- 13 Jan, 2014 37 commits
-
-
Peter Zijlstra authored
The only valid use of preempt_enable_no_resched() is if the very next line is schedule() or if we know preemption cannot actually be enabled by that statement due to known more preempt_count 'refs'. Signed-off-by:
Peter Zijlstra <peterz@infradead.org> Cc: rjw@rjwysocki.net Cc: Eliezer Tamir <eliezer.tamir@linux.intel.com> Cc: rui.zhang@intel.com Cc: jacob.jun.pan@linux.intel.com Cc: Mike Galbraith <bitbucket@online.de> Cc: hpa@zytor.com Cc: Arjan van de Ven <arjan@linux.intel.com> Cc: lenb@kernel.org Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/n/tip-zcfvacdlvlr63qmnn5i58vuj@git.kernel.orgSigned-off-by:
Ingo Molnar <mingo@kernel.org>
-
Peter Zijlstra authored
The only valid use of preempt_enable_no_resched() is if the very next line is schedule() or if we know preemption cannot actually be enabled by that statement due to known more preempt_count 'refs'. This busy_poll stuff looks to be completely and utterly broken, sched_clock() can return utter garbage with interrupts enabled (rare but still) and it can drift unbounded between CPUs. This means that if you get preempted/migrated and your new CPU is years behind on the previous CPU we get to busy spin for a _very_ long time. There is a _REASON_ sched_clock() warns about preemptability - papering over it with a preempt_disable()/preempt_enable_no_resched() is just terminal brain damage on so many levels. Replace sched_clock() usage with local_clock() which has a bounded drift between CPUs (<2 jiffies). There is a further problem with the entire busy wait poll thing in that the spin time is additive to the syscall timeout, not inclusive. Reviewed-by:
Thomas Gleixner <tglx@linutronix.de> Signed-off-by:
-
Peter Zijlstra authored
The only valid use of preempt_enable_no_resched() is if the very next line is schedule() or if we know preemption cannot actually be enabled by that statement due to known more preempt_count 'refs'. Reviewed-by:
-
Peter Zijlstra authored
With various drivers wanting to inject idle time; we get people calling idle routines outside of the idle loop proper. Therefore we need to be extra careful about not missing TIF_NEED_RESCHED -> PREEMPT_NEED_RESCHED propagations. While looking at this, I also realized there's a small window in the existing idle loop where we can miss TIF_NEED_RESCHED; when it hits right after the tif_need_resched() test at the end of the loop but right before the need_resched() test at the start of the loop. So move preempt_fold_need_resched() out of the loop where we're guaranteed to have TIF_NEED_RESCHED set. Signed-off-by:
-
Ingo Molnar authored
Merge these x86 specific bits - we are going to add generic bits as well. Signed-off-by: -
Peter Zijlstra authored
Currently local_bh_disable() is out-of-line for no apparent reason. So inline it to save a few cycles on call/return nonsense, the function body is a single add on x86 (a few loads and store extra on load/store archs). Also expose two new local_bh functions: __local_bh_{dis,en}able_ip(unsigned long ip, unsigned int cnt); Which implement the actual local_bh_{dis,en}able() behaviour. The next patch uses the exposed @cnt argument to optimize bh lock functions. With build fixes from Jacob Pan. Cc: rjw@rjwysocki.net Cc: rui.zhang@intel.com Cc: jacob.jun.pan@linux.intel.com Cc: Mike Galbraith <bitbucket@online.de> Cc: hpa@zytor.com Cc: Arjan van de Ven <arjan@linux.intel.com> Cc: lenb@kernel.org Reviewed-by: -
Peter Zijlstra authored
Use a static_key to avoid touching tsc_disabled and a runtime condition in native_sched_clock() -- less cachelines touched is always better. MAINLINE PRE POST sched_clock_stable: 1 1 1 (cold) sched_clock: 329841 215295 213039 (cold) local_clock: 301773 220773 216084 (warm) sched_clock: 38375 25659 25231 (warm) local_clock: 100371 27242 27601 (warm) rdtsc: 27340 24208 24203 sched_clock_stable: 0 0 0 (cold) sched_clock: 382634 237019 240055 (cold) local_clock: 396890 294819 299942 (warm) sched_clock: 38194 25609 25276 (warm) local_clock: 143452 71232 73232 (warm) rdtsc: 27345 24243 24244 Signed-off-by: -
Peter Zijlstra authored
