zephyr/drivers/timer/mchp_xec_rtos_timer.c

401 lines
10 KiB
C
Raw Normal View History

/*
* Copyright (c) 2019 Intel Corporation
* Copyright (c) 2019 Microchip Technology Incorporated
* SPDX-License-Identifier: Apache-2.0
*/
#define DT_DRV_COMPAT microchip_xec_rtos_timer
#include <soc.h>
#include <drivers/timer/system_timer.h>
#include <sys_clock.h>
#include <spinlock.h>
BUILD_ASSERT(!IS_ENABLED(CONFIG_SMP), "XEC RTOS timer doesn't support SMP");
BUILD_ASSERT(CONFIG_SYS_CLOCK_HW_CYCLES_PER_SEC == 32768,
"XEC RTOS timer HW frequency is fixed at 32768");
#define DEBUG_RTOS_TIMER 0
#if DEBUG_RTOS_TIMER != 0
/* Enable feature to halt timer on JTAG/SWD CPU halt */
#define TIMER_START_VAL (MCHP_RTMR_CTRL_BLK_EN | MCHP_RTMR_CTRL_START \
| MCHP_RTMR_CTRL_HW_HALT_EN)
#else
#define TIMER_START_VAL (MCHP_RTMR_CTRL_BLK_EN | MCHP_RTMR_CTRL_START)
#endif
/*
* Overview:
*
* This driver enables the Microchip XEC 32KHz based RTOS timer as the Zephyr
* system timer. It supports both legacy ("tickful") mode as well as
* TICKLESS_KERNEL. The XEC RTOS timer is a down counter with a fixed
* frequency of 32768 Hz. The driver is based upon the Intel local APIC
* timer driver.
* Configuration:
*
* CONFIG_MCHP_XEC_RTOS_TIMER=y
*
* CONFIG_SYS_CLOCK_HW_CYCLES_PER_SEC=<hz> must be set to 32768.
*
* To reduce truncation errors from accumalating due to conversion
* to/from time, ticks, and HW cycles set ticks per second equal to
* the frequency. With tickless kernel mode enabled the kernel will not
* program a periodic timer at this fast rate.
* CONFIG_SYS_CLOCK_TICKS_PER_SEC=32768
*/
#define CYCLES_PER_TICK \
(CONFIG_SYS_CLOCK_HW_CYCLES_PER_SEC / CONFIG_SYS_CLOCK_TICKS_PER_SEC)
#define TIMER_REGS \
((RTMR_Type *) DT_INST_REG_ADDR(0))
/* Mask off bits[31:28] of 32-bit count */
#define TIMER_MAX 0x0FFFFFFFUL
#define TIMER_COUNT_MASK 0x0FFFFFFFUL
#define TIMER_STOPPED 0xF0000000UL
/* Adjust cycle count programmed into timer for HW restart latency */
#define TIMER_ADJUST_LIMIT 2
#define TIMER_ADJUST_CYCLES 1
/* max number of ticks we can load into the timer in one shot */
#define MAX_TICKS (TIMER_MAX / CYCLES_PER_TICK)
/*
* The spinlock protects all access to the RTMR registers, as well as
* 'total_cycles', 'last_announcement', and 'cached_icr'.
*
* One important invariant that must be observed: `total_cycles` + `cached_icr`
* is always an integral multiple of CYCLE_PER_TICK; this is, timer interrupts
* are only ever scheduled to occur at tick boundaries.
*/
static struct k_spinlock lock;
static u32_t total_cycles;
static u32_t cached_icr = CYCLES_PER_TICK;
static void timer_restart(u32_t countdown)
{
TIMER_REGS->CTRL = 0U;
TIMER_REGS->CTRL = MCHP_RTMR_CTRL_BLK_EN;
TIMER_REGS->PRLD = countdown;
TIMER_REGS->CTRL = TIMER_START_VAL;
}
/*
* Read the RTOS timer counter handling the case where the timer
* has been reloaded within 1 32KHz clock of reading its count register.
* The RTOS timer hardware must synchronize the write to its control register
* on the AHB clock domain with the 32KHz clock domain of its internal logic.
* This synchronization can take from nearly 0 time up to 1 32KHz clock as it
* depends upon which 48MHz AHB clock with a 32KHz period the register write
* was on. We detect the timer is in the load state by checking the read-only
* count register and the START bit in the control register. If count register
* is 0 and the START bit is set then the timer has been started and is in the
* process of moving the preload register value into the count register.
