1#include "param.h"
2#include "types.h"
3#include "defs.h"
4#include "x86.h"
5#include "memlayout.h"
6#include "mmu.h"
7#include "proc.h"
8#include "elf.h"
9
10extern char data[]; // defined by kernel.ld
11pde_t *kpgdir; // for use in scheduler()
12
13// Set up CPU's kernel segment descriptors.
14// Run once on entry on each CPU.
15void
16seginit(void)
17{
18 struct cpu *c;
19
20 // Map "logical" addresses to virtual addresses using identity map.
21 // Cannot share a CODE descriptor for both kernel and user
22 // because it would have to have DPL_USR, but the CPU forbids
23 // an interrupt from CPL=0 to DPL=3.
24 c = &cpus[cpuid()];
25 c->gdt[SEG_KCODE] = SEG(STA_X|STA_R, 0, 0xffffffff, 0);
26 c->gdt[SEG_KDATA] = SEG(STA_W, 0, 0xffffffff, 0);
27 c->gdt[SEG_UCODE] = SEG(STA_X|STA_R, 0, 0xffffffff, DPL_USER);
28 c->gdt[SEG_UDATA] = SEG(STA_W, 0, 0xffffffff, DPL_USER);
29 lgdt(c->gdt, sizeof(c->gdt));
30}
31
32// Return the address of the PTE in page table pgdir
33// that corresponds to virtual address va. If alloc!=0,
34// create any required page table pages.
35static pte_t *
36walkpgdir(pde_t *pgdir, const void *va, int alloc)
37{
38 pde_t *pde;
39 pte_t *pgtab;
40
41 pde = &pgdir[PDX(va)];
42 if(*pde & PTE_P){
43 pgtab = (pte_t*)P2V(PTE_ADDR(*pde));
44 } else {
45 if(!alloc || (pgtab = (pte_t*)kalloc()) == 0)
46 return 0;
47 // Make sure all those PTE_P bits are zero.
48 memset(pgtab, 0, PGSIZE);
49 // The permissions here are overly generous, but they can
50 // be further restricted by the permissions in the page table
51 // entries, if necessary.
52 *pde = V2P(pgtab) | PTE_P | PTE_W | PTE_U;
53 }
54 return &pgtab[PTX(va)];
55}
56
57// Create PTEs for virtual addresses starting at va that refer to
58// physical addresses starting at pa. va and size might not
59// be page-aligned.
60static int
61mappages(pde_t *pgdir, void *va, uint size, uint pa, int perm)
62{
63 char *a, *last;
64 pte_t *pte;
65
66 a = (char*)PGROUNDDOWN((uint)va);
67 last = (char*)PGROUNDDOWN(((uint)va) + size - 1);
68 for(;;){
69 if((pte = walkpgdir(pgdir, a, 1)) == 0)
70 return -1;
71 if(*pte & PTE_P)
72 panic("remap");
73 *pte = pa | perm | PTE_P;
74 if(a == last)
75 break;
76 a += PGSIZE;
77 pa += PGSIZE;
78 }
79 return 0;
80}
81
82// There is one page table per process, plus one that's used when
83// a CPU is not running any process (kpgdir). The kernel uses the
84// current process's page table during system calls and interrupts;
85// page protection bits prevent user code from using the kernel's
86// mappings.
87//
88// setupkvm() and exec() set up every page table like this:
89//
90// 0..KERNBASE: user memory (text+data+stack+heap), mapped to
91// phys memory allocated by the kernel
92// KERNBASE..KERNBASE+EXTMEM: mapped to 0..EXTMEM (for I/O space)
93// KERNBASE+EXTMEM..data: mapped to EXTMEM..V2P(data)
94// for the kernel's instructions and r/o data
95// data..KERNBASE+PHYSTOP: mapped to V2P(data)..PHYSTOP,
96// rw data + free physical memory
97// 0xfe000000..0: mapped direct (devices such as ioapic)
98//
99// The kernel allocates physical memory for its heap and for user memory
100// between V2P(end) and the end of physical memory (PHYSTOP)
101// (directly addressable from end..P2V(PHYSTOP)).
102
103// This table defines the kernel's mappings, which are present in
104// every process's page table.
105static struct kmap {
106 void *virt;
107 uint phys_start;
108 uint phys_end;
109 int perm;
110} kmap[] = {
111 { (void*)KERNBASE, 0, EXTMEM, PTE_W}, // I/O space
112 { (void*)KERNLINK, V2P(KERNLINK), V2P(data), 0}, // kern text+rodata
113 { (void*)data, V2P(data), PHYSTOP, PTE_W}, // kern data+memory
114 { (void*)DEVSPACE, DEVSPACE, 0, PTE_W}, // more devices
115};
116
117// Set up kernel part of a page table.
