn = pfn_to_mfn(pfn);
1495
1496        MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
1497}
1498
1499static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
1500{
1501        struct multicall_space mcs;
1502        unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
1503
1504        mcs = __xen_mc_entry(0);
1505        MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
1506                                pfn_pte(pfn, prot), 0);
1507}
1508
1509/* This needs to make sure the new pte page is pinned iff its being
1510   attached to a pinned pagetable. */
1511static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
1512                                    unsigned level)
1513{
1514        bool pinned = xen_page_pinned(mm->pgd);
1515
1516        trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
1517
1518        if (pinned) {
1519                struct page *page = pfn_to_page(pfn);
1520
1521                if (static_branch_likely(&xen_struct_pages_ready))
1522                        SetPagePinned(page);
1523
1524                xen_mc_batch();
1525
1526                __set_pfn_prot(pfn, PAGE_KERNEL_RO);
1527
1528                if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1529                        __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1530
1531                xen_mc_issue(PARAVIRT_LAZY_MMU);
1532        }
1533}
1534
1535static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1536{
1537        xen_alloc_ptpage(mm, pfn, PT_PTE);
1538}
1539
1540static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1541{
1542        xen_alloc_ptpage(mm, pfn, PT_PMD);
1543}
1544
1545/* This should never happen until we're OK to use struct page */
1546static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
1547{
1548        struct page *page = pfn_to_page(pfn);
1549        bool pinned = PagePinned(page);
1550
1551        trace_xen_mmu_release_ptpage(pfn, level, pinned);
1552
1553        if (pinned) {
1554                xen_mc_batch();
1555
1556                if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1557                        __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1558
1559                __set_pfn_prot(pfn, PAGE_KERNEL);
1560
1561                xen_mc_issue(PARAVIRT_LAZY_MMU);
1562
1563                ClearPagePinned(page);
1564        }
1565}
1566
1567static void xen_release_pte(unsigned long pfn)
1568{
1569        xen_release_ptpage(pfn, PT_PTE);
1570}
1571
1572static void xen_release_pmd(unsigned long pfn)
1573{
1574        xen_release_ptpage(pfn, PT_PMD);
1575}
1576
1577static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1578{
1579        xen_alloc_ptpage(mm, pfn, PT_PUD);
1580}
1581
1582static void xen_release_pud(unsigned long pfn)
1583{
1584        xen_release_ptpage(pfn, PT_PUD);
1585}
1586
1587/*
1588 * Like __va(), but returns address in the kernel mapping (which is
1589 * all we have until the physical memory mapping has been set up.
1590 */
1591static void * __init __ka(phys_addr_t paddr)
1592{
1593        return (void *)(paddr + __START_KERNEL_map);
1594}
1595
1596/* Convert a machine address to physical address */
1597static unsigned long __init m2p(phys_addr_t maddr)
1598{
1599        phys_addr_t paddr;
1600
1601        maddr &= XEN_PTE_MFN_MASK;
1602        paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1603
1604        return paddr;
1605}
1606
1607/* Convert a machine address to kernel virtual */
1608static void * __init m2v(phys_addr_t maddr)
1609{
1610        return __ka(m2p(maddr));
1611}
1612
1613/* Set the page permissions on an identity-mapped pages */
1614static void __init set_page_prot_flags(void *addr, pgprot_t prot,
1615                                       unsigned long flags)
1616{
1617        unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1618        pte_t pte = pfn_pte(pfn, prot);
1619
1620        if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags))
1621                BUG();
1622}
1623static void __init set_page_prot(void *addr, pgprot_t prot)
1624{
1625        return set_page_prot_flags(addr, prot, UVMF_NONE);
1626}
1627
1628void __init xen_setup_machphys_mapping(void)
1629{
1630        struct xen_machphys_mapping mapping;
1631
1632        if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1633                machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1634                machine_to_phys_nr = mapping.max_mfn + 1;
1635        } else {
1636                machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
1637        }
1638}
1639
1640static void __init convert_pfn_mfn(void *v)
1641{
1642        pte_t *pte = v;
1643        int i;
1644
1645        /* All levels are converted the same way, so just treat them
1646           as ptes. */
1647        for (i = 0; i < PTRS_PER_PTE; i++)
1648                pte[i] = xen_make_pte(pte[i].pte);
1649}
1650static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end,
1651                                 unsigned long addr)
1652{
1653        if (*pt_base == PFN_DOWN(__pa(addr))) {
1654                set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1655                clear_page((void *)addr);
1656                (*pt_base)++;
1657        }
1658        if (*pt_end == PFN_DOWN(__pa(addr))) {
1659                set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1660                clear_page((void *)addr);
1661                (*pt_end)--;
1662        }
1663}
1664/*
1665 * Set up the initial kernel pagetable.
