xref: /kvm-unit-tests/lib/x86/vm.c (revision 4363f1d9a646a5c7ea673bee8fc33ca6f2cddbd8)
17d36db35SAvi Kivity #include "vm.h"
27d36db35SAvi Kivity #include "libcflat.h"
3efd8e5aaSPaolo Bonzini #include "vmalloc.h"
45aca024eSPaolo Bonzini #include "alloc_page.h"
57d36db35SAvi Kivity 
6*4363f1d9SPaolo Bonzini pteval_t *install_pte(pgd_t *cr3,
77d36db35SAvi Kivity 		      int pte_level,
87d36db35SAvi Kivity 		      void *virt,
9*4363f1d9SPaolo Bonzini 		      pteval_t pte,
10*4363f1d9SPaolo Bonzini 		      pteval_t *pt_page)
117d36db35SAvi Kivity {
127d36db35SAvi Kivity     int level;
13*4363f1d9SPaolo Bonzini     pteval_t *pt = cr3;
147d36db35SAvi Kivity     unsigned offset;
157d36db35SAvi Kivity 
167d36db35SAvi Kivity     for (level = PAGE_LEVEL; level > pte_level; --level) {
17*4363f1d9SPaolo Bonzini 	offset = PGDIR_OFFSET((uintptr_t)virt, level);
18d10d16e1SAlexander Gordeev 	if (!(pt[offset] & PT_PRESENT_MASK)) {
19*4363f1d9SPaolo Bonzini 	    pteval_t *new_pt = pt_page;
207d36db35SAvi Kivity             if (!new_pt)
217d36db35SAvi Kivity                 new_pt = alloc_page();
227d36db35SAvi Kivity             else
237d36db35SAvi Kivity                 pt_page = 0;
247d36db35SAvi Kivity 	    memset(new_pt, 0, PAGE_SIZE);
25d10d16e1SAlexander Gordeev 	    pt[offset] = virt_to_phys(new_pt) | PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK;
267d36db35SAvi Kivity 	}
27d10d16e1SAlexander Gordeev 	pt = phys_to_virt(pt[offset] & PT_ADDR_MASK);
287d36db35SAvi Kivity     }
29*4363f1d9SPaolo Bonzini     offset = PGDIR_OFFSET((uintptr_t)virt, level);
307d36db35SAvi Kivity     pt[offset] = pte;
3104262816SPaolo Bonzini     return &pt[offset];
327d36db35SAvi Kivity }
337d36db35SAvi Kivity 
341df80b57SPeter Feiner /*
351df80b57SPeter Feiner  * Finds last PTE in the mapping of @virt that's at or above @lowest_level. The
361df80b57SPeter Feiner  * returned PTE isn't necessarily present, but its parent is.
371df80b57SPeter Feiner  */
38*4363f1d9SPaolo Bonzini struct pte_search find_pte_level(pgd_t *cr3, void *virt,
391df80b57SPeter Feiner 				 int lowest_level)
407d36db35SAvi Kivity {
41*4363f1d9SPaolo Bonzini 	pteval_t *pt = cr3, pte;
427d36db35SAvi Kivity 	unsigned offset;
43*4363f1d9SPaolo Bonzini 	unsigned shift;
441df80b57SPeter Feiner 	struct pte_search r;
457d36db35SAvi Kivity 
461df80b57SPeter Feiner 	assert(lowest_level >= 1 && lowest_level <= PAGE_LEVEL);
471df80b57SPeter Feiner 
481df80b57SPeter Feiner 	for (r.level = PAGE_LEVEL;; --r.level) {
491df80b57SPeter Feiner 		shift = (r.level - 1) * PGDIR_WIDTH + 12;
50*4363f1d9SPaolo Bonzini 		offset = ((uintptr_t)virt >> shift) & PGDIR_MASK;
511df80b57SPeter Feiner 		r.pte = &pt[offset];
521df80b57SPeter Feiner 		pte = *r.pte;
531df80b57SPeter Feiner 
54d10d16e1SAlexander Gordeev 		if (!(pte & PT_PRESENT_MASK))
551df80b57SPeter Feiner 			return r;
561df80b57SPeter Feiner 
571df80b57SPeter Feiner 		if ((r.level == 2 || r.level == 3) && (pte & PT_PAGE_SIZE_MASK))
581df80b57SPeter Feiner 			return r;
591df80b57SPeter Feiner 
601df80b57SPeter Feiner 		if (r.level == lowest_level)
611df80b57SPeter Feiner 			return r;
621df80b57SPeter Feiner 
631df80b57SPeter Feiner 		pt = phys_to_virt(pte & 0xffffffffff000ull);
647d36db35SAvi Kivity 	}
651df80b57SPeter Feiner }
661df80b57SPeter Feiner 
671df80b57SPeter Feiner /*
681df80b57SPeter Feiner  * Returns the leaf PTE in the mapping of @virt (i.e., 4K PTE or a present huge
691df80b57SPeter Feiner  * PTE). Returns NULL if no leaf PTE exists.
