1 | /* |
---|
2 | * hal_boot.S - Kernel boot entry point |
---|
3 | * |
---|
4 | * Copyright (c) 2017 Maxime Villard |
---|
5 | * This code is inspired a lot from the NetBSD boot procedure, written by |
---|
6 | * Maxime Villard too. XXX copyright |
---|
7 | * |
---|
8 | * This file is part of ALMOS-MKH. |
---|
9 | * |
---|
10 | * ALMOS-MKH is free software; you can redistribute it and/or modify it |
---|
11 | * under the terms of the GNU General Public License as published by |
---|
12 | * the Free Software Foundation; version 2.0 of the License. |
---|
13 | * |
---|
14 | * ALMOS-MKH is distributed in the hope that it will be useful, but |
---|
15 | * WITHOUT ANY WARRANTY; without even the implied warranty of |
---|
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
---|
17 | * General Public License for more details. |
---|
18 | * |
---|
19 | * You should have received a copy of the GNU General Public License |
---|
20 | * along with ALMOS-MKH.; if not, write to the Free Software Foundation, |
---|
21 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA |
---|
22 | */ |
---|
23 | |
---|
24 | #define x86_ASM |
---|
25 | #include <hal_boot.h> |
---|
26 | #include <hal_register.h> |
---|
27 | #include <hal_segmentation.h> |
---|
28 | |
---|
29 | #define MULTIBOOT_HEADER_MAGIC 0x1BADB002 |
---|
30 | #define MULTIBOOT_HEADER_FLAGS 0x00000000 |
---|
31 | #define MULTIBOOT_INFO_MAGIC 0x2BADB002 |
---|
32 | #define CHECKSUM -(MULTIBOOT_HEADER_MAGIC + MULTIBOOT_HEADER_FLAGS) |
---|
33 | |
---|
34 | #if L2_SLOT_KERNBASE > 0 |
---|
35 | #define TABLE_L2_ENTRIES (2 * (NKL2_KIMG_ENTRIES + 1)) |
---|
36 | #else |
---|
37 | #define TABLE_L2_ENTRIES (NKL2_KIMG_ENTRIES + 1) |
---|
38 | #endif |
---|
39 | |
---|
40 | #if L3_SLOT_KERNBASE > 0 |
---|
41 | #define TABLE_L3_ENTRIES (2 * NKL3_KIMG_ENTRIES) |
---|
42 | #else |
---|
43 | #define TABLE_L3_ENTRIES NKL3_KIMG_ENTRIES |
---|
44 | #endif |
---|
45 | |
---|
46 | #define PROC0_PML4_OFF 0 |
---|
47 | #define PROC0_STK_OFF (PROC0_PML4_OFF + 1 * PAGE_SIZE) |
---|
48 | #define PROC0_PTP3_OFF (PROC0_STK_OFF + STKPAGES * PAGE_SIZE) |
---|
49 | #define PROC0_PTP2_OFF (PROC0_PTP3_OFF + NKL4_KIMG_ENTRIES * PAGE_SIZE) |
---|
50 | #define PROC0_PTP1_OFF (PROC0_PTP2_OFF + TABLE_L3_ENTRIES * PAGE_SIZE) |
---|
51 | #define TABLESIZE \ |
---|
52 | ((NKL4_KIMG_ENTRIES + TABLE_L3_ENTRIES + TABLE_L2_ENTRIES + 1 + STKPAGES) \ |
---|
53 | * PAGE_SIZE) |
---|
54 | |
---|
55 | /* |
---|
56 | * fillkpt - Fill in a kernel page table |
---|
57 | * eax = pte (page frame | control | status) |
---|
58 | * ebx = page table address |
---|
59 | * ecx = number of pages to map |
---|
60 | * |
---|
61 | * Each entry is 8 (PDE_SIZE) bytes long: we must set the 4 upper bytes to 0. |
---|
62 | */ |