1666 *
1667 * We can construct this by grafting the Xen provided pagetable into
1668 * head_64.S's preconstructed pagetables.  We copy the Xen L2's into
1669 * level2_ident_pgt, and level2_kernel_pgt.  This means that only the
1670 * kernel has a physical mapping to start with - but that's enough to
1671 * get __va working.  We need to fill in the rest of the physical
1672 * mapping once some sort of allocator has been set up.
1673 */
1674void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1675{
1676        pud_t *l3;
1677        pmd_t *l2;
1678        unsigned long addr[3];
1679        unsigned long pt_base, pt_end;
1680        unsigned i;
1681
1682        /* max_pfn_mapped is the last pfn mapped in the initial memory
1683         * mappings. Considering that on Xen after the kernel mappings we
1684         * have the mappings of some pages that don't exist in pfn space, we
1685         * set max_pfn_mapped to the last real pfn mapped. */
1686        if (xen_start_info->mfn_list < __START_KERNEL_map)
1687                max_pfn_mapped = xen_start_info->first_p2m_pfn;
1688        else
1689                max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1690
1691        pt_base = PFN_DOWN(__pa(xen_start_info->pt_base));
1692        pt_end = pt_base + xen_start_info->nr_pt_frames;
1693
1694        /* Zap identity mapping */
1695        init_top_pgt[0] = __pgd(0);
1696
1697        /* Pre-constructed entries are in pfn, so convert to mfn */
1698        /* L4[273] -> level3_ident_pgt  */
1699        /* L4[511] -> level3_kernel_pgt */
1700        convert_pfn_mfn(init_top_pgt);
1701
1702        /* L3_i[0] -> level2_ident_pgt */
1703        convert_pfn_mfn(level3_ident_pgt);
1704        /* L3_k[510] -> level2_kernel_pgt */
1705        /* L3_k[511] -> level2_fixmap_pgt */
1706        convert_pfn_mfn(level3_kernel_pgt);
1707
1708        /* L3_k[511][508-FIXMAP_PMD_NUM ... 507] -> level1_fixmap_pgt */
1709        convert_pfn_mfn(level2_fixmap_pgt);
1710
1711        /* We get [511][511] and have Xen's version of level2_kernel_pgt */
1712        l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1713        l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1714
1715        addr[0] = (unsigned long)pgd;
1716        addr[1] = (unsigned long)l3;
1717        addr[2] = (unsigned long)l2;
1718        /* Graft it onto L4[273][0]. Note that we creating an aliasing problem:
1719         * Both L4[273][0] and L4[511][510] have entries that point to the same
1720         * L2 (PMD) tables. Meaning that if you modify it in __va space
1721         * it will be also modified in the __ka space! (But if you just
1722         * modify the PMD table to point to other PTE's or none, then you
1723         * are OK - which is what cleanup_highmap does) */
1724        copy_page(level2_ident_pgt, l2);
1725        /* Graft it onto L4[511][510] */
1726        copy_page(level2_kernel_pgt, l2);
1727
1728        /*
1729         * Zap execute permission from the ident map. Due to the sharing of
1730         * L1 entries we need to do this in the L2.
1731         */
1732        if (__supported_pte_mask & _PAGE_NX) {
1733                for (i = 0; i < PTRS_PER_PMD; ++i) {
1734                        if (pmd_none(level2_ident_pgt[i]))
1735                                continue;
1736                        level2_ident_pgt[i] = pmd_set_flags(level2_ident_pgt[i], _PAGE_NX);
1737                }
1738        }
1739
1740        /* Copy the initial P->M table mappings if necessary. */
1741        i = pgd_index(xen_start_info->mfn_list);
1742        if (i && i < pgd_index(__START_KERNEL_map))
1743                init_top_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i];
1744
1745        /* Make pagetable pieces RO */
1746        set_page_prot(init_top_pgt, PAGE_KERNEL_RO);
1747        set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1748        set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1749        set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1750        set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO);
1751        set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1752        set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1753
1754        for (i = 0; i < FIXMAP_PMD_NUM; i++) {
1755                set_page_prot(level1_fixmap_pgt + i * PTRS_PER_PTE,
1756                              PAGE_KERNEL_RO);
1757        }
1758
1759        /* Pin down new L4 */
1760        pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1761                          PFN_DOWN(__pa_symbol(init_top_pgt)));
1762
1763        /* Unpin Xen-provided one */
1764        pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1765
1766        /*
1767         * At this stage there can be no user pgd, and no page structure to
1768         * attach it to, so make sure we just set kernel pgd.
