On 16/10/23 5:37 pm, Daniel Borkmann wrote:
On 10/12/23 10:03 PM, Hari Bathini wrote:
Most BPF programs are small, but they consume a page each. For systems
with busy traffic and many BPF programs, this may also add significant
pressure on instruction TLB. High iTLB pressure usually slows down the
whole system causing visible performance degradation for production
workloads.
bpf_prog_pack, a customized allocator that packs multiple bpf programs
into preallocated memory chunks, was proposed [1] to address it. This
series extends this support on powerpc.
Both bpf_arch_text_copy() & bpf_arch_text_invalidate() functions,
needed for this support depend on instruction patching in text area.
Currently, patch_instruction() supports patching only one instruction
at a time. The first patch introduces patch_instructions() function
to enable patching more than one instruction at a time. This helps in
avoiding performance degradation while JITing bpf programs.
Patches 2 & 3 implement the above mentioned arch specific functions
using patch_instructions(). Patch 4 fixes a misnomer in bpf JITing
code. The last patch enables the use of BPF prog pack allocator on
powerpc and also, ensures cleanup is handled gracefully.
[1] https://lore.kernel.org/bpf/20220204185742.271030-1-song@xxxxxxxxxx/
Changes in v6:
* No changes in patches 2-5/5 except addition of Acked-by tags from Song.
* Skipped merging code path of patch_instruction() & patch_instructions()
to avoid performance overhead observed on ppc32 with that.
I presume this will be routed via Michael?
Yes, Daniel. This can go via linuxppc tree.
Thanks
Hari
