Hi,
I don't know if there is particular need for this feature, but I think it would be "quite simple" to enable the RR scheduler in tickless mode.
Basic idea:
Whenever a thread gets "activated", the scheduler need to check whether there are multiple threads with same priority that compete for CPU time. If so, it activates a VT that fires after a certain amount of time (probably configurable in chconf.h) to reschedule in a RR manner like in standard (not tickless) mode. This would basically activate the tick system dynamically if there are multiple active threads with same priority level.
Expected results:
Such a solution would bring together the benefits of both tick-worlds by making the system generally more responsible (tickless mode) but also allowing concurrent execution of threads with identical priority levels (standard mode). The only drawback is, that in case RR scheduling is required, the system will become slightly slower than in tickless mode.
As I said before, I don't know if anyone actually needs this right now, since 255 priority levels and dynamic priority boosting can solve any scheduling isues as well, but hey, another feature doesn't hurt
BTW: Another feature might be the introduction of other schedulers or the possibility to override the scheduler by custom code/hooks.
Round-Robin scheduler in tickless mode Topic is solved
Re: Round-Robin scheduler in tickless mode
This is something I raised a little while ago; the current arrangement could potentially cause me a problem. Giovanni explained why things are the way they are, which makes sense in some ways.
A timer to limit the duration of a time slice would solve that problem; however how do you take account of interruptions due to the execution of higher priority threads?
If possible, I would be inclined to have an option to manage this with a hardware timer, since virtual timers are relatively inefficient.
A timer to limit the duration of a time slice would solve that problem; however how do you take account of interruptions due to the execution of higher priority threads?
If possible, I would be inclined to have an option to manage this with a hardware timer, since virtual timers are relatively inefficient.
- Giovanni
- Site Admin
- Posts: 14457
- Joined: Wed May 27, 2009 8:48 am
- Location: Salerno, Italy
- Has thanked: 1076 times
- Been thanked: 922 times
- Contact:
Re: Round-Robin scheduler in tickless mode
Hi,
RR is enabled even in tickless mode but it is cooperative, you need to call yield to make threads at same priority level rotate.
Introducing preemption using virtual timers would be possible, in theory, but I believe it would kill performance, a timer would have to be stopped/restarted at context switch (the time quantum) and consumed time would have to be calculated using time stamps if a switch occurs before the timer is triggered.
I think it should be possible to implement this using hooks, not sure if changes would be required, I am available to, at very least, introduce new hooks if required. If done it could become a library module.
Giovanni
RR is enabled even in tickless mode but it is cooperative, you need to call yield to make threads at same priority level rotate.
Introducing preemption using virtual timers would be possible, in theory, but I believe it would kill performance, a timer would have to be stopped/restarted at context switch (the time quantum) and consumed time would have to be calculated using time stamps if a switch occurs before the timer is triggered.
I think it should be possible to implement this using hooks, not sure if changes would be required, I am available to, at very least, introduce new hooks if required. If done it could become a library module.
Giovanni
-
- Posts: 135
- Joined: Wed Feb 04, 2015 5:03 pm
- Location: CITEC, Bielefeld University, germany
- Has thanked: 15 times
- Been thanked: 24 times
- Contact:
Re: Round-Robin scheduler in tickless mode
Giovanni wrote:Introducing preemption using virtual timers would be possible, in theory, but I believe it would kill performance, a timer would have to be stopped/restarted at context switch (the time quantum) and consumed time would have to be calculated using time stamps if a switch occurs before the timer is triggered.
Actually this process should be similar complex as it is for the non-tickless mode where you have to restart the timer at each tick as well. The only difference is that for the latter you can make several assumptions which again simplify the problem a little. However, I agree that preemptive RR scheduling the way I described it is not optimal regarding performance, but it is only required and would only have any performance impact IF there are multiple threads with identical priority that are also active at the same time. It would also be more intuitive that two threads with same priority will share time 50/50 instead of "first come, first serve" like it is now.
However, if the scheduler would be some kind of module that can be changed by users with "minimal" effort, that should be absolutely fine.
- Thomas
- Giovanni
- Site Admin
- Posts: 14457
- Joined: Wed May 27, 2009 8:48 am
- Location: Salerno, Italy
- Has thanked: 1076 times
- Been thanked: 922 times
- Contact:
Re: Round-Robin scheduler in tickless mode
I am not ruling this out, I will evaluate and probably experiment with it.