118pde_t*
119setupkvm(void)
120{
121 pde_t *pgdir;
122 struct kmap *k;
123
124 if((pgdir = (pde_t*)kalloc()) == 0)
125 return 0;
126 memset(pgdir, 0, PGSIZE);
127 if (P2V(PHYSTOP) > (void*)DEVSPACE)
128 panic("PHYSTOP too high");
129 for(k = kmap; k < &kmap[NELEM(kmap)]; k++)
130 if(mappages(pgdir, k->virt, k->phys_end - k->phys_start,
131 (uint)k->phys_start, k->perm) < 0) {
132 freevm(pgdir);
133 return 0;
134 }
135 return pgdir;
136}
137
138// Allocate one page table for the machine for the kernel address
139// space for scheduler processes.
140void
141kvmalloc(void)
142{
143 kpgdir = setupkvm();
144 switchkvm();
145}
146
147// Switch h/w page table register to the kernel-only page table,
148// for when no process is running.
149void
150switchkvm(void)
151{
152 lcr3(V2P(kpgdir)); // switch to the kernel page table
153}
154
155// Switch TSS and h/w page table to correspond to process p.
156void
157switchuvm(struct proc *p)
158{
159 if(p == 0)
160 panic("switchuvm: no process");
161 if(p->kstack == 0)
162 panic("switchuvm: no kstack");
163 if(p->pgdir == 0)
164 panic("switchuvm: no pgdir");
165
166 pushcli();
167 mycpu()->gdt[SEG_TSS] = SEG16(STS_T32A, &mycpu()->ts,
168 sizeof(mycpu()->ts)-1, 0);
169 mycpu()->gdt[SEG_TSS].s = 0;
170 mycpu()->ts.ss0 = SEG_KDATA << 3;
171 mycpu()->ts.esp0 = (uint)p->kstack + KSTACKSIZE;
172 // setting IOPL=0 in eflags *and* iomb beyond the tss segment limit
173 // forbids I/O instructions (e.g., inb and outb) from user space
174 mycpu()->ts.iomb = (ushort) 0xFFFF;
175 ltr(SEG_TSS << 3);
176 lcr3(V2P(p->pgdir)); // switch to process's address space
177 popcli();
178}
179
180// Load the initcode into address 0 of pgdir.
181// sz must be less than a page.
182void
183inituvm(pde_t *pgdir, char *init, uint sz)
184{
185 char *mem;
186
187 if(sz >= PGSIZE)
188 panic("inituvm: more than a page");
189 mem = kalloc();
190 memset(mem, 0, PGSIZE);
191 mappages(pgdir, 0, PGSIZE, V2P(mem), PTE_W|PTE_U);
192 memmove(mem, init, sz);
193}
194
195// Load a program segment into pgdir. addr must be page-aligned
196// and the pages from addr to addr+sz must already be mapped.
197int
198loaduvm(pde_t *pgdir, char *addr, struct inode *ip, uint offset, uint sz)
199{
200 uint i, pa, n;
201 pte_t *pte;
202
203 if((uint) addr % PGSIZE != 0)
204 panic("loaduvm: addr must be page aligned");
205 for(i = 0; i < sz; i += PGSIZE){
206 if((pte = walkpgdir(pgdir, addr+i, 0)) == 0)
207 panic("loaduvm: address should exist");
208 pa = PTE_ADDR(*pte);
209 if(sz - i < PGSIZE)
210 n = sz - i;
211 else
212 n = PGSIZE;
213 if(readi(ip, P2V(pa), offset+i, n) != n)
214 return -1;
215 }
216 return 0;
217}
218
219// Allocate page tables and physical memory to grow process from oldsz to
220// newsz, which need not be page aligned. Returns new size or 0 on error.
221int
222allocuvm(pde_t *pgdir, uint oldsz, uint newsz)
223{
224 char *mem;
225 uint a;
226
227 if(newsz >= KERNBASE)
228 return 0;
229 if(newsz < oldsz)
230 return oldsz;
231
232 a = PGROUNDUP(oldsz);
233 for(; a < newsz; a += PGSIZE){
234 mem = kalloc();
235 if(mem == 0){
236 cprintf("allocuvm out of memory\n");
237 deallocuvm(pgdir, newsz, oldsz);
238 return 0;
239 }
240 memset(mem, 0, PGSIZE);
241 if(mappages(pgdir, (char*)a, PGSIZE, V2P(mem), PTE_W|PTE_U) < 0){
242 cprintf("allocuvm out of memory (2)\n");
243 deallocuvm(pgdir, newsz, oldsz);
244 kfree(mem);
245 return 0;
246 }
247 }
248 return newsz;
249}
250
251// Deallocate user pages to bring the process size from oldsz to
252// newsz. oldsz and newsz need not be page-aligned, nor does newsz
253// need to be less than oldsz. oldsz can be larger than the actual
254// process size. Returns the new process size.
