From: Oleksandr Andrushchenko <[email protected]>
While disconnecting from the backend we clean up shared resources
(event channel, ring buffer), but never let the backend know about
that. This may lead to inconsistent backend state and/or shared
resources use.
Fix this by explicitly letting the backend know that frontend has
destroyed shared resources by changing its Xen bus state accordingly.
Signed-off-by: Oleksandr Andrushchenko <[email protected]>
---
drivers/input/misc/xen-kbdfront.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/drivers/input/misc/xen-kbdfront.c b/drivers/input/misc/xen-kbdfront.c
index 24bc5c5d876f..ecb6e719e0e2 100644
--- a/drivers/input/misc/xen-kbdfront.c
+++ b/drivers/input/misc/xen-kbdfront.c
@@ -488,6 +488,8 @@ static int xenkbd_connect_backend(struct xenbus_device *dev,
static void xenkbd_disconnect_backend(struct xenkbd_info *info)
{
+ xenbus_switch_state(info->xbdev, XenbusStateClosing);
+
if (info->irq >= 0)
unbind_from_irqhandler(info->irq, info);
info->irq = -1;
--
2.21.0
On 15/03/2019 10:23, Oleksandr Andrushchenko wrote:
> From: Oleksandr Andrushchenko <[email protected]>
>
> While disconnecting from the backend we clean up shared resources
> (event channel, ring buffer), but never let the backend know about
> that. This may lead to inconsistent backend state and/or shared
> resources use.
> Fix this by explicitly letting the backend know that frontend has
> destroyed shared resources by changing its Xen bus state accordingly.
>
> Signed-off-by: Oleksandr Andrushchenko <[email protected]>
> ---
> drivers/input/misc/xen-kbdfront.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/drivers/input/misc/xen-kbdfront.c b/drivers/input/misc/xen-kbdfront.c
> index 24bc5c5d876f..ecb6e719e0e2 100644
> --- a/drivers/input/misc/xen-kbdfront.c
> +++ b/drivers/input/misc/xen-kbdfront.c
> @@ -488,6 +488,8 @@ static int xenkbd_connect_backend(struct xenbus_device *dev,
>
> static void xenkbd_disconnect_backend(struct xenkbd_info *info)
> {
> + xenbus_switch_state(info->xbdev, XenbusStateClosing);
> +
> if (info->irq >= 0)
> unbind_from_irqhandler(info->irq, info);
> info->irq = -1;
>
As already stated for the related netfront patch: I'm not sure this
is really what we want. Have you tested xl save/restore or migration
of the guest with your patch applied?
Juergen
On 3/22/19 1:03 PM, Juergen Gross wrote:
> On 15/03/2019 10:23, Oleksandr Andrushchenko wrote:
>> From: Oleksandr Andrushchenko <[email protected]>
>>
>> While disconnecting from the backend we clean up shared resources
>> (event channel, ring buffer), but never let the backend know about
>> that. This may lead to inconsistent backend state and/or shared
>> resources use.
>> Fix this by explicitly letting the backend know that frontend has
>> destroyed shared resources by changing its Xen bus state accordingly.
>>
>> Signed-off-by: Oleksandr Andrushchenko <[email protected]>
>> ---
>> drivers/input/misc/xen-kbdfront.c | 2 ++
>> 1 file changed, 2 insertions(+)
>>
>> diff --git a/drivers/input/misc/xen-kbdfront.c b/drivers/input/misc/xen-kbdfront.c
>> index 24bc5c5d876f..ecb6e719e0e2 100644
>> --- a/drivers/input/misc/xen-kbdfront.c
>> +++ b/drivers/input/misc/xen-kbdfront.c
>> @@ -488,6 +488,8 @@ static int xenkbd_connect_backend(struct xenbus_device *dev,
>>
>> static void xenkbd_disconnect_backend(struct xenkbd_info *info)
>> {
>> + xenbus_switch_state(info->xbdev, XenbusStateClosing);
>> +
>> if (info->irq >= 0)
>> unbind_from_irqhandler(info->irq, info);
>> info->irq = -1;
>>
> As already stated for the related netfront patch: I'm not sure this
> is really what we want. Have you tested xl save/restore or migration
> of the guest with your patch applied?
Well, it comes out that this is not enough and needs much more work,
so please let's abandon this patch
>
> Juergen
Thank you,
Oleksandr