tag:blogger.com,1999:blog-6469704299235308349.post4296595129517401571..comments2024-02-09T16:24:46.087+11:00Comments on -ck hacking: BFS 0.427 for linux 3.7.xckhttp://www.blogger.com/profile/02904761195451530213noreply@blogger.comBlogger29125tag:blogger.com,1999:blog-6469704299235308349.post-64858723036492754272013-02-26T06:53:16.448+11:002013-02-26T06:53:16.448+11:00Fine to read that you and your family are o.k. ! :...Fine to read that you and your family are o.k. ! :-) I hope, your brother keeps recovering!<br /><br />No need for a hurry for BFS/CK. We all appreciate your well done work, and of course, only when you'd call it that name.<br /><br />Best wishes, ManuelAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-51503470504294367742013-02-25T07:47:19.740+11:002013-02-25T07:47:19.740+11:00We're ok, I'm just too busy right now, sor...We're ok, I'm just too busy right now, sorry. It will come out when it does.ckhttps://www.blogger.com/profile/02904761195451530213noreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-26054513340599012612013-02-25T07:01:42.238+11:002013-02-25T07:01:42.238+11:00@ CK: Do you need help? Is your family well up? Pl...@ CK: Do you need help? Is your family well up? Please tell us, how we as community can support you, whereever needed!<br /><br />Manuel KrauseAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-20478525218018820572013-02-23T05:06:22.763+11:002013-02-23T05:06:22.763+11:00BTW, this was done with kernel 3.7.7. ManuelBTW, this was done with kernel 3.7.7. ManuelAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-82027036735220265232013-02-23T04:58:04.478+11:002013-02-23T04:58:04.478+11:00Regarding the RCU stuff... I've experimented w...Regarding the RCU stuff... I've experimented with it on my unicore uniprocessor. At first there are less kernel config options for me -- o.k. But changing the remaining rest of them did not make any difference in usability. Before always using kernel standard 500:0, I did 257:10, 331:10, last one, 101:99:<br /><br />~ # zcat /proc/config.gz |grep RCU<br /># RCU Subsystem<br />CONFIG_TINY_PREEMPT_RCU=y<br />CONFIG_PREEMPT_RCU=y<br /># CONFIG_TREE_RCU_TRACE is not set<br />CONFIG_RCU_BOOST=y<br />CONFIG_RCU_BOOST_PRIO=99<br />CONFIG_RCU_BOOST_DELAY=101<br /># CONFIG_PROVE_RCU_DELAY is not set<br /># CONFIG_SPARSE_RCU_POINTER is not set<br /># CONFIG_RCU_TRACE is not set<br /><br />Only taking prime numbers in delay/frequency as it prevents reoccurrence. ;-)<br /><br />Varying these values seems to be useless to systems with one CPU/Core. But disabling RCU Boost resulted in hiccups in video and audio.<br /><br />Best regards,<br /><br />Manuel Krause<br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-4443259706601859182013-02-21T09:04:25.527+11:002013-02-21T09:04:25.527+11:00BFQ v6 for the 3.8 kernel is out, as of 20130219, ...BFQ v6 for the 3.8 kernel is out, as of 20130219, too:<br />http://algo.ing.unimo.it/people/paolo/disk_sched/patches/3.8.0-v6/<br /><br />Best regards, <br />Manuel KrauseAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-31467725454891620872013-02-19T23:23:39.608+11:002013-02-19T23:23:39.608+11:00ck are you working on a 3.8 patch ?