255int
256deallocuvm(pde_t *pgdir, uint oldsz, uint newsz)
257{
258 pte_t *pte;
259 uint a, pa;
260
261 if(newsz >= oldsz)
262 return oldsz;
263
264 a = PGROUNDUP(newsz);
265 for(; a < oldsz; a += PGSIZE){
266 pte = walkpgdir(pgdir, (char*)a, 0);
267 if(!pte)
268 a = PGADDR(PDX(a) + 1, 0, 0) - PGSIZE;
269 else if((*pte & PTE_P) != 0){
270 pa = PTE_ADDR(*pte);
271 if(pa == 0)
272 panic("kfree");
273 char *v = P2V(pa);
274 kfree(v);
275 *pte = 0;
276 }
277 }
278 return newsz;
279}
280
281// Free a page table and all the physical memory pages
282// in the user part.
283void
284freevm(pde_t *pgdir)
285{
286 uint i;
287
288 if(pgdir == 0)
289 panic("freevm: no pgdir");
290 deallocuvm(pgdir, KERNBASE, 0);
291 for(i = 0; i < NPDENTRIES; i++){
292 if(pgdir[i] & PTE_P){
293 char * v = P2V(PTE_ADDR(pgdir[i]));
294 kfree(v);
295 }
296 }
297 kfree((char*)pgdir);
298}
299
300// Clear PTE_U on a page. Used to create an inaccessible
301// page beneath the user stack.
302void
303clearpteu(pde_t *pgdir, char *uva)
304{
305 pte_t *pte;
306
307 pte = walkpgdir(pgdir, uva, 0);
308 if(pte == 0)
309 panic("clearpteu");
310 *pte &= ~PTE_U;
311}
312
313// Given a parent process's page table, create a copy
314// of it for a child.
315pde_t*
316copyuvm(pde_t *pgdir, uint sz)
317{
318 pde_t *d;
319 pte_t *pte;
320 uint pa, i, flags;
321 char *mem;
322
323 if((d = setupkvm()) == 0)
324 return 0;
325 for(i = 0; i < sz; i += PGSIZE){
326 if((pte = walkpgdir(pgdir, (void *) i, 0)) == 0)
327 panic("copyuvm: pte should exist");
328 if(!(*pte & PTE_P))
329 panic("copyuvm: page not present");
330 pa = PTE_ADDR(*pte);
331 flags = PTE_FLAGS(*pte);
332 if((mem = kalloc()) == 0)
333 goto bad;
334 memmove(mem, (char*)P2V(pa), PGSIZE);
335 if(mappages(d, (void*)i, PGSIZE, V2P(mem), flags) < 0) {
336 kfree(mem);
337 goto bad;
338 }
339 }
340 return d;
341
342bad:
343 freevm(d);
344 return 0;
345}
346
347//PAGEBREAK!
348// Map user virtual address to kernel address.
349char*
350uva2ka(pde_t *pgdir, char *uva)
351{
352 pte_t *pte;
353
354 pte = walkpgdir(pgdir, uva, 0);
355 if((*pte & PTE_P) == 0)
356 return 0;
357 if((*pte & PTE_U) == 0)
358 return 0;
359 return (char*)P2V(PTE_ADDR(*pte));
360}
361
362// Copy len bytes from p to user address va in page table pgdir.
363// Most useful when pgdir is not the current page table.
364// uva2ka ensures this only works for PTE_U pages.
365int
366copyout(pde_t *pgdir, uint va, void *p, uint len)
367{
368 char *buf, *pa0;
369 uint n, va0;
370
371 buf = (char*)p;
372 while(len > 0){
373 va0 = (uint)PGROUNDDOWN(va);
374 pa0 = uva2ka(pgdir, (char*)va0);
375 if(pa0 == 0)
376 return -1;
377 n = PGSIZE - (va - va0);
378 if(n > len)
379 n = len;
380 memmove(pa0 + (va - va0), buf, n);
381 len -= n;
382 buf += n;
383 va = va0 + PGSIZE;
384 }
385 return 0;
386}
387
388//PAGEBREAK!
389// Blank page.
390//PAGEBREAK!
391// Blank page.
392//PAGEBREAK!
393// Blank page.
394
395