On Thu, Oct 14, 2021 at 10:31:22AM +0100, Will Deacon wrote:

On Sat, Oct 09, 2021 at 12:07:57PM +0200, Peter Zijlstra wrote:

quoted

Add a comment to the PT_flags to indicate their actual value, this
makes it easier to see what bits are used and where there might be a
possible hole to use.

Notable PT_SEIZED was placed wrong, also PT_EVENT_FLAG() space seems
ill defined, as written is seems to be meant to cover the entire
PTRACE_O_ range offset by 3 bits, which would then be 3+[0..21],
however PT_SEIZED is in the middle of that.

Why do you think PT_EVENT_FLAG() should cover all the PTRACE_O_* options?
Just going by the name and current callers, I'd only expect it to cover
the PTRACE_EVENT_* flags, no?

Because PT_EXITKILL and PT_SUSPEND_SECCOMP are also exposed in that same
mapping.

Ideally we'd change PT_OPT_FLAG_SHIFT to 8 or something and have the
high 24 bits for OPT and then use the low 8 bits for SEIZED and the new
flags.