Prior to this update, the systemtap scripts caused the "scheduling while atomic" error when running on the Messaging Real-time Grid kernel. To fix this bug, patches have been applied, and the error no longer occurs.
DescriptionDaniel Bristot de Oliveira
2014-09-29 17:33:06 UTC
Description of problem:
When running on the MRG Realtime kernel, systemtap scripts are causing "BUG: scheduling while atomic" messages.
Version-Release number of selected component (if applicable):
RHEL 6
Red Hat Enterprise MRG 2.5 Realtime Kernel
systemtap-2.5-5.el6.x86_64
How reproducible:
Always
Steps to Reproduce:
1. Install MRG Realtime kernel on RHEL6
2. Save the following script as test.stap:
%<---- test.stap ----
#!/usr/bin/stap
probe vm.pagefault {
printf("%s\n", probefunc());
}
%<---- test.stap ----
3. run test.stap script:
# stap test.stap
Actual results:
Kernel hits "scheduling while atomic" BUG. For example:
BUG: scheduling while atomic: makewhatis/2878/0x00000003
Modules linked in: stap_3e768d9dd19a5493d042da8393d971a0_2841(O) autofs4 ipt_REJECT nf_conntrack_ipv4 nf_defrag_ipv4 iptable_filter ip_tables ip6t_REJECT nf_conntrack_ipv6 nf_defrag_ipv6 xt_state ip6table_filter ip6_tables ipv6 ppdev microcode pcspkr parport_pc parport e1000 joydev sg snd_hda_intel snd_hda_codec snd_hwdep snd_seq snd_seq_device snd_pcm snd_timer snd soundcore snd_page_alloc i2c_piix4 i2c_core ext4 jbd2 mbcache sd_mod crc_t10dif sr_mod cdrom sym53c8xx scsi_transport_spi floppy pata_acpi ata_generic ata_piix dm_mirror dm_region_hash dm_log dm_mod [last unloaded: stap_3e768d9dd19a5493d042da8393d971a0_2831]
CPU: 2 PID: 2878 Comm: makewhatis Tainted: G W O 3.10.33-rt32.45.el6rt.x86_64 #1
Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
0000000000000002 ffff880158509b88 ffffffff81575e1e ffff880158509b98
ffffffff8108362a ffff880158509c28 ffffffff8157737c ffff880158509bc8
ffff880149f6c000 ffff880149f6c000 ffff880149f6c010 ffff880149f6dfd8
Call Trace:
<#DB> [<ffffffff81575e1e>] dump_stack+0x19/0x1b
[<ffffffff8108362a>] __schedule_bug+0x5a/0x70
[<ffffffff8157737c>] __schedule+0x7cc/0x800
[<ffffffff81577594>] schedule+0x34/0xa0
[<ffffffff8157899d>] rt_spin_lock_slowlock+0x10d/0x350
[<ffffffff8114cd21>] ? handle_mm_fault+0x1/0x680
[<ffffffff81579296>] rt_spin_lock+0x26/0x30
[<ffffffffa05aeab3>] stp_print_flush+0x63/0x1d0 [stap_3e768d9dd19a5493d042da8393d971a0_2841]
[<ffffffff8114cd21>] ? handle_mm_fault+0x1/0x680
[<ffffffffa05b8ddf>] probe_2058+0x3bf/0x4e8 [stap_3e768d9dd19a5493d042da8393d971a0_2841]
[<ffffffffa05b5609>] enter_kprobe_probe+0x1d9/0x3f0 [stap_3e768d9dd19a5493d042da8393d971a0_2841]
[<ffffffff8157ca86>] kprobe_exceptions_notify+0x4a6/0x570
[<ffffffff8114cd21>] ? handle_mm_fault+0x1/0x680
[<ffffffff8114cd20>] ? handle_pte_fault+0xbf0/0xbf0
[<ffffffff8157da05>] notifier_call_chain+0x55/0x80
[<ffffffff8157da78>] __atomic_notifier_call_chain+0x48/0x70
[<ffffffff8157dab6>] atomic_notifier_call_chain+0x16/0x20
[<ffffffff8157daee>] notify_die+0x2e/0x30
[<ffffffff8157a7bb>] do_int3+0x4b/0x120
[<ffffffff81579eab>] int3+0x2b/0x40
[<ffffffff8114cd20>] ? handle_pte_fault+0xbf0/0xbf0
<<EOE>> [<ffffffff8157d69c>] ? __do_page_fault+0x24c/0x550
[<ffffffffa05af6d5>] ? __stp_tf_get_map_entry+0xa5/0xd0 [stap_3e768d9dd19a5493d042da8393d971a0_2841]
[<ffffffffa05b66dd>] ? __stp_utrace_task_finder_target_syscall_exit+0x6d/0x280 [stap_3e768d9dd19a5493d042da8393d971a0_2841]
[<ffffffffa05b3860>] ? utrace_report_syscall_exit+0xf0/0x100 [stap_3e768d9dd19a5493d042da8393d971a0_2841]
[<ffffffff8157d9ae>] do_page_fault+0xe/0x10
[<ffffffff8157cd48>] do_async_page_fault+0x28/0xc0
[<ffffffff81579f78>] async_page_fault+0x28/0x30
Expected results:
The "BUG scheduling while atomic" should not happen.
