Date: Fri, 11 Jan 2013 11:44:09 -0500 From: John Baldwin <jhb@freebsd.org> To: freebsd-current@freebsd.org Cc: Ryan Stone <rysto32@gmail.com> Subject: Re: Spurious witness warning when destroying spin mtx Message-ID: <201301111144.10060.jhb@freebsd.org> In-Reply-To: <CAFMmRNyYccyXFh0r2jC2Q5ynYQH09SiZNguLp8X4JWSX4Lua5w@mail.gmail.com> References: <CAFMmRNyYccyXFh0r2jC2Q5ynYQH09SiZNguLp8X4JWSX4Lua5w@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Friday, November 23, 2012 10:08:28 PM Ryan Stone wrote:
> Today I saw a spurious witness warning for "acquiring duplicate lock of
> same type". The root cause is that when running mtx_destroy on a spinlock
> that is held by the current thread, mtx_destroy calls spinlock_exit()
> before calling WITNESS_UNLOCK, which opens up a window in which the CPU can
> be interrupted and attempt to acquire another spinlock of the same type as
> the one being destroyed. This patch should fix it:
>
> diff --git a/sys/kern/kern_mutex.c b/sys/kern/kern_mutex.c
> index 2f13863..96f43f8 100644
> --- a/sys/kern/kern_mutex.c
> +++ b/sys/kern/kern_mutex.c
> @@ -918,16 +918,16 @@ _mtx_destroy(volatile uintptr_t *c)
> else {
> MPASS((m->mtx_lock & (MTX_RECURSED|MTX_CONTESTED)) == 0);
>
> + lock_profile_release_lock(&m->lock_object);
> + /* Tell witness this isn't locked to make it happy. */
> + WITNESS_UNLOCK(&m->lock_object, LOP_EXCLUSIVE, __FILE__,
> + __LINE__);
> +
> /* Perform the non-mtx related part of mtx_unlock_spin().
> */ if (LOCK_CLASS(&m->lock_object) == &lock_class_mtx_spin)
> spinlock_exit();
> else
> curthread->td_locks--;
> -
> - lock_profile_release_lock(&m->lock_object);
> - /* Tell witness this isn't locked to make it happy. */
> - WITNESS_UNLOCK(&m->lock_object, LOP_EXCLUSIVE, __FILE__,
> - __LINE__);
> }
>
> m->mtx_lock = MTX_DESTROYED
Ah, I would tweak this slightly perhaps to match mtx_unlock() and
mtx_unlock_spin():
if (LOCK_CLASS() == &lock_class_mtx_sleep)
curthread->td_locks--;
/* lock profile and witness stuff */
if (LOCK_CLASS() == &lock_class_mtx_spin)
spinlock_exit();
You are correct that it is broken for the spin case.
--
John Baldwin
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201301111144.10060.jhb>