]>
Commit | Line | Data |
---|---|---|
83d290c5 | 1 | // SPDX-License-Identifier: GPL-2.0+ |
be059e88 SG |
2 | /* |
3 | * (C) Copyright 2008-2011 | |
4 | * Graeme Russ, <[email protected]> | |
5 | * | |
6 | * (C) Copyright 2002 | |
7 | * Daniel Engström, Omicron Ceti AB, <[email protected]> | |
8 | * | |
9 | * (C) Copyright 2002 | |
10 | * Sysgo Real-Time Solutions, GmbH <www.elinos.com> | |
11 | * Marius Groeger <[email protected]> | |
12 | * | |
13 | * (C) Copyright 2002 | |
14 | * Sysgo Real-Time Solutions, GmbH <www.elinos.com> | |
15 | * Alex Zuepke <[email protected]> | |
16 | * | |
17 | * Part of this file is adapted from coreboot | |
18 | * src/arch/x86/lib/cpu.c | |
be059e88 SG |
19 | */ |
20 | ||
21 | #include <common.h> | |
9edefc27 | 22 | #include <cpu_func.h> |
35a3f871 | 23 | #include <init.h> |
be059e88 | 24 | #include <malloc.h> |
caca13f6 | 25 | #include <spl.h> |
be059e88 SG |
26 | #include <asm/control_regs.h> |
27 | #include <asm/cpu.h> | |
28 | #include <asm/mp.h> | |
29 | #include <asm/msr.h> | |
30 | #include <asm/mtrr.h> | |
31 | #include <asm/processor-flags.h> | |
32 | ||
33 | DECLARE_GLOBAL_DATA_PTR; | |
34 | ||
35 | /* | |
36 | * Constructor for a conventional segment GDT (or LDT) entry | |
37 | * This is a macro so it can be used in initialisers | |
38 | */ | |
39 | #define GDT_ENTRY(flags, base, limit) \ | |
40 | ((((base) & 0xff000000ULL) << (56-24)) | \ | |
41 | (((flags) & 0x0000f0ffULL) << 40) | \ | |
42 | (((limit) & 0x000f0000ULL) << (48-16)) | \ | |
43 | (((base) & 0x00ffffffULL) << 16) | \ | |
44 | (((limit) & 0x0000ffffULL))) | |
45 | ||
46 | struct gdt_ptr { | |
47 | u16 len; | |
48 | u32 ptr; | |
49 | } __packed; | |
50 | ||
51 | struct cpu_device_id { | |
52 | unsigned vendor; | |
53 | unsigned device; | |
54 | }; | |
55 | ||
56 | struct cpuinfo_x86 { | |
57 | uint8_t x86; /* CPU family */ | |
58 | uint8_t x86_vendor; /* CPU vendor */ | |
59 | uint8_t x86_model; | |
60 | uint8_t x86_mask; | |
61 | }; | |
62 | ||
caca13f6 SG |
63 | /* gcc 7.3 does not wwant to drop x86_vendors, so use #ifdef */ |
64 | #ifndef CONFIG_TPL_BUILD | |
be059e88 SG |
65 | /* |
66 | * List of cpu vendor strings along with their normalized | |
67 | * id values. | |
68 | */ | |
69 | static const struct { | |
70 | int vendor; | |
71 | const char *name; | |
72 | } x86_vendors[] = { | |
73 | { X86_VENDOR_INTEL, "GenuineIntel", }, | |
74 | { X86_VENDOR_CYRIX, "CyrixInstead", }, | |
75 | { X86_VENDOR_AMD, "AuthenticAMD", }, | |
76 | { X86_VENDOR_UMC, "UMC UMC UMC ", }, | |
77 | { X86_VENDOR_NEXGEN, "NexGenDriven", }, | |
78 | { X86_VENDOR_CENTAUR, "CentaurHauls", }, | |
79 | { X86_VENDOR_RISE, "RiseRiseRise", }, | |
80 | { X86_VENDOR_TRANSMETA, "GenuineTMx86", }, | |
81 | { X86_VENDOR_TRANSMETA, "TransmetaCPU", }, | |
82 | { X86_VENDOR_NSC, "Geode by NSC", }, | |
