1================================================================= 2CPU Scheduler implementation hints for architecture specific code 3================================================================= 4 5 Nick Piggin, 2005 6 7Context switch 8============== 91. Runqueue locking 10By default, the switch_to arch function is called with the runqueue 11locked. This is usually not a problem unless switch_to may need to 12take the runqueue lock. This is usually due to a wake up operation in 13the context switch. See arch/ia64/include/asm/switch_to.h for an example. 14 15To request the scheduler call switch_to with the runqueue unlocked, 16you must `#define __ARCH_WANT_UNLOCKED_CTXSW` in a header file 17(typically the one where switch_to is defined). 18 19Unlocked context switches introduce only a very minor performance 20penalty to the core scheduler implementation in the CONFIG_SMP case. 21 22CPU idle 23======== 24Your cpu_idle routines need to obey the following rules: 25 261. Preempt should now disabled over idle routines. Should only 27 be enabled to call schedule() then disabled again. 28 292. need_resched/TIF_NEED_RESCHED is only ever set, and will never 30 be cleared until the running task has called schedule(). Idle 31 threads need only ever query need_resched, and may never set or 32 clear it. 33 343. When cpu_idle finds (need_resched() == 'true'), it should call 35 schedule(). It should not call schedule() otherwise. 36 374. The only time interrupts need to be disabled when checking 38 need_resched is if we are about to sleep the processor until 39 the next interrupt (this doesn't provide any protection of 40 need_resched, it prevents losing an interrupt): 41 42 4a. Common problem with this type of sleep appears to be:: 43 44 local_irq_disable(); 45 if (!need_resched()) { 46 local_irq_enable(); 47 *** resched interrupt arrives here *** 48 __asm__("sleep until next interrupt"); 49 } 50 515. TIF_POLLING_NRFLAG can be set by idle routines that do not 52 need an interrupt to wake them up when need_resched goes high. 53 In other words, they must be periodically polling need_resched, 54 although it may be reasonable to do some background work or enter 55 a low CPU priority. 56 57 - 5a. If TIF_POLLING_NRFLAG is set, and we do decide to enter 58 an interrupt sleep, it needs to be cleared then a memory 59 barrier issued (followed by a test of need_resched with 60 interrupts disabled, as explained in 3). 61 62arch/x86/kernel/process.c has examples of both polling and 63sleeping idle functions. 64 65 66Possible arch/ problems 67======================= 68 69Possible arch problems I found (and either tried to fix or didn't): 70 71ia64 - is safe_halt call racy vs interrupts? (does it sleep?) (See #4a) 72 73sparc - IRQs on at this point(?), change local_irq_save to _disable. 74 - TODO: needs secondary CPUs to disable preempt (See #1) 75