---|
63 | #define fillkpt \ |
---|
64 | cmpl $0,%ecx ; /* zero-sized? */ \ |
---|
65 | je 2f ; \ |
---|
66 | 1: movl $0,(PDE_SIZE-4)(%ebx) ; /* upper 32 bits: 0 */ \ |
---|
67 | movl %eax,(%ebx) ; /* store phys addr */ \ |
---|
68 | addl $PDE_SIZE,%ebx ; /* next PTE/PDE */ \ |
---|
69 | addl $PAGE_SIZE,%eax ; /* next phys page */ \ |
---|
70 | loop 1b ; \ |
---|
71 | 2: ; |
---|
72 | |
---|
73 | /* |
---|
74 | * fillkpt_nox - Same as fillkpt, but sets the NX/XD bit. |
---|
75 | */ |
---|
76 | #define fillkpt_nox \ |
---|
77 | cmpl $0,%ecx ; /* zero-sized? */ \ |
---|
78 | je 2f ; \ |
---|
79 | 1: movl $PG_NX32,(PDE_SIZE-4)(%ebx); /* upper 32 bits: NX */ \ |
---|
80 | movl %eax,(%ebx) ; /* store phys addr */ \ |
---|
81 | addl $PDE_SIZE,%ebx ; /* next PTE/PDE */ \ |
---|
82 | addl $PAGE_SIZE,%eax ; /* next phys page */ \ |
---|
83 | loop 1b ; \ |
---|
84 | 2: ; |
---|
85 | |
---|
86 | /* |
---|
87 | * fillkpt_blank - Fill in a kernel page table with blank entries |
---|
88 | * ebx = page table address |
---|
89 | * ecx = number of pages to map |
---|
90 | */ |
---|
91 | #define fillkpt_blank \ |
---|
92 | cmpl $0,%ecx ; /* zero-sized? */ \ |
---|
93 | je 2f ; \ |
---|
94 | 1: movl $0,(PDE_SIZE-4)(%ebx) ; /* upper 32 bits: 0 */ \ |
---|
95 | movl $0,(%ebx) ; /* lower 32 bits: 0 */ \ |
---|
96 | addl $PDE_SIZE,%ebx ; /* next PTE/PDE */ \ |
---|
97 | loop 1b ; \ |
---|
98 | 2: ; |
---|
99 | |
---|
100 | /* |
---|
101 | * killkpt - Destroy a kernel page table (long mode) |
---|
102 | * rbx = page table address |
---|
103 | * rcx = number of pages to destroy |
---|
104 | */ |
---|
105 | #define killkpt \ |
---|
106 | 1: movq $0,(%rbx) ; \ |
---|
107 | addq $PDE_SIZE,%rbx ; \ |
---|
108 | loop 1b ; |
---|
109 | |
---|
110 | /* 32bit version of PG_NX */ |
---|
111 | #define PG_NX32 0x80000000 |
---|
112 | |
---|
113 | #define RELOC(x) ((x) - KERNBASE) |
---|
114 | |
---|
115 | .globl start_x86_64 |
---|
116 | .globl L4paddr |
---|
117 | .globl iom_base |
---|
118 | |
---|
119 | /* |
---|
120 | * The multiboot header |
---|
121 | */ |
---|
122 | .section .boot,"ax",@progbits |
---|
123 | multiboot_header: |
---|
124 | .align 4 |
---|
125 | .long MULTIBOOT_HEADER_MAGIC |
---|
126 | .long MULTIBOOT_HEADER_FLAGS |
---|
127 | .long CHECKSUM |
---|
128 | |
---|
129 | /* |
---|
130 | * The variables used in the boot procedure. |
---|
131 | */ |
---|
132 | .section .data |
---|
133 | .align 4 |
---|
134 | |
---|
135 | .type L4paddr, @object |
---|
136 | L4paddr: .quad 0 /* paddr of L4 */ |
---|
137 | |
---|
138 | .type iom_base, @object |
---|
139 | iom_base: .quad 0 /* virt. addr. of ISA I/O MEM */ |
---|
140 | |
---|
141 | |
---|
142 | #define GDT64_LIMIT gdt64_end-gdt64_start-1 |
---|
