[dpdk-dev] [PATCH v2] eal: rename state values in rte_lcore_state

[Stephen Hemminger]

On Tue, 2 Apr 2019 23:03:06 +0200
David Marchand <david.marchand at redhat.com> wrote:

> On Tue, Apr 2, 2019 at 10:48 PM Stephen Hemminger <
> stephen at networkplumber.org> wrote:  
> 
> > On Tue, 2 Apr 2019 22:15:40 +0200
> > David Marchand <david.marchand at redhat.com> wrote:
> >  
> > > On Tue, Apr 2, 2019 at 5:57 PM Stephen Hemminger <
> > stephen at networkplumber.org>
> > > wrote:
> > >  
> > > > C language does not really treat enum's as first class symbols.
> > > > The values in an enum live in a global namespace.  That means if
> > > > DPDK defines "RUNNING" it can't be used by another enum in an
> > > > application using DPDK.
> > > >
> > > > To solve this add a prefix "RTE_LCORE_" to the enum values, and
> > > > make them grammatically consistent.
> > > >  
> > >
> > > Well, I understand this is not clean, but this patch breaks the API.  
> >
> > The lcore state was marked as internal in the header file, code that
> > ignores that is going to have problems. The values are the same.
> >
> > We could defer this to 19.11 (next LTS) since it is cosmetic.
> >  
> 
> We must announce it before changing.

I disagree, if an API is marked as internal it can be changed at any
time (as long as ABI is maintained).

> 
> 
> At that time, I am willing to do more work to make lcore_config hidden;
> > ie no inline's to access it.
> >  
> 
> Ah I was looking at that, to see if we could at least shoot the direct
> accesses by calling the existing API.
> I can see we are missing one wrapper for the cpu id...
> 
> And there are quite some odds things in with cpu affinity in dpaa:
> drivers/bus/dpaa/dpaa_bus.c:    cpu = lcore_config[lcore].core_id;
> drivers/bus/dpaa/dpaa_bus.c:
> &lcore_config[lcore].cpuset);
> drivers/bus/fslmc/portal/dpaa2_hw_dpio.c:                       if
> (CPU_ISSET(i, &lcore_config[lcore_id].cpuset)) {
> 
> 

Drivers have no API/ABI restriction, it is only user code.