1 /* Needed so the AMD K8 runs correctly. */
2 /* this should be done by Eric
3 * 2004.11 yhlu add d0 e0 support
4 * 2004.12 yhlu add dual core support
5 * 2005.02 yhlu add e0 memory hole support
8 * 2005.08 yhlu add microcode support
11 #include <console/console.h>
12 #include <cpu/x86/msr.h>
13 #include <cpu/amd/mtrr.h>
14 #include <device/device.h>
15 #include <device/pci.h>
17 #include <cpu/x86/msr.h>
18 #include <cpu/x86/pae.h>
19 #include <pc80/mc146818rtc.h>
20 #include <cpu/x86/lapic.h>
21 #include "northbridge/amd/amdk8/amdk8.h"
22 #include <cpu/amd/model_fxx_rev.h>
23 #include <cpu/amd/microcode.h>
25 #include <cpu/x86/cache.h>
26 #include <cpu/x86/mtrr.h>
27 #include <cpu/x86/smm.h>
28 #include <cpu/amd/multicore.h>
29 #include <cpu/amd/model_fxx_msr.h>
31 #if CONFIG_WAIT_BEFORE_CPUS_INIT
32 void cpus_ready_for_init(void)
34 #if CONFIG_MEM_TRAIN_SEQ == 1
35 struct sys_info *sysinfox = (struct sys_info *)((CONFIG_RAMTOP) - CONFIG_DCACHE_RAM_GLOBAL_VAR_SIZE);
36 // wait for ap memory to trained
37 wait_all_core0_mem_trained(sysinfox);
42 #if CONFIG_K8_REV_F_SUPPORT == 0
43 int is_e0_later_in_bsp(int nodeid)
48 if (nodeid == 0) { // we don't need to do that for node 0 in core0/node0
49 return !is_cpu_pre_e0();
51 // d0 will be treated as e0 with this methods, but the d0 nb_cfg_54 always 0
53 dev = dev_find_slot(0, PCI_DEVFN(0x18 + nodeid, 2));
56 val_old = pci_read_config32(dev, 0x80);
59 pci_write_config32(dev, 0x80, val);
60 val = pci_read_config32(dev, 0x80);
61 e0_later = !!(val & (1 << 3));
62 if (e0_later) { // pre_e0 bit 3 always be 0 and can not be changed
63 pci_write_config32(dev, 0x80, val_old); // restore it
70 #if CONFIG_K8_REV_F_SUPPORT == 1
71 int is_cpu_f0_in_bsp(int nodeid)
75 dev = dev_find_slot(0, PCI_DEVFN(0x18 + nodeid, 3));
76 dword = pci_read_config32(dev, 0xfc);
77 return (dword & 0xfff00) == 0x40f00;
81 #define MCI_STATUS 0x401
83 static inline msr_t rdmsr_amd(u32 index)
88 :"=a"(result.lo), "=d"(result.hi)
89 :"c"(index), "D"(0x9c5a203a)
94 static inline void wrmsr_amd(u32 index, msr_t msr)
99 :"c"(index), "a"(msr.lo), "d"(msr.hi), "D"(0x9c5a203a)
104 #define ZERO_CHUNK_KB 0x800UL /* 2M */
105 #define TOLM_KB 0x400000UL
113 struct mtrr mtrrs[MTRR_COUNT];
114 msr_t top_mem, top_mem2;
118 static void save_mtrr_state(struct mtrr_state *state)
121 for (i = 0; i < MTRR_COUNT; i++) {
122 state->mtrrs[i].base = rdmsr(MTRRphysBase_MSR(i));
123 state->mtrrs[i].mask = rdmsr(MTRRphysMask_MSR(i));
125 state->top_mem = rdmsr(TOP_MEM);
126 state->top_mem2 = rdmsr(TOP_MEM2);
127 state->def_type = rdmsr(MTRRdefType_MSR);
130 static void restore_mtrr_state(struct mtrr_state *state)
135 for (i = 0; i < MTRR_COUNT; i++) {
136 wrmsr(MTRRphysBase_MSR(i), state->mtrrs[i].base);
137 wrmsr(MTRRphysMask_MSR(i), state->mtrrs[i].mask);