143 | /* Temporary gdt64, with base address in low memory */ |
---|
144 | .type gdt64_lo, @object |
---|
145 | gdt64_lo: |
---|
146 | .word GDT64_LIMIT |
---|
147 | .quad RELOC(gdt64_start) |
---|
148 | .align 64 |
---|
149 | |
---|
150 | /* Temporary gdt64, with base address in high memory */ |
---|
151 | .type gdt64_hi, @object |
---|
152 | gdt64_hi: |
---|
153 | .word GDT64_LIMIT |
---|
154 | .quad gdt64_start |
---|
155 | .align 64 |
---|
156 | #undef GDT64_LIMIT |
---|
157 | |
---|
158 | .type gdt64_start, @object |
---|
159 | gdt64_start: |
---|
160 | .quad 0x0000000000000000 /* always empty */ |
---|
161 | .quad 0x00af9a000000ffff /* kernel CS */ |
---|
162 | .quad 0x00cf92000000ffff /* kernel DS */ |
---|
163 | gdt64_end: |
---|
164 | |
---|
165 | .type farjmp64, @object |
---|
166 | farjmp64: |
---|
167 | .long RELOC(longmode) |
---|
168 | .word GDT_FIXED_SEL(GDT_KCODE_SEL, SEL_KPL) |
---|
169 | .align 64 |
---|
170 | |
---|
171 | /* Space for the temporary stack */ |
---|
172 | .size tmpstk, tmpstk - . |
---|
173 | .space 512 |
---|
174 | tmpstk: |
---|
175 | |
---|
176 | .text |
---|
177 | |
---|
178 | start_x86_64: |
---|
179 | .code32 |
---|
180 | |
---|
181 | /* Warm boot */ |
---|
182 | movw $0x1234,0x472 |
---|
183 | |
---|
184 | /* Make sure it is a multiboot-compliant bootloader. */ |
---|
185 | // cmpl $MULTIBOOT_INFO_MAGIC,%eax |
---|
186 | // jne boot_panic |
---|
187 | |
---|
188 | movl $RELOC(tmpstk),%esp |
---|
189 | |
---|
190 | /* Reset the PSL. */ |
---|
191 | pushl $PSL_MBO |
---|
192 | popfl |
---|
193 | |
---|
194 | /* |
---|
195 | * There are four levels of pages in amd64: PML4 -> PDP -> PD -> PT. They will |
---|
196 | * be referred to as: L4 -> L3 -> L2 -> L1. |
---|
197 | * |
---|
198 | * Virtual address space of the kernel: |
---|
199 | * +---------------+------------+-----------------------------------+--- |
---|
200 | * | TEXT + RODATA | DATA + BSS | L4 -> PROC0 STK -> L3 -> L2 -> L1 | |
---|
201 | * +---------------+------------+-----------------------------------+--- |
---|
202 | * (1) |
---|
203 | * |
---|
204 | * ---+-------------+ |
---|
205 | * | ISA I/O MEM | |
---|
206 | * ---+-------------+ |
---|
207 | * (2) |
---|
208 | * |
---|
209 | * PROC0 STK is obviously not linked as a page level. It just happens to be |
---|
210 | * caught between L4 and L3. |
---|
211 | * |
---|
212 | * (PROC0 STK + L4 + L3 + L2 + L1) is later referred to as BOOTSTRAP TABLES. |
---|
213 | * |
---|
214 | * ISA I/O MEM has no physical page allocated here, just virtual addresses. |
---|
215 | * |
---|
216 | * Important note: the kernel segments are properly 4k-aligned |
---|
217 | * (see kern.ldscript), so there's no need to enforce alignment. |
---|
218 | */ |
---|
219 | |
---|