Giovanni
Giovanni
- Giovanni
- Site Admin
- Posts: 14457
- Joined: Wed May 27, 2009 8:48 am
- Location: Salerno, Italy
- Has thanked: 1076 times
- Been thanked: 922 times
- Contact:
Re: Round-Robin scheduler in tickless mode
Hi,
This is what I am thinking to do:
I will work on completing this after some more urgent matter. A new system initialization hook is required, I will add this too.
Giovanni
This is what I am thinking to do:
Code: Select all
/*
ChibiOS - Copyright (C) 2006..2016 Giovanni Di Sirio.
This file is part of ChibiOS.
ChibiOS 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 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/>.
*/
/**
* @file chpreempt.h
* @brief Preemption enforcement structures and macros.
*
* @addtogroup preemption_enforcement
* @details This modules export hook macros required for implementing
* a preemptive round robin mode for threads at the same priority
* level.<br>
* This method is alternative to the ChibiOS/RT native implementation
* which is not compatible with the tick-less mode, however, this
* timers-based solution can decrease threads context-switch
* performance because the added overhead.
* @note This file is not included automatically by @p ch.h, you need
* to:
* - Define @p CH_CFG_ROUND_ROBIN_QUANTUM in chconf.h.
* - Include this file from @p chconf.h.
* - Add all the macros and functions to the appropriate hooks.
* .
* @{
*/
#ifndef CHPREEMPT_H
#define CHPREEMPT_H
/*===========================================================================*/
/* Module constants. */
/*===========================================================================*/
/*===========================================================================*/
/* Module pre-compile time settings. */
/*===========================================================================*/
/*===========================================================================*/
/* Derived constants and error checks. */
/*===========================================================================*/
#if !defined(CH_CFG_ROUND_ROBIN_QUANTUM)
#error "CH_CFG_ROUND_ROBIN_QUANTUM not defined in chconf.h"
#endif
#if CH_CFG_ROUND_ROBIN_QUANTUM <= 0
#error "CH_CFG_ROUND_ROBIN_QUANTUM must be greater than zero"
#endif
#if CH_CFG_TIME_QUANTUM > 0
#error "CH_CFG_ROUND_ROBIN_QUANTUM is not compatible with CH_CFG_TIME_QUANTUM > 0"
#endif
/*===========================================================================*/
/* Module data structures and types. */
/*===========================================================================*/
/*===========================================================================*/
/* Module macros. */
/*===========================================================================*/
/**
* @brief Fields to be added to the @p thread_t structure.
*
* @notapi
*/
#define CH_PREEMPT_THREAD_FIELDS \
virtual_timer_t vt;
/*===========================================================================*/
/* External declarations. */
/*===========================================================================*/
/*===========================================================================*/
/* Module inline functions. */
/*===========================================================================*/
/**
* @brief Hook code for system initialization.
*
* @notapi
*/
static inline void ch_preempt_system_init(void) {
}
/**
* @brief Hook code for thread initialization.
*/
static inline void ch_preempt_thread_init(void) {
}
/**
* @brief Hook code for context switch.
*/
static inline void ch_preempt_thread_switch(thread_t *new, thread_t *old) {
}
#endif /* CHPREEMPT_H */
/** @} */
I will work on completing this after some more urgent matter. A new system initialization hook is required, I will add this too.
Giovanni
- Giovanni
- Site Admin
- Posts: 14457
- Joined: Wed May 27, 2009 8:48 am
- Location: Salerno, Italy
- Has thanked: 1076 times
- Been thanked: 922 times
- Contact:
Re: Round-Robin scheduler in tickless mode
Hi,
It has been not so complex after all...
I committed a new module in OSLIB: chpreempt.c/.h. Apparently it does not harm, the test suite passes but I haven't yet tested the RR scenario. I can do this at a later time but it would be great if you could stress test this a bit.
You need to setup your chconf.h this way:
and add this to the makefile:
There is a call to chVTSetI() for each context switch, performance impact TBD, the test suite benchmarks should tell the truth.
Giovanni
It has been not so complex after all...