The below tells us the static_key conversion has a problem; since the exact point of clearing that flag isn't too important, delay the flip and use a workqueue to process it. [ ] TSC synchronization [CPU#0 -> CPU#22]: [ ] Measured 8 cycles TSC warp between CPUs, turning off TSC clock. [ ] [ ] ====================================================== [ ] [ INFO: possible circular locking dependency detected ] [ ] 3.13.0-rc3-01745-g848b0d0322cb-dirty #637 Not tainted [ ] ------------------------------------------------------- [ ] swapper/0/1 is trying to acquire lock: [ ] (jump_label_mutex){+.+...}, at: [<ffffffff8115a637>] jump_label_lock+0x17/0x20 [ ] [ ] but task is already holding lock: [ ] (cpu_hotplug.lock){+.+.+.}, at: [<ffffffff8109408b>] cpu_hotplug_begin+0x2b/0x60 [ ] [ ] which lock already depends on the new lock. [ ] [ ] [ ] the existing dependency chain (in reverse order) is: [ ] [ ] -> #1 (cpu_hotplug.lock){+.+.+.}: [ ] [<ffffffff810def00>] lock_acquire+0x90/0x130 [ ] [<ffffffff81661f83>] mutex_lock_nested+0x63/0x3e0 [ ] [<ffffffff81093fdc>] get_online_cpus+0x3c/0x60 [ ] [<ffffffff8104cc67>] arch_jump_label_transform+0x37/0x130 [ ] [<ffffffff8115a3cf>] __jump_label_update+0x5f/0x80 [ ] [<ffffffff8115a48d>] jump_label_update+0x9d/0xb0 [ ] [<ffffffff8115aa6d>] static_key_slow_inc+0x9d/0xb0 [ ] [<ffffffff810c0f65>] sched_feat_set+0xf5/0x100 [ ] [<ffffffff810c5bdc>] set_numabalancing_state+0x2c/0x30 [ ] [<ffffffff81d12f3d>] numa_policy_init+0x1af/0x1b7 [ ] [<ffffffff81cebdf4>] start_kernel+0x35d/0x41f [ ] [<ffffffff81ceb5a5>] x86_64_start_reservations+0x2a/0x2c [ ] [<ffffffff81ceb6a2>] x86_64_start_kernel+0xfb/0xfe [ ] [ ] -> #0 (jump_label_mutex){+.+...}: [ ] [<ffffffff810de141>] __lock_acquire+0x1701/0x1eb0 [ ] [<ffffffff810def00>] lock_acquire+0x90/0x130 [ ] [<ffffffff81661f83>] mutex_lock_nested+0x63/0x3e0 [ ] [<ffffffff8115a637>] jump_label_lock+0x17/0x20 [ ] [<ffffffff8115aa3b>] static_key_slow_inc+0x6b/0xb0 [ ] [<ffffffff810ca775>] clear_sched_clock_stable+0x15/0x20 [ ] [<ffffffff810503b3>] mark_tsc_unstable+0x23/0x70 [ ] [<ffffffff810772cb>] check_tsc_sync_source+0x14b/0x150 [ ] [<ffffffff81076612>] native_cpu_up+0x3a2/0x890 [ ] [<ffffffff810941cb>] _cpu_up+0xdb/0x160 [ ] [<ffffffff810942c9>] cpu_up+0x79/0x90 [ ] [<ffffffff81d0af6b>] smp_init+0x60/0x8c [ ] [<ffffffff81cebf42>] kernel_init_freeable+0x8c/0x197 [ ] [<ffffffff8164e32e>] kernel_init+0xe/0x130 [ ] [<ffffffff8166beec>] ret_from_fork+0x7c/0xb0 [ ] [ ] other info that might help us debug this: [ ] [ ] Possible unsafe locking scenario: [ ] [ ] CPU0 CPU1 [ ] ---- ---- [ ] lock(cpu_hotplug.lock); [ ] lock(jump_label_mutex); [ ] lock(cpu_hotplug.lock); [ ] lock(jump_label_mutex); [ ] [ ] *** DEADLOCK *** [ ] [ ] 2 locks held by swapper/0/1: [ ] #0: (cpu_add_remove_lock){+.+.+.}, at: [<ffffffff81094037>] cpu_maps_update_begin+0x17/0x20 [ ] #1: (cpu_hotplug.lock){+.+.+.}, at: [<ffffffff8109408b>] cpu_hotplug_begin+0x2b/0x60 [ ] [ ] stack backtrace: [ ] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 3.13.0-rc3-01745-g848b0d0322cb-dirty #637 [ ] Hardware name: Supermicro X8DTN/X8DTN, BIOS 4.6.3 01/08/2010 [ ] ffffffff82c9c270 ffff880236843bb8 ffffffff8165c5f5 ffffffff82c9c270 [ ] ffff880236843bf8 ffffffff81658c02 ffff880236843c80 ffff8802368586a0 [ ] ffff880236858678 0000000000000001 0000000000000002 ffff880236858000 [ ] Call Trace: [ ] [<ffffffff8165c5f5>] dump_stack+0x4e/0x7a [ ] [<ffffffff81658c02>] print_circular_bug+0x1f9/0x207 [ ] [<ffffffff810de141>] __lock_acquire+0x1701/0x1eb0 [ ] [<ffffffff816680ff>] ? __atomic_notifier_call_chain+0x8f/0xb0 [ ] [<ffffffff810def00>] lock_acquire+0x90/0x130 [ ] [<ffffffff8115a637>] ? jump_label_lock+0x17/0x20 [ ] [<ffffffff8115a637>] ? jump_label_lock+0x17/0x20 [ ] [<ffffffff81661f83>] mutex_lock_nested+0x63/0x3e0 [ ] [<ffffffff8115a637>] ? jump_label_lock+0x17/0x20 [ ] [<ffffffff8115a637>] jump_label_lock+0x17/0x20 [ ] [<ffffffff8115aa3b>] static_key_slow_inc+0x6b/0xb0 [ ] [<ffffffff810ca775>] clear_sched_clock_stable+0x15/0x20 [ ] [<ffffffff810503b3>] mark_tsc_unstable+0x23/0x70 [ ] [<ffffffff810772cb>] check_tsc_sync_source+0x14b/0x150 [ ] [<ffffffff81076612>] native_cpu_up+0x3a2/0x890 [ ] [<ffffffff810941cb>] _cpu_up+0xdb/0x160 [ ] [<ffffffff810942c9>] cpu_up+0x79/0x90 [ ] [<ffffffff81d0af6b>] smp_init+0x60/0x8c [ ] [<ffffffff81cebf42>] kernel_init_freeable+0x8c/0x197 [ ] [<ffffffff8164e320>] ? rest_init+0xd0/0xd0 [ ] [<ffffffff8164e32e>] kernel_init+0xe/0x130 [ ] [<ffffffff8166beec>] ret_from_fork+0x7c/0xb0 [ ] [<ffffffff8164e320>] ? rest_init+0xd0/0xd0 [ ] ------------[ cut here ]------------ [ ] WARNING: CPU: 0 PID: 1 at /usr/src/linux-2.6/kernel/smp.c:374 smp_call_function_many+0xad/0x300() [ ] Modules linked in: [ ] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 3.13.0-rc3-01745-g848b0d0322cb-dirty #637 [ ] Hardware name: Supermicro X8DTN/X8DTN, BIOS 4.6.3 01/08/2010 [ ] 0000000000000009 ffff880236843be0 ffffffff8165c5f5 0000000000000000 [ ] ffff880236843c18 ffffffff81093d8c 0000000000000000 0000000000000000 [ ] ffffffff81ccd1a0 ffffffff810ca951 0000000000000000 ffff880236843c28 [ ] Call Trace: [ ] [<ffffffff8165c5f5>] dump_stack+0x4e/0x7a [ ] [<ffffffff81093d8c>] warn_slowpath_common+0x8c/0xc0 [ ] [<ffffffff810ca951>] ? sched_clock_tick+0x1/0xa0 [ ] [<ffffffff81093dda>] warn_slowpath_null+0x1a/0x20 [ ] [<ffffffff8110b72d>] smp_call_function_many+0xad/0x300 [ ] [<ffffffff8104f200>] ? arch_unregister_cpu+0x30/0x30 [ ] [<ffffffff8104f200>] ? arch_unregister_cpu+0x30/0x30 [ ] [<ffffffff810ca951>] ? sched_clock_tick+0x1/0xa0 [ ] [<ffffffff8110ba96>] smp_call_function+0x46/0x80 [ ] [<ffffffff8104f200>] ? arch_unregister_cpu+0x30/0x30 [ ] [<ffffffff8110bb3c>] on_each_cpu+0x3c/0xa0 [ ] [<ffffffff810ca950>] ? sched_clock_idle_sleep_event+0x20/0x20 [ ] [<ffffffff810ca951>] ? sched_clock_tick+0x1/0xa0 [ ] [<ffffffff8104f964>] text_poke_bp+0x64/0xd0 [ ] [<ffffffff810ca950>] ? sched_clock_idle_sleep_event+0x20/0x20 [ ] [<ffffffff8104ccde>] arch_jump_label_transform+0xae/0x130 [ ] [<ffffffff8115a3cf>] __jump_label_update+0x5f/0x80 [ ] [<ffffffff8115a48d>] jump_label_update+0x9d/0xb0 [ ] [<ffffffff8115aa6d>] static_key_slow_inc+0x9d/0xb0 [ ] [<ffffffff810ca775>] clear_sched_clock_stable+0x15/0x20 [ ] [<ffffffff810503b3>] mark_tsc_unstable+0x23/0x70 [ ] [<ffffffff810772cb>] check_tsc_sync_source+0x14b/0x150 [ ] [<ffffffff81076612>] native_cpu_up+0x3a2/0x890 [ ] [<ffffffff810941cb>] _cpu_up+0xdb/0x160 [ ] [<ffffffff810942c9>] cpu_up+0x79/0x90 [ ] [<ffffffff81d0af6b>] smp_init+0x60/0x8c [ ] [<ffffffff81cebf42>] kernel_init_freeable+0x8c/0x197 [ ] [<ffffffff8164e320>] ? rest_init+0xd0/0xd0 [ ] [<ffffffff8164e32e>] kernel_init+0xe/0x130 [ ] [<ffffffff8166beec>] ret_from_fork+0x7c/0xb0 [ ] [<ffffffff8164e320>] ? rest_init+0xd0/0xd0 [ ] ---[ end trace 6ff1df5620c49d26 ]--- [ ] tsc: Marking TSC unstable due to check_tsc_sync_source failed Signed-off-by: -
Peter Zijlstra authored
In order to avoid the runtime condition and variable load turn sched_clock_stable into a static_key. Also provide a shorter implementation of local_clock() and cpu_clock(int) when sched_clock_stable==1. MAINLINE PRE POST sched_clock_stable: 1 1 1 (cold) sched_clock: 329841 221876 215295 (cold) local_clock: 301773 234692 220773 (warm) sched_clock: 38375 25602 25659 (warm) local_clock: 100371 33265 27242 (warm) rdtsc: 27340 24214 24208 sched_clock_stable: 0 0 0 (cold) sched_clock: 382634 235941 237019 (cold) local_clock: 396890 297017 294819 (warm) sched_clock: 38194 25233 25609 (warm) local_clock: 143452 71234 71232 (warm) rdtsc: 27345 24245 24243 Signed-off-by: -
Peter Zijlstra authored
Now that x86 no longer requires IRQs disabled for sched_clock() and ia64 never had this requirement (it doesn't seem to do cpufreq at all), we can remove the requirement of disabling IRQs. MAINLINE PRE POST sched_clock_stable: 1 1 1 (cold) sched_clock: 329841 257223 221876 (cold) local_clock: 301773 309889 234692 (warm) sched_clock: 38375 25280 25602 (warm) local_clock: 100371 85268 33265 (warm) rdtsc: 27340 24247 24214 sched_clock_stable: 0 0 0 (cold) sched_clock: 382634 301224 235941 (cold) local_clock: 396890 399870 297017 (warm) sched_clock: 38194 25630 25233 (warm) local_clock: 143452 129629 71234 (warm) rdtsc: 27345 24307 24245 Signed-off-by: -
Peter Zijlstra authored
Use a ring-buffer like multi-version object structure which allows always having a coherent object; we use this to avoid having to disable IRQs while reading sched_clock() and avoids a problem when getting an NMI while changing the cyc2ns data. MAINLINE PRE POST sched_clock_stable: 1 1 1 (cold) sched_clock: 329841 331312 257223 (cold) local_clock: 301773 310296 309889 (warm) sched_clock: 38375 38247 25280 (warm) local_clock: 100371 102713 85268 (warm) rdtsc: 27340 27289 24247 sched_clock_stable: 0 0 0 (cold) sched_clock: 382634 372706 301224 (cold) local_clock: 396890 399275 399870 (warm) sched_clock: 38194 38124 25630 (warm) local_clock: 143452 148698 129629 (warm) rdtsc: 27345 27365 24307 Signed-off-by: -
Peter Zijlstra authored
There are no __cycles_2_ns() users outside of arch/x86/kernel/tsc.c, so move it there. There are no cycles_2_ns() users. Signed-off-by:
-
Peter Zijlstra authored
