git-svn-id: svn://svn.code.sf.net/p/chibios/svn/trunk@802 35acf78f-673a-0410-8e92-d51de3d6d3f4

This commit is contained in:
gdisirio 2009-02-24 16:07:42 +00:00
parent d425f2ec12
commit e9274448e9
7 changed files with 227 additions and 16 deletions

View File

@ -21,6 +21,7 @@
* @page articles Articles
* @{
* ChibiOS/RT Articles and Code Examples:
* - @subpage article_stacks
* - @subpage article_mutual_exclusion
* - @subpage article_atomic
* - @subpage article_saveram

106
docs/src/stacks.dox Normal file
View File

@ -0,0 +1,106 @@
/*
ChibiOS/RT - Copyright (C) 2006-2007 Giovanni Di Sirio.
This file is part of ChibiOS/RT.
ChibiOS/RT is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 3 of the License, or
(at your option) any later version.
ChibiOS/RT is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
/**
* @page article_stacks Stacks and stack sizes
* @{
* In a RTOS like ChibiOS/RT there are several dedicated stacks, each stack
* has a dedicated RAM space that must have a correctly sized assigned area.
* <h2>The stacks</h2>
* There are several stacks in the systems, some are always present, some
* others are present only in some architectures:
* - <b>Main stack</b>, this stack is used by the @p main() function and the
* thread that execute it. It is not a normal thread stack because it is
* initialized in the startup code and its size is defined in a port
* dependent way. Details are in the various ports documentation.
* - <b>Interrupt Stack</b>, some architectures have a dedicated interrupt
* stack. This is an important feature in a multithreaded environment,
* without a dedicated interrupt stack each thread has to reserve
* enough space, for interrupts servicing, within its own stack. This space,
* multiplied by the total threads number, can be a significant RAM waste.
* - <b>Thread Stack</b>, each thread has a dedicated stack for its own
* execution and context switch.
* - <b>Other Stacks</b>, some architectures (ARM) can have other stacks but
* the OS does not directly use any of them.
* .
* <h2>Risks</h2>
* The most critical thing when writing an embedded multithreaded application
* is to determine the correct stack size for main, threads and, when present,
* interrupts.<br>
* Assign too much space to a stack wastes RAM, assign too little space
* leads to crashes or, worst scenario, hard to track instability.
*
* <h2>Assign the correct size</h2>
* You may try to examine the asm listings in order to calculate the exact
* stack requirements but this requires much time, experience and patience.<br>
* An alternative way is to use an interactive method. Follow this procedure
* for each thread in the system:
* - Enable the following debug options in the kernel:
* - @p CH_DBG_ENABLE_STACK_CHECK, this enables a stack check before any
* context switch. This option halts the system in @p chSysHalt() just
* before a stack overflow happens.
* - @p CH_DBG_FILL_THREADS, this option fills the threads working area
* with an easily recognizable pattern (0x55).
* - Assign a large and safe size to the thread stack, as example 256 bytes
* on 32 MCUs, 128 bytes on 8/16 bit MCUs. This is almost always too much
* for simple threads.
* - Run the application, if the application crashes or halts then increase
* the stack size and repeat (you know how to use the debugger right?).
* - Let the application run and make sure to trigger the thread in a way to
* make it follow most or all its code paths. If the application crashes or
* halts then increase the stack size and repeat.
* - Stop the application using the debugger and examine the thread working
* area (you know what a map file is, right?). You can see that the thread
* stack overwrote the fill pattern (0x55) from the top of the working area
* downward. You can estimate the excess stack by counting the untouched
* locations.
* - Trim down the stack size and repeat until the application still runs
* correctly and you have a decent margin in the stack.
* - Repeat for all the thread classes in the system.
* - Turn off the debug options.
* - Done.
* .
* <h2>Final Notes</h2>
* Some useful info:
* - Stack overflows are the most common source of problems during development,
* when in trouble with crashes or anomalous behaviors always first verify
* stack sizes.
* - The required stack size can, and very often does change when changing
* compiler vendor, compiler version, compiler options, code type (ARM
* or THUMB as example).
* - Code compiled in THUMB mode uses more stack space compared to the
* same code compiled in ARM mode. In GCC this is related to lack of tail
* calls optimizations in THUMB mode, this is probably true also in other
* compilers.
* - Speed optimized code often requires less stack space compared to space
* optimized code. Be careful when changing optimizations.
* - The interrupts space overhead on the thread stacks (@p INT_REQUIRED_STACK
* defined in @p chcore.h) is included in the total working area size
* by the system macros @p THD_WA_SIZE() and @p WORKING_AREA().<br>
* The correct way to reserve space into the thread stacks for interrupts
* processing is to override the @p INT_REQUIRED_STACK default value.
* Architectures with a dedicated interrupt stack do not require changes
* to this value. Resizing of the global interrupt stack may be required
* instead.
* - Often is a good idea to have some extra space in stacks unless you
* are really starved on RAM. Anyway optimize stack space at the very
* end of your development cycle.
* .
*/
/** @} */

