All of lore.kernel.org
 help / color / mirror / Atom feed
* [Qemu-devel] [PATCH RFC] pci: call PCIDeviceClass->exit on its .realize failure
@ 2016-09-14  9:14 Cao jin
  2016-09-14 11:59 ` Marcel Apfelbaum
  0 siblings, 1 reply; 3+ messages in thread
From: Cao jin @ 2016-09-14  9:14 UTC (permalink / raw)
  To: qemu-devel; +Cc: Michael S. Tsirkin, Marcel Apfelbaum

It cannot guarantee all pci devices will free the allocated resource in
its .realize function on realize failure.

CC: Michael S. Tsirkin <mst@redhat.com>
CC: Marcel Apfelbaum <marcel@redhat.com>
Signed-off-by: Cao jin <caoj.fnst@cn.fujitsu.com>
---

I found not all the devices will free the allocated resources on .realize
failure, and .exit function is the one who take responsibility to free all
the resource. In theory, I think it should be PCIDeviceClass->exit who does
the cleanup on realize failure, with appropriate check whether certain
resources is allocated.
It passed make check, but maybe need more confirmation, so, RFC.

 hw/pci/pci.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/hw/pci/pci.c b/hw/pci/pci.c
index 24fae16..4b63a79 100644
--- a/hw/pci/pci.c
+++ b/hw/pci/pci.c
@@ -1967,6 +1967,9 @@ static void pci_qdev_realize(DeviceState *qdev, Error **errp)
         if (local_err) {
             error_propagate(errp, local_err);
             do_pci_unregister_device(pci_dev);
+            if (pc->exit) {
+                pc->exit(pci_dev);
+            }
             return;
         }
     }
-- 
2.1.0

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [Qemu-devel] [PATCH RFC] pci: call PCIDeviceClass->exit on its .realize failure
  2016-09-14  9:14 [Qemu-devel] [PATCH RFC] pci: call PCIDeviceClass->exit on its .realize failure Cao jin
@ 2016-09-14 11:59 ` Marcel Apfelbaum
  2016-09-19  6:16   ` Cao jin
  0 siblings, 1 reply; 3+ messages in thread
From: Marcel Apfelbaum @ 2016-09-14 11:59 UTC (permalink / raw)
  To: Cao jin, qemu-devel; +Cc: Michael S. Tsirkin

On 09/14/2016 12:14 PM, Cao jin wrote:
> It cannot guarantee all pci devices will free the allocated resource in
> its .realize function on realize failure.
>
> CC: Michael S. Tsirkin <mst@redhat.com>
> CC: Marcel Apfelbaum <marcel@redhat.com>
> Signed-off-by: Cao jin <caoj.fnst@cn.fujitsu.com>
> ---
>
> I found not all the devices will free the allocated resources on .realize
> failure, and .exit function is the one who take responsibility to free all
> the resource. In theory, I think it should be PCIDeviceClass->exit who does
> the cleanup on realize failure, with appropriate check whether certain
> resources is allocated.
> It passed make check, but maybe need more confirmation, so, RFC.
>
>  hw/pci/pci.c | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/hw/pci/pci.c b/hw/pci/pci.c
> index 24fae16..4b63a79 100644
> --- a/hw/pci/pci.c
> +++ b/hw/pci/pci.c
> @@ -1967,6 +1967,9 @@ static void pci_qdev_realize(DeviceState *qdev, Error **errp)
>          if (local_err) {
>              error_propagate(errp, local_err);
>              do_pci_unregister_device(pci_dev);
> +            if (pc->exit) {
> +                pc->exit(pci_dev);
> +            }

Hi,

I think the call to pc->exit is not necessary here.

The call to pc->realize failed, so it is the callee (pc)
responsibility to clean the resources.

Please see in the same method:

      pci_add_option_rom(pci_dev, is_default_rom, &local_err);
         if (local_err) {
             error_propagate(errp, local_err);
             pci_qdev_unrealize(DEVICE(pci_dev), NULL);

              ^^^^^^^^^^^^^^^^^^^

This time we call pci_qdev_unrealize (that will call pc->exit)
because pc->realize succeeded and it is now the caller responsibility
to clean the resources.


Have you found a specific scenario that causes problems?

Thanks,
Marcel

>              return;
>          }
>      }
>

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [Qemu-devel] [PATCH RFC] pci: call PCIDeviceClass->exit on its .realize failure
  2016-09-14 11:59 ` Marcel Apfelbaum
@ 2016-09-19  6:16   ` Cao jin
  0 siblings, 0 replies; 3+ messages in thread
From: Cao jin @ 2016-09-19  6:16 UTC (permalink / raw)
  To: Marcel Apfelbaum, qemu-devel; +Cc: Michael S. Tsirkin

Hi,
     sorry for replying late(was in vacation).

On 09/14/2016 07:59 PM, Marcel Apfelbaum wrote:
> On 09/14/2016 12:14 PM, Cao jin wrote:
>> It cannot guarantee all pci devices will free the allocated resource in
>> its .realize function on realize failure.
>>
>> CC: Michael S. Tsirkin <mst@redhat.com>
>> CC: Marcel Apfelbaum <marcel@redhat.com>
>> Signed-off-by: Cao jin <caoj.fnst@cn.fujitsu.com>
>> ---
>>
>> I found not all the devices will free the allocated resources on .realize
>> failure, and .exit function is the one who take responsibility to free
>> all
>> the resource. In theory, I think it should be PCIDeviceClass->exit who
>> does
>> the cleanup on realize failure, with appropriate check whether certain
>> resources is allocated.
>> It passed make check, but maybe need more confirmation, so, RFC.
>>

>
> Hi,
>
> I think the call to pc->exit is not necessary here.
>
> The call to pc->realize failed, so it is the callee (pc)
> responsibility to clean the resources.
>
> Please see in the same method:
>
>       pci_add_option_rom(pci_dev, is_default_rom, &local_err);
>          if (local_err) {
>              error_propagate(errp, local_err);
>              pci_qdev_unrealize(DEVICE(pci_dev), NULL);
>
>               ^^^^^^^^^^^^^^^^^^^
>
> This time we call pci_qdev_unrealize (that will call pc->exit)
> because pc->realize succeeded and it is now the caller responsibility
> to clean the resources.
>
>
> Have you found a specific scenario that causes problems?

No, it came to my mind when doing the patch "convert to error". Look 
usb_xhci_realize(), it will call usb_xhci_init to do some resource 
allocation, but won't free the resource once return because of 
msi_init's failure(that may be my fault), so the following patch[*] move 
it down.

I agree your view above. We can forget this patch:)

[*]http://lists.nongnu.org/archive/html/qemu-devel/2016-09/msg03063.html

-- 
Yours Sincerely,

Cao jin

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2016-09-19  6:16 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-09-14  9:14 [Qemu-devel] [PATCH RFC] pci: call PCIDeviceClass->exit on its .realize failure Cao jin
2016-09-14 11:59 ` Marcel Apfelbaum
2016-09-19  6:16   ` Cao jin

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.