1769         */
1770        xen_mc_batch();
1771        __xen_write_cr3(true, __pa(init_top_pgt));
1772        xen_mc_issue(PARAVIRT_LAZY_CPU);
1773
1774        /* We can't that easily rip out L3 and L2, as the Xen pagetables are
1775         * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ...  for
1776         * the initial domain. For guests using the toolstack, they are in:
1777         * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only
1778         * rip out the [L4] (pgd), but for guests we shave off three pages.
1779         */
1780        for (i = 0; i < ARRAY_SIZE(addr); i++)
1781                check_pt_base(&pt_base, &pt_end, addr[i]);
1782
1783        /* Our (by three pages) smaller Xen pagetable that we are using */
1784        xen_pt_base = PFN_PHYS(pt_base);
1785        xen_pt_size = (pt_end - pt_base) * PAGE_SIZE;
1786        memblock_reserve(xen_pt_base, xen_pt_size);
1787
1788        /* Revector the xen_start_info */
1789        xen_start_info = (struct start_info *)__va(__pa(xen_start_info));
1790}
1791
1792/*
1793 * Read a value from a physical address.
1794 */
1795static unsigned long __init xen_read_phys_ulong(phys_addr_t addr)
1796{
1797        unsigned long *vaddr;
1798        unsigned long val;
1799
1800        vaddr = early_memremap_ro(addr, sizeof(val));
1801        val = *vaddr;
1802        early_memunmap(vaddr, sizeof(val));
1803        return val;
1804}
1805
1806/*
1807 * Translate a virtual address to a physical one without relying on mapped
1808 * page tables. Don't rely on big pages being aligned in (guest) physical
1809 * space!
1810 */
1811static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr)
1812{
1813        phys_addr_t pa;
1814        pgd_t pgd;
1815        pud_t pud;
1816        pmd_t pmd;
1817        pte_t pte;
1818
1819        pa = read_cr3_pa();
1820        pgd = native_make_pgd(xen_read_phys_ulong(pa + pgd_index(vaddr) *
1821                                                       sizeof(pgd)));
1822        if (!pgd_present(pgd))
1823                return 0;
1824
1825        pa = pgd_val(pgd) & PTE_PFN_MASK;
1826        pud = native_make_pud(xen_read_phys_ulong(pa + pud_index(vaddr) *
1827                                                       sizeof(pud)));
1828        if (!pud_present(pud))
1829                return 0;
1830        pa = pud_val(pud) & PTE_PFN_MASK;
1831        if (pud_large(pud))
1832                return pa + (vaddr & ~PUD_MASK);
1833
1834        pmd = native_make_pmd(xen_read_phys_ulong(pa + pmd_index(vaddr) *
1835                                                       sizeof(pmd)));
1836        if (!pmd_present(pmd))
1837                return 0;
1838        pa = pmd_val(pmd) & PTE_PFN_MASK;
1839        if (pmd_large(pmd))
1840                return pa + (vaddr & ~PMD_MASK);
1841
1842        pte = native_make_pte(xen_read_phys_ulong(pa + pte_index(vaddr) *
1843                                                       sizeof(pte)));
1844        if (!pte_present(pte))
1845                return 0;
1846        pa = pte_pfn(pte) << PAGE_SHIFT;
1847
1848        return pa | (vaddr & ~PAGE_MASK);
1849}
1850
1851/*
1852 * Find a new area for the hypervisor supplied p2m list and relocate the p2m to
1853 * this area.
1854 */
1855void __init xen_relocate_p2m(void)
1856{
1857        phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys;
1858        unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end;
1859        int n_pte, n_pt, n_pmd, n_pud, idx_pte, idx_pt, idx_pmd, idx_pud;
1860        pte_t *pt;
1861        pmd_t *pmd;
1862        pud_t *pud;
1863        pgd_t *pgd;
1864        unsigned long *new_p2m;
1865
1866        size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1867        n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT;
1868        n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT;
1869        n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT;
1870        n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT;
1871        n_frames = n_pte + n_pt + n_pmd + n_pud;
1872
1873        new_area = xen_find_free_area(PFN_PHYS(n_frames));
1874        if (!new_area) {
1875                xen_raw_console_write("Can't find new memory area for p2m needed due to E820 map conflict\n");
1876                BUG();
1877        }
1878
1879        /*
1880         * Setup the page tables for addressing the new p2m list.
