On 8/25/2024 8:27 AM, Michael Kelley wrote:
From: Naman Jain <namjain@xxxxxxxxxxxxxxxxxxx> Sent: Thursday, August 22, 2024 4:09 AM
Rescind offer handling relies on rescind callbacks for some of the
resources cleanup, if they are registered. It does not unregister
vmbus device for the primary channel closure, when callback is
registered.
Add logic to unregister vmbus for the primary channel in rescind callback
to ensure channel removal and relid release, and to ensure rescind flag
is false when driver probe happens again.
Fixes: ca3cda6fcf1e ("uio_hv_generic: add rescind support")
Signed-off-by: Naman Jain <namjain@xxxxxxxxxxxxxxxxxxx>
---
drivers/hv/vmbus_drv.c | 1 +
drivers/uio/uio_hv_generic.c | 7 +++++++
2 files changed, 8 insertions(+)
diff --git a/drivers/hv/vmbus_drv.c b/drivers/hv/vmbus_drv.c
index c857dc3975be..4bae382a3eb4 100644
--- a/drivers/hv/vmbus_drv.c
+++ b/drivers/hv/vmbus_drv.c
@@ -1952,6 +1952,7 @@ void vmbus_device_unregister(struct hv_device
*device_obj)
*/
device_unregister(&device_obj->device);
}
+EXPORT_SYMBOL_GPL(vmbus_device_unregister);
#ifdef CONFIG_ACPI
/*
diff --git a/drivers/uio/uio_hv_generic.c b/drivers/uio/uio_hv_generic.c
index c99890c16d29..ea26c0b460d6 100644
--- a/drivers/uio/uio_hv_generic.c
+++ b/drivers/uio/uio_hv_generic.c
@@ -121,6 +121,13 @@ static void hv_uio_rescind(struct vmbus_channel *channel)
/* Wake up reader */
uio_event_notify(&pdata->info);
+
+ /*
+ * With rescind callback registered, rescind path will not unregister the device
+ * when the primary channel is rescinded. Without it, next onoffer msg does not come.
+ */
+ if (!channel->primary_channel)
+ vmbus_device_unregister(channel->device_obj);
When the rescind callback is *not* set, vmbus_onoffer_rescind() makes the
call to vmbus_device_unregister(). But it does so bracketed with get_device()/
put_device(). Your code here does not do the bracketing. Is there a reason for
the difference? Frankly, I'm not sure why vmbus_onoffer_rescind() does the
bracketing, and I can't definitively say if it is really needed. So I guess I'm
just asking if you know. :-)
Michael
}
/* ysfs API to allow mmap of the ring buffers
--
2.34.1
IMHO, we have already NULL checked channel->device_obj and other couple
of things to make sure we are safe to clean this up. At other places as
well, I don't see the use of put and get device. So I think its not
required. I am open to suggestions.
Regards,
Naman
