zephyr/drivers/timer/apic_tsc.c

323 lines
9.2 KiB
C
Raw Permalink Normal View History

/*
* Copyright (c) 2021 Intel Corporation
* SPDX-License-Identifier: Apache-2.0
*/
#include <cpuid.h> /* Header provided by the toolchain. */
#include <zephyr/init.h>
#include <zephyr/arch/x86/cpuid.h>
#include <zephyr/drivers/timer/system_timer.h>
#include <zephyr/sys_clock.h>
#include <zephyr/spinlock.h>
#include <zephyr/drivers/interrupt_controller/loapic.h>
#include <zephyr/irq.h>
/*
* This driver is selected when either CONFIG_APIC_TIMER_TSC or
* CONFIG_APIC_TSC_DEADLINE_TIMER is selected. The later is preferred over
* the former when the TSC deadline comparator is available.
*/
BUILD_ASSERT((!IS_ENABLED(CONFIG_APIC_TIMER_TSC) &&
IS_ENABLED(CONFIG_APIC_TSC_DEADLINE_TIMER)) ||
(!IS_ENABLED(CONFIG_APIC_TSC_DEADLINE_TIMER) &&
IS_ENABLED(CONFIG_APIC_TIMER_TSC)),
"one of CONFIG_APIC_TIMER_TSC or CONFIG_APIC_TSC_DEADLINE_TIMER must be set");
/*
* If the TSC deadline comparator is not supported then the ICR in one-shot
* mode is used as a fallback method to trigger the next timeout interrupt.
* Those config symbols must then be defined:
*
* CONFIG_APIC_TIMER_TSC_N=<n>
* CONFIG_APIC_TIMER_TSC_M=<m>
*
* These are set to indicate the ratio of the TSC frequency to the local
* APIC timer frequency. This can be found via CPUID 0x15 (n = EBX, m = EAX)
* on most CPUs.
*/
#ifdef CONFIG_APIC_TIMER_TSC
#define APIC_TIMER_TSC_M CONFIG_APIC_TIMER_TSC_M
#define APIC_TIMER_TSC_N CONFIG_APIC_TIMER_TSC_N
#else
#define APIC_TIMER_TSC_M 1
#define APIC_TIMER_TSC_N 1
#endif
#define IA32_TSC_DEADLINE_MSR 0x6e0
#define IA32_TSC_ADJUST_MSR 0x03b
#define CYC_PER_TICK (uint32_t)(CONFIG_SYS_CLOCK_HW_CYCLES_PER_SEC \
/ CONFIG_SYS_CLOCK_TICKS_PER_SEC)
/* the unsigned long cast limits divisors to native CPU register width */
#define cycle_diff_t unsigned long
#define CYCLE_DIFF_MAX (~(cycle_diff_t)0)
/*
* We have two constraints on the maximum number of cycles we can wait for.
*
* 1) sys_clock_announce() accepts at most INT32_MAX ticks.
*
* 2) The number of cycles between two reports must fit in a cycle_diff_t
* variable before converting it to ticks.
*
* Then:
*
* 3) Pick the smallest between (1) and (2).
*
* 4) Take into account some room for the unavoidable IRQ servicing latency.
* Let's use 3/4 of the max range.
*
* Finally let's add the LSB value to the result so to clear out a bunch of
* consecutive set bits coming from the original max values to produce a
* nicer literal for assembly generation.
*/
#define CYCLES_MAX_1 ((uint64_t)INT32_MAX * (uint64_t)CYC_PER_TICK)
#define CYCLES_MAX_2 ((uint64_t)CYCLE_DIFF_MAX)
#define CYCLES_MAX_3 MIN(CYCLES_MAX_1, CYCLES_MAX_2)
#define CYCLES_MAX_4 (CYCLES_MAX_3 / 2 + CYCLES_MAX_3 / 4)
#define CYCLES_MAX (CYCLES_MAX_4 + LSB_GET(CYCLES_MAX_4))
struct apic_timer_lvt {
uint8_t vector : 8;
uint8_t unused0 : 8;
uint8_t masked : 1;
enum { ONE_SHOT, PERIODIC, TSC_DEADLINE } mode: 2;
uint32_t unused2 : 13;
};
static struct k_spinlock lock;
static uint64_t last_cycle;
static uint64_t last_tick;
static uint32_t last_elapsed;
static union { uint32_t val; struct apic_timer_lvt lvt; } lvt_reg;
static ALWAYS_INLINE uint64_t rdtsc(void)
{
uint32_t hi, lo;
__asm__ volatile("rdtsc" : "=d"(hi), "=a"(lo));
return lo + (((uint64_t)hi) << 32);
}
static inline void wrmsr(int32_t msr, uint64_t val)
{
uint32_t hi = (uint32_t) (val >> 32);
uint32_t lo = (uint32_t) val;
__asm__ volatile("wrmsr" :: "d"(hi), "a"(lo), "c"(msr));
}
static void set_trigger(uint64_t deadline)
{
if (IS_ENABLED(CONFIG_APIC_TSC_DEADLINE_TIMER)) {
wrmsr(IA32_TSC_DEADLINE_MSR, deadline);