139 wrmsr(TOP_MEM, state->top_mem);
140 wrmsr(TOP_MEM2, state->top_mem2);
141 wrmsr(MTRRdefType_MSR, state->def_type);
147 static void print_mtrr_state(struct mtrr_state *state)
150 for (i = 0; i < MTRR_COUNT; i++) {
151 printk(BIOS_DEBUG, "var mtrr %d: %08x%08x mask: %08x%08x\n",
153 state->mtrrs[i].base.hi, state->mtrrs[i].base.lo,
154 state->mtrrs[i].mask.hi, state->mtrrs[i].mask.lo);
156 printk(BIOS_DEBUG, "top_mem: %08x%08x\n",
157 state->top_mem.hi, state->top_mem.lo);
158 printk(BIOS_DEBUG, "top_mem2: %08x%08x\n",
159 state->top_mem2.hi, state->top_mem2.lo);
160 printk(BIOS_DEBUG, "def_type: %08x%08x\n",
161 state->def_type.hi, state->def_type.lo);
165 static void set_init_ecc_mtrrs(void)
171 /* First clear all of the msrs to be safe */
172 for (i = 0; i < MTRR_COUNT; i++) {
174 zero.lo = zero.hi = 0;
175 wrmsr(MTRRphysBase_MSR(i), zero);
176 wrmsr(MTRRphysMask_MSR(i), zero);
179 /* Write back cache the first 1MB */
181 msr.lo = 0x00000000 | MTRR_TYPE_WRBACK;
182 wrmsr(MTRRphysBase_MSR(0), msr);
184 msr.lo = ~((CONFIG_RAMTOP) - 1) | 0x800;
185 wrmsr(MTRRphysMask_MSR(0), msr);
187 /* Set the default type to write combining */
189 msr.lo = 0xc00 | MTRR_TYPE_WRCOMB;
190 wrmsr(MTRRdefType_MSR, msr);
192 /* Set TOP_MEM to 4G */
200 static inline void clear_2M_ram(unsigned long basek,
201 struct mtrr_state *mtrr_state)
203 unsigned long limitk;
207 /* Report every 64M */
208 if ((basek % (64 * 1024)) == 0) {
210 /* Restore the normal state */
212 restore_mtrr_state(mtrr_state);
215 /* Print a status message */
216 printk(BIOS_DEBUG, "%c", (basek >= TOLM_KB) ? '+' : '-');
218 /* Return to the initialization state */
219 set_init_ecc_mtrrs();
224 limitk = (basek + ZERO_CHUNK_KB) & ~(ZERO_CHUNK_KB - 1);
226 /* couldn't happen, memory must on 2M boundary */
231 size = (limitk - basek) << 10;
232 addr = map_2M_page(basek >> 11);
233 if (addr == MAPPING_ERROR) {
234 printk(BIOS_ERR, "Cannot map page: %lx\n", basek >> 11);
238 /* clear memory 2M (limitk - basek) */
239 addr = (void *)(((uint32_t) addr) | ((basek & 0x7ff) << 10));
240 memset(addr, 0, size);
243 static void init_ecc_memory(unsigned node_id)
245 unsigned long startk, begink, endk;
247 struct mtrr_state mtrr_state;
249 device_t f1_dev, f2_dev, f3_dev;
250 int enable_scrubbing;
253 f1_dev = dev_find_slot(0, PCI_DEVFN(0x18 + node_id, 1));
255 die("Cannot find cpu function 1\n");
257 f2_dev = dev_find_slot(0, PCI_DEVFN(0x18 + node_id, 2));
259 die("Cannot find cpu function 2\n");
261 f3_dev = dev_find_slot(0, PCI_DEVFN(0x18 + node_id, 3));
263 die("Cannot find cpu function 3\n");
266 /* See if we scrubbing should be enabled */
267 enable_scrubbing = 1;
268 if( get_option(&enable_scrubbing, "hw_scrubber") < 0 )
270 enable_scrubbing = CONFIG_HW_SCRUBBER;
273 /* Enable cache scrubbing at the lowest possible rate */