83 | { X86_VENDOR_SIS, "SiS SiS SiS ", }, | |
84 | }; | |
caca13f6 | 85 | #endif |
be059e88 SG |
86 | |
87 | static void load_ds(u32 segment) | |
88 | { | |
89 | asm volatile("movl %0, %%ds" : : "r" (segment * X86_GDT_ENTRY_SIZE)); | |
90 | } | |
91 | ||
92 | static void load_es(u32 segment) | |
93 | { | |
94 | asm volatile("movl %0, %%es" : : "r" (segment * X86_GDT_ENTRY_SIZE)); | |
95 | } | |
96 | ||
97 | static void load_fs(u32 segment) | |
98 | { | |
99 | asm volatile("movl %0, %%fs" : : "r" (segment * X86_GDT_ENTRY_SIZE)); | |
100 | } | |
101 | ||
102 | static void load_gs(u32 segment) | |
103 | { | |
104 | asm volatile("movl %0, %%gs" : : "r" (segment * X86_GDT_ENTRY_SIZE)); | |
105 | } | |
106 | ||
107 | static void load_ss(u32 segment) | |
108 | { | |
109 | asm volatile("movl %0, %%ss" : : "r" (segment * X86_GDT_ENTRY_SIZE)); | |
110 | } | |
111 | ||
112 | static void load_gdt(const u64 *boot_gdt, u16 num_entries) | |
113 | { | |
114 | struct gdt_ptr gdt; | |
115 | ||
116 | gdt.len = (num_entries * X86_GDT_ENTRY_SIZE) - 1; | |
117 | gdt.ptr = (ulong)boot_gdt; | |
118 | ||
119 | asm volatile("lgdtl %0\n" : : "m" (gdt)); | |
120 | } | |
121 | ||
122 | void arch_setup_gd(gd_t *new_gd) | |
123 | { | |
124 | u64 *gdt_addr; | |
125 | ||
126 | gdt_addr = new_gd->arch.gdt; | |
127 | ||
128 | /* | |
129 | * CS: code, read/execute, 4 GB, base 0 | |
130 | * | |
131 | * Some OS (like VxWorks) requires GDT entry 1 to be the 32-bit CS | |
132 | */ | |
133 | gdt_addr[X86_GDT_ENTRY_UNUSED] = GDT_ENTRY(0xc09b, 0, 0xfffff); | |
134 | gdt_addr[X86_GDT_ENTRY_32BIT_CS] = GDT_ENTRY(0xc09b, 0, 0xfffff); | |
135 | ||
136 | /* DS: data, read/write, 4 GB, base 0 */ | |
137 | gdt_addr[X86_GDT_ENTRY_32BIT_DS] = GDT_ENTRY(0xc093, 0, 0xfffff); | |
138 | ||
2fa863e9 MY |
139 | /* |
140 | * FS: data, read/write, sizeof (Global Data Pointer), | |
141 | * base (Global Data Pointer) | |
142 | */ | |
be059e88 | 143 | new_gd->arch.gd_addr = new_gd; |
2fa863e9 MY |
144 | gdt_addr[X86_GDT_ENTRY_32BIT_FS] = GDT_ENTRY(0x8093, |
145 | (ulong)&new_gd->arch.gd_addr, | |
146 | sizeof(new_gd->arch.gd_addr) - 1); | |
be059e88 SG |
147 | |
148 | /* 16-bit CS: code, read/execute, 64 kB, base 0 */ | |
149 | gdt_addr[X86_GDT_ENTRY_16BIT_CS] = GDT_ENTRY(0x009b, 0, 0x0ffff); | |
150 | ||
151 | /* 16-bit DS: data, read/write, 64 kB, base 0 */ | |
152 | gdt_addr[X86_GDT_ENTRY_16BIT_DS] = GDT_ENTRY(0x0093, 0, 0x0ffff); | |
153 | ||
154 | gdt_addr[X86_GDT_ENTRY_16BIT_FLAT_CS] = GDT_ENTRY(0x809b, 0, 0xfffff); | |
155 | gdt_addr[X86_GDT_ENTRY_16BIT_FLAT_DS] = GDT_ENTRY(0x8093, 0, 0xfffff); | |
156 | ||
157 | load_gdt(gdt_addr, X86_GDT_NUM_ENTRIES); | |
158 | load_ds(X86_GDT_ENTRY_32BIT_DS); | |
159 | load_es(X86_GDT_ENTRY_32BIT_DS); | |
160 | load_gs(X86_GDT_ENTRY_32BIT_DS); | |
161 | load_ss(X86_GDT_ENTRY_32BIT_DS); | |
