C++ rdmsr函数代码示例

static void generate_P_state_entries(int core, int cores_per_package)
{
	int ratio_min, ratio_max, ratio_turbo, ratio_step, ratio_range_2;
	int coord_type, power_max, power_unit, num_entries;
	int ratio, power, clock, clock_max;
	int vid, vid_turbo, vid_min, vid_max, vid_range_2;
	u32 control_status;
	const struct pattrs *pattrs = pattrs_get();
	msr_t msr;

	/* Inputs from CPU attributes */
	ratio_max = pattrs->iacore_ratios[IACORE_MAX];
	ratio_min = pattrs->iacore_ratios[IACORE_LFM];
	vid_max = pattrs->iacore_vids[IACORE_MAX];
	vid_min = pattrs->iacore_vids[IACORE_LFM];

	/* Hardware coordination of P-states */
	coord_type = HW_ALL;

	/* Max Non-Turbo Frequency */
	clock_max = (ratio_max * pattrs->bclk_khz) / 1000;

	/* Calculate CPU TDP in mW */
	msr = rdmsr(MSR_PKG_POWER_SKU_UNIT);
	power_unit = 1 << (msr.lo & 0xf);
	msr = rdmsr(MSR_PKG_POWER_LIMIT);
	power_max = ((msr.lo & 0x7fff) / power_unit) * 1000;

	/* Write _PCT indicating use of FFixedHW */
	acpigen_write_empty_PCT();

	/* Write _PPC with NVS specified limit on supported P-state */
	acpigen_write_PPC_NVS();

	/* Write PSD indicating configured coordination type */
	acpigen_write_PSD_package(core, 1, coord_type);

	/* Add P-state entries in _PSS table */
	acpigen_write_name("_PSS");

	/* Determine ratio points */
	ratio_step = 1;
	num_entries = (ratio_max - ratio_min) / ratio_step;
	while (num_entries > 15) { /* ACPI max is 15 ratios */
		ratio_step <<= 1;
		num_entries >>= 1;
	}

	/* P[T] is Turbo state if enabled */
	if (get_turbo_state() == TURBO_ENABLED) {
		/* _PSS package count including Turbo */
		acpigen_write_package(num_entries + 2);

		ratio_turbo = pattrs->iacore_ratios[IACORE_TURBO];
		vid_turbo = pattrs->iacore_vids[IACORE_TURBO];
		control_status = (ratio_turbo << 8) | vid_turbo;

		/* Add entry for Turbo ratio */
		acpigen_write_PSS_package(
			clock_max + 1,		/*MHz*/
			power_max,		/*mW*/
			10,			/*lat1*/
			10,			/*lat2*/
			control_status,		/*control*/
			control_status);	/*status*/
	} else {
		/* _PSS package count without Turbo */
		acpigen_write_package(num_entries + 1);
		ratio_turbo = ratio_max;
		vid_turbo = vid_max;
	}

	/* First regular entry is max non-turbo ratio */
	control_status = (ratio_max << 8) | vid_max;
	acpigen_write_PSS_package(
		clock_max,		/*MHz*/
		power_max,		/*mW*/
		10,			/*lat1*/
		10,			/*lat2*/
		control_status,		/*control */
		control_status);	/*status*/

	/* Set up ratio and vid ranges for VID calculation */
	ratio_range_2 = (ratio_turbo - ratio_min) * 2;
	vid_range_2 = (vid_turbo - vid_min) * 2;

	/* Generate the remaining entries */
	for (ratio = ratio_min + ((num_entries - 1) * ratio_step);
	     ratio >= ratio_min; ratio -= ratio_step) {

		/* Calculate VID for this ratio */
		vid = ((ratio - ratio_min) * vid_range_2) /
			ratio_range_2 + vid_min;
		/* Round up if remainder */
		if (((ratio - ratio_min) * vid_range_2) % ratio_range_2)
			vid++;

		/* Calculate power at this ratio */
		power = calculate_power(power_max, ratio_max, ratio);
		clock = (ratio * pattrs->bclk_khz) / 1000;
//.........这里部分代码省略.........

/*
 * AP cpu's call this to sync up protected mode.
 *
 * WARNING! %gs is not set up on entry.  This routine sets up %gs.
 */
void
init_secondary(void)
{
	int	gsel_tss;
	int	x, myid = bootAP;
	u_int64_t msr, cr0;
	struct mdglobaldata *md;
	struct privatespace *ps;

	ps = CPU_prvspace[myid];

	gdt_segs[GPROC0_SEL].ssd_base =
		(long) &ps->mdglobaldata.gd_common_tss;
	ps->mdglobaldata.mi.gd_prvspace = ps;

	/* We fill the 32-bit segment descriptors */
	for (x = 0; x < NGDT; x++) {
		if (x != GPROC0_SEL && x != (GPROC0_SEL + 1))
			ssdtosd(&gdt_segs[x], &gdt[myid * NGDT + x]);
	}
	/* And now a 64-bit one */
	ssdtosyssd(&gdt_segs[GPROC0_SEL],
	    (struct system_segment_descriptor *)&gdt[myid * NGDT + GPROC0_SEL]);

	r_gdt.rd_limit = NGDT * sizeof(gdt[0]) - 1;
	r_gdt.rd_base = (long) &gdt[myid * NGDT];
	lgdt(&r_gdt);			/* does magic intra-segment return */

	/* lgdt() destroys the GSBASE value, so we load GSBASE after lgdt() */
	wrmsr(MSR_FSBASE, 0);		/* User value */
	wrmsr(MSR_GSBASE, (u_int64_t)ps);
	wrmsr(MSR_KGSBASE, 0);		/* XXX User value while we're in the kernel */

	lidt(&r_idt_arr[mdcpu->mi.gd_cpuid]);