*/
static INLINE u32_t timer_count(void)
{
u32_t ccr = TIMER_REGS->CNT;
if ((ccr == 0) && (TIMER_REGS->CTRL & MCHP_RTMR_CTRL_START)) {
ccr = cached_icr;
}
return ccr;
}
#ifdef CONFIG_TICKLESS_KERNEL
static u32_t last_announcement; /* last time we called z_clock_announce() */
/*
* Request a timeout n Zephyr ticks in the future from now.
* Requested number of ticks in the future of n <= 1 means the kernel wants
* the tick announced as soon as possible, ideally no more than one tick
* in the future.
*
* Per comment below we don't clear RTMR pending interrupt.
* RTMR counter register is read-only and is loaded from the preload
* register by a 0->1 transition of the control register start bit.
* Writing a new value to preload only takes effect once the count
* register reaches 0.
*/
void z_clock_set_timeout(s32_t n, bool idle)
{
ARG_UNUSED(idle);
u32_t ccr, temp;
int full_ticks; /* number of complete ticks we'll wait */
u32_t full_cycles; /* full_ticks represented as cycles */
u32_t partial_cycles; /* number of cycles to first tick boundary */
kernel/timeout: Make timeout arguments an opaque type Add a k_timeout_t type, and use it everywhere that kernel API functions were accepting a millisecond timeout argument. Instead of forcing milliseconds everywhere (which are often not integrally representable as system ticks), do the conversion to ticks at the point where the timeout is created. This avoids an extra unit conversion in some application code, and allows us to express the timeout in units other than milliseconds to achieve greater precision. The existing K_MSEC() et. al. macros now return initializers for a k_timeout_t. The K_NO_WAIT and K_FOREVER constants have now become k_timeout_t values, which means they cannot be operated on as integers. Applications which have their own APIs that need to inspect these vs. user-provided timeouts can now use a K_TIMEOUT_EQ() predicate to test for equality. Timer drivers, which receive an integer tick count in ther z_clock_set_timeout() functions, now use the integer-valued K_TICKS_FOREVER constant instead of K_FOREVER. For the initial release, to preserve source compatibility, a CONFIG_LEGACY_TIMEOUT_API kconfig is provided. When true, the k_timeout_t will remain a compatible 32 bit value that will work with any legacy Zephyr application. Some subsystems present timeout (or timeout-like) values to their own users as APIs that would re-use the kernel's own constants and conventions. These will require some minor design work to adapt to the new scheme (in most cases just using k_timeout_t directly in their own API), and they have not been changed in this patch, instead selecting CONFIG_LEGACY_TIMEOUT_API via kconfig. These subsystems include: CAN Bus, the Microbit display driver, I2S, LoRa modem drivers, the UART Async API, Video hardware drivers, the console subsystem, and the network buffer abstraction. k_sleep() now takes a k_timeout_t argument, with a k_msleep() variant provided that works identically to the original API. Most of the changes here are just type/configuration management and documentation, but there are logic changes in mempool, where a loop that used a timeout numerically has been reworked using a new z_timeout_end_calc() predicate. Also in queue.c, a (when POLL was enabled) a similar loop was needlessly used to try to retry the k_poll() call after a spurious failure. But k_poll() does not fail spuriously, so the loop was removed. Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
2020-03-06 00:18:14 +01:00
if (idle && (n == K_TICKS_FOREVER)) {
/*
* We are not in a locked section. Are writes to two
* global objects safe from pre-emption?