162 | load_fs(X86_GDT_ENTRY_32BIT_FS); | |
163 | } | |
164 | ||
165 | #ifdef CONFIG_HAVE_FSP | |
166 | /* | |
167 | * Setup FSP execution environment GDT | |
168 | * | |
169 | * Per Intel FSP external architecture specification, before calling any FSP | |
170 | * APIs, we need make sure the system is in flat 32-bit mode and both the code | |
171 | * and data selectors should have full 4GB access range. Here we reuse the one | |
172 | * we used in arch/x86/cpu/start16.S, and reload the segement registers. | |
173 | */ | |
174 | void setup_fsp_gdt(void) | |
175 | { | |
176 | load_gdt((const u64 *)(gdt_rom + CONFIG_RESET_SEG_START), 4); | |
177 | load_ds(X86_GDT_ENTRY_32BIT_DS); | |
178 | load_ss(X86_GDT_ENTRY_32BIT_DS); | |
179 | load_es(X86_GDT_ENTRY_32BIT_DS); | |
180 | load_fs(X86_GDT_ENTRY_32BIT_DS); | |
181 | load_gs(X86_GDT_ENTRY_32BIT_DS); | |
182 | } | |
183 | #endif | |
184 | ||
185 | /* | |
186 | * Cyrix CPUs without cpuid or with cpuid not yet enabled can be detected | |
187 | * by the fact that they preserve the flags across the division of 5/2. | |
188 | * PII and PPro exhibit this behavior too, but they have cpuid available. | |
189 | */ | |
190 | ||
191 | /* | |
192 | * Perform the Cyrix 5/2 test. A Cyrix won't change | |
193 | * the flags, while other 486 chips will. | |
194 | */ | |
195 | static inline int test_cyrix_52div(void) | |
196 | { | |
197 | unsigned int test; | |
198 | ||
199 | __asm__ __volatile__( | |
200 | "sahf\n\t" /* clear flags (%eax = 0x0005) */ | |
201 | "div %b2\n\t" /* divide 5 by 2 */ | |
202 | "lahf" /* store flags into %ah */ | |
203 | : "=a" (test) | |
204 | : "0" (5), "q" (2) | |
205 | : "cc"); | |
206 | ||
207 | /* AH is 0x02 on Cyrix after the divide.. */ | |
208 | return (unsigned char) (test >> 8) == 0x02; | |
209 | } | |
210 | ||
caca13f6 | 211 | #ifndef CONFIG_TPL_BUILD |
be059e88 SG |
212 | /* |
213 | * Detect a NexGen CPU running without BIOS hypercode new enough | |
214 | * to have CPUID. (Thanks to Herbert Oppmann) | |
215 | */ | |
216 | static int deep_magic_nexgen_probe(void) | |
217 | { | |
218 | int ret; | |
219 | ||
220 | __asm__ __volatile__ ( | |
221 | " movw $0x5555, %%ax\n" | |
222 | " xorw %%dx,%%dx\n" | |
223 | " movw $2, %%cx\n" | |
224 | " divw %%cx\n" | |
225 | " movl $0, %%eax\n" | |
226 | " jnz 1f\n" | |
227 | " movl $1, %%eax\n" | |
228 | "1:\n" | |
229 | : "=a" (ret) : : "cx", "dx"); | |
230 | return ret; | |
231 | } | |
caca13f6 | 232 | #endif |
be059e88 SG |
233 | |
234 | static bool has_cpuid(void) | |
235 | { | |
236 | return flag_is_changeable_p(X86_EFLAGS_ID); | |
237 | } | |
238 | ||
239 | static bool has_mtrr(void) | |
240 | { | |
241 | return cpuid_edx(0x00000001) & (1 << 12) ? true : false; | |
242 | } | |
243 | ||
caca13f6 | 244 | #ifndef CONFIG_TPL_BUILD |
be059e88 SG |
245 | static int build_vendor_name(char *vendor_name) |