} else {
/* use the timer ICR to trigger next interrupt */
uint64_t curr_cycle = rdtsc();
uint64_t delta_cycles = deadline - MIN(deadline, curr_cycle);
uint64_t icr = (delta_cycles * APIC_TIMER_TSC_M) / APIC_TIMER_TSC_N;
/* cap icr to 32 bits, and not zero */
icr = CLAMP(icr, 1, UINT32_MAX);
x86_write_loapic(LOAPIC_TIMER_ICR, icr);
}
}
static void isr(const void *arg)
{
ARG_UNUSED(arg);
k_spinlock_key_t key = k_spin_lock(&lock);
uint64_t curr_cycle = rdtsc();
uint64_t delta_cycles = curr_cycle - last_cycle;
uint32_t delta_ticks = (cycle_diff_t)delta_cycles / CYC_PER_TICK;
last_cycle += (cycle_diff_t)delta_ticks * CYC_PER_TICK;
last_tick += delta_ticks;
last_elapsed = 0;
if (!IS_ENABLED(CONFIG_TICKLESS_KERNEL)) {
uint64_t next_cycle = last_cycle + CYC_PER_TICK;
set_trigger(next_cycle);
}
k_spin_unlock(&lock, key);
sys_clock_announce(delta_ticks);
}
void sys_clock_set_timeout(int32_t ticks, bool idle)
{
ARG_UNUSED(idle);
if (!IS_ENABLED(CONFIG_TICKLESS_KERNEL)) {
return;
}
k_spinlock_key_t key = k_spin_lock(&lock);
uint64_t next_cycle;
if (ticks == K_TICKS_FOREVER) {
next_cycle = last_cycle + CYCLES_MAX;
} else {
next_cycle = (last_tick + last_elapsed + ticks) * CYC_PER_TICK;
if ((next_cycle - last_cycle) > CYCLES_MAX) {
next_cycle = last_cycle + CYCLES_MAX;
}
}
/*
* Interpreted strictly, the IA SDM description of the
* TSC_DEADLINE MSR implies that it will trigger an immediate
* interrupt if we try to set an expiration across the 64 bit
* rollover. Unfortunately there's no way to test that as on
* real hardware it requires more than a century of uptime,
* but this is cheap and safe.
*/
if (next_cycle < last_cycle) {
next_cycle = UINT64_MAX;
}
set_trigger(next_cycle);
k_spin_unlock(&lock, key);
}
uint32_t sys_clock_elapsed(void)
{
if (!IS_ENABLED(CONFIG_TICKLESS_KERNEL)) {
return 0;
}
k_spinlock_key_t key = k_spin_lock(&lock);
uint64_t curr_cycle = rdtsc();
uint64_t delta_cycles = curr_cycle - last_cycle;
uint32_t delta_ticks = (cycle_diff_t)delta_cycles / CYC_PER_TICK;
last_elapsed = delta_ticks;
k_spin_unlock(&lock, key);
return delta_ticks;
}
uint32_t sys_clock_cycle_get_32(void)
{
return (uint32_t) rdtsc();
}
uint64_t sys_clock_cycle_get_64(void)
{
return rdtsc();
}
static inline uint32_t timer_irq(void)
{
/* The Zephyr APIC API is... idiosyncratic. The timer is a
* "local vector table" interrupt. These aren't system IRQs
* presented to the IO-APIC, they're indices into a register
* array in the local APIC. By Zephyr convention they come
* after all the external IO-APIC interrupts, but that number
* changes depending on device configuration so we have to
* fetch it at runtime. The timer happens to be the first
* entry in the table.
*/
return z_loapic_irq_base();
}
/* The TSC_ADJUST MSR implements a synchronized offset such that
* multiple CPUs (within a socket, anyway) can synchronize exactly, or
* implement managed timing spaces for guests in a recoverable way,
* etc... We set it to zero on all cores for simplicity, because
* firmware often leaves it in an inconsistent state between cores.
*/
static void clear_tsc_adjust(void)
{
/* But don't touch it on ACRN, where an hypervisor bug
* confuses the APIC emulation and deadline interrupts don't
* arrive.
*/
#ifndef CONFIG_BOARD_ACRN
wrmsr(IA32_TSC_ADJUST_MSR, 0);
#endif
}
void smp_timer_init(void)
{
/* Copy the LVT configuration from CPU0, because IRQ_CONNECT()
* doesn't know how to manage LVT interrupts for anything
* other than the calling/initial CPU. Same fence needed to
* prevent later MSR writes from reordering before the APIC
* configuration write.
