[PATCH v4 0/2] fix & prevent the missing preemption disabling
Steven Rostedt
rostedt at goodmis.org
Mon Oct 25 19:42:33 PDT 2021
On Tue, 26 Oct 2021 10:09:12 +0800
王贇 <yun.wang at linux.alibaba.com> wrote:
> Just a ping, to see if there are any more comments :-P
I guess you missed what went into mainline (and your name found a bug
in my perl script for importing patches ;-)
https://lore.kernel.org/all/20211019091344.65629198@gandalf.local.home/
Which means patch 1 needs to change:
> + /*
> + * Disable preemption to fulfill the promise.
> + *
> + * Don't worry about the bit 0 cases, they indicate
> + * the disabling behaviour has already been done by
> + * internal call previously.
> + */
> + preempt_disable_notrace();
> +
> return bit + 1;
> }
>
> +/*
> + * Preemption will be enabled (if it was previously enabled).
> + */
> static __always_inline void trace_clear_recursion(int bit)
> {
> if (!bit)
> return;
>
> + if (bit > 0)
> + preempt_enable_notrace();
> +
Where this wont work anymore.
Need to preempt disable and enable always.
-- Steve
> barrier();
> bit--;
> trace_recursion_clear(bit);
> @@ -209,7 +227,7 @@ static __always_inline void trace_clear_recursion(int bit)
> * tracing recursed in the same context (normal vs interrupt),
> *
More information about the linux-riscv
mailing list