701df80b57SPeter Feiner  */
71*4363f1d9SPaolo Bonzini pteval_t *get_pte(pgd_t *cr3, void *virt)
721df80b57SPeter Feiner {
731df80b57SPeter Feiner 	struct pte_search search;
741df80b57SPeter Feiner 
751df80b57SPeter Feiner 	search = find_pte_level(cr3, virt, 1);
761df80b57SPeter Feiner 	return found_leaf_pte(search) ? search.pte : NULL;
771df80b57SPeter Feiner }
781df80b57SPeter Feiner 
791df80b57SPeter Feiner /*
801df80b57SPeter Feiner  * Returns the PTE in the mapping of @virt at the given level @pte_level.
811df80b57SPeter Feiner  * Returns NULL if the PT at @pte_level isn't present (i.e., the mapping at
821df80b57SPeter Feiner  * @pte_level - 1 isn't present).
831df80b57SPeter Feiner  */
84*4363f1d9SPaolo Bonzini pteval_t *get_pte_level(pgd_t *cr3, void *virt, int pte_level)
851df80b57SPeter Feiner {
861df80b57SPeter Feiner 	struct pte_search search;
871df80b57SPeter Feiner 
881df80b57SPeter Feiner 	search = find_pte_level(cr3, virt, pte_level);
891df80b57SPeter Feiner 	return search.level == pte_level ? search.pte : NULL;
907d36db35SAvi Kivity }
917d36db35SAvi Kivity 
92*4363f1d9SPaolo Bonzini pteval_t *install_large_page(pgd_t *cr3, phys_addr_t phys, void *virt)
937d36db35SAvi Kivity {
9404262816SPaolo Bonzini     return install_pte(cr3, 2, virt,
95d10d16e1SAlexander Gordeev 		       phys | PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK | PT_PAGE_SIZE_MASK, 0);
967d36db35SAvi Kivity }
977d36db35SAvi Kivity 
98*4363f1d9SPaolo Bonzini pteval_t *install_page(pgd_t *cr3, phys_addr_t phys, void *virt)
997d36db35SAvi Kivity {
100d10d16e1SAlexander Gordeev     return install_pte(cr3, 1, virt, phys | PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK, 0);
1017d36db35SAvi Kivity }
1027d36db35SAvi Kivity 
103*4363f1d9SPaolo Bonzini void install_pages(pgd_t *cr3, phys_addr_t phys, size_t len, void *virt)
1041df80b57SPeter Feiner {
105*4363f1d9SPaolo Bonzini 	phys_addr_t max = (u64)len + (u64)phys;
1061df80b57SPeter Feiner 	assert(phys % PAGE_SIZE == 0);
107*4363f1d9SPaolo Bonzini 	assert((uintptr_t) virt % PAGE_SIZE == 0);
1081df80b57SPeter Feiner 	assert(len % PAGE_SIZE == 0);
1091df80b57SPeter Feiner 
1101df80b57SPeter Feiner 	while (phys + PAGE_SIZE <= max) {
1111df80b57SPeter Feiner 		install_page(cr3, phys, virt);
1121df80b57SPeter Feiner 		phys += PAGE_SIZE;
1131df80b57SPeter Feiner 		virt = (char *) virt + PAGE_SIZE;
1141df80b57SPeter Feiner 	}
1151df80b57SPeter Feiner }
1161df80b57SPeter Feiner 
117*4363f1d9SPaolo Bonzini bool any_present_pages(pgd_t *cr3, void *virt, size_t len)
1181df80b57SPeter Feiner {
119*4363f1d9SPaolo Bonzini 	uintptr_t max = (uintptr_t) virt + len;
