mirror of
https://github.com/cesanta/mongoose.git
synced 2024-12-16 10:29:22 +08:00
250 lines
7.8 KiB
C
250 lines
7.8 KiB
C
|
/*
|
||
|
* Copyright (c) 2015-2016, Freescale Semiconductor, Inc.
|
||
|
* Copyright 2016-2021 NXP
|
||
|
* All rights reserved.
|
||
|
*
|
||
|
* SPDX-License-Identifier: BSD-3-Clause
|
||
|
*/
|
||
|
|
||
|
#include "fsl_common.h"
|
||
|
|
||
|
/* Component ID definition, used by tools. */
|
||
|
#ifndef FSL_COMPONENT_ID
|
||
|
#define FSL_COMPONENT_ID "platform.drivers.common_arm"
|
||
|
#endif
|
||
|
|
||
|
#ifndef __GIC_PRIO_BITS
|
||
|
#if defined(ENABLE_RAM_VECTOR_TABLE)
|
||
|
uint32_t InstallIRQHandler(IRQn_Type irq, uint32_t irqHandler)
|
||
|
{
|
||
|
#ifdef __VECTOR_TABLE
|
||
|
#undef __VECTOR_TABLE
|
||
|
#endif
|
||
|
|
||
|
/* Addresses for VECTOR_TABLE and VECTOR_RAM come from the linker file */
|
||
|
#if defined(__CC_ARM) || defined(__ARMCC_VERSION)
|
||
|
extern uint32_t Image$$VECTOR_ROM$$Base[];
|
||
|
extern uint32_t Image$$VECTOR_RAM$$Base[];
|
||
|
extern uint32_t Image$$RW_m_data$$Base[];
|
||
|
|
||
|
#define __VECTOR_TABLE Image$$VECTOR_ROM$$Base
|
||
|
#define __VECTOR_RAM Image$$VECTOR_RAM$$Base
|
||
|
#define __RAM_VECTOR_TABLE_SIZE (((uint32_t)Image$$RW_m_data$$Base - (uint32_t)Image$$VECTOR_RAM$$Base))
|
||
|
#elif defined(__ICCARM__)
|
||
|
extern uint32_t __RAM_VECTOR_TABLE_SIZE[];
|
||
|
extern uint32_t __VECTOR_TABLE[];
|
||
|
extern uint32_t __VECTOR_RAM[];
|
||
|
#elif defined(__GNUC__)
|
||
|
extern uint32_t __VECTOR_TABLE[];
|
||
|
extern uint32_t __VECTOR_RAM[];
|
||
|
extern uint32_t __RAM_VECTOR_TABLE_SIZE_BYTES[];
|
||
|
uint32_t __RAM_VECTOR_TABLE_SIZE = (uint32_t)(__RAM_VECTOR_TABLE_SIZE_BYTES);
|
||
|
#endif /* defined(__CC_ARM) || defined(__ARMCC_VERSION) */
|
||
|
uint32_t n;
|
||
|
uint32_t ret;
|
||
|
uint32_t irqMaskValue;
|
||
|
|
||
|
irqMaskValue = DisableGlobalIRQ();
|
||
|
if (SCB->VTOR != (uint32_t)__VECTOR_RAM)
|
||
|
{
|
||
|
/* Copy the vector table from ROM to RAM */
|
||
|
for (n = 0; n < ((uint32_t)__RAM_VECTOR_TABLE_SIZE) / sizeof(uint32_t); n++)
|
||
|
{
|
||
|
__VECTOR_RAM[n] = __VECTOR_TABLE[n];
|
||
|
}
|
||
|
/* Point the VTOR to the position of vector table */
|
||
|
SCB->VTOR = (uint32_t)__VECTOR_RAM;
|
||
|
}
|
||
|
|
||
|
ret = __VECTOR_RAM[(int32_t)irq + 16];
|
||
|
/* make sure the __VECTOR_RAM is noncachable */
|
||
|
__VECTOR_RAM[(int32_t)irq + 16] = irqHandler;
|
||
|
|
||
|
EnableGlobalIRQ(irqMaskValue);
|
||
|
|
||
|
return ret;
|
||
|
}
|
||
|
#endif /* ENABLE_RAM_VECTOR_TABLE. */
|
||
|
#endif /* __GIC_PRIO_BITS. */
|
||
|
|
||
|
#if (defined(FSL_FEATURE_SOC_SYSCON_COUNT) && (FSL_FEATURE_SOC_SYSCON_COUNT > 0))
|
||
|
|
||
|
/*
|
||
|
* When FSL_FEATURE_POWERLIB_EXTEND is defined to non-zero value,
|
||
|
* powerlib should be used instead of these functions.