1881         * We have asked the hypervisor to map the p2m list at the user address
1882         * PUD_SIZE. It may have done so, or it may have used a kernel space
1883         * address depending on the Xen version.
1884         * To avoid any possible virtual address collision, just use
1885         * 2 * PUD_SIZE for the new area.
1886         */
1887        pud_phys = new_area;
1888        pmd_phys = pud_phys + PFN_PHYS(n_pud);
1889        pt_phys = pmd_phys + PFN_PHYS(n_pmd);
1890        p2m_pfn = PFN_DOWN(pt_phys) + n_pt;
1891
1892        pgd = __va(read_cr3_pa());
1893        new_p2m = (unsigned long *)(2 * PGDIR_SIZE);
1894        for (idx_pud = 0; idx_pud < n_pud; idx_pud++) {
1895                pud = early_memremap(pud_phys, PAGE_SIZE);
1896                clear_page(pud);
1897                for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD);
1898                                idx_pmd++) {
1899                        pmd = early_memremap(pmd_phys, PAGE_SIZE);
1900                        clear_page(pmd);
1901                        for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD);
1902                                        idx_pt++) {
1903                                pt = early_memremap(pt_phys, PAGE_SIZE);
1904                                clear_page(pt);
1905                                for (idx_pte = 0;
1906                                     idx_pte < min(n_pte, PTRS_PER_PTE);
1907                                     idx_pte++) {
1908                                        pt[idx_pte] = pfn_pte(p2m_pfn,
1909                                                              PAGE_KERNEL);
1910                                        p2m_pfn++;
1911                                }
1912                                n_pte -= PTRS_PER_PTE;
1913                                early_memunmap(pt, PAGE_SIZE);
1914                                make_lowmem_page_readonly(__va(pt_phys));
1915                                pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE,
1916                                                PFN_DOWN(pt_phys));
1917                                pmd[idx_pt] = __pmd(_PAGE_TABLE | pt_phys);
1918                                pt_phys += PAGE_SIZE;
1919                        }
1920                        n_pt -= PTRS_PER_PMD;
1921                        early_memunmap(pmd, PAGE_SIZE);
1922                        make_lowmem_page_readonly(__va(pmd_phys));
1923                        pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE,
1924                                        PFN_DOWN(pmd_phys));
1925                        pud[idx_pmd] = __pud(_PAGE_TABLE | pmd_phys);
1926                        pmd_phys += PAGE_SIZE;
1927                }
1928                n_pmd -= PTRS_PER_PUD;
1929                early_memunmap(pud, PAGE_SIZE);
1930                make_lowmem_page_readonly(__va(pud_phys));
1931                pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys));
1932                set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys));
1933                pud_phys += PAGE_SIZE;
1934        }
1935
1936        /* Now copy the old p2m info to the new area. */
1937        memcpy(new_p2m, xen_p2m_addr, size);
1938        xen_p2m_addr = new_p2m;
1939
1940        /* Release the old p2m list and set new list info. */
1941        p2m_pfn = PFN_DOWN(xen_early_virt_to_phys(xen_start_info->mfn_list));
1942        BUG_ON(!p2m_pfn);
1943        p2m_pfn_end = p2m_pfn + PFN_DOWN(size);
1944
1945        if (xen_start_info->mfn_list < __START_KERNEL_map) {
1946                pfn = xen_start_info->first_p2m_pfn;
1947                pfn_end = xen_start_info->first_p2m_pfn +
1948                          xen_start_info->nr_p2m_frames;
1949                set_pgd(pgd + 1, __pgd(0));
1950        } else {
1951                pfn = p2m_pfn;
1952                pfn_end = p2m_pfn_end;
1953        }
1954
1955        memblock_free(PFN_PHYS(pfn), PAGE_SIZE * (pfn_end - pfn));
1956        while (pfn < pfn_end) {
1957                if (pfn == p2m_pfn) {
1958                        pfn = p2m_pfn_end;
1959                        continue;
1960                }
1961                make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1962                pfn++;
1963        }
1964
1965        xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
1966        xen_start_info->first_p2m_pfn =  PFN_DOWN(new_area);
1967        xen_start_info->nr_p2m_frames = n_frames;
1968}
1969