I failed to me...ck are you working on a 3.8 patch ?<br />I failed to merge the 3.7 patch into the new kernel. There seem to be some bigger changes in the scheduler interface. Probably related to the new rcu stuff... I'm eager to test this with bfs but no way to "brute force" the old code into the new kernel :)Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-37752606338528903092013-02-17T19:46:01.850+11:002013-02-17T19:46:01.850+11:00Hi, CK. For the accounting problem, I am not sure ...Hi, CK. For the accounting problem, I am not sure if this one is the one you trying to solve. But I still find this with 427 with 3.7.7 kernel. To reproduce it:<br />1. compile the kernel source, (I used 'time make -j4')<br />2. use top to monitoring, some as processes will show with 9999 %CPU and a huge TIME+<br /><br />like this<br /><br /> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND<br />15120 root 5 0 18860 5924 1168 S 9999 0.1 5124095h as<br />15111 root 7 0 50556 23m 3784 R 8 0.6 0:00.25 cc1<br />Alfred Chenhttps://www.blogger.com/profile/03164306846702841944noreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-3895251250309935092013-02-17T17:17:38.378+11:002013-02-17T17:17:38.378+11:00[ 0.126666]
[ 0.126666] -> #1 (&grq.l...[ 0.126666]<br />[ 0.126666] -> #1 (&grq.lock){-.....}:<br />[ 0.126666] [] lock_acquire+0x96/0xc0<br />[ 0.126666] [] _raw_spin_lock_irqsave+0x53/0x90<br />[ 0.126666] [] get_page_from_freelist+0x587/0x640<br />[ 0.126666] [] __alloc_pages_nodemask+0xf6/0x820<br />[ 0.126666] [] new_slab+0x67/0x270<br />[ 0.126666] [] __slab_alloc.isra.59.constprop.64+0x14d/0x242<br />[ 0.126666] [] __kmalloc+0x10c/0x150<br />[ 0.126666] [] pcpu_mem_zalloc+0x2b/0x80<br />[ 0.126666] [] pcpu_extend_area_map+0x31/0x120<br />[ 0.126666] [] pcpu_alloc+0x23b/0x9f0<br />[ 0.126666] [] __alloc_percpu+0xb/0x10<br />[ 0.126666] [] build_sched_domains+0x66/0xaf0<br />[ 0.126666] [] sched_init_smp+0x70/0x2b2<br />[ 0.126666] [] kernel_init_freeable+0x86/0x18b<br />[ 0.126666] [] kernel_init+0x9/0x100<br />[ 0.126666] [] ret_from_fork+0x7c/0xb0<br />[ 0.126666]<br />[ 0.126666] -> #0 (sched_domains_mutex){+.+.+.}:<br />[ 0.126666] [] __lock_acquire+0x1cef/0x1db0<br />[ 0.126666] [] lock_acquire+0x96/0xc0<br />[ 0.126666] [] mutex_lock_nested+0x72/0x3d0<br />[ 0.126666] [] sched_init_smp+0x129/0x2b2<br />[ 0.126666] [] kernel_init_freeable+0x86/0x18b<br />[ 0.126666] [] kernel_init+0x9/0x100<br />[ 0.126666] [] ret_from_fork+0x7c/0xb0<br />[ 0.126666]<br />[ 0.126666] other info that might help us debug this:<br />[ 0.126666]<br />[ 0.126666] Possible unsafe locking scenario:<br />[ 0.126666]<br />[ 0.126666] CPU0 CPU1<br />[ 0.126666] ---- ----<br />[ 0.126666] lock(&grq.lock);<br />[ 0.126666] lock(sched_domains_mutex);<br />[ 0.126666] lock(&grq.lock);<br />[ 0.126666] lock(sched_domains_mutex);<br />[ 0.126666]<br />[ 0.126666] *** DEADLOCK ***<br />[ 0.126666]<br />[ 0.126666] 1 lock held by BFS/0/1:<br />[ 0.126666] #0: (&grq.lock){-.....}, at: [] sched_init_smp+0x101/0x2b2<br />[ 0.126666]<br />[ 0.126666] stack backtrace:<br />[ 0.126666] Pid: 1, comm: BFS/0 Not tainted 3.7.7+ #114<br />[ 0.126666] Call Trace:<br />[ 0.126666] [] print_circular_bug+0x1fb/0x20c<br />[ 0.126666] [] __lock_acquire+0x1cef/0x1db0<br />[ 0.126666] [] ? sched_clock_cpu+0xa8/0x120<br />[ 0.126666] [] lock_acquire+0x96/0xc0<br />[ 0.126666] [] ? sched_init_smp+0x129/0x2b2<br />[ 0.126666] [] mutex_lock_nested+0x72/0x3d0<br />[ 0.126666] [] ? sched_init_smp+0x129/0x2b2<br />[ 0.126666] [] ? sched_init_smp+0x129/0x2b2<br />[ 0.126666] [] sched_init_smp+0x129/0x2b2<br />[ 0.126666] [] ? native_smp_cpus_done+0xa2/0xab<br />[ 0.126666] [] kernel_init_freeable+0x86/0x18b<br />[ 0.126666] [] ? schedule_tail+0x87/0x130<br />[ 0.126666] [] ? schedule_tail+0x48/0x130<br />[ 0.126666] [] ? rest_init+0x140/0x140<br />[ 0.126666] [] kernel_init+0x9/0x100<br />[ 0.126666] [] ret_from_fork+0x7c/0xb0<br />[ 0.126666] [] ? rest_init+0x140/0x140<br />Alfred Chenhttps://www.blogger.com/profile/03164306846702841944noreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-45819528676612202072013-02-17T17:15:29.754+11:002013-02-17T17:15:29.754+11:00Thanks for your updates, CK. As I reported suspici...Thanks for your updates, CK. As I reported suspicious_rcu_warn issue before, so I retest it with 427. But another dead lock issue is found. I will just post the log here.