#if 0
	lldt(_default_ldt);
	mdcpu->gd_currentldt = _default_ldt;
#endif

	gsel_tss = GSEL(GPROC0_SEL, SEL_KPL);
	gdt[myid * NGDT + GPROC0_SEL].sd_type = SDT_SYSTSS;

	md = mdcpu;	/* loaded through %gs:0 (mdglobaldata.mi.gd_prvspace)*/

	md->gd_common_tss.tss_rsp0 = 0;	/* not used until after switch */
#if 0 /* JG XXX */
	md->gd_common_tss.tss_ioopt = (sizeof md->gd_common_tss) << 16;
#endif
	md->gd_tss_gdt = &gdt[myid * NGDT + GPROC0_SEL];
	md->gd_common_tssd = *md->gd_tss_gdt;

	/* double fault stack */
	md->gd_common_tss.tss_ist1 =
		(long)&md->mi.gd_prvspace->idlestack[
			sizeof(md->mi.gd_prvspace->idlestack)];

	ltr(gsel_tss);

	/*
	 * Set to a known state:
	 * Set by mpboot.s: CR0_PG, CR0_PE
	 * Set by cpu_setregs: CR0_NE, CR0_MP, CR0_TS, CR0_WP, CR0_AM
	 */
	cr0 = rcr0();
	cr0 &= ~(CR0_CD | CR0_NW | CR0_EM);
	load_cr0(cr0);

	/* Set up the fast syscall stuff */
	msr = rdmsr(MSR_EFER) | EFER_SCE;
	wrmsr(MSR_EFER, msr);
	wrmsr(MSR_LSTAR, (u_int64_t)IDTVEC(fast_syscall));
	wrmsr(MSR_CSTAR, (u_int64_t)IDTVEC(fast_syscall32));
	msr = ((u_int64_t)GSEL(GCODE_SEL, SEL_KPL) << 32) |
	      ((u_int64_t)GSEL(GUCODE32_SEL, SEL_UPL) << 48);
	wrmsr(MSR_STAR, msr);
	wrmsr(MSR_SF_MASK, PSL_NT|PSL_T|PSL_I|PSL_C|PSL_D|PSL_IOPL);

	pmap_set_opt();		/* PSE/4MB pages, etc */
	pmap_init_pat();	/* Page Attribute Table */

	/* set up CPU registers and state */
	cpu_setregs();

	/* set up SSE/NX registers */
	initializecpu(myid);

	/* set up FPU state on the AP */
	npxinit();

	/* disable the APIC, just to be SURE */
	lapic->svr &= ~APIC_SVR_ENABLE;
}

/**
 * This is Black Magic DRAM timing juju[1].
 *
 * DRAM delay depends on CPU clock, memory bus clock, memory bus loading,
 * memory bus termination, your middle initial (ha! caught you!), GeodeLink
 * clock rate, and DRAM timing specifications.
 *
 * From this the code computes a number which is "known to work". No,
 * hardware is not an exact science. And, finally, if an FS2 (JTAG debugger)
 * is hooked up, then just don't do anything. This code was written by a master
 * of the Dark Arts at AMD and should not be modified in any way.
 *
 * [1] (http://www.thefreedictionary.com/juju)
 *
 * @param dimm0 The SMBus address of DIMM 0 (mainboard dependent).
 * @param dimm1 The SMBus address of DIMM 1 (mainboard dependent).
 * @param terminated The bus is terminated. (mainboard dependent).
 */
static void SetDelayControl(u8 dimm0, u8 dimm1, int terminated)
{
	u32 glspeed;
	u8 spdbyte0, spdbyte1, dimms, i;
	msr_t msr;

	glspeed = GeodeLinkSpeed();

	/* Fix delay controls for DM and IM arrays. */
	for (i = 0; i < ARRAY_SIZE(delay_msr_table); i++)
		wrmsr(delay_msr_table[i].index, delay_msr_table[i].msr);

	msr = rdmsr(GLCP_FIFOCTL);
	msr.hi = 0x00000005;
	wrmsr(GLCP_FIFOCTL, msr);

	/* Enable setting. */
	msr.hi = 0;
	msr.lo = 0x00000001;
	wrmsr(CPU_BC_MSS_ARRAY_CTL_ENA, msr);

	/* Debug Delay Control setup check.
	 * Leave it alone if it has been setup. FS2 or something is here.
	 */
	msr = rdmsr(GLCP_DELAY_CONTROLS);
	if (msr.lo & ~(DELAY_LOWER_STATUS_MASK))
		return;

	/* Delay Controls based on DIMM loading. UGH!
	 * Number of devices = module width (SPD 6) / device width (SPD 13)
	 *                     * physical banks (SPD 5)
	 *
	 * Note: We only support a module width of 64.
	 */
	dimms = 0;
	spdbyte0 = spd_read_byte(dimm0, SPD_PRIMARY_SDRAM_WIDTH);
	if (spdbyte0 != 0xFF) {
		dimms++;
		spdbyte0 = (u8)64 / spdbyte0 *
			   (u8)(spd_read_byte(dimm0, SPD_NUM_DIMM_BANKS));
	} else {
		spdbyte0 = 0;
	}

	spdbyte1 = spd_read_byte(dimm1, SPD_PRIMARY_SDRAM_WIDTH);
	if (spdbyte1 != 0xFF) {
		dimms++;
		spdbyte1 = (u8)64 / spdbyte1 *
			   (u8)(spd_read_byte(dimm1, SPD_NUM_DIMM_BANKS));
	} else {
		spdbyte1 = 0;
	}

	/* Zero GLCP_DELAY_CONTROLS MSR */
	msr.hi = msr.lo = 0;

