[SPARC64]: check fork_idle() error
[pohmelfs.git] / arch / i386 / xen / mmu.c
blob874db0cd1d2a5f6fc1c9f9a172863edb9c260339
1 /*
2 * Xen mmu operations
4 * This file contains the various mmu fetch and update operations.
5 * The most important job they must perform is the mapping between the
6 * domain's pfn and the overall machine mfns.
8 * Xen allows guests to directly update the pagetable, in a controlled
9 * fashion. In other words, the guest modifies the same pagetable
10 * that the CPU actually uses, which eliminates the overhead of having
11 * a separate shadow pagetable.
13 * In order to allow this, it falls on the guest domain to map its
14 * notion of a "physical" pfn - which is just a domain-local linear
15 * address - into a real "machine address" which the CPU's MMU can
16 * use.
18 * A pgd_t/pmd_t/pte_t will typically contain an mfn, and so can be
19 * inserted directly into the pagetable. When creating a new
20 * pte/pmd/pgd, it converts the passed pfn into an mfn. Conversely,
21 * when reading the content back with __(pgd|pmd|pte)_val, it converts
22 * the mfn back into a pfn.
24 * The other constraint is that all pages which make up a pagetable
25 * must be mapped read-only in the guest. This prevents uncontrolled
26 * guest updates to the pagetable. Xen strictly enforces this, and
27 * will disallow any pagetable update which will end up mapping a
28 * pagetable page RW, and will disallow using any writable page as a
29 * pagetable.
31 * Naively, when loading %cr3 with the base of a new pagetable, Xen
32 * would need to validate the whole pagetable before going on.
33 * Naturally, this is quite slow. The solution is to "pin" a
34 * pagetable, which enforces all the constraints on the pagetable even
35 * when it is not actively in use. This menas that Xen can be assured
36 * that it is still valid when you do load it into %cr3, and doesn't
37 * need to revalidate it.
39 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
41 #include <linux/sched.h>
42 #include <linux/highmem.h>
43 #include <linux/bug.h>
44 #include <linux/sched.h>
46 #include <asm/pgtable.h>
47 #include <asm/tlbflush.h>
48 #include <asm/mmu_context.h>
49 #include <asm/paravirt.h>
51 #include <asm/xen/hypercall.h>
52 #include <asm/xen/hypervisor.h>
54 #include <xen/page.h>
55 #include <xen/interface/xen.h>
57 #include "multicalls.h"
58 #include "mmu.h"
60 xmaddr_t arbitrary_virt_to_machine(unsigned long address)
62 pte_t *pte = lookup_address(address);
63 unsigned offset = address & PAGE_MASK;
65 BUG_ON(pte == NULL);
67 return XMADDR((pte_mfn(*pte) << PAGE_SHIFT) + offset);
70 void make_lowmem_page_readonly(void *vaddr)
72 pte_t *pte, ptev;
73 unsigned long address = (unsigned long)vaddr;
75 pte = lookup_address(address);
76 BUG_ON(pte == NULL);
78 ptev = pte_wrprotect(*pte);
80 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
81 BUG();
84 void make_lowmem_page_readwrite(void *vaddr)
86 pte_t *pte, ptev;
87 unsigned long address = (unsigned long)vaddr;
89 pte = lookup_address(address);
90 BUG_ON(pte == NULL);
92 ptev = pte_mkwrite(*pte);
94 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
95 BUG();
99 void xen_set_pmd(pmd_t *ptr, pmd_t val)
101 struct multicall_space mcs;
102 struct mmu_update *u;
104 preempt_disable();
106 mcs = xen_mc_entry(sizeof(*u));
107 u = mcs.args;
108 u->ptr = virt_to_machine(ptr).maddr;
109 u->val = pmd_val_ma(val);
110 MULTI_mmu_update(mcs.mc, u, 1, NULL, DOMID_SELF);
112 xen_mc_issue(PARAVIRT_LAZY_MMU);
114 preempt_enable();
118 * Associate a virtual page frame with a given physical page frame
119 * and protection flags for that frame.
