[V2] ethdev: fix eth device released repeatedly
Checks
Commit Message
The rte_eth_dev_pci_generic_remove() will be called to detach an Ethernet
device when App calls rte_dev_remove() to detach a pci device. In addition,
the rte_eth_dev_close() can also detach an Ethernet device.
In secondary process, if App first calls rte_eth_dev_close() and then calls
rte_dev_remove(), because rte_eth_dev_close() doesn't clear "eth_dev->data"
, the address of the released Ethernet device can still be found by device
name. As a result, the Ethernet device will be released repeatedly in this
case. The state of the Ethernet device is equal to RTE_ETH_DEV_UNUSED after
calling rte_eth_dev_close(). Use this state to avoid this problem.
Signed-off-by: Huisong Li <lihuisong@huawei.com>
---
v1 -> v2:
* fix the commit log description.
RFC -> v1:
* fix commit log and add a judgment for secondary process.
---
lib/ethdev/ethdev_pci.h | 15 +++++++++++++++
1 file changed, 15 insertions(+)
Comments
12/10/2021 13:39, Huisong Li:
> The rte_eth_dev_pci_generic_remove() will be called to detach an Ethernet
> device when App calls rte_dev_remove() to detach a pci device. In addition,
> the rte_eth_dev_close() can also detach an Ethernet device.
> In secondary process, if App first calls rte_eth_dev_close() and then calls
> rte_dev_remove(), because rte_eth_dev_close() doesn't clear "eth_dev->data"
It would be clearer if you start this sentence with:
"In secondary process, rte_eth_dev_close() doesn't clear eth_dev->data."
Then you can explain that if calling rte_dev_remove() after rte_eth_dev_close(),
etc...
> , the address of the released Ethernet device can still be found by device
> name. As a result, the Ethernet device will be released repeatedly in this
> case. The state of the Ethernet device is equal to RTE_ETH_DEV_UNUSED after
> calling rte_eth_dev_close(). Use this state to avoid this problem.
>
> Signed-off-by: Huisong Li <lihuisong@huawei.com>
> ---
> + /*
> + * In secondary process, if applications first call rte_eth_dev_close()
> + * and then call this interface, because rte_eth_dev_close() doesn't
> + * clear eth_dev->data, the address of the released Ethernet device can
> + * still be found by device name. As a result, the Ethernet device will
> + * be released repeatedly in this case.
> + * The state of the Ethernet device is equal to RTE_ETH_DEV_UNUSED after
> + * calling rte_eth_dev_close(). Use this state to avoid this problem.
This is a comment for the commit log.
Inside the code, we should be more to the point.
I suggest this comment:
/* A released port can be found by its name in shared memory. */
> + */
> + if (rte_eal_process_type() != RTE_PROC_PRIMARY &&
Better to directly compare with RTE_PROC_SECONDARY
> + eth_dev->state == RTE_ETH_DEV_UNUSED) {
> + RTE_ETHDEV_LOG(INFO, "The ethdev port has been released.");
Not sure we need any log here.
> + return 0;
> + }
在 2021/10/12 23:33, Thomas Monjalon 写道:
> 12/10/2021 13:39, Huisong Li:
>> The rte_eth_dev_pci_generic_remove() will be called to detach an Ethernet
>> device when App calls rte_dev_remove() to detach a pci device. In addition,
>> the rte_eth_dev_close() can also detach an Ethernet device.
>> In secondary process, if App first calls rte_eth_dev_close() and then calls
>> rte_dev_remove(), because rte_eth_dev_close() doesn't clear "eth_dev->data"
> It would be clearer if you start this sentence with:
> "In secondary process, rte_eth_dev_close() doesn't clear eth_dev->data."
> Then you can explain that if calling rte_dev_remove() after rte_eth_dev_close(),
> etc...
Right. Thanks!😁
>> , the address of the released Ethernet device can still be found by device
>> name. As a result, the Ethernet device will be released repeatedly in this
>> case. The state of the Ethernet device is equal to RTE_ETH_DEV_UNUSED after
>> calling rte_eth_dev_close(). Use this state to avoid this problem.
>>
>> Signed-off-by: Huisong Li <lihuisong@huawei.com>
>> ---
>> + /*
>> + * In secondary process, if applications first call rte_eth_dev_close()
>> + * and then call this interface, because rte_eth_dev_close() doesn't
>> + * clear eth_dev->data, the address of the released Ethernet device can
>> + * still be found by device name. As a result, the Ethernet device will
>> + * be released repeatedly in this case.
>> + * The state of the Ethernet device is equal to RTE_ETH_DEV_UNUSED after
>> + * calling rte_eth_dev_close(). Use this state to avoid this problem.
> This is a comment for the commit log.
> Inside the code, we should be more to the point.
> I suggest this comment:
> /* A released port can be found by its name in shared memory. */
ack
>
>> + */
>> + if (rte_eal_process_type() != RTE_PROC_PRIMARY &&
> Better to directly compare with RTE_PROC_SECONDARY
ack
>
>> + eth_dev->state == RTE_ETH_DEV_UNUSED) {
>> + RTE_ETHDEV_LOG(INFO, "The ethdev port has been released.");
> Not sure we need any log here.
ack
>
>> + return 0;
>> + }
>
>
> .
Hi, Thomas
*The commit log:*
In secondary process, rte_eth_dev_close() doesn't clear eth_dev->data.
If calling rte_dev_remove() after rte_eth_dev_close(), in
rte_eth_dev_pci_generic_remove()
function, the released eth device still can be found by its name in
shared memory.
As a result, the eth device will be released repeatedly. The state of
the eth device
is modified to RTE_ETH_DEV_UNUSED after rte_eth_dev_close(). So this
state can
be used to avoid this problem.