85
io_channels.txt Normal file
View File

@ -0,0 +1,85 @@
I/O Channels
- Channels are specific for I/O operations, however, a channel can hide a
complex IPC operation.
- Channels are N-sized not necessarily byte-sized.
- Channels support timeout.
- The IOChannel structure hides a virtualized implementation using a VMT.
- The APIs are macros that hide the VMT.
- Channels must support events, at least 3 events are predefined:
0 - Incoming data event.
1 - Output queue empty.
2 - I/O Status Change (at least one status flag was pended).
X - More events can be defined and are channel specific.
- Read/write functions are non blocking and can transfer no data if the
buffers are empty/full.
- Zero sized read and writes simply returns zero, nothing is queued.
/**
* @brief Returns the channel data unit size.
* @details The channel data unit size is characteristic of the channel and
* cannot be modified.
* @param[in] iop pointer to an IOChannel structure
* @return The channel data unit size in bytes.
*/
size_t chIOGetWidth(const IOChannel *iop);
/**
* @brief Returns the event sources associated to the channel.
* @details A channel can have associated event sources. The event sources are
* identified by a numerical identifier, the following identifiers
* are predefined:
* - CH_IO_EVT_INPUT signaled when some data is queued in the input buffer.
* - CH_IO_EVT_OUTPUT signaled when the output buffer is emptied.
* - CH_IO_EVT_STATUS signaled when a channel related condition happens.
*
* @param[in] iop pointer to an IOChannel structure
* @param[in] n the numerical identifier.
* @return A pointer to the @p EventSource structure associated to the numerical
* identifier.
* @retval NULL there is no event source associated to the specified
* identifier.
*/
EventSource *chIOGetEventSource(const IOChannel *iop, ioevtsrc_t n);
/**
* @brief Returns the channel status flags.
* @details The channel status flags are returned and cleared.
*
* @param[in] iop pointer to an IOChannel structure
* @return The status flags.
* @retval 0 no flags pending.
*/
iosts_t chIOGetAndClearStatus(IOChannel *iop);
/**
* @brief Asynchronous read.
* @details This function reads up to @p n data units into the specified
* buffer without blocking. If there is no data into the input queue
* then the function returns immediatly.
*
* @param[in] iop pointer to an IOChannel structure
* @param[out] buf the buffer where to copy the input data
* @param[in] n the maximum number of data units to transfer
* @return The actual data units number read.
* @retval 0 the input queue is empty, no data transfer was performed.
*/
size_t chIORead(IOChannel *iop, void *buf, size_t n);
/**
* @brief Asynchronous write.
* @details This function writes up to @p n data units from the specified
* buffer without blocking. If there is no space into the output queue
* then the function returns immediatly.
*
* @param[in] iop pointer to an IOChannel structure
* @param[out] buf the buffer with the data to be written
* @param[in] n the maximum number of data units to transfer
* @return The actual data units number written.
* @retval 0 the output queue is full, no data transfer was performed.
*/
size_t chIOWrite(IOChannel *iop, const void *buf, size_t n);
bool_t chIOWaitInput(IOChannel *iop, systime_t timeout);
bool_t chIOWaitOutput(IOChannel *iop, systime_t timeout);