1970void __init xen_reserve_special_pages(void)
1971{
1972        phys_addr_t paddr;
1973
1974        memblock_reserve(__pa(xen_start_info), PAGE_SIZE);
1975        if (xen_start_info->store_mfn) {
1976                paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn));
1977                memblock_reserve(paddr, PAGE_SIZE);
1978        }
1979        if (!xen_initial_domain()) {
1980                paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn));
1981                memblock_reserve(paddr, PAGE_SIZE);
1982        }
1983}
1984
1985void __init xen_pt_check_e820(void)
1986{
1987        if (xen_is_e820_reserved(xen_pt_base, xen_pt_size)) {
1988                xen_raw_console_write("Xen hypervisor allocated page table memory conflicts with E820 map\n");
1989                BUG();
1990        }
1991}
1992
1993static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
1994
1995static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
1996{
1997        pte_t pte;
1998
1999        phys >>= PAGE_SHIFT;
2000
2001        switch (idx) {
2002        case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
2003#ifdef CONFIG_X86_VSYSCALL_EMULATION
2004        case VSYSCALL_PAGE:
2005#endif
2006                /* All local page mappings */
2007                pte = pfn_pte(phys, prot);
2008                break;
2009
2010#ifdef CONFIG_X86_LOCAL_APIC
2011        case FIX_APIC_BASE:     /* maps dummy local APIC */
2012                pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2013                break;
2014#endif
2015
2016#ifdef CONFIG_X86_IO_APIC
2017        case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
2018                /*
2019                 * We just don't map the IO APIC - all access is via
2020                 * hypercalls.  Keep the address in the pte for reference.
2021                 */
2022                pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2023                break;
2024#endif
2025
2026        case FIX_PARAVIRT_BOOTMAP:
2027                /* This is an MFN, but it isn't an IO mapping from the
2028                   IO domain */
2029                pte = mfn_pte(phys, prot);
2030                break;
2031
2032        default:
2033                /* By default, set_fixmap is used for hardware mappings */
2034                pte = mfn_pte(phys, prot);
2035                break;
2036        }
2037
2038        __native_set_fixmap(idx, pte);
2039
2040#ifdef CONFIG_X86_VSYSCALL_EMULATION
2041        /* Replicate changes to map the vsyscall page into the user
2042           pagetable vsyscall mapping. */
2043        if (idx == VSYSCALL_PAGE) {
2044                unsigned long vaddr = __fix_to_virt(idx);
2045                set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
2046        }
2047#endif
2048}
2049
2050static void __init xen_post_allocator_init(void)
2051{
2052        pv_ops.mmu.set_pte = xen_set_pte;
2053        pv_ops.mmu.set_pmd = xen_set_pmd;
2054        pv_ops.mmu.set_pud = xen_set_pud;
2055        pv_ops.mmu.set_p4d = xen_set_p4d;
2056
2057        /* This will work as long as patching hasn't happened yet
2058           (which it hasn't) */
2059        pv_ops.mmu.alloc_pte = xen_alloc_pte;
2060        pv_ops.mmu.alloc_pmd = xen_alloc_pmd;
2061        pv_ops.mmu.release_pte = xen_release_pte;
2062        pv_ops.mmu.release_pmd = xen_release_pmd;
2063        pv_ops.mmu.alloc_pud = xen_alloc_pud;
2064        pv_ops.mmu.release_pud = xen_release_pud;
2065        pv_ops.mmu.make_pte = PV_CALLEE_SAVE(xen_make_pte);
2066
2067        pv_ops.mmu.write_cr3 = &xen_write_cr3;
2068}
2069
2070static void xen_leave_lazy_mmu(void)
2071{
2072        preempt_disable();
2073        xen_mc_flush();
2074        paravirt_leave_lazy_mmu();
2075        preempt_enable();
2076}
2077
2078static const struct pv_mmu_ops xen_mmu_ops __initconst = {
2079        .read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2),
2080        .write_cr2 = xen_write_cr2,
2081
2082        .read_cr3 = xen_read_cr3,
2083        .write_cr3 = xen_write_cr3_init,
2084
2085        .flush_tlb_user = xen_flush_tlb,
2086        .flush_tlb_kernel = xen_flush_tlb,
2087        .flush_tlb_one_user = xen_flush_tlb_one_user,
2088        .flush_tlb_multi = xen_flush_tlb_multi,
2089        .tlb_remove_table = tlb_remove_table,
2090
2091        .pgd_alloc = xen_pgd_alloc,
2092        .pgd_free = xen_pgd_free,
2093
2094        .alloc_pte = xen_alloc_pte_init,
2095        .release_pte = xen_release_pte_init,
2096        .alloc_pmd = xen_alloc_pmd_init,