I committed a new module in OSLIB: chpreempt.c/.h. Apparently it does not harm, the test suite passes but I haven't yet tested the RR scenario. I can do this at a later time but it would be great if you could stress test this a bit.
You need to setup your chconf.h this way:
Code: Select all
/*
ChibiOS - Copyright (C) 2006..2016 Giovanni Di Sirio
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/
/**
* @file templates/chconf.h
* @brief Configuration file template.
* @details A copy of this file must be placed in each project directory, it
* contains the application specific kernel settings.
*
* @addtogroup config
* @details Kernel related settings and hooks.
* @{
*/
#ifndef CHCONF_H
#define CHCONF_H
#define _CHIBIOS_RT_CONF_
#define _CHIBIOS_RT_CONF_VER_5_0_
/*===========================================================================*/
/**
* @name System timers settings
* @{
*/
/*===========================================================================*/
/**
* @brief System time counter resolution.
* @note Allowed values are 16 or 32 bits.
*/
#define CH_CFG_ST_RESOLUTION 32
/**
* @brief System tick frequency.
* @details Frequency of the system timer that drives the system ticks. This
* setting also defines the system tick time unit.
*/
#define CH_CFG_ST_FREQUENCY 10000
/**
* @brief Time intervals data size.
* @note Allowed values are 16, 32 or 64 bits.
*/
#define CH_CFG_INTERVALS_SIZE 32
/**
* @brief Time types data size.
* @note Allowed values are 16 or 32 bits.
*/
#define CH_CFG_TIME_TYPES_SIZE 32
/**
* @brief Time delta constant for the tick-less mode.
* @note If this value is zero then the system uses the classic
* periodic tick. This value represents the minimum number
* of ticks that is safe to specify in a timeout directive.
* The value one is not valid, timeouts are rounded up to
* this value.
*/
#define CH_CFG_ST_TIMEDELTA 2
/** @} */
/*===========================================================================*/
/**
* @name Kernel parameters and options
* @{
*/
/*===========================================================================*/
/**
* @brief Round robin interval.
* @details This constant is the number of system ticks allowed for the
* threads before preemption occurs. Setting this value to zero
* disables the preemption for threads with equal priority and the
* round robin becomes cooperative. Note that higher priority
* threads can still preempt, the kernel is always preemptive.
* @note Disabling the round robin preemption makes the kernel more compact
* and generally faster.
* @note The round robin preemption is not supported in tickless mode and
* must be set to zero in that case.
*/
#define CH_CFG_TIME_QUANTUM 0
/**
* @brief Managed RAM size.
* @details Size of the RAM area to be managed by the OS. If set to zero
* then the whole available RAM is used. The core memory is made
* available to the heap allocator and/or can be used directly through
* the simplified core memory allocator.
*
* @note In order to let the OS manage the whole RAM the linker script must
* provide the @p __heap_base__ and @p __heap_end__ symbols.
* @note Requires @p CH_CFG_USE_MEMCORE.
*/
#define CH_CFG_MEMCORE_SIZE 0
/**
* @brief Idle thread automatic spawn suppression.
* @details When this option is activated the function @p chSysInit()
* does not spawn the idle thread. The application @p main()
* function becomes the idle thread and must implement an
* infinite loop.
*/
#define CH_CFG_NO_IDLE_THREAD FALSE
/** @} */
/*===========================================================================*/
/**
* @name Performance options
* @{
*/
/*===========================================================================*/
/**
* @brief OS optimization.
* @details If enabled then time efficient rather than space efficient code
* is used when two possible implementations exist.
*
* @note This is not related to the compiler optimization options.
* @note The default is @p TRUE.
*/
#define CH_CFG_OPTIMIZE_SPEED TRUE
/** @} */
/*===========================================================================*/
/**
* @name Subsystem options
* @{
*/
/*===========================================================================*/
/**
* @brief Time Measurement APIs.
* @details If enabled then the time measurement APIs are included in
* the kernel.
*
* @note The default is @p TRUE.
*/
#define CH_CFG_USE_TM TRUE
/**
* @brief Threads registry APIs.
* @details If enabled then the registry APIs are included in the kernel.
*
* @note The default is @p TRUE.
*/
#define CH_CFG_USE_REGISTRY TRUE
/**
* @brief Threads synchronization APIs.