120*4363f1d9SPaolo Bonzini 	uintptr_t curr;
1211df80b57SPeter Feiner 
122*4363f1d9SPaolo Bonzini 	for (curr = (uintptr_t) virt; curr < max; curr += PAGE_SIZE) {
123*4363f1d9SPaolo Bonzini 		pteval_t *ptep = get_pte(cr3, (void *) curr);
1241df80b57SPeter Feiner 		if (ptep && (*ptep & PT_PRESENT_MASK))
1251df80b57SPeter Feiner 			return true;
1261df80b57SPeter Feiner 	}
1271df80b57SPeter Feiner 	return false;
1281df80b57SPeter Feiner }
1297d36db35SAvi Kivity 
130*4363f1d9SPaolo Bonzini static void setup_mmu_range(pgd_t *cr3, phys_addr_t start, size_t len)
13163254428SGleb Natapov {
13263254428SGleb Natapov 	u64 max = (u64)len + (u64)start;
13363254428SGleb Natapov 	u64 phys = start;
13463254428SGleb Natapov 
13563254428SGleb Natapov 	while (phys + LARGE_PAGE_SIZE <= max) {
13663254428SGleb Natapov 		install_large_page(cr3, phys, (void *)(ulong)phys);
13763254428SGleb Natapov 		phys += LARGE_PAGE_SIZE;
13863254428SGleb Natapov 	}
1391df80b57SPeter Feiner 	install_pages(cr3, phys, max - phys, (void *)(ulong)phys);
14063254428SGleb Natapov }
14163254428SGleb Natapov 
142937e2392SPaolo Bonzini void *setup_mmu(phys_addr_t end_of_memory)
1437d36db35SAvi Kivity {
144*4363f1d9SPaolo Bonzini     pgd_t *cr3 = alloc_page();
1457d36db35SAvi Kivity 
1467d36db35SAvi Kivity     memset(cr3, 0, PAGE_SIZE);
14763254428SGleb Natapov 
14863254428SGleb Natapov #ifdef __x86_64__
149937e2392SPaolo Bonzini     if (end_of_memory < (1ul << 32))
150937e2392SPaolo Bonzini         end_of_memory = (1ul << 32);  /* map mmio 1:1 */
15163254428SGleb Natapov 
152937e2392SPaolo Bonzini     setup_mmu_range(cr3, 0, end_of_memory);
15363254428SGleb Natapov #else
154937e2392SPaolo Bonzini     if (end_of_memory > (1ul << 31))
155937e2392SPaolo Bonzini 	    end_of_memory = (1ul << 31);
15663254428SGleb Natapov 
15763254428SGleb Natapov     /* 0 - 2G memory, 2G-3G valloc area, 3G-4G mmio */
158937e2392SPaolo Bonzini     setup_mmu_range(cr3, 0, end_of_memory);
15963254428SGleb Natapov     setup_mmu_range(cr3, 3ul << 30, (1ul << 30));
160efd8e5aaSPaolo Bonzini     init_alloc_vpage((void*)(3ul << 30));
16163254428SGleb Natapov #endif
16263254428SGleb Natapov 
1637d36db35SAvi Kivity     write_cr3(virt_to_phys(cr3));
1647d36db35SAvi Kivity #ifndef __x86_64__
1657d36db35SAvi Kivity     write_cr4(X86_CR4_PSE);
1667d36db35SAvi Kivity #endif
16797011120SGleb Natapov     write_cr0(X86_CR0_PG |X86_CR0_PE | X86_CR0_WP);
1687d36db35SAvi Kivity 
1697d36db35SAvi Kivity     printf("paging enabled\n");
170b006d7ebSAndrew Jones     printf("cr0 = %lx\n", read_cr0());
171b006d7ebSAndrew Jones     printf("cr3 = %lx\n", read_cr3());
