All of lore.kernel.org
 help / color / mirror / Atom feed
From: Sheng Yang <sheng@linux.intel.com>
To: Avi Kivity <avi@redhat.com>, Marcelo Tosatti <mtosatti@redhat.com>
Cc: kvm@vger.kernel.org, Sheng Yang <sheng@linux.intel.com>
Subject: [PATCH 15/15] KVM: Fix racy in kvm_free_assigned_irq
Date: Thu, 25 Dec 2008 17:09:39 +0800	[thread overview]
Message-ID: <1230196179-6918-16-git-send-email-sheng@linux.intel.com> (raw)
In-Reply-To: <1230196179-6918-1-git-send-email-sheng@linux.intel.com>

Thanks to Marcelo's observation, The following code have potential issue:

if (cancel_work_sync(&assigned_dev->interrupt_work))
	kvm_put_kvm(kvm);

In fact, cancel_work_sync() would return true either work struct is only
scheduled or the callback of work struct is executed. This code only
consider the former situation.

Also, we have a window between cancel_work_sync() and free_irq. This patch fixs
them two.

Signed-off-by: Sheng Yang <sheng@linux.intel.com>
---
 virt/kvm/kvm_main.c |   26 ++++++++++++++++++++++----
 1 files changed, 22 insertions(+), 4 deletions(-)

diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 1863942..ed10f15 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -186,10 +186,28 @@ static void kvm_free_assigned_irq(struct kvm *kvm,
 	if (!assigned_dev->irq_requested_type)
 		return;
 
-	if (cancel_work_sync(&assigned_dev->interrupt_work))
-		/* We had pending work. That means we will have to take
-		 * care of kvm_put_kvm.
-		 */
+	/*
+	 * We need to ensure: kvm_put_kvm() paired with kvm_get_kvm() in
+	 * kvm_assigned_dev_intr, and no more interrupt after we cancelled
+	 * current one.
+	 *
+	 * Here we have two possiblities for cancel_work_sync() return true:
+	 * 1. The work is scheduled, but callback haven't been called.  We need
+	 * to call kvm_put_kvm() here. And IRQ is already disabled without
+	 * doubt.
+	 *
+	 * 2. The callback have executed, here we don't need to call
+	 * kvm_put_kvm(), but we may need to disable irq(e.g. for MSI).
+	 *
+	 * We judge the two condition according to if we have pending IRQs in
+	 * irq_fifo.
+	 */
+	if (kfifo_len(assigned_dev->irq_fifo) == 0 &&
+	    (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI))
+		disable_irq_nosync(assigned_dev->host_irq);
+
+	if (cancel_work_sync(&assigned_dev->interrupt_work) &&
+	    kfifo_len(assigned_dev->irq_fifo) != 0)
 		kvm_put_kvm(kvm);
 
 	free_irq(assigned_dev->host_irq, (void *)assigned_dev);
-- 
1.5.4.5


  parent reply	other threads:[~2008-12-25  9:09 UTC|newest]

Thread overview: 37+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2008-12-25  9:09 [PATCH 0/15] Device assignment & MSI enhancement Sheng Yang
2008-12-25  9:09 ` [PATCH 01/15] KVM: Add MSI_ACTION flag for assigned irq Sheng Yang
2008-12-25  9:09 ` [PATCH 02/15] KVM: Use kvm_free_assigned_irq() for free irq Sheng Yang
2008-12-25  9:09 ` [PATCH 03/15] KVM: Add support to disable MSI for assigned device Sheng Yang
2008-12-25  9:09 ` [PATCH 04/15] KVM: Add a route layer to convert MSI message to GSI Sheng Yang
2008-12-25  9:09 ` [PATCH 05/15] KVM: Using gsi_msg mapping for MSI device assignment Sheng Yang
2008-12-25  9:09 ` [PATCH 06/15] KVM: Improve MSI dispatch function Sheng Yang
2008-12-25  9:09 ` [PATCH 07/15] KVM: Using ioapic_irqchip() macro for kvm_set_irq Sheng Yang
2008-12-25  9:09 ` [PATCH 08/15] KVM: Merge MSI handling to kvm_set_irq Sheng Yang
2008-12-25  9:09 ` [PATCH 09/15] KVM: Split IOAPIC structure Sheng Yang
2008-12-25  9:09 ` [PATCH 10/15] KVM: Unified the delivery of IOAPIC and MSI Sheng Yang
2008-12-25  9:09 ` [PATCH 11/15] KVM: Change API of kvm_ioapic_get_delivery_bitmask Sheng Yang
2008-12-25  9:09 ` [PATCH 12/15] KVM: Update intr delivery func to accept unsigned long* bitmap Sheng Yang
2008-12-25  9:09 ` [PATCH 13/15] KVM: bit ops for deliver_bitmap Sheng Yang
2008-12-25  9:09 ` [PATCH 14/15] KVM: Using kfifo for irq recording Sheng Yang
2008-12-26  2:29   ` [PATCH 14/15] KVM: Replace host_irq_disable with a new flag Sheng Yang
2008-12-25  9:09 ` Sheng Yang [this message]
2008-12-25 11:56   ` [PATCH 15/15] KVM: Fix racy in kvm_free_assigned_irq Sheng Yang
2008-12-26  2:30     ` Sheng Yang
2008-12-27 20:06       ` Marcelo Tosatti
2008-12-27 20:15         ` Marcelo Tosatti
2008-12-28 11:24         ` Sheng Yang
2008-12-28 12:57           ` Avi Kivity
2008-12-29  5:42           ` Amit Shah
2008-12-29 12:23             ` Sheng Yang
2008-12-29 13:37               ` Avi Kivity
2008-12-29 13:49                 ` Sheng Yang
2008-12-29 15:20               ` Marcelo Tosatti
2008-12-30  2:14                 ` Sheng Yang
2008-12-30 16:45                   ` Marcelo Tosatti
2008-12-31  5:43                     ` Sheng Yang
2009-01-02  0:10                       ` Marcelo Tosatti
2009-01-05  7:07                         ` Sheng Yang
2009-01-05 13:27                           ` Avi Kivity
2009-01-06  1:25                             ` Sheng Yang
2008-12-29 13:20             ` Avi Kivity
2008-12-25  9:13 ` [PATCH 0/15] Device assignment & MSI enhancement Sheng Yang

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1230196179-6918-16-git-send-email-sheng@linux.intel.com \
    --to=sheng@linux.intel.com \
    --cc=avi@redhat.com \
    --cc=kvm@vger.kernel.org \
    --cc=mtosatti@redhat.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.