Commit 99605175c (xen-pt: Fix PCI devices re-attach failed) introduced
a subtle bug. As soon as the guest switches off Bus Mastering on the
device it immediately causes all the BARs be unmapped due to the DMA
address space of the device being changed. This is undesired behavior
because the guest may try to communicate with the device after that
which triggers the following errors in the logs:

[00:05.0] xen_pt_bar_read: Error: Should not read BAR through QEMU. @0x0000000000000200
[00:05.0] xen_pt_bar_write: Error: Should not write BAR through QEMU. @0x0000000000000200

The issue that the original patch tried to workaround (uneven number of
region_add/del calls on device attach/detach) was fixed in later QEMU
versions.

Signed-off-by: Igor Druzhinin <igor.druzhinin@citrix.com>
Reported-by: Ross Lagerwall <ross.lagerwall@citrix.com>
---
 hw/xen/xen_pt.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/hw/xen/xen_pt.c b/hw/xen/xen_pt.c
index 9b7a960..e5a6eff 100644
--- a/hw/xen/xen_pt.c
+++ b/hw/xen/xen_pt.c
@@ -907,7 +907,7 @@ out:
         }
     }
 
-    memory_listener_register(&s->memory_listener, &s->dev.bus_master_as);
+    memory_listener_register(&s->memory_listener, &address_space_memory);
     memory_listener_register(&s->io_listener, &address_space_io);
     s->listener_set = true;
     XEN_PT_LOG(d,
-- 
2.7.4

ping?

On Fri, Apr 06, 2018 at 10:21:23PM +0100, Igor Druzhinin wrote:
> Commit 99605175c (xen-pt: Fix PCI devices re-attach failed) introduced
> a subtle bug. As soon as the guest switches off Bus Mastering on the
> device it immediately causes all the BARs be unmapped due to the DMA
> address space of the device being changed. This is undesired behavior
> because the guest may try to communicate with the device after that
> which triggers the following errors in the logs:
> 
> [00:05.0] xen_pt_bar_read: Error: Should not read BAR through QEMU. @0x0000000000000200
> [00:05.0] xen_pt_bar_write: Error: Should not write BAR through QEMU. @0x0000000000000200
> 
> The issue that the original patch tried to workaround (uneven number of
> region_add/del calls on device attach/detach) was fixed in later QEMU
> versions.

Do you know when the issue was fixed?

-- 
Anthony PERARD

On 17/04/18 15:15, Anthony PERARD wrote:
> On Fri, Apr 06, 2018 at 10:21:23PM +0100, Igor Druzhinin wrote:
>> Commit 99605175c (xen-pt: Fix PCI devices re-attach failed) introduced
>> a subtle bug. As soon as the guest switches off Bus Mastering on the
>> device it immediately causes all the BARs be unmapped due to the DMA
>> address space of the device being changed. This is undesired behavior
>> because the guest may try to communicate with the device after that
>> which triggers the following errors in the logs:
>>
>> [00:05.0] xen_pt_bar_read: Error: Should not read BAR through QEMU. @0x0000000000000200
>> [00:05.0] xen_pt_bar_write: Error: Should not write BAR through QEMU. @0x0000000000000200
>>
>> The issue that the original patch tried to workaround (uneven number of
>> region_add/del calls on device attach/detach) was fixed in later QEMU
>> versions.
> 
> Do you know when the issue was fixed?
> 

I haven't tracked down a particular version but the previous behavior of
memory_listener_unregister() was to remove the listener from the list
without calling the callback. It has changed since then and now the
callback is called in listener_del_address_space().

Igor

On 17/04/18 15:15, Anthony PERARD wrote:
> On Fri, Apr 06, 2018 at 10:21:23PM +0100, Igor Druzhinin wrote:
>> Commit 99605175c (xen-pt: Fix PCI devices re-attach failed) introduced
>> a subtle bug. As soon as the guest switches off Bus Mastering on the
>> device it immediately causes all the BARs be unmapped due to the DMA
>> address space of the device being changed. This is undesired behavior
>> because the guest may try to communicate with the device after that
>> which triggers the following errors in the logs:
>>
>> [00:05.0] xen_pt_bar_read: Error: Should not read BAR through QEMU. @0x0000000000000200
>> [00:05.0] xen_pt_bar_write: Error: Should not write BAR through QEMU. @0x0000000000000200
>>
>> The issue that the original patch tried to workaround (uneven number of
>> region_add/del calls on device attach/detach) was fixed in later QEMU
>> versions.
> 
> Do you know when the issue was fixed?
> 

I think it's this commit:

commit d25836cafd7508090d211e97acfc0abc5ae88daa
Author: Peter Xu <peterx@redhat.com>
Date:   Mon Jan 22 14:02:44 2018 +0800

    memory: do explicit cleanup when remove listeners

Igor

On Tue, Apr 17, 2018 at 03:18:55PM +0100, Igor Druzhinin wrote:
> On 17/04/18 15:15, Anthony PERARD wrote:
> > On Fri, Apr 06, 2018 at 10:21:23PM +0100, Igor Druzhinin wrote:
> >> The issue that the original patch tried to workaround (uneven number of
> >> region_add/del calls on device attach/detach) was fixed in later QEMU
> >> versions.
> > 
> > Do you know when the issue was fixed?
> > 
> 
> I haven't tracked down a particular version but the previous behavior of
> memory_listener_unregister() was to remove the listener from the list
> without calling the callback. It has changed since then and now the
> callback is called in listener_del_address_space().

On Tue, Apr 17, 2018 at 03:29:42PM +0100, Igor Druzhinin wrote:
> I think it's this commit:
> 
> commit d25836cafd7508090d211e97acfc0abc5ae88daa
> Author: Peter Xu <peterx@redhat.com>
> Date:   Mon Jan 22 14:02:44 2018 +0800
> 
>     memory: do explicit cleanup when remove listeners


I think these information ought to be in the commit message, in
particular the fact that the callback wasn't call on detach. And with
the commit message updated, you can add my:
Acked-by: Anthony PERARD <anthony.perard@citrix.com>

Thanks,

-- 
Anthony PERARD