From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752751AbbFVSBW (ORCPT ); Mon, 22 Jun 2015 14:01:22 -0400 Received: from mail-pd0-f182.google.com ([209.85.192.182]:36425 "EHLO mail-pd0-f182.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751810AbbFVSBO (ORCPT ); Mon, 22 Jun 2015 14:01:14 -0400 From: John Stultz To: lkml Cc: John Stultz , Felipe Balbi , Al Viro , Andrzej Pietrasiewicz , Krzysztof Opasiak , Greg Kroah-Hartman , Michal Nazarewicz , Robert Baldyga , linux-usb@vger.kernel.org Subject: [PATCH] functionfs: Avoid aio locking problem Date: Mon, 22 Jun 2015 11:01:04 -0700 Message-Id: <1434996064-20284-1-git-send-email-john.stultz@linaro.org> X-Mailer: git-send-email 1.9.1 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org The functionfs aio logic seems broken. When using functionfs, I was seeing frequent hangs, and enabling spinlock debugging, I got: g_ffs gadget: g_ffs ready ci_hdrc ci_hdrc.0: CI_HDRC_CONTROLLER_RESET_EVENT received BUG: spinlock lockup suspected on CPU#0, adbd/2791 lock: 0xe7764880, .magic: e7764880, .owner: /-1, .owner_cpu: -407539900 CPU: 0 PID: 2791 Comm: adbd Not tainted 4.1.0-rc1-00032-g359b12f #147 Hardware name: Qualcomm (Flattened Device Tree) [] (unwind_backtrace) from [] (show_stack+0x10/0x14) [] (show_stack) from [] (dump_stack+0x70/0xbc) [] (dump_stack) from [] (do_raw_spin_lock+0x114/0x1a0) [] (do_raw_spin_lock) from [] (_raw_spin_lock_irqsave+0x50/0x5c) [] (_raw_spin_lock_irqsave) from [] (kiocb_set_cancel_fn+0x1c/0x60) [] (kiocb_set_cancel_fn) from [] (ffs_epfile_read_iter+0x8c/0x140) [] (ffs_epfile_read_iter) from [] (__vfs_read+0xb0/0xd4) [] (__vfs_read) from [] (vfs_read+0x7c/0x100) [] (vfs_read) from [] (SyS_read+0x40/0x8c) [] (SyS_read) from [] (ret_fast_syscall+0x0/0x4c) INFO: rcu_preempt detected stalls on CPUs/tasks: 0: (1 GPs behind) idle=805/140000000000000/0 softirq=7187/7189 fqs=2601 (detected by 3, t=2603 jiffies, g=3028, c=3027, q=474) Task dump for CPU 0: adbd R running 0 2791 1 0x00000002 [] (__schedule) from [] (0xffffffff) Looking at the code, the __vfs_read() calls new_sync_read(), which allocates a struct kiocb kiocb on the stack and passes it to the ffs_epfile_read_iter() funciton. That then calls kiocb_set_cancel_fn() passing a pointer to that kiocb. However, kiocb_set_cancel_fn() assumes the kiocb is a sub-element of a struct aio_kiocb, and it tries to grab the kioctx from that parent structure. However it seems there is no aio_kiocb structure here, so the spin_lock_irqsave hangs trying to lock random data on the stack. This patch avoids the issue, by only calling kiocb_set_cancel_fn if the aio flag is set. Cc: Felipe Balbi Cc: Al Viro Cc: Andrzej Pietrasiewicz Cc: Krzysztof Opasiak Cc: Greg Kroah-Hartman Cc: Michal Nazarewicz Cc: Robert Baldyga Cc: linux-usb@vger.kernel.org Signed-off-by: John Stultz --- drivers/usb/gadget/function/f_fs.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/drivers/usb/gadget/function/f_fs.c b/drivers/usb/gadget/function/f_fs.c index 3507f88..d2434c9 100644 --- a/drivers/usb/gadget/function/f_fs.c +++ b/drivers/usb/gadget/function/f_fs.c @@ -924,7 +924,8 @@ static ssize_t ffs_epfile_write_iter(struct kiocb *kiocb, struct iov_iter *from) kiocb->private = p; - kiocb_set_cancel_fn(kiocb, ffs_aio_cancel); + if (p->aio) + kiocb_set_cancel_fn(kiocb, ffs_aio_cancel); res = ffs_epfile_io(kiocb->ki_filp, p); if (res == -EIOCBQUEUED) @@ -968,7 +969,8 @@ static ssize_t ffs_epfile_read_iter(struct kiocb *kiocb, struct iov_iter *to) kiocb->private = p; - kiocb_set_cancel_fn(kiocb, ffs_aio_cancel); + if (p->aio) + kiocb_set_cancel_fn(kiocb, ffs_aio_cancel); res = ffs_epfile_io(kiocb->ki_filp, p); if (res == -EIOCBQUEUED) -- 1.9.1 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in Please read the FAQ at http://www.tux.org/lkml/