	/* Save some power, disable clock to second DIMM if it is empty. */
	if (spdbyte1 == 0)
		msr.hi |= DELAY_UPPER_DISABLE_CLK135;

	spdbyte0 += spdbyte1;

	if ((dimms == 1) && (terminated == DRAM_TERMINATED)) {
		msr.hi = 0xF2F100FF;
		msr.lo = 0x56960004;
	} else for (i = 0; i < ARRAY_SIZE(delay_control_table); i++) {
		if ((dimms == delay_control_table[i].dimms) &&
		    (spdbyte0 <= delay_control_table[i].devices)) {
			if (glspeed < 334) {
				msr.hi |= delay_control_table[i].slow_hi;
				msr.lo |= delay_control_table[i].slow_low;
			} else {
				msr.hi |= delay_control_table[i].fast_hi;
				msr.lo |= delay_control_table[i].fast_low;
			}
			break;
		}
	}
	wrmsr(GLCP_DELAY_CONTROLS, msr);
}

void
via_nano_setup(struct cpu_info *ci)
{
	u_int32_t regs[4], val;
	u_int64_t msreg;
	int model = (ci->ci_signature >> 4) & 15;

	if (model >= 9) {
		CPUID(0xC0000000, regs[0], regs[1], regs[2], regs[3]);
		val = regs[0];
		if (val >= 0xC0000001) {
			CPUID(0xC0000001, regs[0], regs[1], regs[2], regs[3]);
			val = regs[3];
		} else
			val = 0;

		if (val & (C3_CPUID_HAS_RNG | C3_CPUID_HAS_ACE))
			printf("%s:", ci->ci_dev->dv_xname);

		/* Enable RNG if present and disabled */
		if (val & C3_CPUID_HAS_RNG) {
			extern int viac3_rnd_present;

			if (!(val & C3_CPUID_DO_RNG)) {
				msreg = rdmsr(0x110B);
				msreg |= 0x40;
				wrmsr(0x110B, msreg);
			}
			viac3_rnd_present = 1;
			printf(" RNG");
		}

		/* Enable AES engine if present and disabled */
		if (val & C3_CPUID_HAS_ACE) {
#ifdef CRYPTO
			if (!(val & C3_CPUID_DO_ACE)) {
				msreg = rdmsr(0x1107);
				msreg |= (0x01 << 28);
				wrmsr(0x1107, msreg);
			}
			amd64_has_xcrypt |= C3_HAS_AES;
#endif /* CRYPTO */
			printf(" AES");
		}

		/* Enable ACE2 engine if present and disabled */
		if (val & C3_CPUID_HAS_ACE2) {
#ifdef CRYPTO
			if (!(val & C3_CPUID_DO_ACE2)) {
				msreg = rdmsr(0x1107);
				msreg |= (0x01 << 28);
				wrmsr(0x1107, msreg);
			}
			amd64_has_xcrypt |= C3_HAS_AESCTR;
#endif /* CRYPTO */
			printf(" AES-CTR");
		}

		/* Enable SHA engine if present and disabled */
		if (val & C3_CPUID_HAS_PHE) {
#ifdef CRYPTO
			if (!(val & C3_CPUID_DO_PHE)) {
				msreg = rdmsr(0x1107);
				msreg |= (0x01 << 28/**/);
				wrmsr(0x1107, msreg);
			}
			amd64_has_xcrypt |= C3_HAS_SHA;
#endif /* CRYPTO */
			printf(" SHA1 SHA256");
		}

		/* Enable MM engine if present and disabled */
		if (val & C3_CPUID_HAS_PMM) {
#ifdef CRYPTO
			if (!(val & C3_CPUID_DO_PMM)) {
				msreg = rdmsr(0x1107);
				msreg |= (0x01 << 28/**/);
				wrmsr(0x1107, msreg);
			}
			amd64_has_xcrypt |= C3_HAS_MM;
#endif /* CRYPTO */
			printf(" RSA");
		}

		printf("\n");
	}
}

static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
{
	u64 misc_enable;

	/* Unmask CPUID levels if masked: */
	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);

		if (misc_enable & MSR_IA32_MISC_ENABLE_LIMIT_CPUID) {
			misc_enable &= ~MSR_IA32_MISC_ENABLE_LIMIT_CPUID;
			wrmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
			c->cpuid_level = cpuid_eax(0);
			get_cpu_cap(c);
		}
	}

	if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
		(c->x86 == 0x6 && c->x86_model >= 0x0e))
		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);

	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64)) {
		unsigned lower_word;

		wrmsr(MSR_IA32_UCODE_REV, 0, 0);
		/* Required by the SDM */
		sync_core();
		rdmsr(MSR_IA32_UCODE_REV, lower_word, c->microcode);
	}

	/*
	 * Atom erratum AAE44/AAF40/AAG38/AAH41:
	 *
	 * A race condition between speculative fetches and invalidating
	 * a large page.  This is worked around in microcode, but we
	 * need the microcode to have already been loaded... so if it is
	 * not, recommend a BIOS update and disable large pages.
	 */
	if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_mask <= 2 &&
	    c->microcode < 0x20e) {
		printk(KERN_WARNING "Atom PSE erratum detected, BIOS microcode update recommended\n");
		clear_cpu_cap(c, X86_FEATURE_PSE);
	}

#ifdef CONFIG_X86_64
	set_cpu_cap(c, X86_FEATURE_SYSENTER32);
#else
	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
	if (c->x86 == 15 && c->x86_cache_alignment == 64)
		c->x86_cache_alignment = 128;
#endif

	/* CPUID workaround for 0F33/0F34 CPU */
	if (c->x86 == 0xF && c->x86_model == 0x3
	    && (c->x86_mask == 0x3 || c->x86_mask == 0x4))
		c->x86_phys_bits = 36;

	/*
	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
	 * with P/T states and does not stop in deep C-states.
	 *
	 * It is also reliable across cores and sockets. (but not across
	 * cabinets - we turn it off in that case explicitly.)
	 */
	if (c->x86_power & (1 << 8)) {
		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
		if (!check_tsc_unstable())
			sched_clock_stable = 1;
	}

	/*
	 * There is a known erratum on Pentium III and Core Solo
	 * and Core Duo CPUs.
	 * " Page with PAT set to WC while associated MTRR is UC
	 *   may consolidate to UC "
	 * Because of this erratum, it is better to stick with
	 * setting WC in MTRR rather than using PAT on these CPUs.
	 *
	 * Enable PAT WC only on P4, Core 2 or later CPUs.
	 */
	if (c->x86 == 6 && c->x86_model < 15)
		clear_cpu_cap(c, X86_FEATURE_PAT);