121 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
123 pgd_t *pgd;
124 pud_t *pud;
125 pmd_t *pmd;
126 pte_t *pte;
128 pgd = swapper_pg_dir + pgd_index(vaddr);
129 if (pgd_none(*pgd)) {
130 BUG();
131 return;
133 pud = pud_offset(pgd, vaddr);
134 if (pud_none(*pud)) {
135 BUG();
136 return;
138 pmd = pmd_offset(pud, vaddr);
139 if (pmd_none(*pmd)) {
140 BUG();
141 return;
143 pte = pte_offset_kernel(pmd, vaddr);
144 /* <mfn,flags> stored as-is, to permit clearing entries */
145 xen_set_pte(pte, mfn_pte(mfn, flags));
148 * It's enough to flush this one mapping.
149 * (PGE mappings get flushed as well)
151 __flush_tlb_one(vaddr);
154 void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
155 pte_t *ptep, pte_t pteval)
157 if (mm == current->mm || mm == &init_mm) {
158 if (xen_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
159 struct multicall_space mcs;
160 mcs = xen_mc_entry(0);
162 MULTI_update_va_mapping(mcs.mc, addr, pteval, 0);
163 xen_mc_issue(PARAVIRT_LAZY_MMU);
164 return;
165 } else
166 if (HYPERVISOR_update_va_mapping(addr, pteval, 0) == 0)
167 return;
169 xen_set_pte(ptep, pteval);
172 #ifdef CONFIG_X86_PAE
173 void xen_set_pud(pud_t *ptr, pud_t val)
175 struct multicall_space mcs;
176 struct mmu_update *u;
178 preempt_disable();
180 mcs = xen_mc_entry(sizeof(*u));
181 u = mcs.args;
182 u->ptr = virt_to_machine(ptr).maddr;
183 u->val = pud_val_ma(val);
184 MULTI_mmu_update(mcs.mc, u, 1, NULL, DOMID_SELF);
186 xen_mc_issue(PARAVIRT_LAZY_MMU);
188 preempt_enable();
191 void xen_set_pte(pte_t *ptep, pte_t pte)
193 ptep->pte_high = pte.pte_high;
194 smp_wmb();
195 ptep->pte_low = pte.pte_low;
198 void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
200 set_64bit((u64 *)ptep, pte_val_ma(pte));
203 void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
205 ptep->pte_low = 0;
206 smp_wmb(); /* make sure low gets written first */
207 ptep->pte_high = 0;
210 void xen_pmd_clear(pmd_t *pmdp)
212 xen_set_pmd(pmdp, __pmd(0));
215 unsigned long long xen_pte_val(pte_t pte)
217 unsigned long long ret = 0;
219 if (pte.pte_low) {
220 ret = ((unsigned long long)pte.pte_high << 32) | pte.pte_low;
221 ret = machine_to_phys(XMADDR(ret)).paddr | 1;
224 return ret;
227 unsigned long long xen_pmd_val(pmd_t pmd)
229 unsigned long long ret = pmd.pmd;
230 if (ret)
231 ret = machine_to_phys(XMADDR(ret)).paddr | 1;
232 return ret;
235 unsigned long long xen_pgd_val(pgd_t pgd)
237 unsigned long long ret = pgd.pgd;
238 if (ret)
239 ret = machine_to_phys(XMADDR(ret)).paddr | 1;
240 return ret;
243 pte_t xen_make_pte(unsigned long long pte)
245 if (pte & 1)
246 pte = phys_to_machine(XPADDR(pte)).maddr;
248 return (pte_t){ pte, pte >> 32 };
251 pmd_t xen_make_pmd(unsigned long long pmd)
253 if (pmd & 1)
254 pmd = phys_to_machine(XPADDR(pmd)).maddr;
256 return (pmd_t){ pmd };
259 pgd_t xen_make_pgd(unsigned long long pgd)
261 if (pgd & _PAGE_PRESENT)
262 pgd = phys_to_machine(XPADDR(pgd)).maddr;
264 return (pgd_t){ pgd };
266 #else /* !PAE */
267 void xen_set_pte(pte_t *ptep, pte_t pte)
269 *ptep = pte;
272 unsigned long xen_pte_val(pte_t pte)
274 unsigned long ret = pte.pte_low;
276 if (ret & _PAGE_PRESENT)
277 ret = machine_to_phys(XMADDR(ret)).paddr;
279 return ret;
282 unsigned long xen_pgd_val(pgd_t pgd)
284 unsigned long ret = pgd.pgd;
285 if (ret)
286 ret = machine_to_phys(XMADDR(ret)).paddr | 1;
287 return ret;
290 pte_t xen_make_pte(unsigned long pte)
292 if (pte & _PAGE_PRESENT)
293 pte = phys_to_machine(XPADDR(pte)).maddr;
295 return (pte_t){ pte };
298 pgd_t xen_make_pgd(unsigned long pgd)
300 if (pgd & _PAGE_PRESENT)
301 pgd = phys_to_machine(XPADDR(pgd)).maddr;
303 return (pgd_t){ pgd };
305 #endif /* CONFIG_X86_PAE */
310 (Yet another) pagetable walker. This one is intended for pinning a
311 pagetable. This means that it walks a pagetable and calls the
312 callback function on each page it finds making up the page table,
313 at every level. It walks the entire pagetable, but it only bothers
314 pinning pte pages which are below pte_limit. In the normal case
315 this will be TASK_SIZE, but at boot we need to pin up to
316 FIXADDR_TOP. But the important bit is that we don't pin beyond
317 there, because then we start getting into Xen's ptes.
