Re: [dpdk-dev] [ovs-dev] OvS DPDK crash when HPET timer enabled

[Tonghao Zhang <xiangxia.m.yue@gmail.com>]

On Tue, Apr 14, 2020 at 7:20 AM Ilya Maximets <i.maximets@ovn.org> wrote:
>
> On 4/13/20 5:27 PM, Ravi Kerur wrote:
> > On Sun, Apr 12, 2020 at 5:05 AM Tonghao Zhang <xiangxia.m.yue@gmail.com>
> > wrote:
> >
> >> On Sun, Apr 12, 2020 at 7:57 AM Ravi Kerur <rkerur@gmail.com> wrote:
> >>>
> >>> Hello OvS DPDK team,
> >>>
> >>> I am using OvS 2.13.90 and DPDK 19.11 and wanted to use HPET timer
> >> instead
> >>> of invariant TSC.
> >>>
> >>> Enabled CONFIG_RTE_LIBEAL_USE_HPET=y in DPDK
> >>>
> >>> Kernel has
> >>> CONFIG_HPET_TIMER=y
> >>> CONFIG_HPET_EMULATE_RTC=y
> >>> CONFIG_HPET=y
> >>> CONFIG_HPET_MMAP=y
> >>> CONFIG_HPET_MMAP_DEFAULT=y
> >>>
> >>> cat /sys/devices/system/clocksource/clocksource0/current_clocksource
> >>> tsc
> >>> # cat /sys/devices/system/clocksource/clocksource0/available_clocksource
> >>> tsc hpet acpi_pm
> >>>
> >>> OvS crashes at
> >>>
> >>> gdb ovsdb-server /core
> >>> GNU gdb (Ubuntu 8.1-0ubuntu3) 8.1.0.20180409-git
> >>> Copyright (C) 2018 Free Software Foundation, Inc.
> >>> License GPLv3+: GNU GPL version 3 or later <
> >> http://gnu.org/licenses/gpl.html
> >>>>
> >>> This is free software: you are free to change and redistribute it.
> >>> There is NO WARRANTY, to the extent permitted by law.  Type "show
> >> copying"
> >>> and "show warranty" for details.
> >>> This GDB was configured as "x86_64-linux-gnu".
> >>> Type "show configuration" for configuration details.
> >>> For bug reporting instructions, please see:
> >>> <http://www.gnu.org/software/gdb/bugs/>.
> >>> Find the GDB manual and other documentation resources online at:
> >>> <http://www.gnu.org/software/gdb/documentation/>.
> >>> For help, type "help".
> >>> Type "apropos word" to search for commands related to "word"...
> >>> Reading symbols from ovsdb-server...done.
> >>> [New LWP 9729]
> >>> [Thread debugging using libthread_db enabled]
> >>> Using host libthread_db library
> >> "/lib/x86_64-linux-gnu/libthread_db.so.1".
> >>> Core was generated by `ovsdb-server /usr/local/etc/openvswitch/conf.db
> >>> -vconsole:emer -vsyslog:err -vf'.
> >>> Program terminated with signal SIGSEGV, Segmentation fault.
> >>> #0  0x0000560fc6db8ac1 in rte_get_hpet_cycles ()
> >> Hi,
> >> This is a bug on dpdk, NOT ovs.
> >> I guess that your kernel does't support getentropy which supported in
> >> 3.17 and your cpu don't support rdseed.
> >> and that cause a dpdk bug. To workaround, you change: Not tested, hope
> >> it's useful for you
> >>
> >>
> > You are correct CPU doesn't support RDSEED instruction but RDRAND is
> > supported.
> >
> > It's a RT kernel based on 4.19 + RT patches.
> >
> > Setting CONFIG_RTE_LIBEAL_USE_HPET=n which is the default,'Invariant TSC'
> > will be used. However, I want to use HPET so should I check with DPDK dev?
>
> Yes.  This is a pure DPDK bug.  There is no way to fix it from the OVS side.
> Adding dpdk-dev in CC, moving ovs-dev to BCC to avoid cross posting.
Hi Ilya
I sent a patch to dpdk, but not applied. If applied, I feedback to this thread.
> Actually, looking at the code in DPDK, it's really old.  I'm wondering if it
> ever worked?  Looks like no-one ever tested this configuration.
>
> Best regards, Ilya Maximets.
>
> >
> > Thanks,
> > Ravi
> >
> >
> >
> >> diff --git a/lib/librte_eal/linux/eal_timer.c
> >> b/lib/librte_eal/linux/eal_timer.c
> >> index a904a8297cd2..4067e46da534 100644
> >> --- a/lib/librte_eal/linux/eal_timer.c
> >> +++ b/lib/librte_eal/linux/eal_timer.c
> >> @@ -26,7 +26,7 @@
> >>  #include "eal_private.h"
> >>  #include "eal_internal_cfg.h"
> >>
> >> -enum timer_source eal_timer_source = EAL_TIMER_HPET;
> >> +enum timer_source eal_timer_source = EAL_TIMER_TSC;
> >>
> >>  #ifdef RTE_LIBEAL_USE_HPET
> >>
> >>
> >>> (gdb) bt
> >>> #0  0x0000560fc6db8ac1 in rte_get_hpet_cycles ()
> >>> #1  0x0000560fc6c482b0 in rte_rand_init ()
> >>> #2  0x0000560fc6e125bd in __libc_csu_init ()
> >>> #3  0x00007f5e1f930b28 in __libc_start_main (main=0x560fc6c45940 <main>,
> >>> argc=14, argv=0x7fff61ae5088, init=0x560fc6e12570 <__libc_csu_init>,
> >>>     fini=<optimized out>, rtld_fini=<optimized out>,
> >>> stack_end=0x7fff61ae5078) at ../csu/libc-start.c:266
> >>> #4  0x0000560fc6c485da in _start ()
> >>> (gdb) quit
> >>>
> >>> RTE_INIT(rte_rand_init)
> >>> {
> >>> ...
> >>> }
> >>>
> >>> rte_rand_init is defined as initializer, following diffs won't help,
> >>>
> >>> diff --git a/lib/dpdk.c b/lib/dpdk.c
> >>> index 31450d470..f3d3989c8 100644
> >>> --- a/lib/dpdk.c
> >>> +++ b/lib/dpdk.c
> >>> @@ -405,6 +405,12 @@ dpdk_init__(const struct smap *ovs_other_config)
> >>>      if (result < 0) {
> >>>          VLOG_EMER("Unable to initialize DPDK: %s",
> >>> ovs_strerror(rte_errno));
> >>>          return false;
> >>> +    } else {
> >>> +#ifdef RTE_LIBEAL_USE_HPET
> >>> +        if (rte_eal_hpet_init(1) < 0) {
> >>> +            VLOG_ERR("HPET is not enabled, using TSC as default
> >> timer\n");
> >>> +        }
> >>> +#endif
> >>>      }
> >>>
> >>>      if (VLOG_IS_DBG_ENABLED()) {
> >>>
> >>> Kindly let me know your inputs on how to overcome this issue.
> >>>
> >>> Thanks,
> >>> Ravi
> >>> _______________________________________________
> >>> dev mailing list
> >>> dev@openvswitch.org
> >>> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
> >>
> >>
> >>
> >> --
> >> Best regards, Tonghao
> >>
> > _______________________________________________
> > dev mailing list
> > dev@openvswitch.org
> > https://mail.openvswitch.org/mailman/listinfo/ovs-dev
> >
>


-- 
Best regards, Tonghao