#ifdef CONFIG_KMEMCHECK
	/*
	 * P4s have a "fast strings" feature which causes single-
	 * stepping REP instructions to only generate a #DB on
	 * cache-line boundaries.
	 *
	 * Ingo Molnar reported a Pentium D (model 6) and a Xeon
	 * (model 2) with the same problem.
	 */
	if (c->x86 == 15) {
		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);

		if (misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING) {
			printk(KERN_INFO "kmemcheck: Disabling fast string operations\n");

			misc_enable &= ~MSR_IA32_MISC_ENABLE_FAST_STRING;
			wrmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
//.........这里部分代码省略.........

void cache_as_ram_main(unsigned long bist, unsigned long cpu_init_detectedx)
{
	struct sys_info *sysinfo = &sysinfo_car;
	static const u8 spd_addr[] = {RC00, DIMM0, DIMM2, 0, 0, DIMM1, DIMM3, 0, 0, };
	u32 bsp_apicid = 0, val;
	msr_t msr;

	timestamp_init(timestamp_get());
	timestamp_add_now(TS_START_ROMSTAGE);

	if (!cpu_init_detectedx && boot_cpu()) {
		/* Nothing special needs to be done to find bus 0 */
		/* Allow the HT devices to be found */
		/* mov bsp to bus 0xff when > 8 nodes */
		set_bsp_node_CHtExtNodeCfgEn();
		enumerate_ht_chain();
		sb7xx_51xx_pci_port80();
	}

	post_code(0x30);

	if (bist == 0) {
		bsp_apicid = init_cpus(cpu_init_detectedx, sysinfo); /* mmconf is inited in init_cpus */
		/* All cores run this but the BSP(node0,core0) is the only core that returns. */
	}

	post_code(0x32);

	enable_rs780_dev8();
	sb7xx_51xx_lpc_init();

	ite_enable_serial(SERIAL_DEV, CONFIG_TTYS0_BASE);
	it8718f_disable_reboot(GPIO_DEV);
	console_init();

//	dump_mem(CONFIG_DCACHE_RAM_BASE+CONFIG_DCACHE_RAM_SIZE-0x200, CONFIG_DCACHE_RAM_BASE+CONFIG_DCACHE_RAM_SIZE);

	/* Halt if there was a built in self test failure */
	report_bist_failure(bist);

	// Load MPB
	val = cpuid_eax(1);
	printk(BIOS_DEBUG, "BSP Family_Model: %08x\n", val);
	printk(BIOS_DEBUG, "*sysinfo range: [%p,%p]\n",sysinfo,sysinfo+1);
	printk(BIOS_DEBUG, "bsp_apicid = %02x\n", bsp_apicid);
	printk(BIOS_DEBUG, "cpu_init_detectedx = %08lx\n", cpu_init_detectedx);

	/* Setup sysinfo defaults */
	set_sysinfo_in_ram(0);

	update_microcode(val);

	post_code(0x33);

	cpuSetAMDMSR(0);
	post_code(0x34);

	amd_ht_init(sysinfo);
	post_code(0x35);

	/* Setup nodes PCI space and start core 0 AP init. */
	finalize_node_setup(sysinfo);

	/* Setup any mainboard PCI settings etc. */
	setup_mb_resource_map();
	post_code(0x36);

	/* wait for all the APs core0 started by finalize_node_setup. */
	/* FIXME: A bunch of cores are going to start output to serial at once.
	   It would be nice to fixup prink spinlocks for ROM XIP mode.
	   I think it could be done by putting the spinlock flag in the cache
	   of the BSP located right after sysinfo.
	 */
	wait_all_core0_started();

#if CONFIG_LOGICAL_CPUS
	/* Core0 on each node is configured. Now setup any additional cores. */
	printk(BIOS_DEBUG, "start_other_cores()\n");
	start_other_cores();
	post_code(0x37);
	wait_all_other_cores_started(bsp_apicid);
#endif

	post_code(0x38);

	/* run _early_setup before soft-reset. */
	rs780_early_setup();
	sb7xx_51xx_early_setup();

#if CONFIG_SET_FIDVID
	msr = rdmsr(0xc0010071);
	printk(BIOS_DEBUG, "\nBegin FIDVID MSR 0xc0010071 0x%08x 0x%08x\n", msr.hi, msr.lo);

	/* FIXME: The sb fid change may survive the warm reset and only
	   need to be done once.*/
	enable_fid_change_on_sb(sysinfo->sbbusn, sysinfo->sbdn);

	post_code(0x39);

	if (!warm_reset_detect(0)) {			// BSP is node 0
//.........这里部分代码省略.........

static void __cpuinit init_amd(struct cpuinfo_x86 *c)
{
	u32 l, h;
	int mbytes = num_physpages >> (20-PAGE_SHIFT);
	int r;

#ifdef CONFIG_SMP
	unsigned long long value;

	/* Disable TLB flush filter by setting HWCR.FFDIS on K8
	 * bit 6 of msr C001_0015
	 *
	 * Errata 63 for SH-B3 steppings
	 * Errata 122 for all steppings (F+ have it disabled by default)
	 */
	if (c->x86 == 15) {
		rdmsrl(MSR_K7_HWCR, value);
		value |= 1 << 6;
		wrmsrl(MSR_K7_HWCR, value);
	}
#endif

	early_init_amd(c);

	/*
	 *	FIXME: We should handle the K5 here. Set up the write
	 *	range and also turn on MSR 83 bits 4 and 31 (write alloc,
	 *	no bus pipeline)
	 */

