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

[Ravi Kerur <rkerur@gmail.com>]

Hi Tonghao, Thanks, your patch works fine with HPET initialization. I will
look into DPDK discussion which got started with your patch.

Thanks,
Ravi


On Mon, Apr 13, 2020 at 5:33 PM Tonghao Zhang <xiangxia.m.yue@gmail.com>
wrote:

> 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
>