274 if (enable_scrubbing) {
275 pci_write_config32(f3_dev, SCRUB_CONTROL,
276 (SCRUB_84ms << 16) | (SCRUB_84ms << 8) |
279 pci_write_config32(f3_dev, SCRUB_CONTROL,
280 (SCRUB_NONE << 16) | (SCRUB_NONE << 8) |
282 printk(BIOS_DEBUG, "Scrubbing Disabled\n");
285 /* If ecc support is not enabled don't touch memory */
286 dcl = pci_read_config32(f2_dev, DRAM_CONFIG_LOW);
287 if (!(dcl & DCL_DimmEccEn)) {
288 printk(BIOS_DEBUG, "ECC Disabled\n");
293 (pci_read_config32(f1_dev, 0x40 + (node_id * 8)) & 0xffff0000) >> 2;
295 ((pci_read_config32(f1_dev, 0x44 + (node_id * 8)) & 0xffff0000) >>
298 #if CONFIG_HW_MEM_HOLE_SIZEK != 0
299 unsigned long hole_startk = 0;
301 #if CONFIG_K8_REV_F_SUPPORT == 0
302 if (!is_cpu_pre_e0()) {
306 val = pci_read_config32(f1_dev, 0xf0);
308 hole_startk = ((val & (0xff << 24)) >> 10);
310 #if CONFIG_K8_REV_F_SUPPORT == 0
315 /* Don't start too early */
317 if (begink < (CONFIG_RAMTOP >> 10)) {
318 begink = (CONFIG_RAMTOP >> 10);
321 printk(BIOS_DEBUG, "Clearing memory %luK - %luK: ", begink, endk);
323 /* Save the normal state */
324 save_mtrr_state(&mtrr_state);
326 /* Switch to the init ecc state */
327 set_init_ecc_mtrrs();
330 /* Walk through 2M chunks and zero them */
331 #if CONFIG_HW_MEM_HOLE_SIZEK != 0
332 /* here hole_startk can not be equal to begink, never. Also hole_startk is in 2M boundary, 64M? */
333 if ((hole_startk != 0)
334 && ((begink < hole_startk) && (endk > (4 * 1024 * 1024)))) {
335 for (basek = begink; basek < hole_startk;
336 basek = ((basek + ZERO_CHUNK_KB) & ~(ZERO_CHUNK_KB - 1))) {
337 clear_2M_ram(basek, &mtrr_state);
339 for (basek = 4 * 1024 * 1024; basek < endk;
340 basek = ((basek + ZERO_CHUNK_KB) & ~(ZERO_CHUNK_KB - 1))) {
341 clear_2M_ram(basek, &mtrr_state);
345 for (basek = begink; basek < endk;
346 basek = ((basek + ZERO_CHUNK_KB) & ~(ZERO_CHUNK_KB - 1))) {
347 clear_2M_ram(basek, &mtrr_state);
350 /* Restore the normal state */
352 restore_mtrr_state(&mtrr_state);
355 /* Set the scrub base address registers */
356 pci_write_config32(f3_dev, SCRUB_ADDR_LOW, startk << 10);
357 pci_write_config32(f3_dev, SCRUB_ADDR_HIGH, startk >> 22);
359 /* Enable the scrubber? */
360 if (enable_scrubbing) {
361 /* Enable scrubbing at the lowest possible rate */
362 pci_write_config32(f3_dev, SCRUB_CONTROL,
363 (SCRUB_84ms << 16) | (SCRUB_84ms << 8) |
367 printk(BIOS_DEBUG, " done\n");
370 static inline void k8_errata(void)
373 #if CONFIG_K8_REV_F_SUPPORT == 0
374 if (is_cpu_pre_c0()) {
376 msr = rdmsr(HWCR_MSR);
378 wrmsr(HWCR_MSR, msr);
381 msr = rdmsr_amd(BU_CFG_MSR);
382 msr.hi |= (1 << (45 - 32));
383 wrmsr_amd(BU_CFG_MSR, msr);
386 msr = rdmsr_amd(DC_CFG_MSR);
388 wrmsr_amd(DC_CFG_MSR, msr);
393 if (!is_cpu_pre_c0() && is_cpu_pre_d0()) {
394 msr = rdmsr_amd(DC_CFG_MSR);
