On 12/16/22 6:17 PM, Eduard Zingerman wrote:
BPF limits stack usage by MAX_BPF_STACK bytes across all call frames,
however this is enforced by function check_max_stack_depth() which is
executed after do_check_{subprogs,main}().
This means that when check_ids() is executed the maximal stack depth is not
yet verified, thus in theory the number of stack spills might be
MAX_CALL_FRAMES * MAX_BPF_STACK / BPF_REG_SIZE.
However, any program with stack usage deeper than
MAX_BPF_STACK / BPF_REG_SIZE would be rejected by verifier.
Hence save some memory by reducing the BPF_ID_MAP_SIZE.
This is a follow up for
https://lore.kernel.org/bpf/CAEf4BzYN1JmY9t03pnCHc4actob80wkBz2vk90ihJCBzi8CT9w@xxxxxxxxxxxxxx/
Signed-off-by: Eduard Zingerman <eddyz87@xxxxxxxxx>
Suggested-by: Andrii Nakryiko <andrii@xxxxxxxxxx>
Acked-by: Yonghong Song <yhs@xxxxxx>
