mirror of
https://github.com/AsahiLinux/u-boot
synced 2024-11-18 10:48:51 +00:00
642e8487ec
Most of the timer-calibration methods are not needed on recent Intel CPUs and just increase code size. Add an option to use the known-good way to get the clock frequency in TPL. Size reduction is about 700 bytes. Note that version 1 of this commit caused bootstage to crash since the CPU was not identified. This is corrected by changes previously applied to make sure that the CPU is identified before spl_init() is called, such as 39146a2e0b x86: Move CPU init to before spl_init() Signed-off-by: Simon Glass <sjg@chromium.org> Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
490 lines
12 KiB
C
490 lines
12 KiB
C
// SPDX-License-Identifier: GPL-2.0+
|
|
/*
|
|
* Copyright (c) 2012 The Chromium OS Authors.
|
|
*
|
|
* TSC calibration codes are adapted from Linux kernel
|
|
* arch/x86/kernel/tsc_msr.c and arch/x86/kernel/tsc.c
|
|
*/
|
|
|
|
#include <common.h>
|
|
#include <dm.h>
|
|
#include <malloc.h>
|
|
#include <time.h>
|
|
#include <timer.h>
|
|
#include <asm/cpu.h>
|
|
#include <asm/io.h>
|
|
#include <asm/i8254.h>
|
|
#include <asm/ibmpc.h>
|
|
#include <asm/msr.h>
|
|
#include <asm/u-boot-x86.h>
|
|
|
|
#define MAX_NUM_FREQS 9
|
|
|
|
#define INTEL_FAM6_SKYLAKE_MOBILE 0x4E
|
|
#define INTEL_FAM6_ATOM_GOLDMONT 0x5C /* Apollo Lake */
|
|
#define INTEL_FAM6_SKYLAKE_DESKTOP 0x5E
|
|
#define INTEL_FAM6_ATOM_GOLDMONT_X 0x5F /* Denverton */
|
|
#define INTEL_FAM6_KABYLAKE_MOBILE 0x8E
|
|
#define INTEL_FAM6_KABYLAKE_DESKTOP 0x9E
|
|
|
|
DECLARE_GLOBAL_DATA_PTR;
|
|
|
|
/*
|
|
* native_calibrate_tsc
|
|
* Determine TSC frequency via CPUID, else return 0.
|
|
*/
|
|
static unsigned long native_calibrate_tsc(void)
|
|
{
|
|
struct cpuid_result tsc_info;
|
|
unsigned int crystal_freq;
|
|
|
|
if (gd->arch.x86_vendor != X86_VENDOR_INTEL)
|
|
return 0;
|
|
|
|
if (cpuid_eax(0) < 0x15)
|
|
return 0;
|
|
|
|
tsc_info = cpuid(0x15);
|
|
|
|
if (tsc_info.ebx == 0 || tsc_info.eax == 0)
|
|
return 0;
|
|
|
|
crystal_freq = tsc_info.ecx / 1000;
|
|
if (!CONFIG_IS_ENABLED(X86_TSC_TIMER_NATIVE) && !crystal_freq) {
|
|
switch (gd->arch.x86_model) {
|
|
case INTEL_FAM6_SKYLAKE_MOBILE:
|
|
case INTEL_FAM6_SKYLAKE_DESKTOP:
|
|
case INTEL_FAM6_KABYLAKE_MOBILE:
|
|
case INTEL_FAM6_KABYLAKE_DESKTOP:
|
|
crystal_freq = 24000; /* 24.0 MHz */
|
|
break;
|
|
case INTEL_FAM6_ATOM_GOLDMONT_X:
|
|
crystal_freq = 25000; /* 25.0 MHz */
|
|
break;
|
|
case INTEL_FAM6_ATOM_GOLDMONT:
|
|
crystal_freq = 19200; /* 19.2 MHz */
|
|
break;
|
|
default:
|
|
return 0;
|
|
}
|
|
}
|
|
|
|
return (crystal_freq * tsc_info.ebx / tsc_info.eax) / 1000;
|
|
}
|
|
|
|
static unsigned long cpu_mhz_from_cpuid(void)
|
|
{
|
|
if (gd->arch.x86_vendor != X86_VENDOR_INTEL)
|
|
return 0;
|
|
|
|
if (cpuid_eax(0) < 0x16)
|
|
return 0;
|
|
|
|
return cpuid_eax(0x16);
|
|
}
|
|
|
|
/*
|
|
* According to Intel 64 and IA-32 System Programming Guide,
|
|
* if MSR_PERF_STAT[31] is set, the maximum resolved bus ratio can be
|
|
* read in MSR_PLATFORM_ID[12:8], otherwise in MSR_PERF_STAT[44:40].