* @details If enabled then the @p chThdWait() function is included in
* the kernel.
*
* @note The default is @p TRUE.
*/
#define CH_CFG_USE_WAITEXIT TRUE
/**
* @brief Semaphores APIs.
* @details If enabled then the Semaphores APIs are included in the kernel.
*
* @note The default is @p TRUE.
*/
#define CH_CFG_USE_SEMAPHORES TRUE
/**
* @brief Semaphores queuing mode.
* @details If enabled then the threads are enqueued on semaphores by
* priority rather than in FIFO order.
*
* @note The default is @p FALSE. Enable this if you have special
* requirements.
* @note Requires @p CH_CFG_USE_SEMAPHORES.
*/
#define CH_CFG_USE_SEMAPHORES_PRIORITY FALSE
/**
* @brief Mutexes APIs.
* @details If enabled then the mutexes APIs are included in the kernel.
*
* @note The default is @p TRUE.
*/
#define CH_CFG_USE_MUTEXES TRUE
/**
* @brief Enables recursive behavior on mutexes.
* @note Recursive mutexes are heavier and have an increased
* memory footprint.
*
* @note The default is @p FALSE.
* @note Requires @p CH_CFG_USE_MUTEXES.
*/
#define CH_CFG_USE_MUTEXES_RECURSIVE FALSE
/**
* @brief Conditional Variables APIs.
* @details If enabled then the conditional variables APIs are included
* in the kernel.
*
* @note The default is @p TRUE.
* @note Requires @p CH_CFG_USE_MUTEXES.
*/
#define CH_CFG_USE_CONDVARS TRUE
/**
* @brief Conditional Variables APIs with timeout.
* @details If enabled then the conditional variables APIs with timeout
* specification are included in the kernel.
*
* @note The default is @p TRUE.
* @note Requires @p CH_CFG_USE_CONDVARS.
*/
#define CH_CFG_USE_CONDVARS_TIMEOUT TRUE
/**
* @brief Events Flags APIs.
* @details If enabled then the event flags APIs are included in the kernel.
*
* @note The default is @p TRUE.
*/
#define CH_CFG_USE_EVENTS TRUE
/**
* @brief Events Flags APIs with timeout.
* @details If enabled then the events APIs with timeout specification
* are included in the kernel.
*
* @note The default is @p TRUE.
* @note Requires @p CH_CFG_USE_EVENTS.
*/
#define CH_CFG_USE_EVENTS_TIMEOUT TRUE
/**
* @brief Synchronous Messages APIs.
* @details If enabled then the synchronous messages APIs are included
* in the kernel.
*
* @note The default is @p TRUE.
*/
#define CH_CFG_USE_MESSAGES TRUE
/**
* @brief Synchronous Messages queuing mode.
* @details If enabled then messages are served by priority rather than in
* FIFO order.
*
* @note The default is @p FALSE. Enable this if you have special
* requirements.
* @note Requires @p CH_CFG_USE_MESSAGES.
*/
#define CH_CFG_USE_MESSAGES_PRIORITY FALSE
/**
* @brief Mailboxes APIs.
* @details If enabled then the asynchronous messages (mailboxes) APIs are
* included in the kernel.
*
* @note The default is @p TRUE.
* @note Requires @p CH_CFG_USE_SEMAPHORES.
*/
#define CH_CFG_USE_MAILBOXES TRUE
/**
* @brief Core Memory Manager APIs.
* @details If enabled then the core memory manager APIs are included
* in the kernel.
*
* @note The default is @p TRUE.
*/
#define CH_CFG_USE_MEMCORE TRUE
/**
* @brief Heap Allocator APIs.
* @details If enabled then the memory heap allocator APIs are included
* in the kernel.
*
* @note The default is @p TRUE.
* @note Requires @p CH_CFG_USE_MEMCORE and either @p CH_CFG_USE_MUTEXES or
* @p CH_CFG_USE_SEMAPHORES.
* @note Mutexes are recommended.
*/
#define CH_CFG_USE_HEAP TRUE
/**
* @brief Memory Pools Allocator APIs.
* @details If enabled then the memory pools allocator APIs are included
* in the kernel.
*
* @note The default is @p TRUE.
*/
#define CH_CFG_USE_MEMPOOLS TRUE
/**
* @brief Objects FIFOs APIs.