|
||
|
*/
|
||
|
#if !(defined(FSL_FEATURE_POWERLIB_EXTEND) && (FSL_FEATURE_POWERLIB_EXTEND != 0))
|
||
|
|
||
|
/*
|
||
|
* When the SYSCON STARTER registers are discontinuous, these functions are
|
||
|
* implemented in fsl_power.c.
|
||
|
*/
|
||
|
#if !(defined(FSL_FEATURE_SYSCON_STARTER_DISCONTINUOUS) && FSL_FEATURE_SYSCON_STARTER_DISCONTINUOUS)
|
||
|
|
||
|
void EnableDeepSleepIRQ(IRQn_Type interrupt)
|
||
|
{
|
||
|
uint32_t intNumber = (uint32_t)interrupt;
|
||
|
|
||
|
uint32_t index = 0;
|
||
|
|
||
|
while (intNumber >= 32u)
|
||
|
{
|
||
|
index++;
|
||
|
intNumber -= 32u;
|
||
|
}
|
||
|
|
||
|
SYSCON->STARTERSET[index] = 1UL << intNumber;
|
||
|
(void)EnableIRQ(interrupt); /* also enable interrupt at NVIC */
|
||
|
}
|
||
|
|
||
|
void DisableDeepSleepIRQ(IRQn_Type interrupt)
|
||
|
{
|
||
|
uint32_t intNumber = (uint32_t)interrupt;
|
||
|
|
||
|
(void)DisableIRQ(interrupt); /* also disable interrupt at NVIC */
|
||
|
uint32_t index = 0;
|
||
|
|
||
|
while (intNumber >= 32u)
|
||
|
{
|
||
|
index++;
|
||
|
intNumber -= 32u;
|
||
|
}
|
||
|
|
||
|
SYSCON->STARTERCLR[index] = 1UL << intNumber;
|
||
|
}
|
||
|
#endif /* FSL_FEATURE_SYSCON_STARTER_DISCONTINUOUS */
|
||
|
#endif /* FSL_FEATURE_POWERLIB_EXTEND */
|
||
|
#endif /* FSL_FEATURE_SOC_SYSCON_COUNT */
|
||
|
|
||
|
#if defined(DWT)
|
||
|
/* Use WDT. */
|
||
|
void MSDK_EnableCpuCycleCounter(void)
|
||
|
{
|
||
|
/* Make sure the DWT trace fucntion is enabled. */
|
||
|
if (CoreDebug_DEMCR_TRCENA_Msk != (CoreDebug_DEMCR_TRCENA_Msk & CoreDebug->DEMCR))
|
||
|
{
|
||
|
CoreDebug->DEMCR |= CoreDebug_DEMCR_TRCENA_Msk;
|
||
|
}
|
||
|
|
||
|
/* CYCCNT not supported on this device. */
|
||
|
assert(DWT_CTRL_NOCYCCNT_Msk != (DWT->CTRL & DWT_CTRL_NOCYCCNT_Msk));
|
||
|
|
||
|
/* Read CYCCNT directly if CYCCENT has already been enabled, otherwise enable CYCCENT first. */
|
||
|
if (DWT_CTRL_CYCCNTENA_Msk != (DWT_CTRL_CYCCNTENA_Msk & DWT->CTRL))
|
||
|
{
|
||
|
DWT->CTRL |= DWT_CTRL_CYCCNTENA_Msk;
|
||
|
}
|
||
|
}
|
||
|
|
||
|
uint32_t MSDK_GetCpuCycleCount(void)
|
||
|
{
|
||
|
return DWT->CYCCNT;
|
||
|
}
|
||
|
#endif /* defined(DWT) */
|
||
|
|
||
|
#if !(defined(SDK_DELAY_USE_DWT) && defined(DWT))
|
||
|
/* Use software loop. */
|
||
|
#if defined(__CC_ARM) /* This macro is arm v5 specific */
|
||
|
/* clang-format off */
|
||
|
__ASM static void DelayLoop(uint32_t count)
|
||
|
{
|
||
|
loop
|
||
|
SUBS R0, R0, #1
|
||
|
CMP R0, #0
|
||
|
BNE loop
|
||
|
BX LR
|
||
|
}
|
||
|
#elif defined(__ARM_ARCH_8A__) /* This macro is ARMv8-A specific */
|
||
|
static void DelayLoop(uint32_t count)
|
||
|
{
|
||
|
__ASM volatile(" MOV X0, %0" : : "r"(count));
|
||
|
__ASM volatile(
|
||
|
"loop: \n"
|
||
|
" SUB X0, X0, #1 \n"
|
||
|
" CMP X0, #0 \n"
|
||
|
|
||
|
" BNE loop \n"
|
||
|
:
|
||
|
:
|
||
|
: "r0");
|
||
|
}
|
||
|
/* clang-format on */
|
||
|
#elif defined(__ARMCC_VERSION) || defined(__ICCARM__) || defined(__GNUC__)