|
|
* Unfortunately some Intel Atom SoCs aren't quite compliant to this,
|
|
* so we need manually differentiate SoC families. This is what the
|
|
* field msr_plat does.
|
|
*/
|
|
struct freq_desc {
|
|
u8 x86_family; /* CPU family */
|
|
u8 x86_model; /* model */
|
|
/* 2: use 100MHz, 1: use MSR_PLATFORM_INFO, 0: MSR_IA32_PERF_STATUS */
|
|
u8 msr_plat;
|
|
u32 freqs[MAX_NUM_FREQS];
|
|
};
|
|
|
|
static struct freq_desc freq_desc_tables[] = {
|
|
/* PNW */
|
|
{ 6, 0x27, 0, { 0, 0, 0, 0, 0, 99840, 0, 83200, 0 } },
|
|
/* CLV+ */
|
|
{ 6, 0x35, 0, { 0, 133200, 0, 0, 0, 99840, 0, 83200, 0 } },
|
|
/* TNG - Intel Atom processor Z3400 series */
|
|
{ 6, 0x4a, 1, { 0, 100000, 133300, 0, 0, 0, 0, 0, 0 } },
|
|
/* VLV2 - Intel Atom processor E3000, Z3600, Z3700 series */
|
|
{ 6, 0x37, 1, { 83300, 100000, 133300, 116700, 80000, 0, 0, 0, 0 } },
|
|
/* ANN - Intel Atom processor Z3500 series */
|
|
{ 6, 0x5a, 1, { 83300, 100000, 133300, 100000, 0, 0, 0, 0, 0 } },
|
|
/* AMT - Intel Atom processor X7-Z8000 and X5-Z8000 series */
|
|
{ 6, 0x4c, 1, { 83300, 100000, 133300, 116700,
|
|
80000, 93300, 90000, 88900, 87500 } },
|
|
/* Ivybridge */
|
|
{ 6, 0x3a, 2, { 0, 0, 0, 0, 0, 0, 0, 0, 0 } },
|
|
};
|
|
|
|
static int match_cpu(u8 family, u8 model)
|
|
{
|
|
int i;
|
|
|
|
for (i = 0; i < ARRAY_SIZE(freq_desc_tables); i++) {
|
|
if ((family == freq_desc_tables[i].x86_family) &&
|
|
(model == freq_desc_tables[i].x86_model))
|
|
return i;
|
|
}
|
|
|
|
return -1;
|
|
}
|
|
|
|
/* Map CPU reference clock freq ID(0-7) to CPU reference clock freq(KHz) */
|
|
#define id_to_freq(cpu_index, freq_id) \
|
|
(freq_desc_tables[cpu_index].freqs[freq_id])
|
|
|
|
/*
|
|
* TSC on Intel Atom SoCs capable of determining TSC frequency by MSR is
|
|
* reliable and the frequency is known (provided by HW).
|
|
*
|
|
* On these platforms PIT/HPET is generally not available so calibration won't
|
|
* work at all and there is no other clocksource to act as a watchdog for the
|
|
* TSC, so we have no other choice than to trust it.
|
|
*
|
|
* Returns the TSC frequency in MHz or 0 if HW does not provide it.