319 static int pgd_walk(pgd_t *pgd_base, int (*func)(struct page *, unsigned),
320 unsigned long limit)
322 pgd_t *pgd = pgd_base;
323 int flush = 0;
324 unsigned long addr = 0;
325 unsigned long pgd_next;
327 BUG_ON(limit > FIXADDR_TOP);
329 if (xen_feature(XENFEAT_auto_translated_physmap))
330 return 0;
332 for (; addr != FIXADDR_TOP; pgd++, addr = pgd_next) {
333 pud_t *pud;
334 unsigned long pud_limit, pud_next;
336 pgd_next = pud_limit = pgd_addr_end(addr, FIXADDR_TOP);
338 if (!pgd_val(*pgd))
339 continue;
341 pud = pud_offset(pgd, 0);
343 if (PTRS_PER_PUD > 1) /* not folded */
344 flush |= (*func)(virt_to_page(pud), 0);
346 for (; addr != pud_limit; pud++, addr = pud_next) {
347 pmd_t *pmd;
348 unsigned long pmd_limit;
350 pud_next = pud_addr_end(addr, pud_limit);
352 if (pud_next < limit)
353 pmd_limit = pud_next;
354 else
355 pmd_limit = limit;
357 if (pud_none(*pud))
358 continue;
360 pmd = pmd_offset(pud, 0);
362 if (PTRS_PER_PMD > 1) /* not folded */
363 flush |= (*func)(virt_to_page(pmd), 0);
365 for (; addr != pmd_limit; pmd++) {
366 addr += (PAGE_SIZE * PTRS_PER_PTE);
367 if ((pmd_limit-1) < (addr-1)) {
368 addr = pmd_limit;
369 break;
372 if (pmd_none(*pmd))
373 continue;
375 flush |= (*func)(pmd_page(*pmd), 0);
380 flush |= (*func)(virt_to_page(pgd_base), UVMF_TLB_FLUSH);
382 return flush;
385 static int pin_page(struct page *page, unsigned flags)
387 unsigned pgfl = test_and_set_bit(PG_pinned, &page->flags);
388 int flush;
390 if (pgfl)
391 flush = 0; /* already pinned */
392 else if (PageHighMem(page))
393 /* kmaps need flushing if we found an unpinned
394 highpage */
395 flush = 1;
396 else {
397 void *pt = lowmem_page_address(page);
398 unsigned long pfn = page_to_pfn(page);
399 struct multicall_space mcs = __xen_mc_entry(0);
401 flush = 0;
403 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
404 pfn_pte(pfn, PAGE_KERNEL_RO),
405 flags);
408 return flush;
411 /* This is called just after a mm has been created, but it has not
412 been used yet. We need to make sure that its pagetable is all
413 read-only, and can be pinned. */
414 void xen_pgd_pin(pgd_t *pgd)
416 struct multicall_space mcs;
417 struct mmuext_op *op;
419 xen_mc_batch();
421 if (pgd_walk(pgd, pin_page, TASK_SIZE)) {
422 /* re-enable interrupts for kmap_flush_unused */
423 xen_mc_issue(0);
424 kmap_flush_unused();
425 xen_mc_batch();
428 mcs = __xen_mc_entry(sizeof(*op));
429 op = mcs.args;
431 #ifdef CONFIG_X86_PAE
432 op->cmd = MMUEXT_PIN_L3_TABLE;
433 #else
434 op->cmd = MMUEXT_PIN_L2_TABLE;
435 #endif
436 op->arg1.mfn = pfn_to_mfn(PFN_DOWN(__pa(pgd)));
437 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
439 xen_mc_issue(0);
442 /* The init_mm pagetable is really pinned as soon as its created, but
443 that's before we have page structures to store the bits. So do all
444 the book-keeping now. */
445 static __init int mark_pinned(struct page *page, unsigned flags)
447 SetPagePinned(page);
448 return 0;
451 void __init xen_mark_init_mm_pinned(void)
453 pgd_walk(init_mm.pgd, mark_pinned, FIXADDR_TOP);