2097        .release_pmd = xen_release_pmd_init,
2098
2099        .set_pte = xen_set_pte_init,
2100        .set_pmd = xen_set_pmd_hyper,
2101
2102        .ptep_modify_prot_start = __ptep_modify_prot_start,
2103        .ptep_modify_prot_commit = __ptep_modify_prot_commit,
2104
2105        .pte_val = PV_CALLEE_SAVE(xen_pte_val),
2106        .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2107
2108        .make_pte = PV_CALLEE_SAVE(xen_make_pte_init),
2109        .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2110
2111        .set_pud = xen_set_pud_hyper,
2112
2113        .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2114        .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2115
2116        .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2117        .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2118        .set_p4d = xen_set_p4d_hyper,
2119
2120        .alloc_pud = xen_alloc_pmd_init,
2121        .release_pud = xen_release_pmd_init,
2122
2123#if CONFIG_PGTABLE_LEVELS >= 5
2124        .p4d_val = PV_CALLEE_SAVE(xen_p4d_val),
2125        .make_p4d = PV_CALLEE_SAVE(xen_make_p4d),
2126#endif
2127
2128        .activate_mm = xen_activate_mm,
2129        .dup_mmap = xen_dup_mmap,
2130        .exit_mmap = xen_exit_mmap,
2131
2132        .lazy_mode = {
2133                .enter = paravirt_enter_lazy_mmu,
2134                .leave = xen_leave_lazy_mmu,
2135                .flush = paravirt_flush_lazy_mmu,
2136        },
2137
2138        .set_fixmap = xen_set_fixmap,
2139};
2140
2141void __init xen_init_mmu_ops(void)
2142{
2143        x86_init.paging.pagetable_init = xen_pagetable_init;
2144        x86_init.hyper.init_after_bootmem = xen_after_bootmem;
2145
2146        pv_ops.mmu = xen_mmu_ops;
2147
2148        memset(dummy_mapping, 0xff, PAGE_SIZE);
2149}
2150
2151/* Protected by xen_reservation_lock. */
2152#define MAX_CONTIG_ORDER 9 /* 2MB */
2153static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2154
2155#define VOID_PTE (mfn_pte(0, __pgprot(0)))
2156static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2157                                unsigned long *in_frames,
2158                                unsigned long *out_frames)
2159{
2160        int i;
2161        struct multicall_space mcs;
2162
2163        xen_mc_batch();
2164        for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2165                mcs = __xen_mc_entry(0);
2166
2167                if (in_frames)
2168                        in_frames[i] = virt_to_mfn(vaddr);
2169
2170                MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2171                __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2172
2173                if (out_frames)
2174                        out_frames[i] = virt_to_pfn(vaddr);
2175        }
2176        xen_mc_issue(0);
2177}
2178
2179/*
2180 * Update the pfn-to-mfn mappings for a virtual address range, either to
2181 * point to an array of mfns, or contiguously from a single starting
2182 * mfn.
2183 */
2184static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2185                                     unsigned long *mfns,
2186                                     unsigned long first_mfn)
2187{
2188        unsigned i, limit;
2189        unsigned long mfn;
2190
2191        xen_mc_batch();
2192
2193        limit = 1u << order;
2194        for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2195                struct multicall_space mcs;
2196                unsigned flags;
2197
2198                mcs = __xen_mc_entry(0);
2199                if (mfns)
2200                        mfn = mfns[i];
2201                else
2202                        mfn = first_mfn + i;
2203
2204                if (i < (limit - 1))
2205                        flags = 0;
2206                else {
2207                        if (order == 0)
2208                                flags = UVMF_INVLPG | UVMF_ALL;
2209                        else
2210                                flags = UVMF_TLB_FLUSH | UVMF_ALL;
2211                }
2212
2213                MULTI_update_va_mapping(mcs.mc, vaddr,
2214                                mfn_pte(mfn, PAGE_KERNEL), flags);
2215
2216                set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2217        }
2218
2219        xen_mc_issue(0);
2220}
2221
2222/*
2223 * Perform the hypercall to exchange a region of our pfns to point to
2224 * memory with the required contiguous alignment.  Takes the pfns as
2225 * input, and populates mfns as output.
