Hi Stephen, Please see inline.
Thanks, Anoob > -----Original Message----- > From: Stephen Hemminger <[email protected]> > Sent: Saturday, May 27, 2023 3:42 AM > To: Anoob Joseph <[email protected]> > Cc: Thomas Monjalon <[email protected]>; Akhil Goyal > <[email protected]>; Jerin Jacob Kollanukkaran <[email protected]>; > Konstantin Ananyev <[email protected]>; Volodymyr Fialko > <[email protected]>; Hemant Agrawal <[email protected]>; > Mattias Rönnblom <[email protected]>; Kiran Kumar > Kokkilagadda <[email protected]>; [email protected]; Olivier Matz > <[email protected]> > Subject: [EXT] Re: [PATCH v4 21/22] pdcp: add thread safe processing > > External Email > > ---------------------------------------------------------------------- > On Sat, 27 May 2023 02:32:13 +0530 > Anoob Joseph <[email protected]> wrote: > > > +static inline uint32_t > > +pdcp_atomic_inc(uint32_t *val, const bool mt_safe) { > > + if (mt_safe) > > + return __atomic_fetch_add(val, 1, __ATOMIC_RELAXED); > > + else > > + return (*val)++; > > +} > > This is a bad pattern. None of the rest of DPDK does this. > Either be thread safe or not. [Anoob] Most protocol implementation would have a similar issue. I've tried to follow the approach taken in lib IPsec (please check lib/ipsec/ipsec_sqn.h). From the discussion we had on v3, I've tried to make it compile time constant to remove any conditional checks in datapath. If you still think, this is an issue, I could drop this patch for now and re-introduce it later once we have test applications that can work with traffic. Please share your thoughts.
