Re: [RFC PATCH v2 13/20] context_tracking: Make context_tracking_key __ro_after_init
From: Josh Poimboeuf <jpoimboe@kernel.org>
Date: 2023-07-28 16:00:31
Also in:
bpf, kvm, linux-doc, linux-kselftest, linux-mm, lkml, rcu
On Thu, Jul 20, 2023 at 05:30:49PM +0100, Valentin Schneider wrote:
objtool now warns about it: vmlinux.o: warning: objtool: enter_from_user_mode+0x4e: Non __ro_after_init static key "context_tracking_key" in .noinstr section vmlinux.o: warning: objtool: enter_from_user_mode+0x50: Non __ro_after_init static key "context_tracking_key" in .noinstr section vmlinux.o: warning: objtool: syscall_enter_from_user_mode+0x60: Non __ro_after_init static key "context_tracking_key" in .noinstr section vmlinux.o: warning: objtool: syscall_enter_from_user_mode+0x62: Non __ro_after_init static key "context_tracking_key" in .noinstr section [...] The key can only be enabled (and not disabled) in the __init function ct_cpu_tracker_user(), so mark it as __ro_after_init. Signed-off-by: Valentin Schneider <vschneid@redhat.com>
It's best to avoid temporarily introducing warnings. Bots will rightfully complain about that. This patch and the next one should come before the objtool patches. Also it would be helpful for the commit log to have a brief justification for the patch beyond "fix the objtool warning". Something roughly like: Soon, runtime-mutable text won't be allowed in .noinstr sections, so that a code patching IPI to a userspace-bound CPU can be safely deferred to the next kernel entry. 'context_tracking_key' is only enabled in __init ct_cpu_tracker_user(). Mark it as __ro_after_init. -- Josh