246 | { | |
247 | struct cpuid_result result; | |
248 | result = cpuid(0x00000000); | |
249 | unsigned int *name_as_ints = (unsigned int *)vendor_name; | |
250 | ||
251 | name_as_ints[0] = result.ebx; | |
252 | name_as_ints[1] = result.edx; | |
253 | name_as_ints[2] = result.ecx; | |
254 | ||
255 | return result.eax; | |
256 | } | |
caca13f6 | 257 | #endif |
be059e88 SG |
258 | |
259 | static void identify_cpu(struct cpu_device_id *cpu) | |
260 | { | |
caca13f6 SG |
261 | cpu->device = 0; /* fix gcc 4.4.4 warning */ |
262 | ||
263 | /* | |
264 | * Do a quick and dirty check to save space - Intel and AMD only and | |
265 | * just the vendor. This is enough for most TPL code. | |
266 | */ | |
267 | if (spl_phase() == PHASE_TPL) { | |
268 | struct cpuid_result result; | |
269 | ||
270 | result = cpuid(0x00000000); | |
271 | switch (result.ecx >> 24) { | |
272 | case 'l': /* GenuineIntel */ | |
273 | cpu->vendor = X86_VENDOR_INTEL; | |
274 | break; | |
275 | case 'D': /* AuthenticAMD */ | |
276 | cpu->vendor = X86_VENDOR_AMD; | |
277 | break; | |
278 | default: | |
279 | cpu->vendor = X86_VENDOR_ANY; | |
280 | break; | |
281 | } | |
282 | return; | |
283 | } | |
284 | ||
285 | /* gcc 7.3 does not want to drop x86_vendors, so use #ifdef */ | |
286 | #ifndef CONFIG_TPL_BUILD | |
be059e88 SG |
287 | char vendor_name[16]; |
288 | int i; | |
289 | ||
290 | vendor_name[0] = '\0'; /* Unset */ | |
be059e88 SG |
291 | |
292 | /* Find the id and vendor_name */ | |
293 | if (!has_cpuid()) { | |
294 | /* Its a 486 if we can modify the AC flag */ | |
295 | if (flag_is_changeable_p(X86_EFLAGS_AC)) | |
296 | cpu->device = 0x00000400; /* 486 */ | |
297 | else | |
298 | cpu->device = 0x00000300; /* 386 */ | |
299 | if ((cpu->device == 0x00000400) && test_cyrix_52div()) { | |
300 | memcpy(vendor_name, "CyrixInstead", 13); | |
301 | /* If we ever care we can enable cpuid here */ | |
302 | } | |
303 | /* Detect NexGen with old hypercode */ | |
304 | else if (deep_magic_nexgen_probe()) | |
305 | memcpy(vendor_name, "NexGenDriven", 13); | |
caca13f6 SG |
306 | } else { |
307 | int cpuid_level; | |
be059e88 SG |
308 | |
309 | cpuid_level = build_vendor_name(vendor_name); | |
310 | vendor_name[12] = '\0'; | |
311 | ||
312 | /* Intel-defined flags: level 0x00000001 */ | |
313 | if (cpuid_level >= 0x00000001) { | |
314 | cpu->device = cpuid_eax(0x00000001); | |
315 | } else { | |
316 | /* Have CPUID level 0 only unheard of */ | |
317 | cpu->device = 0x00000400; | |
318 | } | |
319 | } | |
320 | cpu->vendor = X86_VENDOR_UNKNOWN; | |
321 | for (i = 0; i < ARRAY_SIZE(x86_vendors); i++) { | |
322 | if (memcmp(vendor_name, x86_vendors[i].name, 12) == 0) { | |
323 | cpu->vendor = x86_vendors[i].vendor; | |
324 | break; | |
325 | } | |
326 | } | |
caca13f6 | 327 | #endif |
be059e88 SG |
328 | } |