* @details If enabled then the objects FIFOs APIs are included
* in the kernel.
*
* @note The default is @p TRUE.
*/
#define CH_CFG_USE_OBJ_FIFOS TRUE
/**
* @brief Dynamic Threads APIs.
* @details If enabled then the dynamic threads creation APIs are included
* in the kernel.
*
* @note The default is @p TRUE.
* @note Requires @p CH_CFG_USE_WAITEXIT.
* @note Requires @p CH_CFG_USE_HEAP and/or @p CH_CFG_USE_MEMPOOLS.
*/
#define CH_CFG_USE_DYNAMIC TRUE
/** @} */
/*===========================================================================*/
/**
* @name Objects factory options
* @{
*/
/*===========================================================================*/
/**
* @brief Objects Factory APIs.
* @details If enabled then the objects factory APIs are included in the
* kernel.
*
* @note The default is @p FALSE.
*/
#define CH_CFG_USE_FACTORY TRUE
/**
* @brief Maximum length for object names.
* @details If the specified length is zero then the name is stored by
* pointer but this could have unintended side effects.
*/
#define CH_CFG_FACTORY_MAX_NAMES_LENGTH 8
/**
* @brief Enables the registry of generic objects.
*/
#define CH_CFG_FACTORY_OBJECTS_REGISTRY TRUE
/**
* @brief Enables factory for generic buffers.
*/
#define CH_CFG_FACTORY_GENERIC_BUFFERS TRUE
/**
* @brief Enables factory for semaphores.
*/
#define CH_CFG_FACTORY_SEMAPHORES TRUE
/**
* @brief Enables factory for mailboxes.
*/
#define CH_CFG_FACTORY_MAILBOXES TRUE
/**
* @brief Enables factory for objects FIFOs.
*/
#define CH_CFG_FACTORY_OBJ_FIFOS TRUE
/** @} */
/*===========================================================================*/
/**
* @name Debug options
* @{
*/
/*===========================================================================*/
/**
* @brief Debug option, kernel statistics.
*
* @note The default is @p FALSE.
*/
#define CH_DBG_STATISTICS FALSE
/**
* @brief Debug option, system state check.
* @details If enabled the correct call protocol for system APIs is checked
* at runtime.
*
* @note The default is @p FALSE.
*/
#define CH_DBG_SYSTEM_STATE_CHECK FALSE
/**
* @brief Debug option, parameters checks.
* @details If enabled then the checks on the API functions input
* parameters are activated.
*
* @note The default is @p FALSE.
*/
#define CH_DBG_ENABLE_CHECKS FALSE
/**
* @brief Debug option, consistency checks.
* @details If enabled then all the assertions in the kernel code are
* activated. This includes consistency checks inside the kernel,
* runtime anomalies and port-defined checks.
*
* @note The default is @p FALSE.
*/
#define CH_DBG_ENABLE_ASSERTS FALSE
/**
* @brief Debug option, trace buffer.
* @details If enabled then the trace buffer is activated.
*
* @note The default is @p CH_DBG_TRACE_MASK_DISABLED.
*/
#define CH_DBG_TRACE_MASK CH_DBG_TRACE_MASK_NONE
/**
* @brief Trace buffer entries.
* @note The trace buffer is only allocated if @p CH_DBG_TRACE_MASK is
* different from @p CH_DBG_TRACE_MASK_DISABLED.
*/
#define CH_DBG_TRACE_BUFFER_SIZE 128
/**
* @brief Debug option, stack checks.
* @details If enabled then a runtime stack check is performed.
*
* @note The default is @p FALSE.
* @note The stack check is performed in a architecture/port dependent way.
* It may not be implemented or some ports.
* @note The default failure mode is to halt the system with the global
* @p panic_msg variable set to @p NULL.
*/
#define CH_DBG_ENABLE_STACK_CHECK FALSE
/**
* @brief Debug option, stacks initialization.
* @details If enabled then the threads working area is filled with a byte
* value when a thread is created. This can be useful for the
* runtime measurement of the used stack.
*
* @note The default is @p FALSE.
*/
#define CH_DBG_FILL_THREADS FALSE
/**
* @brief Debug option, threads profiling.
* @details If enabled then a field is added to the @p thread_t structure that
* counts the system ticks occurred while executing the thread.
