Re: [PATCHv2 0/7] eBPF JIT for PPC64

On 2016/06/30 12:34PM, Denis Kirjanov wrote:

On 6/30/16, Andreas Ziegler [off-list ref] wrote:

quoted

Hi Naveen,

this patchset makes a change to arch/powerpc/net/Makefile in order to only
compile the previously existing bpf_jit_comp.c if !CONFIG_PPC64, and use
bpf_jit_comp64.c if CONFIG_PPC64 is enabled.

Inside arch/powerpc/net/bpf_jit_comp.c, however, there is still an #ifdef
CONFIG_PPC64 block at line 667 (linux-next of today, i.e., next-20160630):

  #ifdef CONFIG_PPC64
                  /* Function descriptor nastiness: Address + TOC */
                  ((u64 *)image)[0] = (u64)code_base;
                  ((u64 *)image)[1] = local_paca->kernel_toc;
  #endif

From my understanding of the code, this #ifdef can now be removed, as there
is
no way the file could be compiled with CONFIG_PPC64 enabled. Is this
correct?

That was used for running classic BPF on ppc64. With eBPF on ppc64 the
whole block can be removed.

Yes, that can be removed and so can the many 64-bit related macros in 
bpf_jit32.h. The reason I didn't remove those just yet was so that we 
could easily try out the classic BPF JIT for ppc64 BE, for 
testing/performance comparison and so on. I think it would be good to 
retain this for one kernel release.

- Naveen