329 | ||
330 | static inline void get_fms(struct cpuinfo_x86 *c, uint32_t tfms) | |
331 | { | |
332 | c->x86 = (tfms >> 8) & 0xf; | |
333 | c->x86_model = (tfms >> 4) & 0xf; | |
334 | c->x86_mask = tfms & 0xf; | |
335 | if (c->x86 == 0xf) | |
336 | c->x86 += (tfms >> 20) & 0xff; | |
337 | if (c->x86 >= 0x6) | |
338 | c->x86_model += ((tfms >> 16) & 0xF) << 4; | |
339 | } | |
340 | ||
341 | u32 cpu_get_family_model(void) | |
342 | { | |
343 | return gd->arch.x86_device & 0x0fff0ff0; | |
344 | } | |
345 | ||
346 | u32 cpu_get_stepping(void) | |
347 | { | |
348 | return gd->arch.x86_mask; | |
349 | } | |
350 | ||
c0069e9a SG |
351 | /* initialise FPU, reset EM, set MP and NE */ |
352 | static void setup_cpu_features(void) | |
be059e88 SG |
353 | { |
354 | const u32 em_rst = ~X86_CR0_EM; | |
355 | const u32 mp_ne_set = X86_CR0_MP | X86_CR0_NE; | |
356 | ||
c0069e9a SG |
357 | asm ("fninit\n" \ |
358 | "movl %%cr0, %%eax\n" \ | |
359 | "andl %0, %%eax\n" \ | |
360 | "orl %1, %%eax\n" \ | |
361 | "movl %%eax, %%cr0\n" \ | |
362 | : : "i" (em_rst), "i" (mp_ne_set) : "eax"); | |
363 | } | |
be059e88 | 364 | |
c0069e9a SG |
365 | static void setup_identity(void) |
366 | { | |
be059e88 SG |
367 | /* identify CPU via cpuid and store the decoded info into gd->arch */ |
368 | if (has_cpuid()) { | |
369 | struct cpu_device_id cpu; | |
370 | struct cpuinfo_x86 c; | |
371 | ||
372 | identify_cpu(&cpu); | |
373 | get_fms(&c, cpu.device); | |
374 | gd->arch.x86 = c.x86; | |
375 | gd->arch.x86_vendor = cpu.vendor; | |
376 | gd->arch.x86_model = c.x86_model; | |
377 | gd->arch.x86_mask = c.x86_mask; | |
378 | gd->arch.x86_device = cpu.device; | |
379 | ||
380 | gd->arch.has_mtrr = has_mtrr(); | |
381 | } | |
c0069e9a SG |
382 | } |
383 | ||
384 | /* Don't allow PCI region 3 to use memory in the 2-4GB memory hole */ | |
385 | static void setup_pci_ram_top(void) | |
386 | { | |
be059e88 | 387 | gd->pci_ram_top = 0x80000000U; |
c0069e9a SG |
388 | } |
389 | ||
390 | static void setup_mtrr(void) | |
391 | { | |
392 | u64 mtrr_cap; | |
be059e88 SG |
393 | |
394 | /* Configure fixed range MTRRs for some legacy regions */ | |
c0069e9a SG |
395 | if (!gd->arch.has_mtrr) |
396 | return; | |
397 | ||
398 | mtrr_cap = native_read_msr(MTRR_CAP_MSR); | |
399 | if (mtrr_cap & MTRR_CAP_FIX) { | |
400 | /* Mark the VGA RAM area as uncacheable */ | |
401 | native_write_msr(MTRR_FIX_16K_A0000_MSR, | |
402 | MTRR_FIX_TYPE(MTRR_TYPE_UNCACHEABLE), | |
403 | MTRR_FIX_TYPE(MTRR_TYPE_UNCACHEABLE)); | |
404 | ||
405 | /* | |
406 | * Mark the PCI ROM area as cacheable to improve ROM | |
407 | * execution performance. | |
408 | */ | |
409 | native_write_msr(MTRR_FIX_4K_C0000_MSR, | |
410 | MTRR_FIX_TYPE(MTRR_TYPE_WRBACK), | |
411 | MTRR_FIX_TYPE(MTRR_TYPE_WRBACK)); | |
412 | native_write_msr(MTRR_FIX_4K_C8000_MSR, | |
