On 8/21/24 5:38 PM, Eduard Zingerman wrote:
On Wed, 2024-08-21 at 17:34 -0700, Alexei Starovoitov wrote: [...]Something like insn_0 ... r1 = 0 if rX == .. goto insn_0 this jmp will be rewritten to point to newly added *(u64*)(r10 - ..) = r1 so at run time it will overwrite that slot with zero and epilogue will read zero from it instead of ctx.That's exactly what I tried on paper and jmp target was just moving down.
Thanks for bringing this up. I will try a test case to confirm.