396 wrmsr_amd(DC_CFG_MSR, msr);
400 if (is_cpu_pre_d0()) {
401 msr = rdmsr_amd(IC_CFG_MSR);
403 wrmsr_amd(IC_CFG_MSR, msr);
406 /* Erratum 91 prefetch miss is handled in the kernel */
408 /* Erratum 106 ... */
409 msr = rdmsr_amd(LS_CFG_MSR);
411 wrmsr_amd(LS_CFG_MSR, msr);
413 /* Erratum 107 ... */
414 msr = rdmsr_amd(BU_CFG_MSR);
415 msr.hi |= 1 << (43 - 32);
416 wrmsr_amd(BU_CFG_MSR, msr);
419 /* This erratum applies to D0 thru E6 revisions
420 * Revision F and later are unaffected. There are two fixes
421 * depending on processor revision.
424 /* Erratum 110 ... */
425 msr = rdmsr_amd(CPU_ID_HYPER_EXT_FEATURES);
427 wrmsr_amd(CPU_ID_HYPER_EXT_FEATURES, msr);
430 if (!is_cpu_pre_e0())
432 /* Erratum 110 ... */
433 msr = rdmsr_amd(CPU_ID_EXT_FEATURES_MSR);
435 wrmsr_amd(CPU_ID_EXT_FEATURES_MSR, msr);
440 #if CONFIG_K8_REV_F_SUPPORT == 0
441 /* I can't touch this msr on early buggy cpus */
442 if (!is_cpu_pre_b3())
445 msr = rdmsr(NB_CFG_MSR);
447 #if CONFIG_K8_REV_F_SUPPORT == 0
448 if (!is_cpu_pre_c0() && is_cpu_pre_d0()) {
449 /* D0 later don't need it */
450 /* Erratum 86 Disable data masking on C0 and
451 * later processor revs.
452 * FIXME this is only needed if ECC is enabled.
454 msr.hi |= 1 << (36 - 32);
458 /* Erratum 89 is mistakenly labeled as 88 in AMD pub #25759
459 * It is correctly labeled as 89 on page 49 of the document
460 * and in AMD pub#33610
464 /* This supersedes erratum 131; 131 should not be applied with 169
465 * We also need to set some bits in the northbridge, handled in src/northbridge/amdk8/
469 wrmsr(NB_CFG_MSR, msr);
472 msr = rdmsr(HWCR_MSR);
474 wrmsr(HWCR_MSR, msr);
480 static unsigned ehci_debug_addr;
483 static void model_fxx_init(device_t dev)
487 struct node_core_id id;
490 if (!ehci_debug_addr)
491 ehci_debug_addr = get_ehci_debug();
495 /* Turn on caching if we haven't already */
501 set_ehci_debug(ehci_debug_addr);
504 /* Update the microcode */
505 model_fxx_update_microcode(dev->device);
509 /* zero the machine check error status registers */
512 for (i = 0; i < 5; i++) {
513 wrmsr(MCI_STATUS + (i * 4), msr);
520 /* Set the processor name string */
521 init_processor_name();
523 /* Enable the local cpu apics */
526 #if CONFIG_LOGICAL_CPUS == 1
527 u32 siblings = cpuid_ecx(0x80000008) & 0xff;
530 msr = rdmsr_amd(CPU_ID_FEATURES_MSR);
532 wrmsr_amd(CPU_ID_FEATURES_MSR, msr);
534 msr = rdmsr_amd(LOGICAL_CPUS_NUM_MSR);
535 msr.lo = (siblings + 1) << 16;
536 wrmsr_amd(LOGICAL_CPUS_NUM_MSR, msr);
538 msr = rdmsr_amd(CPU_ID_EXT_FEATURES_MSR);
539 msr.hi |= 1 << (33 - 32);
540 wrmsr_amd(CPU_ID_EXT_FEATURES_MSR, msr);
544 id = get_node_core_id(read_nb_cfg_54()); // pre e0 nb_cfg_54 can not be set
546 /* Is this a bad location? In particular can another node prefecth
547 * data from this node before we have initialized it?