220 | /* Find end of kernel image; brings us on (1). */ |
---|
221 | movl $RELOC(__kernel_end),%edi |
---|
222 | |
---|
223 | /* Align up for BOOTSTRAP TABLES. */ |
---|
224 | movl %edi,%esi |
---|
225 | addl $PGOFSET,%esi |
---|
226 | andl $~PGOFSET,%esi |
---|
227 | |
---|
228 | /* We are on the BOOTSTRAP TABLES. Save L4's physical address. */ |
---|
229 | movl $RELOC(L4paddr),%ebp |
---|
230 | movl %esi,(%ebp) |
---|
231 | movl $0,4(%ebp) |
---|
232 | |
---|
233 | /* Now, zero out the BOOTSTRAP TABLES (before filling them in). */ |
---|
234 | movl %esi,%edi |
---|
235 | xorl %eax,%eax |
---|
236 | cld |
---|
237 | movl $TABLESIZE,%ecx |
---|
238 | shrl $2,%ecx |
---|
239 | rep |
---|
240 | stosl /* copy eax -> edi */ |
---|
241 | |
---|
242 | /* |
---|
243 | * Build the page tables and levels. We go from L1 to L4, and link the levels |
---|
244 | * together. Note: RELOC computes &addr - KERNBASE in 32 bits; the value can't |
---|
245 | * be > 4G, or we can't deal with it anyway, since we are in 32bit mode. |
---|
246 | */ |
---|
247 | /* |
---|
248 | * Build L1. |
---|
249 | */ |
---|
250 | leal (PROC0_PTP1_OFF)(%esi),%ebx |
---|
251 | |
---|
252 | /* Skip the area below the kernel text. */ |
---|
253 | movl $(KERNTEXTOFF_LO - KERNBASE_LO),%ecx |
---|
254 | shrl $PGSHIFT,%ecx |
---|
255 | fillkpt_blank |
---|
256 | |
---|
257 | /* Map the kernel code RX. */ |
---|
258 | movl $(KERNTEXTOFF_LO - KERNBASE_LO),%eax /* start of TEXT */ |
---|
259 | movl $RELOC(__kernel_data_start),%ecx |
---|
260 | subl %eax,%ecx |
---|
261 | shrl $PGSHIFT,%ecx |
---|
262 | orl $(PG_V|PG_KR|PG_G),%eax |
---|
263 | fillkpt |
---|
264 | |
---|
265 | /* Map the kernel data RW. */ |
---|
266 | movl $RELOC(__kernel_data_start),%eax |
---|
267 | movl $RELOC(__kernel_end),%ecx |
---|
268 | subl %eax,%ecx |
---|
269 | shrl $PGSHIFT,%ecx |
---|
270 | orl $(PG_V|PG_KW|PG_G),%eax |
---|
271 | fillkpt_nox |
---|
272 | |
---|
273 | /* Map the BOOTSTRAP TABLES RW. */ |
---|
274 | movl $RELOC(__kernel_end),%eax /* start of BOOTSTRAP TABLES */ |
---|
275 | movl $TABLESIZE,%ecx /* length of BOOTSTRAP TABLES */ |
---|
276 | shrl $PGSHIFT,%ecx |
---|
277 | orl $(PG_V|PG_KW|PG_G),%eax |
---|
278 | fillkpt_nox |
---|
279 | |
---|
280 | /* We are on (2). Map ISA I/O MEM RW. */ |
---|
281 | movl $IOM_BEGIN,%eax |
---|
282 | movl $IOM_SIZE,%ecx /* size of ISA I/O MEM */ |
---|
283 | shrl $PGSHIFT,%ecx |
---|
284 | orl $(PG_V|PG_KW|PG_G),%eax |
---|
285 | fillkpt_nox |
---|
286 | |
---|
287 | /* |
---|
288 | * Build L2. Linked to L1. |
---|
289 | */ |
---|
290 | leal (PROC0_PTP2_OFF)(%esi),%ebx |
---|
291 | leal (PROC0_PTP1_OFF)(%esi),%eax |
---|
292 | orl $(PG_V|PG_KW),%eax |
---|
293 | movl $(NKL2_KIMG_ENTRIES+1),%ecx |
---|
294 | fillkpt |
---|
295 | |
---|
