xref: /openbmc/u-boot/arch/x86/cpu/cpu.c (revision 63d54a67)
1 /*
2  * (C) Copyright 2008-2011
3  * Graeme Russ, <graeme.russ@gmail.com>
4  *
5  * (C) Copyright 2002
6  * Daniel Engström, Omicron Ceti AB, <daniel@omicron.se>
7  *
8  * (C) Copyright 2002
9  * Sysgo Real-Time Solutions, GmbH <www.elinos.com>
10  * Marius Groeger <mgroeger@sysgo.de>
11  *
12  * (C) Copyright 2002
13  * Sysgo Real-Time Solutions, GmbH <www.elinos.com>
14  * Alex Zuepke <azu@sysgo.de>
15  *
16  * Part of this file is adapted from coreboot
17  * src/arch/x86/lib/cpu.c
18  *
19  * SPDX-License-Identifier:	GPL-2.0+
20  */
21 
22 #include <common.h>
23 #include <command.h>
24 #include <dm.h>
25 #include <errno.h>
26 #include <malloc.h>
27 #include <asm/control_regs.h>
28 #include <asm/cpu.h>
29 #include <asm/lapic.h>
30 #include <asm/mp.h>
31 #include <asm/post.h>
32 #include <asm/processor.h>
33 #include <asm/processor-flags.h>
34 #include <asm/interrupt.h>
35 #include <asm/tables.h>
36 #include <linux/compiler.h>
37 
38 DECLARE_GLOBAL_DATA_PTR;
39 
40 /*
41  * Constructor for a conventional segment GDT (or LDT) entry
42  * This is a macro so it can be used in initialisers
43  */
44 #define GDT_ENTRY(flags, base, limit)			\
45 	((((base)  & 0xff000000ULL) << (56-24)) |	\
46 	 (((flags) & 0x0000f0ffULL) << 40) |		\
47 	 (((limit) & 0x000f0000ULL) << (48-16)) |	\
48 	 (((base)  & 0x00ffffffULL) << 16) |		\
49 	 (((limit) & 0x0000ffffULL)))
50 
51 struct gdt_ptr {
52 	u16 len;
53 	u32 ptr;
54 } __packed;
55 
56 struct cpu_device_id {
57 	unsigned vendor;
58 	unsigned device;
59 };
60 
61 struct cpuinfo_x86 {
62 	uint8_t x86;            /* CPU family */
63 	uint8_t x86_vendor;     /* CPU vendor */
64 	uint8_t x86_model;
65 	uint8_t x86_mask;
66 };
67 
68 /*
69  * List of cpu vendor strings along with their normalized
70  * id values.
71  */
72 static struct {
73 	int vendor;
74 	const char *name;
75 } x86_vendors[] = {
76 	{ X86_VENDOR_INTEL,     "GenuineIntel", },
77 	{ X86_VENDOR_CYRIX,     "CyrixInstead", },
78 	{ X86_VENDOR_AMD,       "AuthenticAMD", },
79 	{ X86_VENDOR_UMC,       "UMC UMC UMC ", },
80 	{ X86_VENDOR_NEXGEN,    "NexGenDriven", },
81 	{ X86_VENDOR_CENTAUR,   "CentaurHauls", },
82 	{ X86_VENDOR_RISE,      "RiseRiseRise", },
83 	{ X86_VENDOR_TRANSMETA, "GenuineTMx86", },
84 	{ X86_VENDOR_TRANSMETA, "TransmetaCPU", },
85 	{ X86_VENDOR_NSC,       "Geode by NSC", },
86 	{ X86_VENDOR_SIS,       "SiS SiS SiS ", },
87 };
88 
89 static const char *const x86_vendor_name[] = {
90 	[X86_VENDOR_INTEL]     = "Intel",
91 	[X86_VENDOR_CYRIX]     = "Cyrix",
92 	[X86_VENDOR_AMD]       = "AMD",
93 	[X86_VENDOR_UMC]       = "UMC",
94 	[X86_VENDOR_NEXGEN]    = "NexGen",
