lkml.org 
[lkml]   [2012]   [Jun]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [PATCH] xen/mce: Add mutex lock and buffer to avoid sleep in atomic context
On Mon, Jun 11, 2012 at 03:55:00AM +0000, Liu, Jinsong wrote:
> Liu, Jinsong wrote:
> > From a9c5f29330a056291356b912816b5b2e0e061a30 Mon Sep 17 00:00:00 2001
> > From: Liu, Jinsong <jinsong.liu@intel.com>
> > Date: Sat, 9 Jun 2012 00:56:46 +0800
> > Subject: [PATCH] xen/mce: Add mutex lock and buffer to avoid sleep in
> > atomic context
> >
>
> Sorry, I update the patch a little, for spinlock to avoid deadlock.
>
> Thanks,
> Jinsong
>
> ====================
> >From db6c0ac9372c6fbc3637ec4216830e7ee01b31aa Mon Sep 17 00:00:00 2001
> From: Liu, Jinsong <jinsong.liu@intel.com>
> Date: Mon, 11 Jun 2012 19:21:24 +0800
> Subject: [PATCH] xen/mce: Add mutex lock and buffer to avoid sleep in atomic context
>
> copy_to_user might sleep and print a stack trace if it is executed
> in an atomic spinlock context. This patch add a mutex lock and a
> buffer to avoid the issue.
>
> This patch also change the manipulation of mcelog_lock from
> spin_lock_irqsave to spin_trylock to avoid deadlock, since
> mcelog_lock is used at normal process context and
> mce context (which is async exception context that could

Could you explain in more details what is 'async exception
context' and 'mce context' ?

> not protected by spin_lock_irqsave). When fail to get spinlock,
> mc_info would be transferred by hypervisor next time.

What does that mean? How would 'mcelog' program get the data?

>
> Reported-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
> Signed-off-by: Liu, Jinsong <jinsong.liu@intel.com>
> ---
> drivers/xen/mcelog.c | 38 +++++++++++++++++++++++++++++++-------
> 1 files changed, 31 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/xen/mcelog.c b/drivers/xen/mcelog.c
> index 72e87d2..fac29e4 100644
> --- a/drivers/xen/mcelog.c
> +++ b/drivers/xen/mcelog.c
> @@ -56,12 +56,14 @@ static struct mcinfo_logical_cpu *g_physinfo;
> static uint32_t ncpus;
>
> static DEFINE_SPINLOCK(mcelog_lock);
> +static DEFINE_MUTEX(xen_mce_chrdev_read_mutex);
>
> static struct xen_mce_log xen_mcelog = {
> .signature = XEN_MCE_LOG_SIGNATURE,
> .len = XEN_MCE_LOG_LEN,
> .recordlen = sizeof(struct xen_mce),
> };
> +static struct xen_mce_log xen_mcelog_u;
>
> static DEFINE_SPINLOCK(xen_mce_chrdev_state_lock);
> static int xen_mce_chrdev_open_count; /* #times opened */
> @@ -106,9 +108,19 @@ static ssize_t xen_mce_chrdev_read(struct file *filp, char __user *ubuf,
> unsigned num;
> int i, err;
>
> + /*
> + * copy_to_user might sleep and print a stack trace
> + * if it is executed in an atomic spinlock context
> + */
> + mutex_lock(&xen_mce_chrdev_read_mutex);
> +
> spin_lock(&mcelog_lock);
> + memcpy(&xen_mcelog_u, &xen_mcelog, sizeof(struct xen_mce_log));
>
> num = xen_mcelog.next;
> + memset(xen_mcelog.entry, 0, num * sizeof(struct xen_mce));
> + xen_mcelog.next = 0;
> + spin_unlock(&mcelog_lock);
>
> /* Only supports full reads right now */
> err = -EINVAL;
> @@ -117,20 +129,20 @@ static ssize_t xen_mce_chrdev_read(struct file *filp, char __user *ubuf,
>
> err = 0;
> for (i = 0; i < num; i++) {
> - struct xen_mce *m = &xen_mcelog.entry[i];
> + struct xen_mce *m = &xen_mcelog_u.entry[i];
>
> err |= copy_to_user(buf, m, sizeof(*m));
> buf += sizeof(*m);
> }
>
> - memset(xen_mcelog.entry, 0, num * sizeof(struct xen_mce));
> - xen_mcelog.next = 0;
> + memset(xen_mcelog_u.entry, 0, num * sizeof(struct xen_mce));
> + xen_mcelog_u.next = 0;
>
> if (err)
> err = -EFAULT;
>
> out:
> - spin_unlock(&mcelog_lock);
> + mutex_unlock(&xen_mce_chrdev_read_mutex);
>
> return err ? err : buf - ubuf;
> }
> @@ -313,9 +325,21 @@ static int mc_queue_handle(uint32_t flags)
> static irqreturn_t xen_mce_interrupt(int irq, void *dev_id)
> {
> int err;
> - unsigned long tmp;
>
> - spin_lock_irqsave(&mcelog_lock, tmp);
> + /*
> + * mcelog_lock is used at normal process context and
> + * mce context (which is async exception context that could
> + * not protected by spin_lock_irqsave).
> + *
> + * use spin_trylock to avoid deadlock. When fail to get spinlock,
> + * mc_info would be transferred by hypervisor next time.
> + */
> + if (unlikely(!spin_trylock(&mcelog_lock))) {
> + pr_err(XEN_MCELOG
> + "Failed to get mcelog_lock, mc_info would "
> + "be transferred by hypervisor next time.\n");

Ugh. Why the printk? How does this benefit the user? If it
recovers - which I presume "..next time" means then it should be OK?

What does 'transferred by hypervisor' mean actually?

Would it be better to schedule a workqueue to poll the data? Perhaps that
is how this whole IRQ handler should be done - it kicks of an IRQ handler
that de-spolls the data?

> + return IRQ_NONE;
> + }
>
> /* urgent mc_info */
> err = mc_queue_handle(XEN_MC_URGENT);
> @@ -330,7 +354,7 @@ static irqreturn_t xen_mce_interrupt(int irq, void *dev_id)
> pr_err(XEN_MCELOG
> "Failed to handle nonurgent mc_info queue.\n");
>
> - spin_unlock_irqrestore(&mcelog_lock, tmp);
> + spin_unlock(&mcelog_lock);
>
> return IRQ_HANDLED;
> }
> --
> 1.7.1




\
 
 \ /
  Last update: 2012-06-11 22:21    [W:0.049 / U:0.128 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site