<br /><br />[ 0.124430] ======================================================<br />[ 0.125331] [ INFO: possible circular locking dependency detected ]<br />[ 0.126233] 3.7.7+ #114 Not tainted<br />[ 0.126666] -------------------------------------------------------<br />[ 0.126666] BFS/0/1 is trying to acquire lock:<br />[ 0.126666] (sched_domains_mutex){+.+.+.}, at: [] sched_init_smp+0x129/0x2b2<br />[ 0.126666]<br />[ 0.126666] but task is already holding lock:<br />[ 0.126666] (&grq.lock){-.....}, at: [] sched_init_smp+0x101/0x2b2<br />[ 0.126666]<br />[ 0.126666] which lock already depends on the new lock.<br />[ 0.126666]<br />[ 0.126666]<br />[ 0.126666] the existing dependency chain (in reverse order) is:<br />Alfred Chenhttps://www.blogger.com/profile/03164306846702841944noreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-29367829193087510662013-02-12T00:30:02.088+11:002013-02-12T00:30:02.088+11:00I tried many .config variations, but get a time ac...I tried many .config variations, but get a time accounting overflow anyway after some runtime. No luck so far.<br /><br />As of curiousity I tried to Bfs patch the new Linux-3.8rc:<br />- many line moves<br />- 5 fuzzy patches going through<br />- 2 rejects at ondemand.c<br /><br />I can observe this overflow happening when compiling staff for my Gentoo installation. Activity in the ondemand module might be involved while the time accounting failure occures.<br />Maybe the next Linux version is error free :)<br /><br />Greetings from Hamburg, Ralph UlrichAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-18018390573361378632013-02-09T21:40:29.396+11:002013-02-09T21:40:29.396+11:00The oldest version of BFS is 0.414 and is for kern...The oldest version of BFS is 0.414 and is for kernel 3.1. There's nothing available for older kernels.Anonymoushttps://www.blogger.com/profile/11469174621439712081noreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-25414497810447359112013-02-09T17:56:39.542+11:002013-02-09T17:56:39.542+11:00is there a patch anywhere for kernel 2.6.22?is there a patch anywhere for kernel 2.6.22?Mangixhttps://www.blogger.com/profile/06816602571637076369noreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-66990636535175400902013-02-02T04:01:41.464+11:002013-02-02T04:01:41.464+11:00I am just doing fine without time account overflow...I am just doing fine without time account overflows by tuning further RCU:<br /><br />zcat /proc/config.gz |grep RCU<br /># RCU Subsystem<br />CONFIG_TREE_PREEMPT_RCU=y<br />CONFIG_PREEMPT_RCU=y<br /># CONFIG_RCU_USER_QS is not set<br />CONFIG_RCU_FANOUT=32<br />CONFIG_RCU_FANOUT_LEAF=4<br /># CONFIG_RCU_FANOUT_EXACT is not set<br /># CONFIG_TREE_RCU_TRACE is not set<br />CONFIG_RCU_BOOST=y<br />CONFIG_RCU_BOOST_PRIO=14<br />CONFIG_RCU_BOOST_DELAY=320<br />CONFIG_HAVE_RCU_USER_QS=y<br /># CONFIG_PROVE_RCU_DELAY is not set<br /># CONFIG_SPARSE_RCU_POINTER is not set<br />CONFIG_RCU_CPU_STALL_TIMEOUT=32<br />CONFIG_RCU_CPU_STALL_VERBOSE=y<br />CONFIG_RCU_CPU_STALL_INFO=y<br /># CONFIG_RCU_TRACE is not set<br /><br />Con, if you want to reproduce you could try RCU config variations in the opposite direction. Beside I also use io deadline queue and not set NO_HZ.<br />Ralph UlrichAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-81827588761081462812013-02-02T02:16:42.450+11:002013-02-02T02:16:42.450+11:00I have sent you my .config