95 	[X86_VENDOR_CENTAUR]   = "Centaur",
96 	[X86_VENDOR_RISE]      = "Rise",
97 	[X86_VENDOR_TRANSMETA] = "Transmeta",
98 	[X86_VENDOR_NSC]       = "NSC",
99 	[X86_VENDOR_SIS]       = "SiS",
100 };
101 
102 static void load_ds(u32 segment)
103 {
104 	asm volatile("movl %0, %%ds" : : "r" (segment * X86_GDT_ENTRY_SIZE));
105 }
106 
107 static void load_es(u32 segment)
108 {
109 	asm volatile("movl %0, %%es" : : "r" (segment * X86_GDT_ENTRY_SIZE));
110 }
111 
112 static void load_fs(u32 segment)
113 {
114 	asm volatile("movl %0, %%fs" : : "r" (segment * X86_GDT_ENTRY_SIZE));
115 }
116 
117 static void load_gs(u32 segment)
118 {
119 	asm volatile("movl %0, %%gs" : : "r" (segment * X86_GDT_ENTRY_SIZE));
120 }
121 
122 static void load_ss(u32 segment)
123 {
124 	asm volatile("movl %0, %%ss" : : "r" (segment * X86_GDT_ENTRY_SIZE));
125 }
126 
127 static void load_gdt(const u64 *boot_gdt, u16 num_entries)
128 {
129 	struct gdt_ptr gdt;
130 
131 	gdt.len = (num_entries * X86_GDT_ENTRY_SIZE) - 1;
132 	gdt.ptr = (u32)boot_gdt;
133 
134 	asm volatile("lgdtl %0\n" : : "m" (gdt));
135 }
136 
137 void setup_gdt(gd_t *id, u64 *gdt_addr)
138 {
139 	id->arch.gdt = gdt_addr;
140 	/* CS: code, read/execute, 4 GB, base 0 */
141 	gdt_addr[X86_GDT_ENTRY_32BIT_CS] = GDT_ENTRY(0xc09b, 0, 0xfffff);
142 
143 	/* DS: data, read/write, 4 GB, base 0 */
144 	gdt_addr[X86_GDT_ENTRY_32BIT_DS] = GDT_ENTRY(0xc093, 0, 0xfffff);
145 
146 	/* FS: data, read/write, 4 GB, base (Global Data Pointer) */
147 	id->arch.gd_addr = id;
148 	gdt_addr[X86_GDT_ENTRY_32BIT_FS] = GDT_ENTRY(0xc093,
149 		     (ulong)&id->arch.gd_addr, 0xfffff);
150 
151 	/* 16-bit CS: code, read/execute, 64 kB, base 0 */
152 	gdt_addr[X86_GDT_ENTRY_16BIT_CS] = GDT_ENTRY(0x009b, 0, 0x0ffff);
153 
154 	/* 16-bit DS: data, read/write, 64 kB, base 0 */
155 	gdt_addr[X86_GDT_ENTRY_16BIT_DS] = GDT_ENTRY(0x0093, 0, 0x0ffff);
156 
157 	gdt_addr[X86_GDT_ENTRY_16BIT_FLAT_CS] = GDT_ENTRY(0x809b, 0, 0xfffff);
158 	gdt_addr[X86_GDT_ENTRY_16BIT_FLAT_DS] = GDT_ENTRY(0x8093, 0, 0xfffff);
159 
160 	load_gdt(gdt_addr, X86_GDT_NUM_ENTRIES);
161 	load_ds(X86_GDT_ENTRY_32BIT_DS);
162 	load_es(X86_GDT_ENTRY_32BIT_DS);
163 	load_gs(X86_GDT_ENTRY_32BIT_DS);
164 	load_ss(X86_GDT_ENTRY_32BIT_DS);
165 	load_fs(X86_GDT_ENTRY_32BIT_FS);
166 }
167 
168 #ifdef CONFIG_HAVE_FSP
169 /*
170  * Setup FSP execution environment GDT
171  *
172  * Per Intel FSP external architecture specification, before calling any FSP
173  * APIs, we need make sure the system is in flat 32-bit mode and both the code
174  * and data selectors should have full 4GB access range. Here we reuse the one
175  * we used in arch/x86/cpu/start16.S, and reload the segement registers.