Use mul_u64_u32_shr() so that x86_64 can use a single 64x64->128 mul. Before: 0000000000000560 <native_sched_clock>: 560: 44 8b 1d 00 00 00 00 mov 0x0(%rip),%r11d # 567 <native_sched_clock+0x7> 567: 55 push %rbp 568: 48 89 e5 mov %rsp,%rbp 56b: 45 85 db test %r11d,%r11d 56e: 75 4f jne 5bf <native_sched_clock+0x5f> 570: 0f 31 rdtsc 572: 89 c0 mov %eax,%eax 574: 48 c1 e2 20 shl $0x20,%rdx 578: 48 c7 c1 00 00 00 00 mov $0x0,%rcx 57f: 48 09 c2 or %rax,%rdx 582: 48 c7 c7 00 00 00 00 mov $0x0,%rdi 589: 65 8b 04 25 00 00 00 mov %gs:0x0,%eax 590: 00 591: 48 98 cltq 593: 48 8b 34 c5 00 00 00 mov 0x0(,%rax,8),%rsi 59a: 00 59b: 48 89 d0 mov %rdx,%rax 59e: 81 e2 ff 03 00 00 and $0x3ff,%edx 5a4: 48 c1 e8 0a shr $0xa,%rax 5a8: 48 0f af 14 0e imul (%rsi,%rcx,1),%rdx 5ad: 48 0f af 04 0e imul (%rsi,%rcx,1),%rax 5b2: 5d pop %rbp 5b3: 48 03 04 3e add (%rsi,%rdi,1),%rax 5b7: 48 c1 ea 0a shr $0xa,%rdx 5bb: 48 01 d0 add %rdx,%rax 5be: c3 retq After: 0000000000000550 <native_sched_clock>: 550: 8b 3d 00 00 00 00 mov 0x0(%rip),%edi # 556 <native_sched_clock+0x6> 556: 55 push %rbp 557: 48 89 e5 mov %rsp,%rbp 55a: 48 83 e4 f0 and $0xfffffffffffffff0,%rsp 55e: 85 ff test %edi,%edi 560: 75 2c jne 58e <native_sched_clock+0x3e> 562: 0f 31 rdtsc 564: 89 c0 mov %eax,%eax 566: 48 c1 e2 20 shl $0x20,%rdx 56a: 48 09 c2 or %rax,%rdx 56d: 65 48 8b 04 25 00 00 mov %gs:0x0,%rax 574: 00 00 576: 89 c0 mov %eax,%eax 578: 48 f7 e2 mul %rdx 57b: 65 48 8b 0c 25 00 00 mov %gs:0x0,%rcx 582: 00 00 584: c9 leaveq 585: 48 0f ac d0 0a shrd $0xa,%rdx,%rax 58a: 48 01 c8 add %rcx,%rax 58d: c3 retq MAINLINE POST sched_clock_stable: 1 1 (cold) sched_clock: 329841 331312 (cold) local_clock: 301773 310296 (warm) sched_clock: 38375 38247 (warm) local_clock: 100371 102713 (warm) rdtsc: 27340 27289 sched_clock_stable: 0 0 (cold) sched_clock: 382634 372706 (cold) local_clock: 396890 399275 (warm) sched_clock: 38194 38124 (warm) local_clock: 143452 148698 (warm) rdtsc: 27345 27365 Signed-off-by: -
Peter Zijlstra authored
Discourage drivers/modules to be creative with preemption. Sadly all is implemented in macros and inline so if they want to do evil they still can, but at least try and discourage some. Reviewed-by:
-
Peter Zijlstra authored
Currently all _bh_ lock functions do two preempt_count operations: local_bh_disable(); preempt_disable(); and for the unlock: preempt_enable_no_resched(); local_bh_enable(); Since its a waste of perfectly good cycles to modify the same variable twice when you can do it in one go; use the new __local_bh_{dis,en}able_ip() functions that allow us to provide a preempt_count value to add/sub. So define SOFTIRQ_LOCK_OFFSET as the offset a _bh_ lock needs to add/sub to be done in one go. As a bonus it gets rid of the preempt_enable_no_resched() usage. This reduces a 1000 loops of: spin_lock_bh(&bh_lock); spin_unlock_bh(&bh_lock); from 53596 cycles to 51995 cycles. I didn't do enough measurements to say for absolute sure that the result is significant but the the few runs I did for each suggest it is so. Reviewed-by: -
Daniel Lezcano authored
The test on_null_domain is done twice in the trigger_load_balance function. Move the test at the begin of the function, so there is only one check. Signed-off-by:
Daniel Lezcano <daniel.lezcano@linaro.org> Signed-off-by:
-
Daniel Lezcano authored
The cpu information is stored in the struct rq. Pass the struct rq to nohz_idle_balance, so all the functions called in run_rebalance_domains have the same parameters and the 'this_cpu' variable becomes pointless. Signed-off-by:
-
Daniel Lezcano authored
The cpu information is stored in the struct rq and the caller of the rebalance_domains function pass the cpu to retrieve the struct rq but it already has the struct rq info. Replace the cpu parameter with the struct rq. Signed-off-by:
-
Daniel Lezcano authored
The cpu parameter is no longer needed in nohz_balancer_kick, let's remove the parameter. Reviewed-by:
Preeti U Murthy <preeti@linux.vnet.ibm.com> Signed-off-by:
-
Daniel Lezcano authored
The 'call_cpu' is never used in the function. Remove it. Reviewed-by:
-
Daniel Lezcano authored
The on_null_domain() function is getting the cpu to retrieve the struct rq associated with it. Pass 'struct rq' directly to the function as the caller already has the info. Signed-off-by:
-
Daniel Lezcano authored
The cpu information is already stored in the struct rq, so no need to pass it as parameter to the nohz_kick_needed function. The caller of this function just called idle_cpu() before to fill the rq->idle_balance field. Use rq->cpu and rq->idle_balance. Signed-off-by:
-
Daniel Lezcano authored
The cpu information is already stored in the struct rq, so no need to pass it as parameter to the trigger_load_balance function. Cc: linaro-kernel@lists.linaro.org Cc: preeti.lkml@gmail.com Cc: mingo@redhat.com Cc: peterz@infradead.org Signed-off-by:
-
Peter Zijlstra authored
The current hotplug admission control is broken because: CPU_DYING -> migration_call() -> migrate_tasks() -> __migrate_task() cannot fail and hard assumes it _will_ move all tasks off of the dying cpu, failing this will break hotplug. The much simpler solution is a DOWN_PREPARE handler that fails when removing one CPU gets us below the total allocated bandwidth. Signed-off-by:
-
Peter Zijlstra authored
Remove the deadline specific sysctls for now. The problem with them is that the interaction with the exisiting rt knobs is nearly impossible to get right. The current (as per before this patch) situation is that the rt and dl bandwidth is completely separate and we enforce rt+dl < 100%. This is undesirable because this means that the rt default of 95% leaves us hardly any room, even though dl tasks are saver than rt tasks. Another proposed solution was (a discarted patch) to have the dl bandwidth be a fraction of the rt bandwidth. This is highly confusing imo. Furthermore neither proposal is consistent with the situation we actually want; which is rt tasks ran from a dl server. In which case the rt bandwidth is a direct subset of dl. So whichever way we go, the introduction of dl controls at this point is painful. Therefore remove them and instead share the rt budget. This means that for now the rt knobs are used for dl admission control and the dl runtime is accounted against the rt runtime. I realise that this isn't entirely desirable either; but whatever we do we appear to need to change the interface later, so better have a small interface for now. Signed-off-by:
-
Peter Zijlstra authored
For now deadline tasks are not allowed to set smp affinity; however the current tests are wrong, cure this. The test in __sched_setscheduler() also uses an on-stack cpumask_t which is a no-no. Change both tests to use cpumask_subset() such that we test the root domain span to be a subset of the cpus_allowed mask. This way we're sure the tasks can always run on all CPUs they can be balanced over, and have no effective affinity constraints. Signed-off-by:
-
Juri Lelli authored
Data from tests confirmed that the original active load balancing logic didn't scale neither in the number of CPU nor in the number of tasks (as sched_rt does). Here we provide a global data structure to keep track of deadlines of the running tasks in the system. The structure is composed by a bitmask showing the free CPUs and a max-heap, needed when the system is heavily loaded. The implementation and concurrent access scheme are kept simple by design. However, our measurements show that we can compete with sched_rt on large multi-CPUs machines [1]. Only the push path is addressed, the extension to use this structure also for pull decisions is straightforward. However, we are currently evaluating different (in order to decrease/avoid contention) data structures to solve possibly both problems. We are also going to re-run tests considering recent changes inside cpupri [2]. [1] http://retis.sssup.it/~jlelli/papers/Ospert11Lelli.pdf [2] http://www.spinics.net/lists/linux-rt-users/msg06778.htmlSigned-off-by:
Juri Lelli <juri.lelli@gmail.com> Signed-off-by:
-
Dario Faggioli authored
In order of deadline scheduling to be effective and useful, it is important that some method of having the allocation of the available CPU bandwidth to tasks and task groups under control. This is usually called "admission control" and if it is not performed at all, no guarantee can be given on the actual scheduling of the -deadline tasks. Since when RT-throttling has been introduced each task group have a bandwidth associated to itself, calculated as a certain amount of runtime over a period. Moreover, to make it possible to manipulate such bandwidth, readable/writable controls have been added to both procfs (for system wide settings) and cgroupfs (for per-group settings). Therefore, the same interface is being used for controlling the bandwidth distrubution to -deadline tasks and task groups, i.e., new controls but with similar names, equivalent meaning and with the same usage paradigm are added. However, more discussion is needed in order to figure out how we want to manage SCHED_DEADLINE bandwidth at the task group level. Therefore, this patch adds a less sophisticated, but actually very sensible, mechanism to ensure that a certain utilization cap is not overcome per each root_domain (the single rq for !SMP configurations). Another main difference between deadline bandwidth management and RT-throttling is that -deadline tasks have bandwidth on their own (while -rt ones doesn't!), and thus we don't need an higher level throttling mechanism to enforce the desired bandwidth. This patch, therefore: - adds system wide deadline bandwidth management by means of: * /proc/sys/kernel/sched_dl_runtime_us, * /proc/sys/kernel/sched_dl_period_us, that determine (i.e., runtime / period) the total bandwidth available on each CPU of each root_domain for -deadline tasks; - couples the RT and deadline bandwidth management, i.e., enforces that the sum of how much bandwidth is being devoted to -rt -deadline tasks to stay below 100%. This means that, for a root_domain comprising M CPUs, -deadline tasks can be created until the sum of their bandwidths stay below: M * (sched_dl_runtime_us / sched_dl_period_us) It is also possible to disable this bandwidth management logic, and be thus free of oversubscribing the system up to any arbitrary level. Signed-off-by:Dario Faggioli <raistlin@linux.it> Signed-off-by:
-
Dario Faggioli authored
Some method to deal with rt-mutexes and make sched_dl interact with the current PI-coded is needed, raising all but trivial issues, that needs (according to us) to be solved with some restructuring of the pi-code (i.e., going toward a proxy execution-ish implementation). This is under development, in the meanwhile, as a temporary solution, what this commits does is: - ensure a pi-lock owner with waiters is never throttled down. Instead, when it runs out of runtime, it immediately gets replenished and it's deadline is postponed; - the scheduling parameters (relative deadline and default runtime) used for that replenishments --during the whole period it holds the pi-lock-- are the ones of the waiting task with earliest deadline. Acting this way, we provide some kind of boosting to the lock-owner, still by using the existing (actually, slightly modified by the previous commit) pi-architecture. We would stress the fact that this is only a surely needed, all but clean solution to the problem. In the end it's only a way to re-start discussion within the community. So, as always, comments, ideas, rants, etc.. are welcome! :-) Signed-off-by:
-
Peter Zijlstra authored
Turn the pi-chains from plist to rb-tree, in the rt_mutex code, and provide a proper comparison function for -deadline and -priority tasks. This is done mainly because: - classical prio field of the plist is just an int, which might not be enough for representing a deadline; - manipulating such a list would become O(nr_deadline_tasks), which might be to much, as the number of -deadline task increases. Therefore, an rb-tree is used, and tasks are queued in it according to the following logic: - among two -priority (i.e., SCHED_BATCH/OTHER/RR/FIFO) tasks, the one with the higher (lower, actually!) prio wins; - among a -priority and a -deadline task, the latter always wins; - among two -deadline tasks, the one with the earliest deadline wins. Queueing and dequeueing functions are changed accordingly, for both the list of a task's pi-waiters and the list of tasks blocked on a pi-lock. Signed-off-by:
-
Dario Faggioli authored
It is very likely that systems that wants/needs to use the new SCHED_DEADLINE policy also want to have the scheduling latency of the -deadline tasks under control. For this reason a new version of the scheduling wakeup latency, called "wakeup_dl", is introduced. As a consequence of applying this patch there will be three wakeup latency tracer: * "wakeup", that deals with all tasks in the system; * "wakeup_rt", that deals with -rt and -deadline tasks only; * "wakeup_dl", that deals with -deadline tasks only. Signed-off-by:
-
Harald Gustafsson authored
Make it possible to specify a period (different or equal than deadline) for -deadline tasks. Relative deadlines (D_i) are used on task arrivals to generate new scheduling (absolute) deadlines as "d = t + D_i", and periods (P_i) to postpone the scheduling deadlines as "d = d + P_i" when the budget is zero. This is in general useful to model (and schedule) tasks that have slow activation rates (long periods), but have to be scheduled soon once activated (short deadlines). Signed-off-by:
Harald Gustafsson <harald.gustafsson@ericsson.com> Signed-off-by:
-
Dario Faggioli authored
Make the core scheduler and load balancer aware of the load produced by -deadline tasks, by updating the moving average like for sched_rt. Signed-off-by:
-
Juri Lelli authored