*
* @note The default is @p FALSE.
* @note This debug option is not currently compatible with the
* tickless mode.
*/
#define CH_DBG_THREADS_PROFILING FALSE
/** @} */
/*===========================================================================*/
/**
* @name Kernel hooks
* @{
*/
/*===========================================================================*/
/**
* @brief Enforced round robin time quantum.
*/
#define CH_CFG_ROUND_ROBIN_QUANTUM (CH_CFG_ST_FREQUENCY / 100)
#include "chpreempt.h"
/**
* @brief System structure extension.
* @details User fields added to the end of the @p ch_system_t structure.
*/
#define CH_CFG_SYSTEM_EXTRA_FIELDS \
/* Add threads custom fields here.*/ \
CH_PREEMPT_SYSTEM_FIELDS
/**
* @brief System initialization hook.
* @details User initialization code added to the @p chSysInit() function
* just before interrupts are enabled globally.
*/
#define CH_CFG_SYSTEM_INIT_HOOK(tp) { \
/* Add threads initialization code here.*/ \
ch_preempt_system_init(); \
}
/**
* @brief Threads descriptor structure extension.
* @details User fields added to the end of the @p thread_t structure.
*/
#define CH_CFG_THREAD_EXTRA_FIELDS \
/* Add threads custom fields here.*/
/**
* @brief Threads initialization hook.
* @details User initialization code added to the @p _thread_init() function.
*
* @note It is invoked from within @p _thread_init() and implicitly from all
* the threads creation APIs.
*/
#define CH_CFG_THREAD_INIT_HOOK(tp) { \
/* Add threads initialization code here.*/ \
}
/**
* @brief Threads finalization hook.
* @details User finalization code added to the @p chThdExit() API.
*/
#define CH_CFG_THREAD_EXIT_HOOK(tp) { \
/* Add threads finalization code here.*/ \
}
/**
* @brief Context switch hook.
* @details This hook is invoked just before switching between threads.
*/
#define CH_CFG_CONTEXT_SWITCH_HOOK(ntp, otp) { \
/* Context switch code here.*/ \
ch_preempt_thread_switch(); \
}
/**
* @brief ISR enter hook.
*/
#define CH_CFG_IRQ_PROLOGUE_HOOK() { \
/* IRQ prologue code here.*/ \
}
/**
* @brief ISR exit hook.
*/
#define CH_CFG_IRQ_EPILOGUE_HOOK() { \
/* IRQ epilogue code here.*/ \
}
/**
* @brief Idle thread enter hook.
* @note This hook is invoked within a critical zone, no OS functions
* should be invoked from here.
* @note This macro can be used to activate a power saving mode.
*/
#define CH_CFG_IDLE_ENTER_HOOK() { \
/* Idle-enter code here.*/ \
}
/**
* @brief Idle thread leave hook.
* @note This hook is invoked within a critical zone, no OS functions
* should be invoked from here.
* @note This macro can be used to deactivate a power saving mode.
*/
#define CH_CFG_IDLE_LEAVE_HOOK() { \
/* Idle-leave code here.*/ \
}
/**
* @brief Idle Loop hook.
* @details This hook is continuously invoked by the idle thread loop.
*/
#define CH_CFG_IDLE_LOOP_HOOK() { \
/* Idle loop code here.*/ \
}
/**
* @brief System tick event hook.
* @details This hook is invoked in the system tick handler immediately
* after processing the virtual timers queue.
*/
#define CH_CFG_SYSTEM_TICK_HOOK() { \
/* System tick event code here.*/ \
}
/**
* @brief System halt hook.
* @details This hook is invoked in case to a system halting error before
* the system is halted.
*/
#define CH_CFG_SYSTEM_HALT_HOOK(reason) { \
/* System halt code here.*/ \
}
/**
* @brief Trace hook.
* @details This hook is invoked each time a new record is written in the
* trace buffer.
*/
#define CH_CFG_TRACE_HOOK(tep) { \
/* Trace code here.*/ \
}
/** @} */
/*===========================================================================*/
/* Port-specific settings (override port settings defaulted in chcore.h). */
/*===========================================================================*/
#endif /* CHCONF_H */
/** @} */
and add this to the makefile:
Code: Select all
$(CHIBIOS)/os/common/oslib/src/chpreempt.c \
There is a call to chVTSetI() for each context switch, performance impact TBD, the test suite benchmarks should tell the truth.