456 static int unpin_page(struct page *page, unsigned flags)
458 unsigned pgfl = test_and_clear_bit(PG_pinned, &page->flags);
460 if (pgfl && !PageHighMem(page)) {
461 void *pt = lowmem_page_address(page);
462 unsigned long pfn = page_to_pfn(page);
463 struct multicall_space mcs = __xen_mc_entry(0);
465 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
466 pfn_pte(pfn, PAGE_KERNEL),
467 flags);
470 return 0; /* never need to flush on unpin */
473 /* Release a pagetables pages back as normal RW */
474 static void xen_pgd_unpin(pgd_t *pgd)
476 struct mmuext_op *op;
477 struct multicall_space mcs;
479 xen_mc_batch();
481 mcs = __xen_mc_entry(sizeof(*op));
483 op = mcs.args;
484 op->cmd = MMUEXT_UNPIN_TABLE;
485 op->arg1.mfn = pfn_to_mfn(PFN_DOWN(__pa(pgd)));
487 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
489 pgd_walk(pgd, unpin_page, TASK_SIZE);
491 xen_mc_issue(0);
494 void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
496 spin_lock(&next->page_table_lock);
497 xen_pgd_pin(next->pgd);
498 spin_unlock(&next->page_table_lock);
501 void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
503 spin_lock(&mm->page_table_lock);
504 xen_pgd_pin(mm->pgd);
505 spin_unlock(&mm->page_table_lock);
509 #ifdef CONFIG_SMP
510 /* Another cpu may still have their %cr3 pointing at the pagetable, so
511 we need to repoint it somewhere else before we can unpin it. */
512 static void drop_other_mm_ref(void *info)
514 struct mm_struct *mm = info;
516 if (__get_cpu_var(cpu_tlbstate).active_mm == mm)
517 leave_mm(smp_processor_id());
520 static void drop_mm_ref(struct mm_struct *mm)
522 if (current->active_mm == mm) {
523 if (current->mm == mm)
524 load_cr3(swapper_pg_dir);
525 else
526 leave_mm(smp_processor_id());
529 if (!cpus_empty(mm->cpu_vm_mask))
530 xen_smp_call_function_mask(mm->cpu_vm_mask, drop_other_mm_ref,
531 mm, 1);
533 #else
534 static void drop_mm_ref(struct mm_struct *mm)
536 if (current->active_mm == mm)
537 load_cr3(swapper_pg_dir);
539 #endif
542 * While a process runs, Xen pins its pagetables, which means that the
543 * hypervisor forces it to be read-only, and it controls all updates
544 * to it. This means that all pagetable updates have to go via the
545 * hypervisor, which is moderately expensive.
547 * Since we're pulling the pagetable down, we switch to use init_mm,
548 * unpin old process pagetable and mark it all read-write, which
549 * allows further operations on it to be simple memory accesses.
551 * The only subtle point is that another CPU may be still using the
552 * pagetable because of lazy tlb flushing. This means we need need to
553 * switch all CPUs off this pagetable before we can unpin it.
555 void xen_exit_mmap(struct mm_struct *mm)
557 get_cpu(); /* make sure we don't move around */
558 drop_mm_ref(mm);
559 put_cpu();
561 spin_lock(&mm->page_table_lock);
563 /* pgd may not be pinned in the error exit path of execve */
564 if (PagePinned(virt_to_page(mm->pgd)))
565 xen_pgd_unpin(mm->pgd);
566 spin_unlock(&mm->page_table_lock);