Introduces data structures relevant for implementing dynamic migration of -deadline tasks and the logic for checking if runqueues are overloaded with -deadline tasks and for choosing where a task should migrate, when it is the case. Adds also dynamic migrations to SCHED_DEADLINE, so that tasks can be moved among CPUs when necessary. It is also possible to bind a task to a (set of) CPU(s), thus restricting its capability of migrating, or forbidding migrations at all. The very same approach used in sched_rt is utilised: - -deadline tasks are kept into CPU-specific runqueues, - -deadline tasks are migrated among runqueues to achieve the following: * on an M-CPU system the M earliest deadline ready tasks are always running; * affinity/cpusets settings of all the -deadline tasks is always respected. Therefore, this very special form of "load balancing" is done with an active method, i.e., the scheduler pushes or pulls tasks between runqueues when they are woken up and/or (de)scheduled. IOW, every time a preemption occurs, the descheduled task might be sent to some other CPU (depending on its deadline) to continue executing (push). On the other hand, every time a CPU becomes idle, it might pull the second earliest deadline ready task from some other CPU. To enforce this, a pull operation is always attempted before taking any scheduling decision (pre_schedule()), as well as a push one after each scheduling decision (post_schedule()). In addition, when a task arrives or wakes up, the best CPU where to resume it is selected taking into account its affinity mask, the system topology, but also its deadline. E.g., from the scheduling point of view, the best CPU where to wake up (and also where to push) a task is the one which is running the task with the latest deadline among the M executing ones. In order to facilitate these decisions, per-runqueue "caching" of the deadlines of the currently running and of the first ready task is used. Queued but not running tasks are also parked in another rb-tree to speed-up pushes. Signed-off-by: -
Dario Faggioli authored
Introduces the data structures, constants and symbols needed for SCHED_DEADLINE implementation. Core data structure of SCHED_DEADLINE are defined, along with their initializers. Hooks for checking if a task belong to the new policy are also added where they are needed. Adds a scheduling class, in sched/dl.c and a new policy called SCHED_DEADLINE. It is an implementation of the Earliest Deadline First (EDF) scheduling algorithm, augmented with a mechanism (called Constant Bandwidth Server, CBS) that makes it possible to isolate the behaviour of tasks between each other. The typical -deadline task will be made up of a computation phase (instance) which is activated on a periodic or sporadic fashion. The expected (maximum) duration of such computation is called the task's runtime; the time interval by which each instance need to be completed is called the task's relative deadline. The task's absolute deadline is dynamically calculated as the time instant a task (better, an instance) activates plus the relative deadline. The EDF algorithms selects the task with the smallest absolute deadline as the one to be executed first, while the CBS ensures each task to run for at most its runtime every (relative) deadline length time interval, avoiding any interference between different tasks (bandwidth isolation). Thanks to this feature, also tasks that do not strictly comply with the computational model sketched above can effectively use the new policy. To summarize, this patch: - introduces the data structures, constants and symbols needed; - implements the core logic of the scheduling algorithm in the new scheduling class file; - provides all the glue code between the new scheduling class and the core scheduler and refines the interactions between sched/dl and the other existing scheduling classes. Signed-off-by:
Michael Trimarchi <michael@amarulasolutions.com> Signed-off-by:
Fabio Checconi <fchecconi@gmail.com> Signed-off-by:
-
Dario Faggioli authored
Add the syscalls needed for supporting scheduling algorithms with extended scheduling parameters (e.g., SCHED_DEADLINE). In general, it makes possible to specify a periodic/sporadic task, that executes for a given amount of runtime at each instance, and is scheduled according to the urgency of their own timing constraints, i.e.: - a (maximum/typical) instance execution time, - a minimum interval between consecutive instances, - a time constraint by which each instance must be completed. Thus, both the data structure that holds the scheduling parameters of the tasks and the system calls dealing with it must be extended. Unfortunately, modifying the existing struct sched_param would break the ABI and result in potentially serious compatibility issues with legacy binaries. For these reasons, this patch: - defines the new struct sched_attr, containing all the fields that are necessary for specifying a task in the computational model described above; - defines and implements the new scheduling related syscalls that manipulate it, i.e., sched_setattr() and sched_getattr(). Syscalls are introduced for x86 (32 and 64 bits) and ARM only, as a proof of concept and for developing and testing purposes. Making them available on other architectures is straightforward. Since no "user" for these new parameters is introduced in this patch, the implementation of the new system calls is just identical to their already existing counterpart. Future patches that implement scheduling policies able to exploit the new data structure must also take care of modifying the sched_*attr() calls accordingly with their own purposes. Signed-off-by:
-
Ingo Molnar authored
Pick up the latest fixes before applying new changes. Signed-off-by:
-
- 12 Jan, 2014 1 commit
-
-
Rik van Riel authored
Thomas Hellstrom bisected a regression where erratic 3D performance is experienced on virtual machines as measured by glxgears. It identified commit 58d081b5 ("sched/numa: Avoid overloading CPUs on a preferred NUMA node") as the problem which had modified the behaviour of effective_load. Effective load calculates the difference to the system-wide load if a scheduling entity was moved to another CPU. The task group is not heavier as a result of the move but overall system load can increase/decrease as a result of the change. Commit 58d081b5 ("sched/numa: Avoid overloading CPUs on a preferred NUMA node") changed effective_load to make it suitable for calculating if a particular NUMA node was compute overloaded. To reduce the cost of the function, it assumed that a current sched entity weight of 0 was uninteresting but that is not the case. wake_affine() uses a weight of 0 for sync wakeups on the grounds that it is assuming the waking task will sleep and not contribute to load in the near future. In this case, we still want to calculate the effective load of the sched entity hierarchy. As effective_load is no longer used by task_numa_compare since commit fb13c7ee (sched/numa: Use a system-wide search to find swap/migration candidates), this patch simply restores the historical behaviour. Reported-and-tested-by:
Thomas Hellstrom <thellstrom@vmware.com> Signed-off-by:
Rik van Riel <riel@redhat.com> [ Wrote changelog] Signed-off-by:
Mel Gorman <mgorman@suse.de> Signed-off-by:
-
- 10 Jan, 2014 2 commits
-
-
git://git.kernel.org/pub/scm/linux/kernel/git/davem/netLinus Torvalds authored
Pull networking fixes from David Miller: "Famouse last words: "final pull request" :-) I'm sending this because Jason Wang's fixes are pretty important 1) Add missing per-cpu stats initialization to ip6_vti. Otherwise lockdep spits out a call trace. From Li RongQing. 2) Fix NULL oops in wireless hwsim, from Javier Lopez 3) TIPC deferred packet queue unlink must NULL out skb->next to avoid crashes. From Erik Hugne 4) Fix access to uninitialized buffer in nf_nat netfilter code, from Daniel Borkmann 5) Fix lifetime of ipv6 loopback and SIT tunnel addresses, otherwise they basically timeout immediately. From Hannes Frederic Sowa 6) Fix DMA unmapping of TSO packets in bnx2x driver, from Michal Schmidt 7) Do not allow L2 forwarding offload via macvtap device, the way things are now it will not end up being forwaded at all. From Jason Wang 8) Fix transmit queue selection via ndo_dfwd_start_xmit(), fixing things like applying NETIF_F_LLTX to the wrong device (!!) and eliding the proper transmit watchdog handling 9) qlcnic driver was not updating tx statistics at all, from Manish Chopra" * git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: qlcnic: Fix ethtool statistics length calculation qlcnic: Fix bug in TX statistics net: core: explicitly select a txq before doing l2 forwarding macvlan: forbid L2 fowarding offload for macvtap bnx2x: fix DMA unmapping of TSO split BDs ipv6: add link-local, sit and loopback address with INFINITY_LIFE_TIME bnx2x: prevent WARN during driver unload tipc: correctly unlink packets from deferred packet queue ipv6: pcpu_tstats.syncp should be initialised in ip6_vti.c netfilter: only warn once on wrong seqadj usage netfilter: nf_nat: fix access to uninitialized buffer in IRC NAT helper NFC: Fix target mode p2p link establishment iwlwifi: add new devices for 7265 series mac80211: move "bufferable MMPDU" check to fix AP mode scan mac80211_hwsim: Fix NULL pointer dereference -
git://oss.sgi.com/xfs/xfsLinus Torvalds authored
Pull xfs bugfixes from Ben Myers: "Here we have a bugfix for an off-by-one in the remote attribute verifier that results in a forced shutdown which you can hit with v5 superblock by creating a 64k xattr, and a fix for a missing destroy_work_on_stack() in the allocation worker. It's a bit late, but they are both fairly straightforward" * tag 'xfs-for-linus-v3.13-rc8' of git://oss.sgi.com/xfs/xfs: xfs: Calling destroy_work_on_stack() to pair with INIT_WORK_ONSTACK() xfs: fix off-by-one error in xfs_attr3_rmt_verify
-