	/* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
	   3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
	clear_bit(0*32+31, c->x86_capability);
	
	r = get_model_name(c);

	switch(c->x86)
	{
		case 4:
		/*
		 * General Systems BIOSen alias the cpu frequency registers
		 * of the Elan at 0x000df000. Unfortuantly, one of the Linux
		 * drivers subsequently pokes it, and changes the CPU speed.
		 * Workaround : Remove the unneeded alias.
		 */
#define CBAR		(0xfffc) /* Configuration Base Address  (32-bit) */
#define CBAR_ENB	(0x80000000)
#define CBAR_KEY	(0X000000CB)
			if (c->x86_model==9 || c->x86_model == 10) {
				if (inl (CBAR) & CBAR_ENB)
					outl (0 | CBAR_KEY, CBAR);
			}
			break;
		case 5:
			if( c->x86_model < 6 )
			{
				/* Based on AMD doc 20734R - June 2000 */
				if ( c->x86_model == 0 ) {
					clear_bit(X86_FEATURE_APIC, c->x86_capability);
					set_bit(X86_FEATURE_PGE, c->x86_capability);
				}
				break;
			}
			
			if ( c->x86_model == 6 && c->x86_mask == 1 ) {
				const int K6_BUG_LOOP = 1000000;
				int n;
				void (*f_vide)(void);
				unsigned long d, d2;
				
				printk(KERN_INFO "AMD K6 stepping B detected - ");
				
				/*
				 * It looks like AMD fixed the 2.6.2 bug and improved indirect 
				 * calls at the same time.
				 */

				n = K6_BUG_LOOP;
				f_vide = vide;
				rdtscl(d);
				while (n--) 
					f_vide();
				rdtscl(d2);
				d = d2-d;

				if (d > 20*K6_BUG_LOOP) 
					printk("system stability may be impaired when more than 32 MB are used.\n");
				else 
					printk("probably OK (after B9730xxxx).\n");
				printk(KERN_INFO "Please see http://membres.lycos.fr/poulot/k6bug.html\n");
			}

			/* K6 with old style WHCR */
			if (c->x86_model < 8 ||
			   (c->x86_model== 8 && c->x86_mask < 8)) {
				/* We can only write allocate on the low 508Mb */
				if(mbytes>508)
					mbytes=508;

				rdmsr(MSR_K6_WHCR, l, h);
//.........这里部分代码省略.........

static int __init detect_init_APIC (void)
{
	u32 h, l, features;
	extern void get_cpu_vendor(struct cpuinfo_x86*);

	/* Disabled by DMI scan or kernel option? */
	if (dont_enable_local_apic)
		return -1;

	/* Workaround for us being called before identify_cpu(). */
	get_cpu_vendor(&boot_cpu_data);

	switch (boot_cpu_data.x86_vendor) {
	case X86_VENDOR_AMD:
		if (boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1)
			break;
		if (boot_cpu_data.x86 == 15 && cpu_has_apic)
			break;
		goto no_apic;
	case X86_VENDOR_INTEL:
		if (boot_cpu_data.x86 == 6 ||
		    (boot_cpu_data.x86 == 15 && cpu_has_apic) ||
		    (boot_cpu_data.x86 == 5 && cpu_has_apic))
			break;
		goto no_apic;
	default:
		goto no_apic;
	}

	if (!cpu_has_apic) {
		/*
		 * Some BIOSes disable the local APIC in the
		 * APIC_BASE MSR. This can only be done in
		 * software for Intel P6 and AMD K7 (Model > 1).
		 */
		rdmsr(MSR_IA32_APICBASE, l, h);
		if (!(l & MSR_IA32_APICBASE_ENABLE)) {
			printk("Local APIC disabled by BIOS -- reenabling.\n");
			l &= ~MSR_IA32_APICBASE_BASE;
			l |= MSR_IA32_APICBASE_ENABLE | APIC_DEFAULT_PHYS_BASE;
			wrmsr(MSR_IA32_APICBASE, l, h);
		}
	}
	/*
	 * The APIC feature bit should now be enabled
	 * in `cpuid'
	 */
	features = cpuid_edx(1);
	if (!(features & (1 << X86_FEATURE_APIC))) {
		printk("Could not enable APIC!\n");
		return -1;
	}
	set_bit(X86_FEATURE_APIC, &boot_cpu_data.x86_capability);
	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
	if (nmi_watchdog != NMI_NONE)
		nmi_watchdog = NMI_LOCAL_APIC;

	printk("Found and enabled local APIC!\n");

	apic_pm_init1();

	return 0;

no_apic:
	printk("No local APIC present or hardware disabled\n");
	return -1;
}

/*
 *  All traps come here.  It is slower to have all traps call trap()
 *  rather than directly vectoring the handler.  However, this avoids a
 *  lot of code duplication and possible bugs.  The only exception is
 *  VectorSYSCALL.
 *  Trap is called with interrupts disabled via interrupt-gates.
 */
void
trap(Ureg* ureg)
{
    int clockintr, vno, user;
    // cache the previous vno to see what might be causing
    // trouble
    vno = ureg->type;
    uint64_t gsbase = rdmsr(GSbase);
    //if (sce > scx) iprint("====================");
    lastvno = vno;
    if (gsbase < 1ULL<<63)
        die("bogus gsbase");
    Proc *up = externup();
    char buf[ERRMAX];
    Vctl *ctl, *v;

    machp()->perf.intrts = perfticks();
    user = userureg(ureg);
    if(user && (machp()->NIX.nixtype == NIXTC)) {
        up->dbgreg = ureg;
        cycles(&up->kentry);
    }

    clockintr = 0;