176  */
177 void setup_fsp_gdt(void)
178 {
179 	load_gdt((const u64 *)(gdt_rom + CONFIG_RESET_SEG_START), 4);
180 	load_ds(X86_GDT_ENTRY_32BIT_DS);
181 	load_ss(X86_GDT_ENTRY_32BIT_DS);
182 	load_es(X86_GDT_ENTRY_32BIT_DS);
183 	load_fs(X86_GDT_ENTRY_32BIT_DS);
184 	load_gs(X86_GDT_ENTRY_32BIT_DS);
185 }
186 #endif
187 
188 int __weak x86_cleanup_before_linux(void)
189 {
190 #ifdef CONFIG_BOOTSTAGE_STASH
191 	bootstage_stash((void *)CONFIG_BOOTSTAGE_STASH_ADDR,
192 			CONFIG_BOOTSTAGE_STASH_SIZE);
193 #endif
194 
195 	return 0;
196 }
197 
198 /*
199  * Cyrix CPUs without cpuid or with cpuid not yet enabled can be detected
200  * by the fact that they preserve the flags across the division of 5/2.
201  * PII and PPro exhibit this behavior too, but they have cpuid available.
202  */
203 
204 /*
205  * Perform the Cyrix 5/2 test. A Cyrix won't change
206  * the flags, while other 486 chips will.
207  */
208 static inline int test_cyrix_52div(void)
209 {
210 	unsigned int test;
211 
212 	__asm__ __volatile__(
213 	     "sahf\n\t"		/* clear flags (%eax = 0x0005) */
214 	     "div %b2\n\t"	/* divide 5 by 2 */
215 	     "lahf"		/* store flags into %ah */
216 	     : "=a" (test)
217 	     : "0" (5), "q" (2)
218 	     : "cc");
219 
220 	/* AH is 0x02 on Cyrix after the divide.. */
221 	return (unsigned char) (test >> 8) == 0x02;
222 }
223 
224 /*
225  *	Detect a NexGen CPU running without BIOS hypercode new enough
226  *	to have CPUID. (Thanks to Herbert Oppmann)
227  */
228 
229 static int deep_magic_nexgen_probe(void)
230 {
231 	int ret;
232 
233 	__asm__ __volatile__ (
234 		"	movw	$0x5555, %%ax\n"
235 		"	xorw	%%dx,%%dx\n"
236 		"	movw	$2, %%cx\n"
237 		"	divw	%%cx\n"
238 		"	movl	$0, %%eax\n"
239 		"	jnz	1f\n"
240 		"	movl	$1, %%eax\n"
241 		"1:\n"
242 		: "=a" (ret) : : "cx", "dx");
243 	return  ret;
244 }
245 
246 static bool has_cpuid(void)
247 {
248 	return flag_is_changeable_p(X86_EFLAGS_ID);
249 }
250 
251 static bool has_mtrr(void)
252 {
253 	return cpuid_edx(0x00000001) & (1 << 12) ? true : false;
254 }
255 
256 static int build_vendor_name(char *vendor_name)
257 {
258 	struct cpuid_result result;
259 	result = cpuid(0x00000000);
260 	unsigned int *name_as_ints = (unsigned int *)vendor_name;
261 
262 	name_as_ints[0] = result.ebx;
263 	name_as_ints[1] = result.edx;
264 	name_as_ints[2] = result.ecx;
265 
266 	return result.eax;
267 }
268 
269 static void identify_cpu(struct cpu_device_id *cpu)
270 {
271 	char vendor_name[16];
272 	int i;
273 
274 	vendor_name[0] = '\0'; /* Unset */
275 	cpu->device = 0; /* fix gcc 4.4.4 warning */
276 
277 	/* Find the id and vendor_name */
278 	if (!has_cpuid()) {
279 		/* Its a 486 if we can modify the AC flag */
280 		if (flag_is_changeable_p(X86_EFLAGS_AC))
281 			cpu->device = 0x00000400; /* 486 */
282 		else
283 			cpu->device = 0x00000300; /* 386 */
284 		if ((cpu->device == 0x00000400) && test_cyrix_52div()) {
285 			memcpy(vendor_name, "CyrixInstead", 13);
286 			/* If we ever care we can enable cpuid here */
287 		}