2226 *
2227 * Returns a success code indicating whether the hypervisor was able to
2228 * satisfy the request or not.
2229 */
2230static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2231                               unsigned long *pfns_in,
2232                               unsigned long extents_out,
2233                               unsigned int order_out,
2234                               unsigned long *mfns_out,
2235                               unsigned int address_bits)
2236{
2237        long rc;
2238        int success;
2239
2240        struct xen_memory_exchange exchange = {
2241                .in = {
2242                        .nr_extents   = extents_in,
2243                        .extent_order = order_in,
2244                        .extent_start = pfns_in,
2245                        .domid        = DOMID_SELF
2246                },
2247                .out = {
2248                        .nr_extents   = extents_out,
2249                        .extent_order = order_out,
2250                        .extent_start = mfns_out,
2251                        .address_bits = address_bits,
2252                        .domid        = DOMID_SELF
2253                }
2254        };
2255
2256        BUG_ON(extents_in << order_in != extents_out << order_out);
2257
2258        rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2259        success = (exchange.nr_exchanged == extents_in);
2260
2261        BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2262        BUG_ON(success && (rc != 0));
2263
2264        return success;
2265}
2266
2267int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order,
2268                                 unsigned int address_bits,
2269                                 dma_addr_t *dma_handle)
2270{
2271        unsigned long *in_frames = discontig_frames, out_frame;
2272        unsigned long  flags;
2273        int            success;
2274        unsigned long vstart = (unsigned long)phys_to_virt(pstart);
2275
2276        /*
2277         * Currently an auto-translated guest will not perform I/O, nor will
2278         * it require PAE page directories below 4GB. Therefore any calls to
2279         * this function are redundant and can be ignored.
2280         */
2281
2282        if (unlikely(order > MAX_CONTIG_ORDER))
2283                return -ENOMEM;
2284
2285        memset((void *) vstart, 0, PAGE_SIZE << order);
2286
2287        spin_lock_irqsave(&xen_reservation_lock, flags);
2288
2289        /* 1. Zap current PTEs, remembering MFNs. */
2290        xen_zap_pfn_range(vstart, order, in_frames, NULL);
2291
2292        /* 2. Get a new contiguous memory extent. */
2293        out_frame = virt_to_pfn(vstart);
2294        success = xen_exchange_memory(1UL << order, 0, in_frames,
2295                                      1, order, &out_frame,
2296                                      address_bits);
2297
2298        /* 3. Map the new extent in place of old pages. */
2299        if (success)
2300                xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2301        else
2302                xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2303
2304        spin_unlock_irqrestore(&xen_reservation_lock, flags);
2305
2306        *dma_handle = virt_to_machine(vstart).maddr;
2307        return success ? 0 : -ENOMEM;
2308}
2309
2310void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order)
2311{
2312        unsigned long *out_frames = discontig_frames, in_frame;
2313        unsigned long  flags;
2314        int success;
2315        unsigned long vstart;
2316
2317        if (unlikely(order > MAX_CONTIG_ORDER))
2318                return;
2319
2320        vstart = (unsigned long)phys_to_virt(pstart);
2321        memset((void *) vstart, 0, PAGE_SIZE << order);
2322
2323        spin_lock_irqsave(&xen_reservation_lock, flags);
2324
2325        /* 1. Find start MFN of contiguous extent. */
2326        in_frame = virt_to_mfn(vstart);
2327
2328        /* 2. Zap current PTEs. */
2329        xen_zap_pfn_range(vstart, order, NULL, out_frames);
2330
2331        /* 3. Do the exchange for non-contiguous MFNs. */
2332        success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2333                                        0, out_frames, 0);
2334
2335        /* 4. Map new pages in place of old pages. */
2336        if (success)
2337                xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2338        else
2339                xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2340
2341        spin_unlock_irqrestore(&xen_reservation_lock, flags);
2342}
2343
2344static noinline void xen_flush_tlb_all(void)
2345{
2346        struct mmuext_op *op;
2347        struct multicall_space mcs;
2348
2349        preempt_disable();
2350
2351        mcs = xen_mc_entry(sizeof(*op));
2352
2353        op = mcs.args;
2354        op->cmd = MMUEXT_TLB_FLUSH_ALL;
2355        MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
2356
2357        xen_mc_issue(PARAVIRT_LAZY_MMU);
2358
2359        preempt_enable();
2360}
2361
2362#define REMAP_BATCH_SIZE 16
2363
2364struct remap_data {
2365        xen_pfn_t *pfn;
2366        bool contiguous;
2367        bool no_translate;
2368        pgprot_t prot;
2369        struct mmu_update *mmu_update;
2370};
2371
2372static int remap_area_pfn_pte_fn(pte_t *ptep, unsigned long addr, void *data)
2373{
2374        struct remap_data *rmd = data;
2375        pte_t pte = pte_mkspecial(mfn_pte(*rmd->pfn, rmd->prot));
2376
2377        /*
2378         * If we have a contiguous range, just update the pfn itself,
2379         * else update pointer to be "next pfn".