|
|
*/
|
|
static unsigned long __maybe_unused cpu_mhz_from_msr(void)
|
|
{
|
|
u32 lo, hi, ratio, freq_id, freq;
|
|
unsigned long res;
|
|
int cpu_index;
|
|
|
|
if (gd->arch.x86_vendor != X86_VENDOR_INTEL)
|
|
return 0;
|
|
|
|
cpu_index = match_cpu(gd->arch.x86, gd->arch.x86_model);
|
|
if (cpu_index < 0)
|
|
return 0;
|
|
|
|
if (freq_desc_tables[cpu_index].msr_plat) {
|
|
rdmsr(MSR_PLATFORM_INFO, lo, hi);
|
|
ratio = (lo >> 8) & 0xff;
|
|
} else {
|
|
rdmsr(MSR_IA32_PERF_STATUS, lo, hi);
|
|
ratio = (hi >> 8) & 0x1f;
|
|
}
|
|
debug("Maximum core-clock to bus-clock ratio: 0x%x\n", ratio);
|
|
|
|
if (freq_desc_tables[cpu_index].msr_plat == 2) {
|
|
/* TODO: Figure out how best to deal with this */
|
|
freq = 100000;
|
|
debug("Using frequency: %u KHz\n", freq);
|
|
} else {
|
|
/* Get FSB FREQ ID */
|
|
rdmsr(MSR_FSB_FREQ, lo, hi);
|
|
freq_id = lo & 0x7;
|
|
freq = id_to_freq(cpu_index, freq_id);
|
|
debug("Resolved frequency ID: %u, frequency: %u KHz\n",
|
|
freq_id, freq);
|
|
}
|
|
|
|
/* TSC frequency = maximum resolved freq * maximum resolved bus ratio */
|
|
res = freq * ratio / 1000;
|
|
debug("TSC runs at %lu MHz\n", res);
|
|
|
|
return res;
|
|
}
|
|
|
|
/*
|
|
* This reads the current MSB of the PIT counter, and
|
|
* checks if we are running on sufficiently fast and
|
|
* non-virtualized hardware.
|
|
*
|
|
* Our expectations are:
|
|
*
|
|
* - the PIT is running at roughly 1.19MHz
|
|
*
|
|
* - each IO is going to take about 1us on real hardware,
|
|
* but we allow it to be much faster (by a factor of 10) or
|
|
* _slightly_ slower (ie we allow up to a 2us read+counter
|
|
* update - anything else implies a unacceptably slow CPU
|
|
* or PIT for the fast calibration to work.
|
|
*
|
|
* - with 256 PIT ticks to read the value, we have 214us to
|
|
* see the same MSB (and overhead like doing a single TSC
|
|
* read per MSB value etc).
|
|
*
|
|
* - We're doing 2 reads per loop (LSB, MSB), and we expect
|
|
* them each to take about a microsecond on real hardware.
|
|
* So we expect a count value of around 100. But we'll be
|
|
* generous, and accept anything over 50.
|
|
*
|
|
* - if the PIT is stuck, and we see *many* more reads, we
|
|
* return early (and the next caller of pit_expect_msb()
|
|
* then consider it a failure when they don't see the
|
|
* next expected value).
|
|
*
|
|
* These expectations mean that we know that we have seen the
|
|
* transition from one expected value to another with a fairly
|
|
* high accuracy, and we didn't miss any events. We can thus
|
|
* use the TSC value at the transitions to calculate a pretty
|
|
* good value for the TSC frequencty.
|
|
*/
|
|
static inline int pit_verify_msb(unsigned char val)
|
|
{
|
|
/* Ignore LSB */
|
|
inb(0x42);
|
|
return inb(0x42) == val;
|
|
}
|
|
|
|
static inline int pit_expect_msb(unsigned char val, u64 *tscp,
|
|
unsigned long *deltap)
|
|
{
|
|
int count;
|
|
u64 tsc = 0, prev_tsc = 0;
|
|
|
|
for (count = 0; count < 50000; count++) {
|
|
if (!pit_verify_msb(val))
|
|
break;
|
|
prev_tsc = tsc;
|
|
tsc = rdtsc();
|
|
}
|
|
*deltap = rdtsc() - prev_tsc;
|
|
*tscp = tsc;
|
|
|
|
/*
|
|
* We require _some_ success, but the quality control
|
|
* will be based on the error terms on the TSC values.
|
|
*/
|
|
return count > 5;
|
|
}
|
|
|
|
/*
|
|
* How many MSB values do we want to see? We aim for
|
|
* a maximum error rate of 500ppm (in practice the
|
|
* real error is much smaller), but refuse to spend
|
|
* more than 50ms on it.