    //_pmcupdate(machp());

    if(ctl = vctl[vno]) {
        if(ctl->isintr) {
            machp()->intr++;
            if(vno >= VectorPIC && vno != VectorSYSCALL)
                machp()->lastintr = ctl->Vkey.irq;
        } else if(up)
            up->nqtrap++;

        if(ctl->isr) {
            ctl->isr(vno);
            if(islo())print("trap %d: isr %p enabled interrupts\n", vno, ctl->isr);
        }
        for(v = ctl; v != nil; v = v->next) {
            if(v->f) {
                v->f(ureg, v->a);
                if(islo())print("trap %d: ctlf %p enabled interrupts\n", vno, v->f);
            }
        }
        if(ctl->eoi) {
            ctl->eoi(vno);
            if(islo())print("trap %d: eoi %p enabled interrupts\n", vno, ctl->eoi);
        }

        intrtime(vno);
        if(ctl->isintr) {
            if(ctl->Vkey.irq == IrqCLOCK || ctl->Vkey.irq == IrqTIMER)
                clockintr = 1;

            if (ctl->Vkey.irq == IrqTIMER)
                oprof_alarm_handler(ureg);

            if(up && !clockintr)
                preempted();
        }
    }
    else if(vno < nelem(excname) && user) {
        spllo();
        snprint(buf, sizeof buf, "sys: trap: %s", excname[vno]);
        postnote(up, 1, buf, NDebug);
    }
    else if(vno >= VectorPIC && vno != VectorSYSCALL) {
        /*
         * An unknown interrupt.
         * Check for a default IRQ7. This can happen when
         * the IRQ input goes away before the acknowledge.
         * In this case, a 'default IRQ7' is generated, but
         * the corresponding bit in the ISR isn't set.
         * In fact, just ignore all such interrupts.
         */

        /* clear the interrupt */
        i8259isr(vno);

        iprint("cpu%d: spurious interrupt %d, last %d\n",
               machp()->machno, vno, machp()->lastintr);
        intrtime(vno);
        if(user)
            kexit(ureg);
        return;
    }
    else {
        if(vno == VectorNMI) {
            nmienable();
            if(machp()->machno != 0) {
                iprint("cpu%d: PC %#llx\n",
                       machp()->machno, ureg->ip);
//.........这里部分代码省略.........

//.........这里部分代码省略.........
		iosf_port5a_write(step->reg, step->value);
		break;
	case IOSF_PORT_USHPHY:
		iosf_ushphy_write(step->reg, step->value);
		break;
	case IOSF_PORT_SCC:
		iosf_scc_write(step->reg, step->value);
		break;
	case IOSF_PORT_LPSS:
		iosf_lpss_write(step->reg, step->value);
		break;
	case IOSF_PORT_0xa2:
		iosf_porta2_write(step->reg, step->value);
		break;
	case IOSF_PORT_CCU:
		iosf_ccu_write(step->reg, step->value);
		break;
	case IOSF_PORT_SSUS:
		iosf_ssus_write(step->reg, step->value);
		break;
	default:
		printk(BIOS_DEBUG, "No write support for IOSF port 0x%x.\n",
		       step->id);
		break;
	}
}
#endif


static uint64_t reg_script_read_msr(struct reg_script_context *ctx)
{
#if CONFIG_ARCH_X86
	const struct reg_script *step = reg_script_get_step(ctx);
	msr_t msr = rdmsr(step->reg);
	uint64_t value = msr.hi;
	value = msr.hi;
	value <<= 32;
	value |= msr.lo;
	return value;
#endif
}

static void reg_script_write_msr(struct reg_script_context *ctx)
{
#if CONFIG_ARCH_X86
	const struct reg_script *step = reg_script_get_step(ctx);
	msr_t msr;
	msr.hi = step->value >> 32;
	msr.lo = step->value & 0xffffffff;
	wrmsr(step->reg, msr);
#endif
}

#ifndef __PRE_RAM__
/* Default routine provided for systems without platform specific busses */
const struct reg_script_bus_entry *__attribute__((weak))
	platform_bus_table(size_t *table_entries)
{
	/* No platform bus type table supplied */
	*table_entries = 0;
	return NULL;
}

/* Locate the structure containing the platform specific bus access routines */
static const struct reg_script_bus_entry
	*find_bus(const struct reg_script *step)

static void p4_setup_ctrs(struct op_msrs const * const msrs)
{
	unsigned int i;
	unsigned int low, high;
	unsigned int addr;
	unsigned int stag;

	stag = get_stagger();

	rdmsr(MSR_IA32_MISC_ENABLE, low, high);
	if (! MISC_PMC_ENABLED_P(low)) {
		printk(KERN_ERR "oprofile: P4 PMC not available\n");
		return;
	}

	/* clear the cccrs we will use */
	for (i = 0 ; i < num_counters ; i++) {
		rdmsr(p4_counters[VIRT_CTR(stag, i)].cccr_address, low, high);
		CCCR_CLEAR(low);
		CCCR_SET_REQUIRED_BITS(low);
		wrmsr(p4_counters[VIRT_CTR(stag, i)].cccr_address, low, high);
	}

	/* clear cccrs outside our concern */
	for (i = stag ; i < NUM_UNUSED_CCCRS ; i += addr_increment()) {
		rdmsr(p4_unused_cccr[i], low, high);
		CCCR_CLEAR(low);
		CCCR_SET_REQUIRED_BITS(low);
		wrmsr(p4_unused_cccr[i], low, high);
	}

	/* clear all escrs (including those outside our concern) */
	for (addr = MSR_P4_BSU_ESCR0 + stag;
	     addr <= MSR_P4_SSU_ESCR0; addr += addr_increment()) { 
		wrmsr(addr, 0, 0);
	}
	
	for (addr = MSR_P4_MS_ESCR0 + stag;
	     addr <= MSR_P4_TC_ESCR1; addr += addr_increment()){ 
		wrmsr(addr, 0, 0);
	}
	
	for (addr = MSR_P4_IX_ESCR0 + stag;
	     addr <= MSR_P4_CRU_ESCR3; addr += addr_increment()){ 
		wrmsr(addr, 0, 0);
	}

	if (num_counters == NUM_COUNTERS_NON_HT) {		
		wrmsr(MSR_P4_CRU_ESCR4, 0, 0);
		wrmsr(MSR_P4_CRU_ESCR5, 0, 0);
	} else if (stag == 0) {
		wrmsr(MSR_P4_CRU_ESCR4, 0, 0);
	} else {
		wrmsr(MSR_P4_CRU_ESCR5, 0, 0);
	}		
	
