Re: [dpdk-dev] [PATCH v5 02/21] eal: list acceptable init priorities

[Neil Horman <nhorman@tuxdriver.com>]

On Thu, Apr 12, 2018 at 11:57:47PM +0200, Gaëtan Rivet wrote:
> Hello Neil,
> 
> On Thu, Apr 12, 2018 at 07:28:26AM -0400, Neil Horman wrote:
> > On Wed, Apr 11, 2018 at 02:04:03AM +0200, Gaetan Rivet wrote:
> > > Build a central list to quickly see each used priorities for
> > > constructors, allowing to verify that they are both above 100 and in the
> > > proper order.
> > > 
> > > Signed-off-by: Gaetan Rivet <gaetan.rivet@6wind.com>
> > > Acked-by: Neil Horman <nhorman@tuxdriver.com>
> > > Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>
> > > ---
> > >  lib/librte_eal/common/eal_common_log.c     | 2 +-
> > >  lib/librte_eal/common/include/rte_bus.h    | 2 +-
> > >  lib/librte_eal/common/include/rte_common.h | 8 +++++++-
> > >  3 files changed, 9 insertions(+), 3 deletions(-)
> > > 
> > > diff --git a/lib/librte_eal/common/eal_common_log.c b/lib/librte_eal/common/eal_common_log.c
> > > index a27192620..36b9d6e08 100644
> > > --- a/lib/librte_eal/common/eal_common_log.c
> > > +++ b/lib/librte_eal/common/eal_common_log.c
> > > @@ -260,7 +260,7 @@ static const struct logtype logtype_strings[] = {
> > >  };
> > >  
> > >  /* Logging should be first initializer (before drivers and bus) */
> > > -RTE_INIT_PRIO(rte_log_init, 101);
> > > +RTE_INIT_PRIO(rte_log_init, LOG);
> > >  static void
> > >  rte_log_init(void)
> > >  {
> > > diff --git a/lib/librte_eal/common/include/rte_bus.h b/lib/librte_eal/common/include/rte_bus.h
> > > index 6fb08341a..eb9eded4e 100644
> > > --- a/lib/librte_eal/common/include/rte_bus.h
> > > +++ b/lib/librte_eal/common/include/rte_bus.h
> > > @@ -325,7 +325,7 @@ enum rte_iova_mode rte_bus_get_iommu_class(void);
> > >   * The constructor has higher priority than PMD constructors.
> > >   */
> > >  #define RTE_REGISTER_BUS(nm, bus) \
> > > -RTE_INIT_PRIO(businitfn_ ##nm, 110); \
> > > +RTE_INIT_PRIO(businitfn_ ##nm, BUS); \
> > >  static void businitfn_ ##nm(void) \
> > >  {\
> > >  	(bus).name = RTE_STR(nm);\
> > > diff --git a/lib/librte_eal/common/include/rte_common.h b/lib/librte_eal/common/include/rte_common.h
> > > index 6c5bc5a76..8f04518f7 100644
> > > --- a/lib/librte_eal/common/include/rte_common.h
> > > +++ b/lib/librte_eal/common/include/rte_common.h
> > > @@ -81,6 +81,12 @@ typedef uint16_t unaligned_uint16_t;
> > >   */
> > >  #define RTE_SET_USED(x) (void)(x)
> > >  
> > > +#define RTE_PRIORITY_LOG 101
> > > +#define RTE_PRIORITY_BUS 110
> > > +
> > > +#define RTE_PRIO(prio) \
> > > +	RTE_PRIORITY_ ## prio
> > > +
> > >  /**
> > >   * Run function before main() with low priority.
> > >   *
> > > @@ -102,7 +108,7 @@ static void __attribute__((constructor, used)) func(void)
> > >   *   Lowest number is the first to run.
> > >   */
> > >  #define RTE_INIT_PRIO(func, prio) \
> > > -static void __attribute__((constructor(prio), used)) func(void)
> > > +static void __attribute__((constructor(RTE_PRIO(prio)), used)) func(void)
> > >  
> > It just occured to me, that perhaps you should add a RTE_PRORITY_LAST priority,
> > and redefine RTE_INIT to RTE_INIT_PRIO(func, RTE_PRIORITY_LAST) for clarity.  I
> > presume that constructors with no explicit priority run last, but the gcc
> > manual doesn't explicitly say that.  It would be a heck of a bug to track down
> > if somehow unprioritized constructors ran early.
> > 
> > Neil
> > 
> 
> While certainly poorly documented, the behavior is well-defined. I don't see
> a situation where the bug you describe could arise.
> 
> Adding RTE_PRIORITY_LAST is pretty harmless, but I'm not sure it's
> justified to add it. If you still think it is useful, I will do it.
> 
It was more just a way to unify the macros is all, probably not important.

> I'd be curious to hear if anyone has had issues of this kind.
> 
I've not had any, but I was suprised to see that the gcc manual didn't
explicitly call out the implied priority of unprioritized constructors

Neil

> -- 
> Gaëtan Rivet
> 6WIND
>