550 init_ecc_memory(id.nodeid); // only do it for core 0
552 /* Set SMM base address for this CPU */
553 msr = rdmsr(SMM_BASE_MSR);
554 msr.lo = SMM_BASE - (lapicid() * 0x400);
555 wrmsr(SMM_BASE_MSR, msr);
557 /* Enable the SMM memory window */
558 msr = rdmsr(SMM_MASK_MSR);
559 msr.lo |= (1 << 0); /* Enable ASEG SMRAM Range */
560 wrmsr(SMM_MASK_MSR, msr);
562 /* Set SMMLOCK to avoid exploits messing with SMM */
563 msr = rdmsr(HWCR_MSR);
565 wrmsr(HWCR_MSR, msr);
568 static struct device_operations cpu_dev_ops = {
569 .init = model_fxx_init,
572 static struct cpu_device_id cpu_table[] = {
573 #if CONFIG_K8_REV_F_SUPPORT == 0
574 { X86_VENDOR_AMD, 0xf40 }, /* SH-B0 (socket 754) */
575 { X86_VENDOR_AMD, 0xf50 }, /* SH-B0 (socket 940) */
576 { X86_VENDOR_AMD, 0xf51 }, /* SH-B3 (socket 940) */
577 { X86_VENDOR_AMD, 0xf58 }, /* SH-C0 (socket 940) */
578 { X86_VENDOR_AMD, 0xf48 }, /* SH-C0 (socket 754) */
579 { X86_VENDOR_AMD, 0xf5a }, /* SH-CG (socket 940) */
580 { X86_VENDOR_AMD, 0xf4a }, /* SH-CG (socket 754) */
581 { X86_VENDOR_AMD, 0xf7a }, /* SH-CG (socket 939) */
582 { X86_VENDOR_AMD, 0xfc0 }, /* DH-CG (socket 754) */
583 { X86_VENDOR_AMD, 0xfe0 }, /* DH-CG (socket 754) */
584 { X86_VENDOR_AMD, 0xff0 }, /* DH-CG (socket 939) */
585 { X86_VENDOR_AMD, 0xf82 }, /* CH-CG (socket 754) */
586 { X86_VENDOR_AMD, 0xfb2 }, /* CH-CG (socket 939) */
589 { X86_VENDOR_AMD, 0x10f50 }, /* SH-D0 (socket 940) */
590 { X86_VENDOR_AMD, 0x10f40 }, /* SH-D0 (socket 754) */
591 { X86_VENDOR_AMD, 0x10f70 }, /* SH-D0 (socket 939) */
592 { X86_VENDOR_AMD, 0x10fc0 }, /* DH-D0 (socket 754) */
593 { X86_VENDOR_AMD, 0x10ff0 }, /* DH-D0 (socket 939) */
594 { X86_VENDOR_AMD, 0x10f80 }, /* CH-D0 (socket 754) */
595 { X86_VENDOR_AMD, 0x10fb0 }, /* CH-D0 (socket 939) */
598 { X86_VENDOR_AMD, 0x20f50 }, /* SH-E0 */
599 { X86_VENDOR_AMD, 0x20f40 },
600 { X86_VENDOR_AMD, 0x20f70 },
601 { X86_VENDOR_AMD, 0x20fc0 }, /* DH-E3 (socket 754) */
602 { X86_VENDOR_AMD, 0x20ff0 }, /* DH-E3 (socket 939) */
603 { X86_VENDOR_AMD, 0x20f10 }, /* JH-E1 (socket 940) */
604 { X86_VENDOR_AMD, 0x20f51 }, /* SH-E4 (socket 940) */
605 { X86_VENDOR_AMD, 0x20f71 }, /* SH-E4 (socket 939) */
606 { X86_VENDOR_AMD, 0x20fb1 }, /* BH-E4 (socket 939) */
607 { X86_VENDOR_AMD, 0x20f42 }, /* SH-E5 (socket 754) */
608 { X86_VENDOR_AMD, 0x20ff2 }, /* DH-E6 (socket 939) */
609 { X86_VENDOR_AMD, 0x20fc2 }, /* DH-E6 (socket 754) */
610 { X86_VENDOR_AMD, 0x20f12 }, /* JH-E6 (socket 940) */
611 { X86_VENDOR_AMD, 0x20f32 }, /* JH-E6 (socket 939) */
612 { X86_VENDOR_AMD, 0x30ff2 }, /* E4 ? */
615 #if CONFIG_K8_REV_F_SUPPORT == 1
619 * See Revision Guide for AMD NPT Family 0Fh Processors,
620 * Publication #33610, Revision: 3.30, February 2008.