413 | MTRR_FIX_TYPE(MTRR_TYPE_WRBACK), | |
414 | MTRR_FIX_TYPE(MTRR_TYPE_WRBACK)); | |
415 | native_write_msr(MTRR_FIX_4K_D0000_MSR, | |
416 | MTRR_FIX_TYPE(MTRR_TYPE_WRBACK), | |
417 | MTRR_FIX_TYPE(MTRR_TYPE_WRBACK)); | |
418 | native_write_msr(MTRR_FIX_4K_D8000_MSR, | |
419 | MTRR_FIX_TYPE(MTRR_TYPE_WRBACK), | |
420 | MTRR_FIX_TYPE(MTRR_TYPE_WRBACK)); | |
421 | ||
422 | /* Enable the fixed range MTRRs */ | |
423 | msr_setbits_64(MTRR_DEF_TYPE_MSR, MTRR_DEF_TYPE_FIX_EN); | |
be059e88 | 424 | } |
c0069e9a SG |
425 | } |
426 | ||
ece3a460 SG |
427 | int x86_cpu_init_tpl(void) |
428 | { | |
429 | setup_cpu_features(); | |
430 | setup_identity(); | |
431 | ||
432 | return 0; | |
433 | } | |
434 | ||
c0069e9a SG |
435 | int x86_cpu_init_f(void) |
436 | { | |
437 | if (ll_boot_init()) | |
438 | setup_cpu_features(); | |
439 | setup_identity(); | |
440 | setup_mtrr(); | |
441 | setup_pci_ram_top(); | |
be059e88 | 442 | |
be059e88 | 443 | /* Set up the i8254 timer if required */ |
c0069e9a SG |
444 | if (IS_ENABLED(CONFIG_I8254_TIMER)) |
445 | i8254_init(); | |
446 | ||
447 | return 0; | |
448 | } | |
449 | ||
33139a0b SG |
450 | long detect_coreboot_table_at(ulong start, ulong size) |
451 | { | |
452 | u32 *ptr, *end; | |
453 | ||
454 | size /= 4; | |
455 | for (ptr = (void *)start, end = ptr + size; ptr < end; ptr += 4) { | |
456 | if (*ptr == 0x4f49424c) /* "LBIO" */ | |
457 | return (long)ptr; | |
458 | } | |
459 | ||
460 | return -ENOENT; | |
461 | } | |
462 | ||
463 | long locate_coreboot_table(void) | |
464 | { | |
465 | long addr; | |
466 | ||
467 | /* We look for LBIO in the first 4K of RAM and again at 960KB */ | |
468 | addr = detect_coreboot_table_at(0x0, 0x1000); | |
469 | if (addr < 0) | |
470 | addr = detect_coreboot_table_at(0xf0000, 0x1000); | |
471 | ||
472 | return addr; | |
473 | } | |
474 | ||
c0069e9a SG |
475 | int x86_cpu_reinit_f(void) |
476 | { | |
477 | setup_identity(); | |
478 | setup_pci_ram_top(); | |
cfe7a106 SG |
479 | if (locate_coreboot_table() >= 0) |
480 | gd->flags |= GD_FLG_SKIP_LL_INIT; | |
be059e88 SG |
481 | |
482 | return 0; | |
483 | } | |
484 | ||
485 | void x86_enable_caches(void) | |
486 | { | |
487 | unsigned long cr0; | |
488 | ||
489 | cr0 = read_cr0(); | |
490 | cr0 &= ~(X86_CR0_NW | X86_CR0_CD); | |
491 | write_cr0(cr0); | |
492 | wbinvd(); | |
493 | } | |
494 | void enable_caches(void) __attribute__((weak, alias("x86_enable_caches"))); | |
495 | ||
496 | void x86_disable_caches(void) | |
497 | { | |
498 | unsigned long cr0; | |
499 | ||
500 | cr0 = read_cr0(); | |
501 | cr0 |= X86_CR0_NW | X86_CR0_CD; | |
502 | wbinvd(); | |
503 | write_cr0(cr0); | |
504 | wbinvd(); | |
505 | } | |
506 | void disable_caches(void) __attribute__((weak, alias("x86_disable_caches"))); | |
507 | ||
508 | int dcache_status(void) | |
509 | { | |