*/
TIMER_REGS->CTRL = 0U; /* stop timer */
cached_icr = TIMER_STOPPED;
return;
}
if (n < 1) {
full_ticks = 0;
kernel/timeout: Make timeout arguments an opaque type Add a k_timeout_t type, and use it everywhere that kernel API functions were accepting a millisecond timeout argument. Instead of forcing milliseconds everywhere (which are often not integrally representable as system ticks), do the conversion to ticks at the point where the timeout is created. This avoids an extra unit conversion in some application code, and allows us to express the timeout in units other than milliseconds to achieve greater precision. The existing K_MSEC() et. al. macros now return initializers for a k_timeout_t. The K_NO_WAIT and K_FOREVER constants have now become k_timeout_t values, which means they cannot be operated on as integers. Applications which have their own APIs that need to inspect these vs. user-provided timeouts can now use a K_TIMEOUT_EQ() predicate to test for equality. Timer drivers, which receive an integer tick count in ther z_clock_set_timeout() functions, now use the integer-valued K_TICKS_FOREVER constant instead of K_FOREVER. For the initial release, to preserve source compatibility, a CONFIG_LEGACY_TIMEOUT_API kconfig is provided. When true, the k_timeout_t will remain a compatible 32 bit value that will work with any legacy Zephyr application. Some subsystems present timeout (or timeout-like) values to their own users as APIs that would re-use the kernel's own constants and conventions. These will require some minor design work to adapt to the new scheme (in most cases just using k_timeout_t directly in their own API), and they have not been changed in this patch, instead selecting CONFIG_LEGACY_TIMEOUT_API via kconfig. These subsystems include: CAN Bus, the Microbit display driver, I2S, LoRa modem drivers, the UART Async API, Video hardware drivers, the console subsystem, and the network buffer abstraction. k_sleep() now takes a k_timeout_t argument, with a k_msleep() variant provided that works identically to the original API. Most of the changes here are just type/configuration management and documentation, but there are logic changes in mempool, where a loop that used a timeout numerically has been reworked using a new z_timeout_end_calc() predicate. Also in queue.c, a (when POLL was enabled) a similar loop was needlessly used to try to retry the k_poll() call after a spurious failure. But k_poll() does not fail spuriously, so the loop was removed. Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
2020-03-06 00:18:14 +01:00
} else if ((n == K_TICKS_FOREVER) || (n > MAX_TICKS)) {
full_ticks = MAX_TICKS - 1;
} else {
full_ticks = n - 1;
}
full_cycles = full_ticks * CYCLES_PER_TICK;
k_spinlock_key_t key = k_spin_lock(&lock);
ccr = timer_count();
/* turn off to clear any pending interrupt status */
TIMER_REGS->CTRL = 0U;
MCHP_GIRQ_SRC(DT_INST_PROP(0, girq)) =
BIT(DT_INST_PROP(0, girq_bit));
NVIC_ClearPendingIRQ(RTMR_IRQn);
temp = total_cycles;
temp += (cached_icr - ccr);
temp &= TIMER_COUNT_MASK;
total_cycles = temp;
partial_cycles = CYCLES_PER_TICK - (total_cycles % CYCLES_PER_TICK);
cached_icr = full_cycles + partial_cycles;
/* adjust for up to one 32KHz cycle startup time */
temp = cached_icr;
if (temp > TIMER_ADJUST_LIMIT) {
temp -= TIMER_ADJUST_CYCLES;
}
timer_restart(temp);
k_spin_unlock(&lock, key);
}
/*
* Return the number of Zephyr ticks elapsed from last call to
* z_clock_announce in the ISR. The caller casts u32_t to s32_t.
* We must make sure bit[31] is 0 in the return value.
*/
u32_t z_clock_elapsed(void)
{
u32_t ccr;
u32_t ticks;
s32_t elapsed;
k_spinlock_key_t key = k_spin_lock(&lock);
ccr = timer_count();
/* It may not look efficient but the compiler does a good job */
elapsed = (s32_t)total_cycles - (s32_t)last_announcement;
if (elapsed < 0) {
elapsed = -1 * elapsed;
}
ticks = (u32_t)elapsed;
ticks += cached_icr - ccr;
ticks /= CYCLES_PER_TICK;
ticks &= TIMER_COUNT_MASK;
k_spin_unlock(&lock, key);
return ticks;
}
static void xec_rtos_timer_isr(void *arg)
{
ARG_UNUSED(arg);
#ifdef CONFIG_EXECUTION_BENCHMARKING
extern void read_timer_start_of_tick_handler(void);
read_timer_start_of_tick_handler();
#endif
u32_t cycles;
s32_t ticks;
k_spinlock_key_t key = k_spin_lock(&lock);
MCHP_GIRQ_SRC(DT_INST_PROP(0, girq)) =
BIT(DT_INST_PROP(0, girq_bit));
/* Restart the timer as early as possible to minimize drift... */
timer_restart(MAX_TICKS * CYCLES_PER_TICK);
cycles = cached_icr;
cached_icr = MAX_TICKS * CYCLES_PER_TICK;
total_cycles += cycles;
total_cycles &= TIMER_COUNT_MASK;
/* handle wrap by using (power of 2) - 1 mask */
ticks = total_cycles - last_announcement;
ticks &= TIMER_COUNT_MASK;
ticks /= CYCLES_PER_TICK;
last_announcement = total_cycles;
k_spin_unlock(&lock, key);
z_clock_announce(ticks);
#ifdef CONFIG_EXECUTION_BENCHMARKING
extern void read_timer_end_of_tick_handler(void);
read_timer_end_of_tick_handler();
#endif
}
#else
/* Non-tickless kernel build. */
static void xec_rtos_timer_isr(void *arg)
{
ARG_UNUSED(arg);
k_spinlock_key_t key = k_spin_lock(&lock);
MCHP_GIRQ_SRC(DT_INST_PROP(0, girq)) =
BIT(DT_INST_PROP(0, girq_bit));
/* Restart the timer as early as possible to minimize drift... */
timer_restart(cached_icr);
u32_t temp = total_cycles + CYCLES_PER_TICK;
total_cycles = temp & TIMER_COUNT_MASK;
k_spin_unlock(&lock, key);
z_clock_announce(1);
}
u32_t z_clock_elapsed(void)
{
return 0U;
}
#endif /* CONFIG_TICKLESS_KERNEL */
/*
* Warning RTOS timer resolution is 30.5 us.