Additional info:
This error occurs because the module created by systemtap is calling spin_lock() in the atomic context. Although it's correct for RHEL6 kernel (non-rt kernel), as spinlocks are converted to rt_mutex on MRG Realtime Kernel, when a contention happens the rt_mutex puts the task to sleep, causing the error. For the RT kernel, these spinlocks should be converted to raw spinlocks.
Fortunately, this problem was fixed on mainstream systemtap's repo:
GIT rego: git://sourceware.org/git/systemtap.git
# git format-patch -9 d4e048442b3aea10dfc63b1f4b7cd8f8deb58a62
0001-stp-rt-locking-helper-api.patch
0002-stp-rt-replace-__stp_tf_map_lock-rd-wr-lock-with-stp.patch
0003-stp-rt-replace-__stp_tf_vma_lock-rd-wr-lock-with-stp.patch
0004-stp-rt-replace-utrace-lock-with-stp-style-raw-lock.patch
0005-stp-rt-replace-utrace_struct-lock-to-stp-style-raw-l.patch
0006-stp-rt-replace-__stp_tf_task_work_list_lock-to-stp-r.patch
0007-stp-rt-replace-addr_map_lock-rd-wr-lock-with-stp-typ.patch
0008-stp-rt-replace-stp_print-lock-with-stp-style-lock.patch
0009-stp-rt-fix-preemptible-bug_on-for-STAT_GET_CPU.patch
I backported this patches to the RHEL6 system tap, and the error stopped to occur. Although some patches does not applies directly, they are easy to fix with small changes.
Comment 1Daniel Bristot de Oliveira
2014-10-24 12:39:42 UTC
Hi Frank!
There's a customer asking for a z-stream of this BZ.
Regarding the version, the customer is using the MRG-RT on RHEL6.5, so we will need a z-stream for 6.5. But is worth generate a z-stream for RHEL6.6 as well.
Thanks!
Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.
For information on the advisory, and where to find the updated
files, follow the link below.
If the solution does not work for you, open a new bug report.
https://rhn.redhat.com/errata/RHBA-2015-1333.html
Description of problem: When running on the MRG Realtime kernel, systemtap scripts are causing "BUG: scheduling while atomic" messages. Version-Release number of selected component (if applicable): RHEL 6 Red Hat Enterprise MRG 2.5 Realtime Kernel systemtap-2.5-5.el6.x86_64 How reproducible: Always Steps to Reproduce: 1. Install MRG Realtime kernel on RHEL6 2. Save the following script as test.stap: %<---- test.stap ---- #!/usr/bin/stap probe vm.pagefault { printf("%s\n", probefunc()); } %<---- test.stap ---- 3. run test.stap script: # stap test.stap Actual results: Kernel hits "scheduling while atomic" BUG. For example: BUG: scheduling while atomic: makewhatis/2878/0x00000003 Modules linked in: stap_3e768d9dd19a5493d042da8393d971a0_2841(O) autofs4 ipt_REJECT nf_conntrack_ipv4 nf_defrag_ipv4 iptable_filter ip_tables ip6t_REJECT nf_conntrack_ipv6 nf_defrag_ipv6 xt_state ip6table_filter ip6_tables ipv6 ppdev microcode pcspkr parport_pc parport e1000 joydev sg snd_hda_intel snd_hda_codec snd_hwdep snd_seq snd_seq_device snd_pcm snd_timer snd soundcore snd_page_alloc i2c_piix4 i2c_core ext4 jbd2 mbcache sd_mod crc_t10dif sr_mod cdrom sym53c8xx scsi_transport_spi floppy pata_acpi ata_generic ata_piix dm_mirror dm_region_hash dm_log dm_mod [last unloaded: stap_3e768d9dd19a5493d042da8393d971a0_2831] CPU: 2 PID: 2878 Comm: makewhatis Tainted: G W O 3.10.33-rt32.45.el6rt.x86_64 #1 Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011 