510 | return !(read_cr0() & X86_CR0_CD); | |
511 | } | |
512 | ||
513 | void cpu_enable_paging_pae(ulong cr3) | |
514 | { | |
515 | __asm__ __volatile__( | |
516 | /* Load the page table address */ | |
517 | "movl %0, %%cr3\n" | |
518 | /* Enable pae */ | |
519 | "movl %%cr4, %%eax\n" | |
520 | "orl $0x00000020, %%eax\n" | |
521 | "movl %%eax, %%cr4\n" | |
522 | /* Enable paging */ | |
523 | "movl %%cr0, %%eax\n" | |
524 | "orl $0x80000000, %%eax\n" | |
525 | "movl %%eax, %%cr0\n" | |
526 | : | |
527 | : "r" (cr3) | |
528 | : "eax"); | |
529 | } | |
530 | ||
531 | void cpu_disable_paging_pae(void) | |
532 | { | |
533 | /* Turn off paging */ | |
534 | __asm__ __volatile__ ( | |
535 | /* Disable paging */ | |
536 | "movl %%cr0, %%eax\n" | |
537 | "andl $0x7fffffff, %%eax\n" | |
538 | "movl %%eax, %%cr0\n" | |
539 | /* Disable pae */ | |
540 | "movl %%cr4, %%eax\n" | |
541 | "andl $0xffffffdf, %%eax\n" | |
542 | "movl %%eax, %%cr4\n" | |
543 | : | |
544 | : | |
545 | : "eax"); | |
546 | } | |
547 | ||
548 | static bool can_detect_long_mode(void) | |
549 | { | |
550 | return cpuid_eax(0x80000000) > 0x80000000UL; | |
551 | } | |
552 | ||
553 | static bool has_long_mode(void) | |
554 | { | |
555 | return cpuid_edx(0x80000001) & (1 << 29) ? true : false; | |
556 | } | |
557 | ||
558 | int cpu_has_64bit(void) | |
559 | { | |
560 | return has_cpuid() && can_detect_long_mode() && | |
561 | has_long_mode(); | |
562 | } | |
563 | ||
dbb0696b | 564 | #define PAGETABLE_BASE 0x80000 |
be059e88 SG |
565 | #define PAGETABLE_SIZE (6 * 4096) |
566 | ||
567 | /** | |
568 | * build_pagetable() - build a flat 4GiB page table structure for 64-bti mode | |
569 | * | |
570 | * @pgtable: Pointer to a 24iKB block of memory | |
571 | */ | |
572 | static void build_pagetable(uint32_t *pgtable) | |
573 | { | |
574 | uint i; | |
575 | ||
576 | memset(pgtable, '\0', PAGETABLE_SIZE); | |
577 | ||
578 | /* Level 4 needs a single entry */ | |
579 | pgtable[0] = (ulong)&pgtable[1024] + 7; | |
580 | ||
581 | /* Level 3 has one 64-bit entry for each GiB of memory */ | |
582 | for (i = 0; i < 4; i++) | |
583 | pgtable[1024 + i * 2] = (ulong)&pgtable[2048] + 0x1000 * i + 7; | |
584 | ||
585 | /* Level 2 has 2048 64-bit entries, each repesenting 2MiB */ | |
586 | for (i = 0; i < 2048; i++) | |
587 | pgtable[2048 + i * 2] = 0x183 + (i << 21UL); | |
588 | } | |
589 | ||
590 | int cpu_jump_to_64bit(ulong setup_base, ulong target) | |
591 | { | |
592 | uint32_t *pgtable; | |
593 | ||
594 | pgtable = memalign(4096, PAGETABLE_SIZE); | |
595 | if (!pgtable) | |
596 | return -ENOMEM; | |
597 | ||
598 | build_pagetable(pgtable); | |
599 | cpu_call64((ulong)pgtable, setup_base, target); | |
600 | free(pgtable); | |
601 | ||
602 | return -EFAULT; | |
603 | } | |
604 | ||
fa5fcb3b SG |
605 | /* |
606 | * Jump from SPL to U-Boot | |