288 		/* Detect NexGen with old hypercode */
289 		else if (deep_magic_nexgen_probe())
290 			memcpy(vendor_name, "NexGenDriven", 13);
291 	}
292 	if (has_cpuid()) {
293 		int  cpuid_level;
294 
295 		cpuid_level = build_vendor_name(vendor_name);
296 		vendor_name[12] = '\0';
297 
298 		/* Intel-defined flags: level 0x00000001 */
299 		if (cpuid_level >= 0x00000001) {
300 			cpu->device = cpuid_eax(0x00000001);
301 		} else {
302 			/* Have CPUID level 0 only unheard of */
303 			cpu->device = 0x00000400;
304 		}
305 	}
306 	cpu->vendor = X86_VENDOR_UNKNOWN;
307 	for (i = 0; i < ARRAY_SIZE(x86_vendors); i++) {
308 		if (memcmp(vendor_name, x86_vendors[i].name, 12) == 0) {
309 			cpu->vendor = x86_vendors[i].vendor;
310 			break;
311 		}
312 	}
313 }
314 
315 static inline void get_fms(struct cpuinfo_x86 *c, uint32_t tfms)
316 {
317 	c->x86 = (tfms >> 8) & 0xf;
318 	c->x86_model = (tfms >> 4) & 0xf;
319 	c->x86_mask = tfms & 0xf;
320 	if (c->x86 == 0xf)
321 		c->x86 += (tfms >> 20) & 0xff;
322 	if (c->x86 >= 0x6)
323 		c->x86_model += ((tfms >> 16) & 0xF) << 4;
324 }
325 
326 int x86_cpu_init_f(void)
327 {
328 	const u32 em_rst = ~X86_CR0_EM;
329 	const u32 mp_ne_set = X86_CR0_MP | X86_CR0_NE;
330 
331 	/* initialize FPU, reset EM, set MP and NE */
332 	asm ("fninit\n" \
333 	     "movl %%cr0, %%eax\n" \
334 	     "andl %0, %%eax\n" \
335 	     "orl  %1, %%eax\n" \
336 	     "movl %%eax, %%cr0\n" \
337 	     : : "i" (em_rst), "i" (mp_ne_set) : "eax");
338 
339 	/* identify CPU via cpuid and store the decoded info into gd->arch */
340 	if (has_cpuid()) {
341 		struct cpu_device_id cpu;
342 		struct cpuinfo_x86 c;
343 
344 		identify_cpu(&cpu);
345 		get_fms(&c, cpu.device);
346 		gd->arch.x86 = c.x86;
347 		gd->arch.x86_vendor = cpu.vendor;
348 		gd->arch.x86_model = c.x86_model;
349 		gd->arch.x86_mask = c.x86_mask;
350 		gd->arch.x86_device = cpu.device;
351 
352 		gd->arch.has_mtrr = has_mtrr();
353 	}
354 
355 	return 0;
356 }
357 
358 void x86_enable_caches(void)
359 {
360 	unsigned long cr0;
361 
362 	cr0 = read_cr0();
363 	cr0 &= ~(X86_CR0_NW | X86_CR0_CD);
364 	write_cr0(cr0);
365 	wbinvd();
366 }
367 void enable_caches(void) __attribute__((weak, alias("x86_enable_caches")));
368 
369 void x86_disable_caches(void)
370 {
371 	unsigned long cr0;
372 
373 	cr0 = read_cr0();
374 	cr0 |= X86_CR0_NW | X86_CR0_CD;
375 	wbinvd();
376 	write_cr0(cr0);
377 	wbinvd();
378 }
379 void disable_caches(void) __attribute__((weak, alias("x86_disable_caches")));
380 
381 int x86_init_cache(void)
382 {
383 	enable_caches();
384 
385 	return 0;
386 }
387 int init_cache(void) __attribute__((weak, alias("x86_init_cache")));
388 
389 int do_reset(cmd_tbl_t *cmdtp, int flag, int argc, char * const argv[])
390 {
391 	printf("resetting ...\n");
392 
393 	/* wait 50 ms */
394 	udelay(50000);
395 	disable_interrupts();
396 	reset_cpu(0);
397 
398 	/*NOTREACHED*/
399 	return 0;
400 }
401 
402 void  flush_cache(unsigned long dummy1, unsigned long dummy2)
403 {
404 	asm("wbinvd\n");
405 }
406 
