On 2022/11/25 22:15, Jason Gunthorpe wrote:
On Fri, Nov 25, 2022 at 10:06:13PM +0800, Yi Liu wrote:
On 2022/11/25 20:37, Jason Gunthorpe wrote:
On Fri, Nov 25, 2022 at 04:57:27PM +0800, Yi Liu wrote:
+static int vfio_device_group_open(struct vfio_device *device)
+{
+ int ret;
+
+ mutex_lock(&device->group->group_lock);
now the group path holds group_lock first, and then device_set->lock.
this is different with existing code. is it acceptable? I had a quick
check with this change, basic test is good. no a-b-b-a locking issue.
I looked for a while and couldn't find a reason why it wouldn't be OK
ok, so updated this commit as below:
Yeah, can you update the cdev series too, lets see how it looks
yes, it's wip, not finished yet. I'll update tomorrow.
--
Regards,
Yi Liu