|
||
|
/* Cortex-M0 has a smaller instruction set, SUBS isn't supported in thumb-16 mode reported from __GNUC__ compiler,
|
||
|
* use SUB and CMP here for compatibility */
|
||
|
static void DelayLoop(uint32_t count)
|
||
|
{
|
||
|
__ASM volatile(" MOV R0, %0" : : "r"(count));
|
||
|
__ASM volatile(
|
||
|
"loop: \n"
|
||
|
#if defined(__GNUC__) && !defined(__ARMCC_VERSION)
|
||
|
" SUB R0, R0, #1 \n"
|
||
|
#else
|
||
|
" SUBS R0, R0, #1 \n"
|
||
|
#endif
|
||
|
" CMP R0, #0 \n"
|
||
|
|
||
|
" BNE loop \n"
|
||
|
:
|
||
|
:
|
||
|
: "r0");
|
||
|
}
|
||
|
#endif /* defined(__CC_ARM) */
|
||
|
#endif /* defined(SDK_DELAY_USE_DWT) && defined(DWT) */
|
||
|
|
||
|
/*!
|
||
|
* @brief Delay at least for some time.
|
||
|
* Please note that, if not uses DWT, this API will use while loop for delay, different run-time environments have
|
||
|
* effect on the delay time. If precise delay is needed, please enable DWT delay. The two parmeters delayTime_us and
|
||
|
* coreClock_Hz have limitation. For example, in the platform with 1GHz coreClock_Hz, the delayTime_us only supports
|
||
|
* up to 4294967 in current code. If long time delay is needed, please implement a new delay function.
|
||
|
*
|
||
|
* @param delayTime_us Delay time in unit of microsecond.
|
||
|
* @param coreClock_Hz Core clock frequency with Hz.
|
||
|
*/
|
||
|
void SDK_DelayAtLeastUs(uint32_t delayTime_us, uint32_t coreClock_Hz)
|
||
|
{
|
||
|
uint64_t count;
|
||
|
|
||
|
if (delayTime_us > 0U)
|
||
|
{
|
||
|
count = USEC_TO_COUNT(delayTime_us, coreClock_Hz);
|
||
|
|
||
|
assert(count <= UINT32_MAX);
|
||
|
|
||
|
#if defined(SDK_DELAY_USE_DWT) && defined(DWT) /* Use DWT for better accuracy */
|
||
|
|
||
|
MSDK_EnableCpuCycleCounter();
|
||
|
/* Calculate the count ticks. */
|
||
|
count += MSDK_GetCpuCycleCount();
|
||
|
|
||
|
if (count > UINT32_MAX)
|
||
|
{
|
||
|
count -= UINT32_MAX;
|
||
|
/* Wait for cyccnt overflow. */
|
||
|
while (count < MSDK_GetCpuCycleCount())
|
||
|
{
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/* Wait for cyccnt reach count value. */
|
||
|
while (count > MSDK_GetCpuCycleCount())
|
||
|
{
|
||
|
}
|
||
|
#else
|
||
|
/* Divide value may be different in various environment to ensure delay is precise.
|
||
|
* Every loop count includes three instructions, due to Cortex-M7 sometimes executes
|
||
|
* two instructions in one period, through test here set divide 1.5. Other M cores use
|
||
|
* divide 4. By the way, divide 1.5 or 4 could let the count lose precision, but it does
|
||
|
* not matter because other instructions outside while loop is enough to fill the time.
|
||
|
*/
|
||
|
#if (__CORTEX_M == 7)
|
||
|
count = count / 3U * 2U;
|
||
|
#else
|
||
|
count = count / 4U;
|
||
|
#endif
|
||
|
DelayLoop((uint32_t)count);
|
||
|
#endif /* defined(SDK_DELAY_USE_DWT) && defined(DWT) */
|
||
|
}
|
||
|
}
|