407 __weak void reset_cpu(ulong addr)
408 {
409 	/* Do a hard reset through the chipset's reset control register */
410 	outb(SYS_RST | RST_CPU, PORT_RESET);
411 	for (;;)
412 		cpu_hlt();
413 }
414 
415 void x86_full_reset(void)
416 {
417 	outb(FULL_RST | SYS_RST | RST_CPU, PORT_RESET);
418 }
419 
420 int dcache_status(void)
421 {
422 	return !(read_cr0() & 0x40000000);
423 }
424 
425 /* Define these functions to allow ehch-hcd to function */
426 void flush_dcache_range(unsigned long start, unsigned long stop)
427 {
428 }
429 
430 void invalidate_dcache_range(unsigned long start, unsigned long stop)
431 {
432 }
433 
434 void dcache_enable(void)
435 {
436 	enable_caches();
437 }
438 
439 void dcache_disable(void)
440 {
441 	disable_caches();
442 }
443 
444 void icache_enable(void)
445 {
446 }
447 
448 void icache_disable(void)
449 {
450 }
451 
452 int icache_status(void)
453 {
454 	return 1;
455 }
456 
457 void cpu_enable_paging_pae(ulong cr3)
458 {
459 	__asm__ __volatile__(
460 		/* Load the page table address */
461 		"movl	%0, %%cr3\n"
462 		/* Enable pae */
463 		"movl	%%cr4, %%eax\n"
464 		"orl	$0x00000020, %%eax\n"
465 		"movl	%%eax, %%cr4\n"
466 		/* Enable paging */
467 		"movl	%%cr0, %%eax\n"
468 		"orl	$0x80000000, %%eax\n"
469 		"movl	%%eax, %%cr0\n"
470 		:
471 		: "r" (cr3)
472 		: "eax");
473 }
474 
475 void cpu_disable_paging_pae(void)
476 {
477 	/* Turn off paging */
478 	__asm__ __volatile__ (
479 		/* Disable paging */
480 		"movl	%%cr0, %%eax\n"
481 		"andl	$0x7fffffff, %%eax\n"
482 		"movl	%%eax, %%cr0\n"
483 		/* Disable pae */
484 		"movl	%%cr4, %%eax\n"
485 		"andl	$0xffffffdf, %%eax\n"
486 		"movl	%%eax, %%cr4\n"
487 		:
488 		:
489 		: "eax");
490 }
491 
492 static bool can_detect_long_mode(void)
493 {
494 	return cpuid_eax(0x80000000) > 0x80000000UL;
495 }
496 
497 static bool has_long_mode(void)
498 {
499 	return cpuid_edx(0x80000001) & (1 << 29) ? true : false;
500 }
501 
502 int cpu_has_64bit(void)
503 {
504 	return has_cpuid() && can_detect_long_mode() &&
505 		has_long_mode();
506 }
507 
508 const char *cpu_vendor_name(int vendor)
509 {
510 	const char *name;
511 	name = "<invalid cpu vendor>";
512 	if ((vendor < (ARRAY_SIZE(x86_vendor_name))) &&
513 	    (x86_vendor_name[vendor] != 0))
514 		name = x86_vendor_name[vendor];
515 
516 	return name;
517 }
518 
519 char *cpu_get_name(char *name)
520 {
521 	unsigned int *name_as_ints = (unsigned int *)name;
522 	struct cpuid_result regs;
523 	char *ptr;
524 	int i;
525 
526 	/* This bit adds up to 48 bytes */
527 	for (i = 0; i < 3; i++) {
528 		regs = cpuid(0x80000002 + i);
529 		name_as_ints[i * 4 + 0] = regs.eax;
530 		name_as_ints[i * 4 + 1] = regs.ebx;
531 		name_as_ints[i * 4 + 2] = regs.ecx;
532 		name_as_ints[i * 4 + 3] = regs.edx;
533 	}
534 	name[CPU_MAX_NAME_LEN - 1] = '\0';
535 
536 	/* Skip leading spaces. */
537 	ptr = name;
538 	while (*ptr == ' ')
539 		ptr++;
540 
541 	return ptr;
542 }
543 
544 int default_print_cpuinfo(void)
545 {
546 	printf("CPU: %s, vendor %s, device %xh\n",