|
|
*/
|
|
#define MAX_QUICK_PIT_MS 50
|
|
#define MAX_QUICK_PIT_ITERATIONS (MAX_QUICK_PIT_MS * PIT_TICK_RATE / 1000 / 256)
|
|
|
|
static unsigned long __maybe_unused quick_pit_calibrate(void)
|
|
{
|
|
int i;
|
|
u64 tsc, delta;
|
|
unsigned long d1, d2;
|
|
|
|
/* Set the Gate high, disable speaker */
|
|
outb((inb(0x61) & ~0x02) | 0x01, 0x61);
|
|
|
|
/*
|
|
* Counter 2, mode 0 (one-shot), binary count
|
|
*
|
|
* NOTE! Mode 2 decrements by two (and then the
|
|
* output is flipped each time, giving the same
|
|
* final output frequency as a decrement-by-one),
|
|
* so mode 0 is much better when looking at the
|
|
* individual counts.
|
|
*/
|
|
outb(0xb0, 0x43);
|
|
|
|
/* Start at 0xffff */
|
|
outb(0xff, 0x42);
|
|
outb(0xff, 0x42);
|
|
|
|
/*
|
|
* The PIT starts counting at the next edge, so we
|
|
* need to delay for a microsecond. The easiest way
|
|
* to do that is to just read back the 16-bit counter
|
|
* once from the PIT.
|
|
*/
|
|
pit_verify_msb(0);
|
|
|
|
if (pit_expect_msb(0xff, &tsc, &d1)) {
|
|
for (i = 1; i <= MAX_QUICK_PIT_ITERATIONS; i++) {
|
|
if (!pit_expect_msb(0xff-i, &delta, &d2))
|
|
break;
|
|
|
|
/*
|
|
* Iterate until the error is less than 500 ppm
|
|
*/
|
|
delta -= tsc;
|
|
if (d1+d2 >= delta >> 11)
|
|
continue;
|
|
|
|
/*
|
|
* Check the PIT one more time to verify that
|
|
* all TSC reads were stable wrt the PIT.
|
|
*
|
|
* This also guarantees serialization of the
|
|
* last cycle read ('d2') in pit_expect_msb.
|
|
*/
|
|
if (!pit_verify_msb(0xfe - i))
|
|
break;
|
|
goto success;
|
|
}
|
|
}
|
|
debug("Fast TSC calibration failed\n");
|
|
return 0;
|
|
|
|
success:
|
|
/*
|
|
* Ok, if we get here, then we've seen the
|
|
* MSB of the PIT decrement 'i' times, and the
|
|
* error has shrunk to less than 500 ppm.
|
|
*
|
|
* As a result, we can depend on there not being
|
|
* any odd delays anywhere, and the TSC reads are
|
|
* reliable (within the error).
|
|
*
|
|
* kHz = ticks / time-in-seconds / 1000;
|
|
* kHz = (t2 - t1) / (I * 256 / PIT_TICK_RATE) / 1000
|
|
* kHz = ((t2 - t1) * PIT_TICK_RATE) / (I * 256 * 1000)
|
|
*/
|
|
delta *= PIT_TICK_RATE;
|
|
delta /= (i*256*1000);
|
|
debug("Fast TSC calibration using PIT\n");
|
|
return delta / 1000;
|
|
}
|
|
|
|
/* Get the speed of the TSC timer in MHz */
|
|
unsigned notrace long get_tbclk_mhz(void)
|
|
{
|
|
return get_tbclk() / 1000000;
|
|
}
|
|
|
|
static ulong get_ms_timer(void)
|
|
{
|
|
return (get_ticks() * 1000) / get_tbclk();
|
|
}
|
|
|
|
ulong get_timer(ulong base)
|
|
{
|
|
return get_ms_timer() - base;
|
|
}
|
|
|
|
ulong notrace timer_get_us(void)
|
|
{
|
|
return get_ticks() / get_tbclk_mhz();
|
|
}
|
|
|
|
ulong timer_get_boot_us(void)
|
|
{
|
|
return timer_get_us();
|
|
}
|
|
|
|
void __udelay(unsigned long usec)
|
|
{
|
|
u64 now = get_ticks();
|
|
u64 stop;
|
|
|
|
stop = now + usec * get_tbclk_mhz();
|
|
|
|
while ((int64_t)(stop - get_ticks()) > 0)
|
|
#if defined(CONFIG_QEMU) && defined(CONFIG_SMP)
|
|
/*
|
|
* Add a 'pause' instruction on qemu target,
|
|
* to give other VCPUs a chance to run.