2380         */
2381        if (rmd->contiguous)
2382                (*rmd->pfn)++;
2383        else
2384                rmd->pfn++;
2385
2386        rmd->mmu_update->ptr = virt_to_machine(ptep).maddr;
2387        rmd->mmu_update->ptr |= rmd->no_translate ?
2388                MMU_PT_UPDATE_NO_TRANSLATE :
2389                MMU_NORMAL_PT_UPDATE;
2390        rmd->mmu_update->val = pte_val_ma(pte);
2391        rmd->mmu_update++;
2392
2393        return 0;
2394}
2395
2396int xen_remap_pfn(struct vm_area_struct *vma, unsigned long addr,
2397                  xen_pfn_t *pfn, int nr, int *err_ptr, pgprot_t prot,
2398                  unsigned int domid, bool no_translate, struct page **pages)
2399{
2400        int err = 0;
2401        struct remap_data rmd;
2402        struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2403        unsigned long range;
2404        int mapped = 0;
2405
2406        BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_IO)) == (VM_PFNMAP | VM_IO)));
2407
2408        rmd.pfn = pfn;
2409        rmd.prot = prot;
2410        /*
2411         * We use the err_ptr to indicate if there we are doing a contiguous
2412         * mapping or a discontiguous mapping.
2413         */
2414        rmd.contiguous = !err_ptr;
2415        rmd.no_translate = no_translate;
2416
2417        while (nr) {
2418                int index = 0;
2419                int done = 0;
2420                int batch = min(REMAP_BATCH_SIZE, nr);
2421                int batch_left = batch;
2422
2423                range = (unsigned long)batch << PAGE_SHIFT;
2424
2425                rmd.mmu_update = mmu_update;
2426                err = apply_to_page_range(vma->vm_mm, addr, range,
2427                                          remap_area_pfn_pte_fn, &rmd);
2428                if (err)
2429                        goto out;
2430
2431                /*
2432                 * We record the error for each page that gives an error, but
2433                 * continue mapping until the whole set is done
2434                 */
2435                do {
2436                        int i;
2437
2438                        err = HYPERVISOR_mmu_update(&mmu_update[index],
2439                                                    batch_left, &done, domid);
2440
2441                        /*
2442                         * @err_ptr may be the same buffer as @gfn, so
2443                         * only clear it after each chunk of @gfn is
2444                         * used.
2445                         */
2446                        if (err_ptr) {
2447                                for (i = index; i < index + done; i++)
2448                                        err_ptr[i] = 0;
2449                        }
2450                        if (err < 0) {
2451                                if (!err_ptr)
2452                                        goto out;
2453                                err_ptr[i] = err;
2454                                done++; /* Skip failed frame. */
2455                        } else
2456                                mapped += done;
2457                        batch_left -= done;
2458                        index += done;
2459                } while (batch_left);
2460
2461                nr -= batch;
2462                addr += range;
2463                if (err_ptr)
2464                        err_ptr += batch;
2465                cond_resched();
2466        }
2467out:
2468
2469        xen_flush_tlb_all();
2470
2471        return err < 0 ? err : mapped;
2472}
2473EXPORT_SYMBOL_GPL(xen_remap_pfn);
2474
2475#ifdef CONFIG_KEXEC_CORE
2476phys_addr_t paddr_vmcoreinfo_note(void)
2477{
2478        if (xen_pv_domain())
2479                return virt_to_machine(vmcoreinfo_note).maddr;
2480        else
2481                return __pa(vmcoreinfo_note);
2482}
2483#endif /* CONFIG_KEXEC_CORE */
2484
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.