296 | #if L2_SLOT_KERNBASE > 0 |
---|
297 | /* If needed, set up level 2 entries for actual kernel mapping */ |
---|
298 | leal (PROC0_PTP2_OFF + L2_SLOT_KERNBASE * PDE_SIZE)(%esi),%ebx |
---|
299 | leal (PROC0_PTP1_OFF)(%esi),%eax |
---|
300 | orl $(PG_V|PG_KW),%eax |
---|
301 | movl $(NKL2_KIMG_ENTRIES+1),%ecx |
---|
302 | fillkpt |
---|
303 | #endif |
---|
304 | |
---|
305 | /* |
---|
306 | * Build L3. Linked to L2. |
---|
307 | */ |
---|
308 | leal (PROC0_PTP3_OFF)(%esi),%ebx |
---|
309 | leal (PROC0_PTP2_OFF)(%esi),%eax |
---|
310 | orl $(PG_V|PG_KW),%eax |
---|
311 | movl $NKL3_KIMG_ENTRIES,%ecx |
---|
312 | fillkpt |
---|
313 | |
---|
314 | #if L3_SLOT_KERNBASE > 0 |
---|
315 | /* If needed, set up level 3 entries for actual kernel mapping */ |
---|
316 | leal (PROC0_PTP3_OFF + L3_SLOT_KERNBASE * PDE_SIZE)(%esi),%ebx |
---|
317 | leal (PROC0_PTP2_OFF)(%esi),%eax |
---|
318 | orl $(PG_V|PG_KW),%eax |
---|
319 | movl $NKL3_KIMG_ENTRIES,%ecx |
---|
320 | fillkpt |
---|
321 | #endif |
---|
322 | |
---|
323 | /* |
---|
324 | * Build L4 for identity mapping. Linked to L3. |
---|
325 | */ |
---|
326 | leal (PROC0_PML4_OFF)(%esi),%ebx |
---|
327 | leal (PROC0_PTP3_OFF)(%esi),%eax |
---|
328 | orl $(PG_V|PG_KW),%eax |
---|
329 | movl $NKL4_KIMG_ENTRIES,%ecx |
---|
330 | fillkpt |
---|
331 | |
---|
332 | /* Set up L4 entries for actual kernel mapping */ |
---|
333 | leal (PROC0_PML4_OFF + L4_SLOT_KERNBASE * PDE_SIZE)(%esi),%ebx |
---|
334 | leal (PROC0_PTP3_OFF)(%esi),%eax |
---|
335 | orl $(PG_V|PG_KW),%eax |
---|
336 | movl $NKL4_KIMG_ENTRIES,%ecx |
---|
337 | fillkpt |
---|
338 | |
---|
339 | /* Install recursive top level PDE (one entry) */ |
---|
340 | leal (PROC0_PML4_OFF + PDIR_SLOT_PTE * PDE_SIZE)(%esi),%ebx |
---|
341 | leal (PROC0_PML4_OFF)(%esi),%eax |
---|
342 | orl $(PG_V|PG_KW),%eax |
---|
343 | movl $1,%ecx |
---|
344 | fillkpt_nox |
---|
345 | |
---|
346 | /* |
---|
347 | * Startup checklist: |
---|
348 | * 1. Enable PAE (and SSE while here). |
---|
349 | */ |
---|
350 | movl %cr4,%eax |
---|
351 | orl $(CR4_PAE|CR4_OSFXSR|CR4_OSXMMEXCPT|CR4_PGE),%eax |
---|
352 | movl %eax,%cr4 |
---|
353 | |
---|
354 | /* |
---|
355 | * 2. Set Long Mode Enable in EFER. Also enable the syscall extensions, |
---|
356 | * and NOX. |
---|
357 | */ |
---|
358 | movl $MSR_EFER,%ecx |
---|
359 | rdmsr |
---|
360 | xorl %eax,%eax /* XXX */ |
---|
361 | orl $(EFER_LME|EFER_SCE|EFER_NXE),%eax |
---|
362 | wrmsr |
---|
363 | |
---|
364 | /* |
---|
365 | * 3. Load %cr3 with pointer to PML4. |
---|
366 | */ |
---|
367 | movl %esi,%eax |
---|
368 | movl %eax,%cr3 |
---|
369 | |
---|
370 | /* |
---|
371 | * 4. Enable paging and the rest of it. |
---|
372 | */ |
---|
373 | movl %cr0,%eax |
---|
