Changeset 624 for trunk/kernel/libk/busylock.h

Timestamp:

Mar 12, 2019, 1:37:38 PM (5 years ago)

Author:

alain

Message:

Fix several bugs to use the instruction MMU in kernel mode
in replacement of the instruction address extension register,
and remove the "kentry" segment.

This version is running on the tsar_generic_iob" platform.

One interesting bug: the cp0_ebase defining the kernel entry point
(for interrupts, exceptions and syscalls) must be initialized
early in kernel_init(), because the VFS initialisation done by
kernel_ini() uses RPCs, and RPCs uses Inter-Processor-Interrup.

File:

• trunk/kernel/libk/busylock.h (modified) (3 diffs)

trunk/kernel/libk/busylock.h

@@ -2 +2 @@
  * busylock.h: local kernel busy-waiting lock definition.     
  * 
- * Authors  Alain Greiner (2016,2017,2018)
+ * Authors  Alain Greiner (2016,2017,2018,2019)
  *
  * Copyright (c) UPMC Sorbonne Universites
@@ -34 +34 @@
  * a shared object located in a given cluster, made by thread(s) running in same cluster.
  * It uses a busy waiting policy when the lock is taken by another thread, and should
- * be used to execute very short actions, such as accessing basic allocators, or higher 
+ * be used to execute very short actions, such as accessing basic allocators or higher 
  * level synchronisation objects (barriers, queuelocks, or rwlocks).
- * WARNING: a thread cannot yield when it is owning a busylock.
  * 
  * - To acquire the lock, we use a ticket policy to avoid starvation: the calling thread
@@ -42 +41 @@
  *   value  until current == ticket. 
  *
- * - To release the lock, the owner thread increments the "current" value,
- *   decrements its busylocks counter. 
+ * - To release the lock, the owner thread increments the "current" value.
  *
- * - When a thread takes a busylock, it enters a critical section: the busylock_acquire() 
+ * - When a thread takes a busylock, it enters a critical section: the acquire() 
  *   function disables the IRQs, takes the lock, increments the thread busylocks counter,
- *   and save the SR in lock descriptor and returns.
+ *   save the SR in lock descriptor and returns.
  *
- * - The busylock_release() function releases the lock, decrements the thread busylock
- *   counter, restores the SR to exit the critical section, and returns
+ * - The release() function releases the lock, decrements the thread busylock
+ *   counter, restores the SR to exit the critical section, and returns.
  *
- * - If a thread owning a busylock (local or remote) tries to deschedule, the scheduler
- *   signals a kernel panic. 
+ * WARNING: a thread cannot yield when it is holding a busylock (local or remote).
+ *
+ * This rule is checked by all functions containing a thread_yield() AND by the scheduler,
+ * thanks to the busylocks counter stored in the calling thread descriptor.
+ * 1) all functions call "thread_assert_can_yield()" before calling "thread_yield()".
+ * 2) The scheduler checks that the calling thread does not hold any busylock.
+ * In case of violation the core goes to sleep after a [PANIC] message on TXT0.
  ******************************************************************************************/
 
 /*******************************************************************************************
  * This structure defines a busylock.
- * The <type> and <xlist> fields are used for debug.
- * The type defines the lock usage as detailed in the kernel_config.h file. 
+ * The <xlist> field is only used when DEBUG_BUSYLOCK is set.
+ * The <type> field defines the lock usage as detailed in the kernel_config.h file. 
 ******************************************************************************************/