607 | * | |
608 | * This function is work-in-progress with many issues to resolve. | |
609 | * | |
610 | * It works by setting up several regions: | |
611 | * ptr - a place to put the code that jumps into 64-bit mode | |
612 | * gdt - a place to put the global descriptor table | |
613 | * pgtable - a place to put the page tables | |
614 | * | |
615 | * The cpu_call64() code is copied from ROM and then manually patched so that | |
616 | * it has the correct GDT address in RAM. U-Boot is copied from ROM into | |
617 | * its pre-relocation address. Then we jump to the cpu_call64() code in RAM, | |
618 | * which changes to 64-bit mode and starts U-Boot. | |
619 | */ | |
620 | int cpu_jump_to_64bit_uboot(ulong target) | |
621 | { | |
622 | typedef void (*func_t)(ulong pgtable, ulong setup_base, ulong target); | |
623 | uint32_t *pgtable; | |
624 | func_t func; | |
91683260 | 625 | char *ptr; |
fa5fcb3b | 626 | |
dbb0696b | 627 | pgtable = (uint32_t *)PAGETABLE_BASE; |
fa5fcb3b SG |
628 | |
629 | build_pagetable(pgtable); | |
630 | ||
91683260 BM |
631 | extern long call64_stub_size; |
632 | ptr = malloc(call64_stub_size); | |
633 | if (!ptr) { | |
634 | printf("Failed to allocate the cpu_call64 stub\n"); | |
635 | return -ENOMEM; | |
636 | } | |
91683260 | 637 | memcpy(ptr, cpu_call64, call64_stub_size); |
fa5fcb3b | 638 | |
fa5fcb3b | 639 | func = (func_t)ptr; |
fa5fcb3b SG |
640 | |
641 | /* | |
642 | * Copy U-Boot from ROM | |
643 | * TODO([email protected]): Figure out a way to get the text base | |
644 | * correctly here, and in the device-tree binman definition. | |
645 | * | |
646 | * Also consider using FIT so we get the correct image length and | |
647 | * parameters. | |
648 | */ | |
649 | memcpy((char *)target, (char *)0xfff00000, 0x100000); | |
650 | ||
651 | /* Jump to U-Boot */ | |
652 | func((ulong)pgtable, 0, (ulong)target); | |
653 | ||
654 | return -EFAULT; | |
655 | } | |
656 | ||
be059e88 SG |
657 | #ifdef CONFIG_SMP |
658 | static int enable_smis(struct udevice *cpu, void *unused) | |
659 | { | |
660 | return 0; | |
661 | } | |
662 | ||
663 | static struct mp_flight_record mp_steps[] = { | |
664 | MP_FR_BLOCK_APS(mp_init_cpu, NULL, mp_init_cpu, NULL), | |
665 | /* Wait for APs to finish initialization before proceeding */ | |
666 | MP_FR_BLOCK_APS(NULL, NULL, enable_smis, NULL), | |
667 | }; | |
668 | ||
669 | int x86_mp_init(void) | |
670 | { | |
671 | struct mp_params mp_params; | |
672 | ||
673 | mp_params.parallel_microcode_load = 0, | |
674 | mp_params.flight_plan = &mp_steps[0]; | |
675 | mp_params.num_records = ARRAY_SIZE(mp_steps); | |
676 | mp_params.microcode_pointer = 0; | |
677 | ||
678 | if (mp_init(&mp_params)) { | |
679 | printf("Warning: MP init failure\n"); | |
680 | return -EIO; | |
681 | } | |
682 | ||
683 | return 0; | |
684 | } | |
685 | #endif |