|
|
*/
|
|
asm volatile("pause");
|
|
#else
|
|
;
|
|
#endif
|
|
}
|
|
|
|
static int tsc_timer_get_count(struct udevice *dev, u64 *count)
|
|
{
|
|
u64 now_tick = rdtsc();
|
|
|
|
*count = now_tick - gd->arch.tsc_base;
|
|
|
|
return 0;
|
|
}
|
|
|
|
static void tsc_timer_ensure_setup(bool early)
|
|
{
|
|
if (gd->arch.tsc_inited)
|
|
return;
|
|
if (IS_ENABLED(CONFIG_X86_TSC_READ_BASE))
|
|
gd->arch.tsc_base = rdtsc();
|
|
|
|
if (!gd->arch.clock_rate) {
|
|
unsigned long fast_calibrate;
|
|
|
|
fast_calibrate = native_calibrate_tsc();
|
|
if (fast_calibrate)
|
|
goto done;
|
|
|
|
/* Reduce code size by dropping other methods */
|
|
if (CONFIG_IS_ENABLED(X86_TSC_TIMER_NATIVE))
|
|
panic("no timer");
|
|
|
|
fast_calibrate = cpu_mhz_from_cpuid();
|
|
if (fast_calibrate)
|
|
goto done;
|
|
|
|
fast_calibrate = cpu_mhz_from_msr();
|
|
if (fast_calibrate)
|
|
goto done;
|
|
|
|
fast_calibrate = quick_pit_calibrate();
|
|
if (fast_calibrate)
|
|
goto done;
|
|
|
|
if (early)
|
|
fast_calibrate = CONFIG_X86_TSC_TIMER_EARLY_FREQ;
|
|
else
|
|
return;
|
|
|
|
done:
|
|
gd->arch.clock_rate = fast_calibrate * 1000000;
|
|
}
|
|
gd->arch.tsc_inited = true;
|
|
}
|
|
|
|
static int tsc_timer_probe(struct udevice *dev)
|
|
{
|
|
struct timer_dev_priv *uc_priv = dev_get_uclass_priv(dev);
|
|
|
|
/* Try hardware calibration first */
|
|
tsc_timer_ensure_setup(false);
|
|
if (!gd->arch.clock_rate) {
|
|
/*
|
|
* Use the clock frequency specified in the
|
|
* device tree as last resort
|
|
*/
|
|
if (!uc_priv->clock_rate)
|
|
panic("TSC frequency is ZERO");
|
|
} else {
|
|
uc_priv->clock_rate = gd->arch.clock_rate;
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
unsigned long notrace timer_early_get_rate(void)
|
|
{
|
|
/*
|
|
* When TSC timer is used as the early timer, be warned that the timer
|
|
* clock rate can only be calibrated via some hardware ways. Specifying
|
|
* it in the device tree won't work for the early timer.
|
|
*/
|
|
tsc_timer_ensure_setup(true);
|
|
|
|
return gd->arch.clock_rate;
|
|
}
|
|
|
|
u64 notrace timer_early_get_count(void)
|
|
{
|
|
tsc_timer_ensure_setup(true);
|
|
|
|
return rdtsc() - gd->arch.tsc_base;
|
|
}
|
|
|
|
static const struct timer_ops tsc_timer_ops = {
|
|
.get_count = tsc_timer_get_count,
|
|
};
|
|
|
|
static const struct udevice_id tsc_timer_ids[] = {
|
|
{ .compatible = "x86,tsc-timer", },
|
|
{ }
|
|
};
|
|
|
|
U_BOOT_DRIVER(tsc_timer) = {
|
|
.name = "tsc_timer",
|
|
.id = UCLASS_TIMER,
|
|
.of_match = tsc_timer_ids,
|
|
.probe = tsc_timer_probe,
|
|
.ops = &tsc_timer_ops,
|
|
};
|