622 * http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/33610.pdf
624 { X86_VENDOR_AMD, 0x40f50 }, /* SH-F0 (socket F/1207) */
625 { X86_VENDOR_AMD, 0x40f70 }, /* SH-F0 (socket AM2) */
626 { X86_VENDOR_AMD, 0x40f40 }, /* SH-F0 (socket S1g1) */
627 { X86_VENDOR_AMD, 0x40f11 }, /* JH-F1 (socket F/1207) */
628 { X86_VENDOR_AMD, 0x40f31 }, /* JH-F1 (socket AM2) */
629 { X86_VENDOR_AMD, 0x40f01 }, /* JH-F1 (socket S1g1) */
631 { X86_VENDOR_AMD, 0x40f12 }, /* JH-F2 (socket F/1207) */
632 { X86_VENDOR_AMD, 0x40f32 }, /* JH-F2 (socket AM2) */
633 { X86_VENDOR_AMD, 0x40fb2 }, /* BH-F2 (socket AM2) */
634 { X86_VENDOR_AMD, 0x40f82 }, /* BH-F2 (socket S1g1) */
635 { X86_VENDOR_AMD, 0x40ff2 }, /* DH-F2 (socket AM2) */
636 { X86_VENDOR_AMD, 0x50ff2 }, /* DH-F2 (socket AM2) */
637 { X86_VENDOR_AMD, 0x40fc2 }, /* DH-F2 (socket S1g1) */
638 { X86_VENDOR_AMD, 0x40f13 }, /* JH-F3 (socket F/1207) */
639 { X86_VENDOR_AMD, 0x40f33 }, /* JH-F3 (socket AM2) */
640 { X86_VENDOR_AMD, 0x50fd3 }, /* JH-F3 (socket F/1207) */
641 { X86_VENDOR_AMD, 0xc0f13 }, /* JH-F3 (socket F/1207) */
642 { X86_VENDOR_AMD, 0x50ff3 }, /* DH-F3 (socket AM2) */
643 { X86_VENDOR_AMD, 0x60fb1 }, /* BH-G1 (socket AM2) */
644 { X86_VENDOR_AMD, 0x60f81 }, /* BH-G1 (socket S1g1) */
645 { X86_VENDOR_AMD, 0x60fb2 }, /* BH-G2 (socket AM2) */
646 { X86_VENDOR_AMD, 0x60f82 }, /* BH-G2 (socket S1g1) */
647 { X86_VENDOR_AMD, 0x70ff1 }, /* DH-G1 (socket AM2) */
648 { X86_VENDOR_AMD, 0x60ff2 }, /* DH-G2 (socket AM2) */
649 { X86_VENDOR_AMD, 0x70ff2 }, /* DH-G2 (socket AM2) */
650 { X86_VENDOR_AMD, 0x60fc2 }, /* DH-G2 (socket S1g1) */
651 { X86_VENDOR_AMD, 0x70fc2 }, /* DH-G2 (socket S1g1) */
657 static const struct cpu_driver model_fxx __cpu_driver = {
659 .id_table = cpu_table,