0000000000000002 ffff880158509b88 ffffffff81575e1e ffff880158509b98 ffffffff8108362a ffff880158509c28 ffffffff8157737c ffff880158509bc8 ffff880149f6c000 ffff880149f6c000 ffff880149f6c010 ffff880149f6dfd8 Call Trace: <#DB> [<ffffffff81575e1e>] dump_stack+0x19/0x1b [<ffffffff8108362a>] __schedule_bug+0x5a/0x70 [<ffffffff8157737c>] __schedule+0x7cc/0x800 [<ffffffff81577594>] schedule+0x34/0xa0 [<ffffffff8157899d>] rt_spin_lock_slowlock+0x10d/0x350 [<ffffffff8114cd21>] ? handle_mm_fault+0x1/0x680 [<ffffffff81579296>] rt_spin_lock+0x26/0x30 [<ffffffffa05aeab3>] stp_print_flush+0x63/0x1d0 [stap_3e768d9dd19a5493d042da8393d971a0_2841] [<ffffffff8114cd21>] ? handle_mm_fault+0x1/0x680 [<ffffffffa05b8ddf>] probe_2058+0x3bf/0x4e8 [stap_3e768d9dd19a5493d042da8393d971a0_2841] [<ffffffffa05b5609>] enter_kprobe_probe+0x1d9/0x3f0 [stap_3e768d9dd19a5493d042da8393d971a0_2841] [<ffffffff8157ca86>] kprobe_exceptions_notify+0x4a6/0x570 [<ffffffff8114cd21>] ? handle_mm_fault+0x1/0x680 [<ffffffff8114cd20>] ? handle_pte_fault+0xbf0/0xbf0 [<ffffffff8157da05>] notifier_call_chain+0x55/0x80 [<ffffffff8157da78>] __atomic_notifier_call_chain+0x48/0x70 [<ffffffff8157dab6>] atomic_notifier_call_chain+0x16/0x20 [<ffffffff8157daee>] notify_die+0x2e/0x30 [<ffffffff8157a7bb>] do_int3+0x4b/0x120 [<ffffffff81579eab>] int3+0x2b/0x40 [<ffffffff8114cd20>] ? handle_pte_fault+0xbf0/0xbf0 <<EOE>> [<ffffffff8157d69c>] ? __do_page_fault+0x24c/0x550 [<ffffffffa05af6d5>] ? __stp_tf_get_map_entry+0xa5/0xd0 [stap_3e768d9dd19a5493d042da8393d971a0_2841] [<ffffffffa05b66dd>] ? __stp_utrace_task_finder_target_syscall_exit+0x6d/0x280 [stap_3e768d9dd19a5493d042da8393d971a0_2841] [<ffffffffa05b3860>] ? utrace_report_syscall_exit+0xf0/0x100 [stap_3e768d9dd19a5493d042da8393d971a0_2841] [<ffffffff8157d9ae>] do_page_fault+0xe/0x10 [<ffffffff8157cd48>] do_async_page_fault+0x28/0xc0 [<ffffffff81579f78>] async_page_fault+0x28/0x30 Expected results: The "BUG scheduling while atomic" should not happen. Additional info: This error occurs because the module created by systemtap is calling spin_lock() in the atomic context. Although it's correct for RHEL6 kernel (non-rt kernel), as spinlocks are converted to rt_mutex on MRG Realtime Kernel, when a contention happens the rt_mutex puts the task to sleep, causing the error. For the RT kernel, these spinlocks should be converted to raw spinlocks. Fortunately, this problem was fixed on mainstream systemtap's repo: GIT rego: git://sourceware.org/git/systemtap.git # git format-patch -9 d4e048442b3aea10dfc63b1f4b7cd8f8deb58a62 0001-stp-rt-locking-helper-api.patch 0002-stp-rt-replace-__stp_tf_map_lock-rd-wr-lock-with-stp.patch 0003-stp-rt-replace-__stp_tf_vma_lock-rd-wr-lock-with-stp.patch 0004-stp-rt-replace-utrace-lock-with-stp-style-raw-lock.patch 0005-stp-rt-replace-utrace_struct-lock-to-stp-style-raw-l.patch 0006-stp-rt-replace-__stp_tf_task_work_list_lock-to-stp-r.patch 0007-stp-rt-replace-addr_map_lock-rd-wr-lock-with-stp-typ.patch 0008-stp-rt-replace-stp_print-lock-with-stp-style-lock.patch 0009-stp-rt-fix-preemptible-bug_on-for-STAT_GET_CPU.patch I backported this patches to the RHEL6 system tap, and the error stopped to occur. Although some patches does not applies directly, they are easy to fix with small changes.