Giovanni
-
- Posts: 135
- Joined: Wed Feb 04, 2015 5:03 pm
- Location: CITEC, Bielefeld University, germany
- Has thanked: 15 times
- Been thanked: 24 times
- Contact:
Re: Round-Robin scheduler in tickless mode
Hi,
thanks for realizing the feature so quickly! Unfortunately, I can not get it running for my setup Here is the issue:
I am working with the latest version of the stable_17.6.x branch because my code is not compatible with the changes in trunk. I thus patched my copy with your changes for the RR scheduler and at first glance it seemed fine - no compilation issues. But when I wanted to run it on my custom STM32F1 board, it would not start at all. I think I identified the issue, though: The new implementation of chSchdIsPreemptionRequired() in chpreempt.c calls chVTIsArmed(), which again call chSysLock() and chSysUnlock(). These calls probably fail as chSchdIsPreemptionRequired() is called by _port_irq_epilogue() in chcore_v7m.c from within an ISR locked context. So my guess is that this function is called during system startup and fails. I am not sure if this is an issue with the 17.6 branch or with my port, though.
BTW, the line to be added to the Makefile must be
(you missed the "src/")
Best regards,
Thomas
thanks for realizing the feature so quickly! Unfortunately, I can not get it running for my setup Here is the issue:
I am working with the latest version of the stable_17.6.x branch because my code is not compatible with the changes in trunk. I thus patched my copy with your changes for the RR scheduler and at first glance it seemed fine - no compilation issues. But when I wanted to run it on my custom STM32F1 board, it would not start at all. I think I identified the issue, though: The new implementation of chSchdIsPreemptionRequired() in chpreempt.c calls chVTIsArmed(), which again call chSysLock() and chSysUnlock(). These calls probably fail as chSchdIsPreemptionRequired() is called by _port_irq_epilogue() in chcore_v7m.c from within an ISR locked context. So my guess is that this function is called during system startup and fails. I am not sure if this is an issue with the 17.6 branch or with my port, though.
BTW, the line to be added to the Makefile must be
Code: Select all
$(CHIBIOS)/os/common/oslib/src/chpreempt.c \
(you missed the "src/")
Best regards,
Thomas
- Giovanni
- Site Admin
- Posts: 14457
- Joined: Wed May 27, 2009 8:48 am
- Location: Salerno, Italy
- Has thanked: 1076 times
- Been thanked: 922 times
- Contact:
-
- Posts: 135
- Joined: Wed Feb 04, 2015 5:03 pm
- Location: CITEC, Bielefeld University, germany
- Has thanked: 15 times
- Been thanked: 24 times
- Contact:
Re: Round-Robin scheduler in tickless mode
I just edited my last post, when you replied. I could fix the issue:
First, I just replaced the call of chVTIsArmed(&ch.preempt_vt) by its content (ch.preempt.func != NULL). SECOND, I had To fix another issue in chSchDoReschedule() where you called chVTIsArmed(&ch.preempt_vt) instead of chVTIsArmedI(&ch.preempt_vt). Now my system boots up and - voila! - it works =)
I have two threads executing an infinite loop (without sleep or such) running with identical priorities and they are both responsive. If I decrease the priority of one thread the other one is virtually dead, as expected. Even though the implementation is not the cleanest yet and can probably be optimized, it works!
Nice job, Giovanni, and thanks again!
- Thomas
First, I just replaced the call of chVTIsArmed(&ch.preempt_vt) by its content (ch.preempt.func != NULL). SECOND, I had To fix another issue in chSchDoReschedule() where you called chVTIsArmed(&ch.preempt_vt) instead of chVTIsArmedI(&ch.preempt_vt). Now my system boots up and - voila! - it works =)
I have two threads executing an infinite loop (without sleep or such) running with identical priorities and they are both responsive. If I decrease the priority of one thread the other one is virtually dead, as expected. Even though the implementation is not the cleanest yet and can probably be optimized, it works!
Nice job, Giovanni, and thanks again!
- Thomas
Return to “Small Change Requests”
Who is online
Users browsing this forum: No registered users and 30 guests