	/* setup all counters */
	for (i = 0 ; i < num_counters ; ++i) {
		if (counter_config[i].event) {
			reset_value[i] = counter_config[i].count;
			pmc_setup_one_p4_counter(i);
			CTR_WRITE(counter_config[i].count, VIRT_CTR(stag, i));
		} else {
			reset_value[i] = 0;
		}
	}
}

static uint64_t
getctr(uint32_t regno)
{
	return rdmsr(regno + PerfCtrbase);
}

static int
pn_decode_acpi(device_t dev, device_t perf_dev)
{
	int i, j, n;
	uint64_t status;
	uint32_t ctrl;
	u_int cpuid;
	u_int regs[4];
	struct pn_softc *sc;
	struct powernow_state state;
	struct cf_setting sets[POWERNOW_MAX_STATES];
	int count = POWERNOW_MAX_STATES;
	int type;
	int rv;

	if (perf_dev == NULL)
		return (ENXIO);

	rv = CPUFREQ_DRV_SETTINGS(perf_dev, sets, &count);
	if (rv)
		return (ENXIO);
	rv = CPUFREQ_DRV_TYPE(perf_dev, &type);
	if (rv || (type & CPUFREQ_FLAG_INFO_ONLY) == 0)
		return (ENXIO);

	sc = device_get_softc(dev);

	do_cpuid(0x80000001, regs);
	cpuid = regs[0];
	if ((cpuid & 0xfff) == 0x760)
		sc->errata |= A0_ERRATA;

	ctrl = 0;
	sc->sgtc = 0;
	for (n = 0, i = 0; i < count; ++i) {
		ctrl = sets[i].spec[PX_SPEC_CONTROL];
		switch (sc->pn_type) {
		case PN7_TYPE:
			state.fid = ACPI_PN7_CTRL_TO_FID(ctrl);
			state.vid = ACPI_PN7_CTRL_TO_VID(ctrl);
			if ((sc->errata & A0_ERRATA) &&
			    (pn7_fid_to_mult[state.fid] % 10) == 5)
				continue;
			break;
		case PN8_TYPE:
			state.fid = ACPI_PN8_CTRL_TO_FID(ctrl);
			state.vid = ACPI_PN8_CTRL_TO_VID(ctrl);
			break;
		}
		state.freq = sets[i].freq * 1000;
		state.power = sets[i].power;

		j = n;
		while (j > 0 && sc->powernow_states[j - 1].freq < state.freq) {
			memcpy(&sc->powernow_states[j],
			    &sc->powernow_states[j - 1],
			    sizeof(struct powernow_state));
			--j;
		}
		memcpy(&sc->powernow_states[j], &state,
		    sizeof(struct powernow_state));
		++n;
	}

	sc->powernow_max_states = n;
	state = sc->powernow_states[0];
	status = rdmsr(MSR_AMDK7_FIDVID_STATUS);

	switch (sc->pn_type) {
	case PN7_TYPE:
		sc->sgtc = ACPI_PN7_CTRL_TO_SGTC(ctrl);
		/*
		 * XXX Some bios forget the max frequency!
		 * This maybe indicates we have the wrong tables.  Therefore,
		 * don't implement a quirk, but fallback to BIOS legacy
		 * tables instead.
		 */
		if (PN7_STA_MFID(status) != state.fid) {
			device_printf(dev, "ACPI MAX frequency not found\n");
			return (EINVAL);
		}
		sc->fsb = state.freq / 100 / pn7_fid_to_mult[state.fid];
		break;
	case PN8_TYPE:
		sc->vst = ACPI_PN8_CTRL_TO_VST(ctrl),
		sc->mvs = ACPI_PN8_CTRL_TO_MVS(ctrl),
		sc->pll = ACPI_PN8_CTRL_TO_PLL(ctrl),
		sc->rvo = ACPI_PN8_CTRL_TO_RVO(ctrl),
		sc->irt = ACPI_PN8_CTRL_TO_IRT(ctrl);
		sc->low = 0; /* XXX */

		/*
		 * powernow k8 supports only one low frequency.
		 */
		if (sc->powernow_max_states >= 2 &&
		    (sc->powernow_states[sc->powernow_max_states - 2].fid < 8))
			return (EINVAL);
		sc->fsb = state.freq / 100 / pn8_fid_to_mult[state.fid];
		break;
	}
//.........这里部分代码省略.........

void cache_as_ram_main(unsigned long bist, unsigned long cpu_init_detectedx)
{

	struct sys_info *sysinfo = (struct sys_info *)(CONFIG_DCACHE_RAM_BASE + CONFIG_DCACHE_RAM_SIZE - CONFIG_DCACHE_RAM_GLOBAL_VAR_SIZE);
	u32 bsp_apicid = 0;
	u32 val;
	msr_t msr;

	if (!cpu_init_detectedx && boot_cpu()) {
		/* Nothing special needs to be done to find bus 0 */
		/* Allow the HT devices to be found */
		/* mov bsp to bus 0xff when > 8 nodes */
		set_bsp_node_CHtExtNodeCfgEn();
		enumerate_ht_chain();

		/* Setup the rom access for 4M */
		amd8111_enable_rom();
	}

	post_code(0x30);

	if (bist == 0) {
		bsp_apicid = init_cpus(cpu_init_detectedx, sysinfo); /* mmconf is inited in init_cpus */
		/* All cores run this but the BSP(node0,core0) is the only core that returns. */
	}

	post_code(0x32);

	w83627hf_enable_serial(SERIAL_DEV, CONFIG_TTYS0_BASE);
	uart_init();
	console_init();
	printk(BIOS_DEBUG, "\n");

//	dump_mem(CONFIG_DCACHE_RAM_BASE+CONFIG_DCACHE_RAM_SIZE-0x200, CONFIG_DCACHE_RAM_BASE+CONFIG_DCACHE_RAM_SIZE);