* This is called by two code paths:
* 1. Kernel call to k_cycle_get_32() -> arch_k_cycle_get_32() -> here.
* The kernel is casting return to (int) and using it uncasted in math
* expressions with int types. Expression result is stored in an int.
* 2. If CONFIG_ARCH_HAS_CUSTOM_BUSY_WAIT is not defined then
* z_impl_k_busy_wait calls here. This code path uses the value as u32_t.
*
*/
u32_t z_timer_cycle_get_32(void)
{
u32_t ret;
u32_t ccr;
k_spinlock_key_t key = k_spin_lock(&lock);
ccr = timer_count();
ret = (total_cycles + (cached_icr - ccr)) & TIMER_COUNT_MASK;
k_spin_unlock(&lock, key);
return ret;
}
void z_clock_idle_exit(void)
{
if (cached_icr == TIMER_STOPPED) {
cached_icr = CYCLES_PER_TICK;
timer_restart(cached_icr);
}
}
void sys_clock_disable(void)
{
TIMER_REGS->CTRL = 0U;
}
int z_clock_driver_init(struct device *device)
{
ARG_UNUSED(device);
mchp_pcr_periph_slp_ctrl(PCR_RTMR, MCHP_PCR_SLEEP_DIS);
#ifdef CONFIG_TICKLESS_KERNEL
cached_icr = MAX_TICKS;
#endif
TIMER_REGS->CTRL = 0U;
MCHP_GIRQ_SRC(DT_INST_PROP(0, girq)) =
BIT(DT_INST_PROP(0, girq_bit));
NVIC_ClearPendingIRQ(RTMR_IRQn);
IRQ_CONNECT(RTMR_IRQn,
DT_INST_IRQ(0, priority),
xec_rtos_timer_isr, 0, 0);
MCHP_GIRQ_ENSET(DT_INST_PROP(0, girq)) =
BIT(DT_INST_PROP(0, girq_bit));
irq_enable(RTMR_IRQn);
#ifdef CONFIG_ARCH_HAS_CUSTOM_BUSY_WAIT
u32_t btmr_ctrl = B32TMR0_REGS->CTRL = (MCHP_BTMR_CTRL_ENABLE
| MCHP_BTMR_CTRL_AUTO_RESTART
| MCHP_BTMR_CTRL_COUNT_UP
| (47UL << MCHP_BTMR_CTRL_PRESCALE_POS));
B32TMR0_REGS->CTRL = MCHP_BTMR_CTRL_SOFT_RESET;
B32TMR0_REGS->CTRL = btmr_ctrl;
B32TMR0_REGS->PRLD = 0xFFFFFFFFUL;
btmr_ctrl |= MCHP_BTMR_CTRL_START;
timer_restart(cached_icr);
/* wait for Hibernation timer to load count register from preload */
while (TIMER_REGS->CNT == 0)
;
B32TMR0_REGS->CTRL = btmr_ctrl;
#else
timer_restart(cached_icr);
#endif
return 0;
}
#ifdef CONFIG_ARCH_HAS_CUSTOM_BUSY_WAIT
/*
* We implement custom busy wait using a MEC1501 basic timer running on
* the 48MHz clock domain. This code is here for future power management
* save/restore of the timer context.
*/
/*
* 32-bit basic timer 0 configured for 1MHz count up, auto-reload,
* and no interrupt generation.
*/
void arch_busy_wait(u32_t usec_to_wait)
{
if (usec_to_wait == 0) {
return;
}
u32_t start = B32TMR0_REGS->CNT;
for (;;) {
u32_t curr = B32TMR0_REGS->CNT;
if ((curr - start) >= usec_to_wait) {
break;
}
}
}
#endif