547 	       cpu_has_64bit() ? "x86_64" : "x86",
548 	       cpu_vendor_name(gd->arch.x86_vendor), gd->arch.x86_device);
549 
550 	return 0;
551 }
552 
553 #define PAGETABLE_SIZE		(6 * 4096)
554 
555 /**
556  * build_pagetable() - build a flat 4GiB page table structure for 64-bti mode
557  *
558  * @pgtable: Pointer to a 24iKB block of memory
559  */
560 static void build_pagetable(uint32_t *pgtable)
561 {
562 	uint i;
563 
564 	memset(pgtable, '\0', PAGETABLE_SIZE);
565 
566 	/* Level 4 needs a single entry */
567 	pgtable[0] = (uint32_t)&pgtable[1024] + 7;
568 
569 	/* Level 3 has one 64-bit entry for each GiB of memory */
570 	for (i = 0; i < 4; i++) {
571 		pgtable[1024 + i * 2] = (uint32_t)&pgtable[2048] +
572 							0x1000 * i + 7;
573 	}
574 
575 	/* Level 2 has 2048 64-bit entries, each repesenting 2MiB */
576 	for (i = 0; i < 2048; i++)
577 		pgtable[2048 + i * 2] = 0x183 + (i << 21UL);
578 }
579 
580 int cpu_jump_to_64bit(ulong setup_base, ulong target)
581 {
582 	uint32_t *pgtable;
583 
584 	pgtable = memalign(4096, PAGETABLE_SIZE);
585 	if (!pgtable)
586 		return -ENOMEM;
587 
588 	build_pagetable(pgtable);
589 	cpu_call64((ulong)pgtable, setup_base, target);
590 	free(pgtable);
591 
592 	return -EFAULT;
593 }
594 
595 void show_boot_progress(int val)
596 {
597 #if MIN_PORT80_KCLOCKS_DELAY
598 	/*
599 	 * Scale the time counter reading to avoid using 64 bit arithmetics.
600 	 * Can't use get_timer() here becuase it could be not yet
601 	 * initialized or even implemented.
602 	 */
603 	if (!gd->arch.tsc_prev) {
604 		gd->arch.tsc_base_kclocks = rdtsc() / 1000;
605 		gd->arch.tsc_prev = 0;
606 	} else {
607 		uint32_t now;
608 
609 		do {
610 			now = rdtsc() / 1000 - gd->arch.tsc_base_kclocks;
611 		} while (now < (gd->arch.tsc_prev + MIN_PORT80_KCLOCKS_DELAY));
612 		gd->arch.tsc_prev = now;
613 	}
614 #endif
615 	outb(val, POST_PORT);
616 }
617 
618 #ifndef CONFIG_SYS_COREBOOT
619 int last_stage_init(void)
620 {
621 	write_tables();
622 
623 	return 0;
624 }
625 #endif
626 
627 #ifdef CONFIG_SMP
628 static int enable_smis(struct udevice *cpu, void *unused)
629 {
630 	return 0;
631 }
632 
633 static struct mp_flight_record mp_steps[] = {
634 	MP_FR_BLOCK_APS(mp_init_cpu, NULL, mp_init_cpu, NULL),
635 	/* Wait for APs to finish initialization before proceeding */
636 	MP_FR_BLOCK_APS(NULL, NULL, enable_smis, NULL),
637 };
638 
639 static int x86_mp_init(void)
640 {
641 	struct mp_params mp_params;
642 
643 	mp_params.parallel_microcode_load = 0,
644 	mp_params.flight_plan = &mp_steps[0];
645 	mp_params.num_records = ARRAY_SIZE(mp_steps);
646 	mp_params.microcode_pointer = 0;
647 
648 	if (mp_init(&mp_params)) {
649 		printf("Warning: MP init failure\n");
650 		return -EIO;
651 	}
652 
653 	return 0;
654 }
655 #endif
656 
657 __weak int x86_init_cpus(void)
658 {
659 #ifdef CONFIG_SMP
660 	debug("Init additional CPUs\n");
661 	x86_mp_init();
662 #endif
663 
664 	return 0;
665 }
666 
667 int cpu_init_r(void)
668 {
669 	return x86_init_cpus();
670 }
671