	/* Halt if there was a built in self test failure */
	report_bist_failure(bist);

	// Load MPB
	val = cpuid_eax(1);
	printk(BIOS_DEBUG, "BSP Family_Model: %08x \n", val);
	printk(BIOS_DEBUG, "*sysinfo range: [%p,%p]\n",sysinfo,sysinfo+1);
	printk(BIOS_DEBUG, "bsp_apicid = %02x \n", bsp_apicid);
	printk(BIOS_DEBUG, "cpu_init_detectedx = %08lx \n", cpu_init_detectedx);

	/* Setup sysinfo defaults */
	set_sysinfo_in_ram(0);

	update_microcode(val);
	post_code(0x33);

	cpuSetAMDMSR();
	post_code(0x34);

	amd_ht_init(sysinfo);
	post_code(0x35);

	/* Setup nodes PCI space and start core 0 AP init. */
	finalize_node_setup(sysinfo);

	/* Setup any mainboard PCI settings etc. */
	setup_mb_resource_map();
	post_code(0x36);

	/* wait for all the APs core0 started by finalize_node_setup. */
	/* FIXME: A bunch of cores are going to start output to serial at once.
	   It would be nice to fixup prink spinlocks for ROM XIP mode.
	   I think it could be done by putting the spinlock flag in the cache
	   of the BSP located right after sysinfo.
	 */
	wait_all_core0_started();

 #if CONFIG_LOGICAL_CPUS==1
	/* Core0 on each node is configured. Now setup any additional cores. */
	printk(BIOS_DEBUG, "start_other_cores()\n");
	start_other_cores();
	post_code(0x37);
	wait_all_other_cores_started(bsp_apicid);
 #endif

	post_code(0x38);

 #if SET_FIDVID == 1
	msr = rdmsr(0xc0010071);
	printk(BIOS_DEBUG, "\nBegin FIDVID MSR 0xc0010071 0x%08x 0x%08x \n", msr.hi, msr.lo);

	/* FIXME: The sb fid change may survive the warm reset and only
	   need to be done once.*/
	enable_fid_change_on_sb(sysinfo->sbbusn, sysinfo->sbdn);

	post_code(0x39);

	if (!warm_reset_detect(0)) {			// BSP is node 0
		init_fidvid_bsp(bsp_apicid, sysinfo->nodes);
	} else {
		init_fidvid_stage2(bsp_apicid, 0);	// BSP is node 0
	}

	post_code(0x3A);

//.........这里部分代码省略.........

void cache_as_ram_main(unsigned long bist, unsigned long cpu_init_detectedx)
{
	static const u16 spd_addr[] = { DIMM0, 0, 0, 0, DIMM1, 0, 0, 0, };
	int needs_reset = 0;
	u32 bsp_apicid = 0;
	msr_t msr;
	struct cpuid_result cpuid1;
	struct sys_info *sysinfo = &sysinfo_car;

	if (!cpu_init_detectedx && boot_cpu()) {
		/* Nothing special needs to be done to find bus 0 */
		/* Allow the HT devices to be found */
		enumerate_ht_chain();
		/* sb600_lpc_port80(); */
		sb600_pci_port80();
	}

	if (bist == 0)
		bsp_apicid = init_cpus(cpu_init_detectedx, sysinfo);

	enable_rs690_dev8();
	sb600_lpc_init();

	ite_enable_serial(SERIAL_DEV, CONFIG_TTYS0_BASE);
	it8712f_kill_watchdog();

	console_init();

	/* Halt if there was a built in self test failure */
	report_bist_failure(bist);
	printk(BIOS_DEBUG, "bsp_apicid=0x%x\n", bsp_apicid);

	setup_tim8690_resource_map();

	setup_coherent_ht_domain();

#if CONFIG_LOGICAL_CPUS
	/* It is said that we should start core1 after all core0 launched */
	wait_all_core0_started();
	start_other_cores();
#endif
	wait_all_aps_started(bsp_apicid);

	ht_setup_chains_x(sysinfo);

	/* run _early_setup before soft-reset. */
	rs690_early_setup();
	sb600_early_setup();

	/* Check to see if processor is capable of changing FIDVID  */
	/* otherwise it will throw a GP# when reading FIDVID_STATUS */
	cpuid1 = cpuid(0x80000007);
	if ((cpuid1.edx & 0x6) == 0x6 ) {
		/* Read FIDVID_STATUS */
		msr=rdmsr(0xc0010042);
		printk(BIOS_DEBUG, "begin msr fid, vid: hi=0x%x, lo=0x%x\n", msr.hi, msr.lo);

		enable_fid_change();
		enable_fid_change_on_sb(sysinfo->sbbusn, sysinfo->sbdn);
		init_fidvid_bsp(bsp_apicid);

		/* show final fid and vid */
		msr=rdmsr(0xc0010042);
		printk(BIOS_DEBUG, "end msr fid, vid: hi=0x%x, lo=0x%x\n", msr.hi, msr.lo);
	} else {
		printk(BIOS_DEBUG, "Changing FIDVID not supported\n");
	}

	needs_reset = optimize_link_coherent_ht();
	needs_reset |= optimize_link_incoherent_ht(sysinfo);
	rs690_htinit();
	printk(BIOS_DEBUG, "needs_reset=0x%x\n", needs_reset);

	if (needs_reset) {
		print_info("ht reset -\n");
		soft_reset();
	}

	allow_all_aps_stop(bsp_apicid);

	/* It's the time to set ctrl now; */
	printk(BIOS_DEBUG, "sysinfo->nodes: %2x  sysinfo->ctrl: %p  spd_addr: %p\n",
		     sysinfo->nodes, sysinfo->ctrl, spd_addr);
	fill_mem_ctrl(sysinfo->nodes, sysinfo->ctrl, spd_addr);
	sdram_initialize(sysinfo->nodes, sysinfo->ctrl, sysinfo);

	rs690_before_pci_init();
	sb600_before_pci_init();

	post_cache_as_ram();
}