K. Y.,
here are 3 additional patches for your "[PATCH 0/5] Drivers: hv: vmbus: Cleanup
the vmbus unload path" series (with the fix I suggested). I tested the whole
set on Gen2 Win2012R2 guest, load/unload path seems being stable. Can you
please take a look?
Thanks,
Vitaly Kuznetsov (3):
Drivers: hv: vmbus: do cleanup on all vmbus_open() failure paths
Drivers: hv: vmbus: kill tasklets on module unload
Drivers: hv: vmbus: setup irq after synic is initialized
drivers/hv/channel.c | 7 ++++---
drivers/hv/vmbus_drv.c | 10 +++++++---
2 files changed, 11 insertions(+), 6 deletions(-)
--
1.9.3
In case there was an error reported in the response to the CHANNELMSG_OPENCHANNEL
call we need to do the cleanup as a vmbus_open() user won't be doing it after
receiving an error. The cleanup should be done on all failure paths.
Signed-off-by: Vitaly Kuznetsov <[email protected]>
---
drivers/hv/channel.c | 7 ++++---
1 file changed, 4 insertions(+), 3 deletions(-)
diff --git a/drivers/hv/channel.c b/drivers/hv/channel.c
index 54da66d..836386f 100644
--- a/drivers/hv/channel.c
+++ b/drivers/hv/channel.c
@@ -186,11 +186,12 @@ int vmbus_open(struct vmbus_channel *newchannel, u32 send_ringbuffer_size,
list_del(&open_info->msglistentry);
spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
- if (err == 0)
- newchannel->state = CHANNEL_OPENED_STATE;
+ if (err != 0)
+ goto error_gpadl;
+ newchannel->state = CHANNEL_OPENED_STATE;
kfree(open_info);
- return err;
+ return 0;
error1:
spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
--
1.9.3
Explicitly kill tasklets we create on module unload.
Signed-off-by: Vitaly Kuznetsov <[email protected]>
---
drivers/hv/vmbus_drv.c | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)
diff --git a/drivers/hv/vmbus_drv.c b/drivers/hv/vmbus_drv.c
index 2b56260..cf20400 100644
--- a/drivers/hv/vmbus_drv.c
+++ b/drivers/hv/vmbus_drv.c
@@ -1108,6 +1108,7 @@ static void __exit vmbus_exit(void)
hv_synic_clockevents_cleanup();
vmbus_disconnect();
hv_remove_vmbus_irq();
+ tasklet_kill(&msg_dpc);
vmbus_free_channels();
if (ms_hyperv.features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
atomic_notifier_chain_unregister(&panic_notifier_list,
@@ -1115,8 +1116,10 @@ static void __exit vmbus_exit(void)
}
bus_unregister(&hv_bus);
hv_cleanup();
- for_each_online_cpu(cpu)
+ for_each_online_cpu(cpu) {
+ tasklet_kill(hv_context.event_dpc[cpu]);
smp_call_function_single(cpu, hv_synic_cleanup, NULL, 1);
+ }
acpi_bus_unregister_driver(&vmbus_acpi_driver);
hv_cpu_hotplug_quirk(false);
}
--
1.9.3
vmbus_isr() uses synic pages which are being setup in hv_synic_alloc(), we
need to register this irq handler only after we allocate all the required
pages.
Signed-off-by: Vitaly Kuznetsov <[email protected]>
---
drivers/hv/vmbus_drv.c | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/drivers/hv/vmbus_drv.c b/drivers/hv/vmbus_drv.c
index cf20400..e922804 100644
--- a/drivers/hv/vmbus_drv.c
+++ b/drivers/hv/vmbus_drv.c
@@ -822,11 +822,12 @@ static int vmbus_bus_init(int irq)
if (ret)
goto err_cleanup;
- hv_setup_vmbus_irq(vmbus_isr);
-
ret = hv_synic_alloc();
if (ret)
goto err_alloc;
+
+ hv_setup_vmbus_irq(vmbus_isr);
+
/*
* Initialize the per-cpu interrupt state and
* connect to the host.
--
1.9.3
> -----Original Message-----
> From: Vitaly Kuznetsov [mailto:[email protected]]
> Sent: Tuesday, April 21, 2015 16:18
> To: KY Srinivasan
> Cc: Haiyang Zhang; [email protected]; linux-
> [email protected]; Dexuan Cui
> Subject: [PATCH 1/3] Drivers: hv: vmbus: do cleanup on all vmbus_open()
> failure paths
>
> In case there was an error reported in the response to the
> CHANNELMSG_OPENCHANNEL
> call we need to do the cleanup as a vmbus_open() user won't be doing it
> after
> receiving an error. The cleanup should be done on all failure paths.