View File

@ -175,16 +175,16 @@ struct context {
* This code tricks the compiler to save all the specified registers by
* "touching" them.
*/
#define PORT_IRQ_PROLOGUE() { \
asm ("" : : : "r18", "r19", "r20", "r21", "r22", "r23", "r24", \
"r25", "r26", "r27", "r30", "r31"); \
#define PORT_IRQ_PROLOGUE() { \
asm ("" : : : "r18", "r19", "r20", "r21", "r22", "r23", "r24", \
"r25", "r26", "r27", "r30", "r31"); \
}
/**
* IRQ epilogue code, inserted at the end of all IRQ handlers enabled to
* invoke system APIs.
*/
#define PORT_IRQ_EPILOGUE() { \
#define PORT_IRQ_EPILOGUE() { \
if (chSchRescRequiredI()) \
chSchDoRescheduleI(); \
}
@ -240,7 +240,7 @@ asm ("" : : : "r18", "r19", "r20", "r21", "r22", "r23", "r24", \
* This port function is implemented as inlined code for performance reasons.
*/
#if ENABLE_WFI_IDLE != 0
#define port_wait_for_interrupt() { \
#define port_wait_for_interrupt() { \
asm volatile ("sleep"); \
}
#else

View File

@ -59,10 +59,10 @@ Thread *init_thread(Thread *tp, tprio_t prio) {
}
#if CH_DBG_FILL_THREADS
static void memfill(uint8_t *p, uint32_t n, uint8_t v) {
static void memfill(uint8_t *startp, uint8_t *endp, uint8_t v) {
while (n)
*p++ = v, n--;
while (startp < endp)
*startp++ = v;
}
#endif
@ -95,7 +95,12 @@ Thread *chThdInit(void *workspace, size_t wsize,
(prio <= HIGHPRIO) && (pf != NULL),
"chThdInit");
#if CH_DBG_FILL_THREADS
memfill(workspace, wsize, MEM_FILL_PATTERN);
memfill(workspace,
(uint8_t)workspace + sizeof(Thread),
THREAD_FILL_VALUE);
memfill((uint8_t)workspace + sizeof(Thread),
(uint8_t)workspace + wsize
STACK_FILL_VALUE);
#endif
SETUP_CONTEXT(workspace, wsize, pf, arg);
return init_thread(tp, prio);

View File

@ -35,10 +35,21 @@
#endif
/**
* @brief Fill value for threads working area in debug mode.
* @brief Fill value for thread stack area in debug mode.
*/
#ifndef MEM_FILL_PATTERN
#define MEM_FILL_PATTERN 0x55
#ifndef STACK_FILL_VALUE
#define STACK_FILL_VALUE 0x55
#endif
/**
* @brief Fill value for thread area in debug mode.
* @note The chosen default value is 0xFF in order to make evident which
* thread fields were not initialized when inspecting the memory with
* a debugger. A uninitialized field is not an error in itself but it
* better to know it.
*/
#ifndef THREAD_FILL_VALUE
#define THREAD_FILL_VALUE 0xFF
#endif
/**

View File

@ -17,11 +17,17 @@ After 1.0.0:
* Add checks to all APIs.
* Stack checks option.
* Threads profiling option.
- Registers clearing on thread start.
* Idle loop hook macro.
* Switch the configuration options to TRUE/FALSE rather than def/undef.
* Remove port_puts() from all the ports.
- Stack sizes article into the documentation.
- Find out and document main stack settings in MSP430 and AVR runtimes.
After 1.2.0:
X Abstract I/O channels rather than just serial ports.
- Move the serial drivers implementations in library. Better keep the core
as compact as possible.
- Threads Pools manager in the library.
- New chThdCreate() that takes just two parameters, a pointer to a thread
descriptor and the tread parameter. It could wrap the current variants
@ -30,11 +36,8 @@ After 1.2.0:
- OSEK-style simple tasks within the idle thread.
? Think to something for threads restart.
? Multiple heaps, disjoint heaps, heaps in heaps.
- Abstract I/O channels rather than just serial ports.
- Move the serial drivers implementations in library al keep the I/O channel
interface as part of the kernel. Better keep the core as compact as
possible.
- Update C++ wrapper (Heap, Pools, Mailboxes and any new feature).
- Think about making threads return void.
Ideas for 2.x.x:
- Reference counter for threads, concept of detached threads, threads