172b006d7ebSAndrew Jones     printf("cr4 = %lx\n", read_cr4());
173937e2392SPaolo Bonzini     return cr3;
1747d36db35SAvi Kivity }
1757d36db35SAvi Kivity 
1767d36db35SAvi Kivity void *vmalloc(unsigned long size)
1777d36db35SAvi Kivity {
1787d36db35SAvi Kivity     void *mem, *p;
1797d36db35SAvi Kivity     unsigned pages;
1807d36db35SAvi Kivity 
1817d36db35SAvi Kivity     size += sizeof(unsigned long);
1827d36db35SAvi Kivity 
1837d36db35SAvi Kivity     size = (size + PAGE_SIZE - 1) & ~(PAGE_SIZE - 1);
1847d36db35SAvi Kivity     pages = size / PAGE_SIZE;
185efd8e5aaSPaolo Bonzini     mem = p = alloc_vpages(pages);
1867d36db35SAvi Kivity     while (pages--) {
1877d36db35SAvi Kivity 	install_page(phys_to_virt(read_cr3()), virt_to_phys(alloc_page()), p);
1887d36db35SAvi Kivity 	p += PAGE_SIZE;
1897d36db35SAvi Kivity     }
1907d36db35SAvi Kivity     *(unsigned long *)mem = size;
1917d36db35SAvi Kivity     mem += sizeof(unsigned long);
1927d36db35SAvi Kivity     return mem;
1937d36db35SAvi Kivity }
1947d36db35SAvi Kivity 
195334cd2bfSGleb Natapov uint64_t virt_to_phys_cr3(void *mem)
196334cd2bfSGleb Natapov {
197d10d16e1SAlexander Gordeev     return (*get_pte(phys_to_virt(read_cr3()), mem) & PT_ADDR_MASK) + ((ulong)mem & (PAGE_SIZE - 1));
198334cd2bfSGleb Natapov }
199334cd2bfSGleb Natapov 
2007d36db35SAvi Kivity void vfree(void *mem)
2017d36db35SAvi Kivity {
2027d36db35SAvi Kivity     unsigned long size = ((unsigned long *)mem)[-1];
2037d36db35SAvi Kivity 
2047d36db35SAvi Kivity     while (size) {
205d10d16e1SAlexander Gordeev 	free_page(phys_to_virt(*get_pte(phys_to_virt(read_cr3()), mem) & PT_ADDR_MASK));
2067d36db35SAvi Kivity 	mem += PAGE_SIZE;
2077d36db35SAvi Kivity 	size -= PAGE_SIZE;
2087d36db35SAvi Kivity     }
2097d36db35SAvi Kivity }
2107d36db35SAvi Kivity 
2117d36db35SAvi Kivity void *vmap(unsigned long long phys, unsigned long size)
2127d36db35SAvi Kivity {
2137d36db35SAvi Kivity     void *mem, *p;
2147d36db35SAvi Kivity     unsigned pages;
2157d36db35SAvi Kivity 
2167d36db35SAvi Kivity     size = (size + PAGE_SIZE - 1) & ~(PAGE_SIZE - 1);
2177d36db35SAvi Kivity     pages = size / PAGE_SIZE;
218efd8e5aaSPaolo Bonzini     mem = p = alloc_vpages(pages);
219efd8e5aaSPaolo Bonzini 
220efd8e5aaSPaolo Bonzini     phys &= ~(unsigned long long)(PAGE_SIZE - 1);
2217d36db35SAvi Kivity     while (pages--) {
2227d36db35SAvi Kivity 	install_page(phys_to_virt(read_cr3()), phys, p);
2237d36db35SAvi Kivity 	phys += PAGE_SIZE;
2247d36db35SAvi Kivity 	p += PAGE_SIZE;
2257d36db35SAvi Kivity     }
2267d36db35SAvi Kivity     return mem;
2277d36db35SAvi Kivity }
228