Is that will be more clear?
/*
* A released eth device can be found by its name in shared memory.
* If the state of the eth device is RTE_ETH_DEV_UNUSED, which means
* the eth device has been released.
*/
Is it ok to use the above description as a comment in the code?
Hope for your reply. Thanks.
在 2021/10/12 23:33, Thomas Monjalon 写道:
> 12/10/2021 13:39, Huisong Li:
>> The rte_eth_dev_pci_generic_remove() will be called to detach an Ethernet
>> device when App calls rte_dev_remove() to detach a pci device. In addition,
>> the rte_eth_dev_close() can also detach an Ethernet device.
>> In secondary process, if App first calls rte_eth_dev_close() and then calls
>> rte_dev_remove(), because rte_eth_dev_close() doesn't clear "eth_dev->data"
> It would be clearer if you start this sentence with:
> "In secondary process, rte_eth_dev_close() doesn't clear eth_dev->data."
> Then you can explain that if calling rte_dev_remove() after rte_eth_dev_close(),
> etc...
>
>> , the address of the released Ethernet device can still be found by device
>> name. As a result, the Ethernet device will be released repeatedly in this
>> case. The state of the Ethernet device is equal to RTE_ETH_DEV_UNUSED after
>> calling rte_eth_dev_close(). Use this state to avoid this problem.
>>
>> Signed-off-by: Huisong Li<lihuisong@huawei.com>
>> ---
>> + /*
>> + * In secondary process, if applications first call rte_eth_dev_close()
>> + * and then call this interface, because rte_eth_dev_close() doesn't
>> + * clear eth_dev->data, the address of the released Ethernet device can
>> + * still be found by device name. As a result, the Ethernet device will
>> + * be released repeatedly in this case.
>> + * The state of the Ethernet device is equal to RTE_ETH_DEV_UNUSED after
>> + * calling rte_eth_dev_close(). Use this state to avoid this problem.
> This is a comment for the commit log.
> Inside the code, we should be more to the point.
> I suggest this comment:
> /* A released port can be found by its name in shared memory. */
>
>> + */
>> + if (rte_eal_process_type() != RTE_PROC_PRIMARY &&
> Better to directly compare with RTE_PROC_SECONDARY
>
>> + eth_dev->state == RTE_ETH_DEV_UNUSED) {
>> + RTE_ETHDEV_LOG(INFO, "The ethdev port has been released.");
> Not sure we need any log here.
>
>> + return 0;
>> + }
>
>
> .
14/10/2021 14:32, lihuisong (C):
> Hi, Thomas
>
> *The commit log:*
> In secondary process, rte_eth_dev_close() doesn't clear eth_dev->data.
> If calling rte_dev_remove() after rte_eth_dev_close(), in
> rte_eth_dev_pci_generic_remove()
> function, the released eth device still can be found by its name in
> shared memory.
> As a result, the eth device will be released repeatedly. The state of
> the eth device
> is modified to RTE_ETH_DEV_UNUSED after rte_eth_dev_close(). So this
> state can
> be used to avoid this problem.
>
> Is that will be more clear?
Yes, that's clear (at least for me).
> /*
> * A released eth device can be found by its name in shared memory.
> * If the state of the eth device is RTE_ETH_DEV_UNUSED, which means
> * the eth device has been released.
> */
>
> Is it ok to use the above description as a comment in the code?
Yes. One small change, I think "which" should be "it".
> Hope for your reply. Thanks.
Thanks
在 2021/10/14 20:50, Thomas Monjalon 写道:
> 14/10/2021 14:32, lihuisong (C):
>> Hi, Thomas
>>
>> *The commit log:*
>> In secondary process, rte_eth_dev_close() doesn't clear eth_dev->data.
>> If calling rte_dev_remove() after rte_eth_dev_close(), in
>> rte_eth_dev_pci_generic_remove()
>> function, the released eth device still can be found by its name in
>> shared memory.
>> As a result, the eth device will be released repeatedly. The state of
>> the eth device
>> is modified to RTE_ETH_DEV_UNUSED after rte_eth_dev_close(). So this
>> state can
>> be used to avoid this problem.
>>
>> Is that will be more clear?
> Yes, that's clear (at least for me).
>
>> /*
>> * A released eth device can be found by its name in shared memory.
>> * If the state of the eth device is RTE_ETH_DEV_UNUSED, which means
>> * the eth device has been released.
>> */
>>
>> Is it ok to use the above description as a comment in the code?
> Yes. One small change, I think "which" should be "it".
Thanks. I will fix it.
>
>> Hope for your reply. Thanks.
> Thanks
>
>
>
>
> .
@@ -151,6 +151,21 @@ rte_eth_dev_pci_generic_remove(struct rte_pci_device *pci_dev,
if (!eth_dev)
return 0;
+ /*
+ * In secondary process, if applications first call rte_eth_dev_close()
+ * and then call this interface, because rte_eth_dev_close() doesn't
+ * clear eth_dev->data, the address of the released Ethernet device can
+ * still be found by device name. As a result, the Ethernet device will
+ * be released repeatedly in this case.
+ * The state of the Ethernet device is equal to RTE_ETH_DEV_UNUSED after
+ * calling rte_eth_dev_close(). Use this state to avoid this problem.
+ */
+ if (rte_eal_process_type() != RTE_PROC_PRIMARY &&
+ eth_dev->state == RTE_ETH_DEV_UNUSED) {
+ RTE_ETHDEV_LOG(INFO, "The ethdev port has been released.");
+ return 0;
+ }
+
if (dev_uninit) {
ret = dev_uninit(eth_dev);
if (ret)