Re: [PATCH] nfp: convert nfp_eth_set_bit_config() into a macro

[Matthias Kaehlcke]

El Tue, Oct 03, 2017 at 02:50:00PM -0700 Jakub Kicinski ha dit:

> On Tue,  3 Oct 2017 13:05:46 -0700, Matthias Kaehlcke wrote:
> > nfp_eth_set_bit_config() is marked as __always_inline to allow gcc to
> > identify the 'mask' parameter as known to be constant at compile time,
> > which is required to use the FIELD_GET() macro.
> > 
> > The forced inlining does the trick for gcc, but for kernel builds with
> > clang it results in undefined symbols:
> > 
> > drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp_eth.o: In function
> >   `__nfp_eth_set_aneg':
> > drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp_eth.c:(.text+0x787):
> >   undefined reference to `__compiletime_assert_492'
> > drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp_eth.c:(.text+0x7b1):
> >   undefined reference to `__compiletime_assert_496'
> > 
> > These __compiletime_assert_xyx() calls would have been optimized away if
> > the compiler had seen 'mask' as a constant.
> > 
> > Convert nfp_eth_set_bit_config() into a macro, which allows both gcc and
> > clang to identify 'mask' as a compile time constant.
> > 
> > Signed-off-by: Matthias Kaehlcke <mka@xxxxxxxxxxxx>
> 
> :(
> 
> Is there no chance of fixing the constant propagation in the compiler?

LLVM developers are reluctant and would like us kernel folks to
evaluate possible alternatives for the affected code first:
https://bugs.llvm.org/show_bug.cgi?id=4898

Given that this doesn't seem to be a widespread issue in the kernel
personally I would consider the conversion to a macro in this case an
acceptable solution, though it is definitely ugly. However I'm not the
owner of the driver or the subsystem, so my opinion doesn't really
carry much weight here ;-)

Any ideas about other, less ugly alternatives?

Matthias