I have boosted RCU and ...I have sent you my .config<br />I have boosted RCU and can reduce time accounting overflows.<br />Ralph Ulrich Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-90745716057921043972013-02-01T02:16:56.617+11:002013-02-01T02:16:56.617+11:00Understood, thanks.Understood, thanks.Oleksandr Natalenkohttps://www.blogger.com/profile/12098091624630953604noreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-87754771707995960602013-01-31T12:05:28.746+11:002013-01-31T12:05:28.746+11:00It's a new addon patch for the BFQ I/O schedul...It's a new addon patch for the BFQ I/O scheduler. You can find it here, if you're interested:<br /><br />https://groups.google.com/forum/?fromgroups=#!topic/bfq-iosched/H7PCRpreLAQ<br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-55016250939706797942013-01-31T08:38:29.749+11:002013-01-31T08:38:29.749+11:00No idea what you're talking about.
1) BFQ is ...No idea what you're talking about.<br /><br />1) BFQ is an I/O scheduler not a CPU scheduler.<br />2) There has been no new release of BFQ as far as I can tell.grayskyhttp://repo-ck.comnoreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-30371046538082752182013-01-31T08:27:24.504+11:002013-01-31T08:27:24.504+11:00By the way if I can't reproduce it, it's v...By the way if I can't reproduce it, it's very difficult to debug a problem, and time is always an issue. However since I have had time recently, if you could email me with more details we might be able to debug it together if you're willing to try various patches.ckhttps://www.blogger.com/profile/02904761195451530213noreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-30825906631183387642013-01-31T06:54:02.040+11:002013-01-31T06:54:02.040+11:00If I could tell you, I could fix it.If I could tell you, I could fix it.ckhttps://www.blogger.com/profile/02904761195451530213noreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-19278440108504858092013-01-31T06:44:16.102+11:002013-01-31T06:44:16.102+11:00Uups, got a time accounting overflow again: began ...Uups, got a time accounting overflow again: began at a kworker thread and then consecutive new threads.<br /><br />Is there a correctly working initializing of times spent of new jobs in place already? Perhaps it initializes with a random memory value, which would tell me why this occures after some hours?<br />Ralph UlrichAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-35841750674479687942013-01-31T02:34:57.125+11:002013-01-31T02:34:57.125+11:00Excellent work! And I love having correctly accoun...Excellent work! And I love having correctly accounting of process times again!<br /><br />I observe the nvidia-proprietary settings gui showing less heated graphics processor in my little mac mini pc !? <br /><br />Greeting from cold and cloudy Hamburg,<br />Ralph Ulirch<br /><br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-90452783441107464322013-01-31T02:19:20.997+11:002013-01-31T02:19:20.997+11:00@graysky: Could you please also benchmark the late...@graysky: Could you please also benchmark the latest BFQ released today, which includes the new Early Queue Merge feature? ThanksAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-71486382630544886512013-01-30T18:17:31.750+11:002013-01-30T18:17:31.750+11:00That is precisely the bug fixed in this release.
...That is precisely the bug fixed in this release.<br /><br />This is the patch that fixes it:<br /><a href="http://ck.kolivas.org/patches/bfs/3.0/3.7/incremental/bfs426-fix_cpu_posix_timers.patch" rel="nofollow">bfs426-fix_cpu_posix_timers.patch</a><br /><br />It should apply to older BFS as well.ckhttps://www.blogger.com/profile/02904761195451530213noreply@blogger.comtag:blogger.com,1999:blog-6469704299235308349.post-73917097751115587742013-01-30T18:10:41.031+11:002013-01-30T18:10:41.031+11:00People complain about glibc inability to be compil...People complain about glibc inability to be compiled under -ck kernels: https://bbs.archlinux.org/viewtopic.php?id=154594<br /><br />Any idea?Oleksandr Natalenkohttps://www.blogger.com/profile/12098091624630953604noreply@blogger.com