374 | orl $(CR0_PE|CR0_PG|CR0_NE|CR0_TS|CR0_MP|CR0_WP|CR0_AM),%eax |
---|
375 | movl %eax,%cr0 |
---|
376 | jmp compat |
---|
377 | compat: |
---|
378 | |
---|
379 | /* |
---|
380 | * 5. Not quite done yet, we're now in a compatibility segment, in |
---|
381 | * legacy mode. We must jump to a long mode segment. Need to set up |
---|
382 | * a temporary GDT with a long mode segment in it to do that. |
---|
383 | */ |
---|
384 | movl $RELOC(gdt64_lo),%eax |
---|
385 | lgdt (%eax) |
---|
386 | movl $RELOC(farjmp64),%eax |
---|
387 | ljmp *(%eax) |
---|
388 | |
---|
389 | .code64 |
---|
390 | longmode: |
---|
391 | /* |
---|
392 | * 6. Finally, we're in long mode. However, we're still in the identity |
---|
393 | * mapped area (could not jump out of that earlier because it would |
---|
394 | * have been a > 32bit jump). We can do that now, so here we go. |
---|
395 | */ |
---|
396 | movabsq $longmode_hi,%rax |
---|
397 | jmp *%rax |
---|
398 | |
---|
399 | longmode_hi: |
---|
400 | /* |
---|
401 | * We left the identity mapped area. Base address of |
---|
402 | * the temporary gdt64 should now be in high memory. |
---|
403 | */ |
---|
404 | movq $RELOC(gdt64_hi),%rax |
---|
405 | lgdt (%rax) |
---|
406 | |
---|
407 | /* |
---|
408 | * We have arrived. There's no need anymore for the identity mapping in |
---|
409 | * low memory, remove it. |
---|
410 | */ |
---|
411 | movq $KERNBASE,%r8 |
---|
412 | |
---|
413 | #if L2_SLOT_KERNBASE > 0 |
---|
414 | movq $(NKL2_KIMG_ENTRIES+1),%rcx |
---|
415 | leaq (PROC0_PTP2_OFF)(%rsi),%rbx /* old, phys address */ |
---|
416 | addq %r8,%rbx /* new, virt address */ |
---|
417 | killkpt |
---|
418 | #endif |
---|
419 | |
---|
420 | #if L3_SLOT_KERNBASE > 0 |
---|
421 | movq $NKL3_KIMG_ENTRIES,%rcx |
---|
422 | leaq (PROC0_PTP3_OFF)(%rsi),%rbx /* old, phys address */ |
---|
423 | addq %r8,%rbx /* new, virt address */ |
---|
424 | killkpt |
---|
425 | #endif |
---|
426 | |
---|
427 | movq $NKL4_KIMG_ENTRIES,%rcx |
---|
428 | leaq (PROC0_PML4_OFF)(%rsi),%rbx /* old, phys address of PML4 */ |
---|
429 | addq %r8,%rbx /* new, virt address of PML4 */ |
---|
430 | killkpt |
---|
431 | |
---|
432 | /* Relocate atdevbase. */ |
---|
433 | movq $(TABLESIZE+KERNBASE),%rdx |
---|
434 | addq %rsi,%rdx |
---|
435 | movq %rdx,iom_base(%rip) |
---|
436 | |
---|
437 | /* Set up bootstrap stack. */ |
---|
438 | leaq (PROC0_STK_OFF)(%rsi),%rax |
---|
439 | addq %r8,%rax |
---|
440 | leaq (STKSIZE)(%rax),%rsp |
---|
441 | xorq %rbp,%rbp /* mark end of frames */ |
---|
442 | |
---|
443 | xorw %ax,%ax |
---|
444 | movw %ax,%gs |
---|
445 | movw %ax,%fs |
---|
446 | |
---|
447 | /* The first physical page available. */ |
---|
448 | leaq (TABLESIZE)(%rsi),%rdi |
---|
449 | |
---|
450 | call init_x86_64 |
---|
451 | |
---|