*/
x86_write_loapic(LOAPIC_TIMER, lvt_reg.val);
__asm__ volatile("mfence" ::: "memory");
clear_tsc_adjust();
irq_enable(timer_irq());
}
init: remove the need for a dummy device pointer in SYS_INIT functions The init infrastructure, found in `init.h`, is currently used by: - `SYS_INIT`: to call functions before `main` - `DEVICE_*`: to initialize devices They are all sorted according to an initialization level + a priority. `SYS_INIT` calls are really orthogonal to devices, however, the required function signature requires a `const struct device *dev` as a first argument. The only reason for that is because the same init machinery is used by devices, so we have something like: ```c struct init_entry { int (*init)(const struct device *dev); /* only set by DEVICE_*, otherwise NULL */ const struct device *dev; } ``` As a result, we end up with such weird/ugly pattern: ```c static int my_init(const struct device *dev) { /* always NULL! add ARG_UNUSED to avoid compiler warning */ ARG_UNUSED(dev); ... } ``` This is really a result of poor internals isolation. This patch proposes a to make init entries more flexible so that they can accept sytem initialization calls like this: ```c static int my_init(void) { ... } ``` This is achieved using a union: ```c union init_function { /* for SYS_INIT, used when init_entry.dev == NULL */ int (*sys)(void); /* for DEVICE*, used when init_entry.dev != NULL */ int (*dev)(const struct device *dev); }; struct init_entry { /* stores init function (either for SYS_INIT or DEVICE*) union init_function init_fn; /* stores device pointer for DEVICE*, NULL for SYS_INIT. Allows * to know which union entry to call. */ const struct device *dev; } ``` This solution **does not increase ROM usage**, and allows to offer clean public APIs for both SYS_INIT and DEVICE*. Note that however, init machinery keeps a coupling with devices. **NOTE**: This is a breaking change! All `SYS_INIT` functions will need to be converted to the new signature. See the script offered in the following commit. Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no> init: convert SYS_INIT functions to the new signature Conversion scripted using scripts/utils/migrate_sys_init.py. Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no> manifest: update projects for SYS_INIT changes Update modules with updated SYS_INIT calls: - hal_ti - lvgl - sof - TraceRecorderSource Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no> tests: devicetree: devices: adjust test Adjust test according to the recently introduced SYS_INIT infrastructure. Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no> tests: kernel: threads: adjust SYS_INIT call Adjust to the new signature: int (*init_fn)(void); Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no>
2022-10-19 09:33:44 +02:00
static int sys_clock_driver_init(void)
{
#ifdef CONFIG_ASSERT
uint32_t eax, ebx, ecx, edx;
if (IS_ENABLED(CONFIG_APIC_TSC_DEADLINE_TIMER)) {
ecx = 0; /* prevent compiler warning */
__get_cpuid(CPUID_BASIC_INFO_1, &eax, &ebx, &ecx, &edx);
__ASSERT((ecx & BIT(24)) != 0, "No TSC Deadline support");
}
edx = 0; /* prevent compiler warning */
__get_cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
__ASSERT((edx & BIT(8)) != 0, "No Invariant TSC support");
if (IS_ENABLED(CONFIG_SMP)) {
ebx = 0; /* prevent compiler warning */
__get_cpuid_count(CPUID_EXTENDED_FEATURES_LVL, 0, &eax, &ebx, &ecx, &edx);
__ASSERT((ebx & BIT(1)) != 0, "No TSC_ADJUST MSR support");
}
#endif
if (IS_ENABLED(CONFIG_SMP)) {
clear_tsc_adjust();
}
/* Timer interrupt number is runtime-fetched, so can't use
* static IRQ_CONNECT()
*/
irq_connect_dynamic(timer_irq(), CONFIG_APIC_TIMER_IRQ_PRIORITY, isr, 0, 0);
if (IS_ENABLED(CONFIG_APIC_TIMER_TSC)) {
uint32_t timer_conf;
timer_conf = x86_read_loapic(LOAPIC_TIMER_CONFIG);
timer_conf &= ~0x0f; /* clear divider bits */
timer_conf |= 0x0b; /* divide by 1 */
x86_write_loapic(LOAPIC_TIMER_CONFIG, timer_conf);
}
lvt_reg.val = x86_read_loapic(LOAPIC_TIMER);
lvt_reg.lvt.mode = IS_ENABLED(CONFIG_APIC_TSC_DEADLINE_TIMER) ?
TSC_DEADLINE : ONE_SHOT;
lvt_reg.lvt.masked = 0;
x86_write_loapic(LOAPIC_TIMER, lvt_reg.val);
/* Per the SDM, the TSC_DEADLINE MSR is not serializing, so
* this fence is needed to be sure that an upcoming MSR write
* (i.e. a timeout we're about to set) cannot possibly reorder
* around the initialization we just did.
*/
__asm__ volatile("mfence" ::: "memory");
last_tick = rdtsc() / CYC_PER_TICK;
last_cycle = last_tick * CYC_PER_TICK;
if (!IS_ENABLED(CONFIG_TICKLESS_KERNEL)) {
set_trigger(last_cycle + CYC_PER_TICK);
}
irq_enable(timer_irq());
return 0;
}
SYS_INIT(sys_clock_driver_init, PRE_KERNEL_2,
CONFIG_SYSTEM_CLOCK_INIT_PRIORITY);