>
> Signed-off-by: Vitaly Kuznetsov <[email protected]>
> ---
> drivers/hv/channel.c | 7 ++++---
> 1 file changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/hv/channel.c b/drivers/hv/channel.c
> index 54da66d..836386f 100644
> --- a/drivers/hv/channel.c
> +++ b/drivers/hv/channel.c
> @@ -186,11 +186,12 @@ int vmbus_open(struct vmbus_channel
> *newchannel, u32 send_ringbuffer_size,
> list_del(&open_info->msglistentry);
> spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock,
> flags);
>
> - if (err == 0)
> - newchannel->state = CHANNEL_OPENED_STATE;
> + if (err != 0)
> + goto error_gpadl;
>
> + newchannel->state = CHANNEL_OPENED_STATE;
> kfree(open_info);
> - return err;
> + return 0;
>
> error1:
> spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
> --
Thanks for the patch, Vitaly!
The patch looks good to me except for 1 small issue:
I suppose open_info->response.open_result.status is different from
Linux-style -EXXX error codes.
Maybe here we can return -EAGAIN if err != 0 ?
-- Dexuan
On Tue, Apr 21, 2015 at 10:17:52AM +0200, Vitaly Kuznetsov wrote:
> diff --git a/drivers/hv/channel.c b/drivers/hv/channel.c
> index 54da66d..836386f 100644
> --- a/drivers/hv/channel.c
> +++ b/drivers/hv/channel.c
> @@ -186,11 +186,12 @@ int vmbus_open(struct vmbus_channel *newchannel, u32 send_ringbuffer_size,
> list_del(&open_info->msglistentry);
> spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
>
> - if (err == 0)
> - newchannel->state = CHANNEL_OPENED_STATE;
> + if (err != 0)
Doesn't the double negative not make it slightly more confusing?
regards,
dan carpenter
Dexuan Cui <[email protected]> writes:
>> -----Original Message-----
>> From: Vitaly Kuznetsov [mailto:[email protected]]
>> Sent: Tuesday, April 21, 2015 16:18
>> To: KY Srinivasan
>> Cc: Haiyang Zhang; [email protected]; linux-
>> [email protected]; Dexuan Cui
>> Subject: [PATCH 1/3] Drivers: hv: vmbus: do cleanup on all vmbus_open()
>> failure paths
>>
>> In case there was an error reported in the response to the
>> CHANNELMSG_OPENCHANNEL
>> call we need to do the cleanup as a vmbus_open() user won't be doing it
>> after
>> receiving an error. The cleanup should be done on all failure paths.
>>
>> Signed-off-by: Vitaly Kuznetsov <[email protected]>
>> ---
>> drivers/hv/channel.c | 7 ++++---
>> 1 file changed, 4 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/hv/channel.c b/drivers/hv/channel.c
>> index 54da66d..836386f 100644
>> --- a/drivers/hv/channel.c
>> +++ b/drivers/hv/channel.c
>> @@ -186,11 +186,12 @@ int vmbus_open(struct vmbus_channel
>> *newchannel, u32 send_ringbuffer_size,
>> list_del(&open_info->msglistentry);
>> spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock,
>> flags);
>>
>> - if (err == 0)
>> - newchannel->state = CHANNEL_OPENED_STATE;
>> + if (err != 0)
>> + goto error_gpadl;
>>
>> + newchannel->state = CHANNEL_OPENED_STATE;
>> kfree(open_info);
>> - return err;
>> + return 0;
>>
>> error1:
>> spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
>> --
>
> Thanks for the patch, Vitaly!
>
> The patch looks good to me except for 1 small issue:
> I suppose open_info->response.open_result.status is different from
> Linux-style -EXXX error codes.
>
> Maybe here we can return -EAGAIN if err != 0 ?
I think I inspected all vmbus_open() callers and they all just check ret
!= 0 but I agree we'd better return some -EXXXX code here. I'm not
exactly sure if open_result.status != 0 usually means a permanent or a
temporary failure but if it's temporary -EAGAIN here seems reasonable.
I'll send v2, thanks for the review!
>
> -- Dexuan
--
Vitaly
Dan Carpenter <[email protected]> writes:
> On Tue, Apr 21, 2015 at 10:17:52AM +0200, Vitaly Kuznetsov wrote:
>> diff --git a/drivers/hv/channel.c b/drivers/hv/channel.c
>> index 54da66d..836386f 100644
>> --- a/drivers/hv/channel.c
>> +++ b/drivers/hv/channel.c
>> @@ -186,11 +186,12 @@ int vmbus_open(struct vmbus_channel *newchannel, u32 send_ringbuffer_size,
>> list_del(&open_info->msglistentry);
>> spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
>>
>> - if (err == 0)
>> - newchannel->state = CHANNEL_OPENED_STATE;
>> + if (err != 0)
>
> Doesn't the double negative not make it slightly more confusing?
>
Point taken, will fix in v2 :-)
> regards,
> dan carpenter
--
Vitaly