[dpdk-dev] [PATCH v3] eal/arm64: fix rdtsc precise version

[Jerin Jacob]

On Wed, Mar 11, 2020 at 3:35 PM Linhaifeng <haifeng.lin at huawei.com> wrote:
>
>
>
> > -----Original Message-----
> > From: Jerin Jacob [mailto:jerinjacobk at gmail.com]
> > Sent: Tuesday, March 10, 2020 6:47 PM
> > To: Linhaifeng <haifeng.lin at huawei.com>
> > Cc: Gavin Hu <Gavin.Hu at arm.com>; dev at dpdk.org; thomas at monjalon.net;
> > chenchanghu <chenchanghu at huawei.com>; xudingke
> > <xudingke at huawei.com>; Lilijun (Jerry) <jerry.lilijun at huawei.com>; Honnappa
> > Nagarahalli <Honnappa.Nagarahalli at arm.com>; Steve Capper
> > <Steve.Capper at arm.com>; nd <nd at arm.com>
> > Subject: Re: [PATCH v3] eal/arm64: fix rdtsc precise version
> >
> > On Tue, Mar 10, 2020 at 3:09 PM Linhaifeng <haifeng.lin at huawei.com> wrote:
> > >
> > > In order to get more accurate the cntvct_el0 reading, SW must invoke
> > > isb and arch_counter_enforce_ordering.
> > >
> > > Reference of linux kernel:
> > > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tre
> > > e/arch/arm64/include/asm/arch_timer.h?h=v5.5#n220
> > >
> > > Signed-off-by: Haifeng Lin <haifeng.lin at huawei.com>
> >
> > Not addressed Fixes: comment
> > http://mails.dpdk.org/archives/dev/2020-March/159547.html
> >
> > > ---
> > >  .../common/include/arch/arm/rte_atomic_64.h   |  3 +++
> > >  .../common/include/arch/arm/rte_cycles_64.h   | 20
> > +++++++++++++++++--
> > >  2 files changed, 21 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/lib/librte_eal/common/include/arch/arm/rte_atomic_64.h
> > > b/lib/librte_eal/common/include/arch/arm/rte_atomic_64.h
> > > index 859ae129d..2587f98a2 100644
> > > --- a/lib/librte_eal/common/include/arch/arm/rte_atomic_64.h
> > > +++ b/lib/librte_eal/common/include/arch/arm/rte_atomic_64.h
> > > @@ -21,6 +21,7 @@ extern "C" {
> > >
> > >  #define dsb(opt) asm volatile("dsb " #opt : : : "memory")  #define
> > > dmb(opt) asm volatile("dmb " #opt : : : "memory")
> > > +#define isb()    (asm volatile("isb" : : : "memory"))
> > >
> > >  #define rte_mb() dsb(sy)
> > >
> > > @@ -44,6 +45,8 @@ extern "C" {
> > >
> > >  #define rte_cio_rmb() dmb(oshld)
> > >
> > > +#define rte_isb() isb()
> >
> > Not addressed comment
> > http://mails.dpdk.org/archives/dev/2020-March/159547.html
> >
> >
> > > +
> > >  /*------------------------ 128 bit atomic operations
> > > -------------------------*/
> > >
> > >  #if defined(__ARM_FEATURE_ATOMICS) ||
> > > defined(RTE_ARM_FEATURE_ATOMICS) diff --git
> > > a/lib/librte_eal/common/include/arch/arm/rte_cycles_64.h
> > > b/lib/librte_eal/common/include/arch/arm/rte_cycles_64.h
> > > index 68e7c7338..bc4e3f8e6 100644
> > > --- a/lib/librte_eal/common/include/arch/arm/rte_cycles_64.h
> > > +++ b/lib/librte_eal/common/include/arch/arm/rte_cycles_64.h
> > > @@ -18,6 +18,7 @@ extern "C" {
> > >   *   The time base for this lcore.
> > >   */
> > >  #ifndef RTE_ARM_EAL_RDTSC_USE_PMU
> > > +
> > >  /**
> > >   * This call is portable to any ARMv8 architecture, however, typically
> > >   * cntvct_el0 runs at <= 100MHz and it may be imprecise for some tasks.
> > > @@ -59,11 +60,26 @@ rte_rdtsc(void)
> > >  }
> > >  #endif
> > >
> > > +#define arch_counter_enforce_ordering(val) do
> > {                                                \
> > > +       uint64_t tmp, _val = (val);
> > \
> > > +
> > \
> > > +       asm
> > volatile(
> >                                           \
> > > +       "       eor  %0, %1, %1\n"
> > \
> > > +       "       add  %0, sp, %0\n"
> > \
> > > +       "       ldr  xzr, [%0]"
> > \
> > > +       : "=r" (tmp) : "r" (_val));
> > \
> > > +} while (0)
> > > +
> >
> > Not addressed the comments in
> > http://mails.dpdk.org/archives/dev/2020-March/159547.html
> >
> > Gavin, Linhaifeng,
> > I don't think, this ordering is valid in the DPDK context.
> >
> > See the patch from Will.
> > https://lore.kernel.org/patchwork/patch/1076132/
> >
> > Thoughts?
>
> I think arch_counter_enforce_ordering maybe is invalid but isb must be valid for DPDK context.

Yes. Only isb required in the DPDK context. That would translate to
the following change

 -       rte_mb();
+ asm volatile("isb" : : : "memory")

With the above change and fixing(removing
arch_counter_enforce_ordering) in git commit log:
Acked-by: Jerin Jacob <jerinj at marvell.com>