According to the TODO item,
> * the flow control implementation in firmware is buggy (sends a flood of pause
> frames, resets the link, device and driver buffer queues become
> desynchronized), disable it by default
Currently, qlge_mpi_port_cfg_work calls qlge_mb_get_port_cfg which gets
the link config from the firmware and saves it to qdev->link_config. By
default, flow control is enabled. This commit writes the
save the pause parameter of qdev->link_config and don't let it
overwritten by link settings of current port. Since qdev->link_config=0
when qdev is initialized, this could disable flow control by default and
the pause parameter value could also survive MPI resetting,
$ ethtool -a enp94s0f0
Pause parameters for enp94s0f0:
Autonegotiate: off
RX: off
TX: off
The follow control can be enabled manually,
$ ethtool -A enp94s0f0 rx on tx on
$ ethtool -a enp94s0f0
Pause parameters for enp94s0f0:
Autonegotiate: off
RX: on
TX: on
Signed-off-by: Coiby Xu <[email protected]>
---
drivers/staging/qlge/TODO | 3 ---
drivers/staging/qlge/qlge_mpi.c | 11 ++++++++++-
2 files changed, 10 insertions(+), 4 deletions(-)
diff --git a/drivers/staging/qlge/TODO b/drivers/staging/qlge/TODO
index b7a60425fcd2..8c84160b5993 100644
--- a/drivers/staging/qlge/TODO
+++ b/drivers/staging/qlge/TODO
@@ -4,9 +4,6 @@
ql_build_rx_skb(). That function is now used exclusively to handle packets
that underwent header splitting but it still contains code to handle non
split cases.
-* the flow control implementation in firmware is buggy (sends a flood of pause
- frames, resets the link, device and driver buffer queues become
- desynchronized), disable it by default
* some structures are initialized redundantly (ex. memset 0 after
alloc_etherdev())
* the driver has a habit of using runtime checks where compile time checks are
diff --git a/drivers/staging/qlge/qlge_mpi.c b/drivers/staging/qlge/qlge_mpi.c
index 2630ebf50341..0f1c7da80413 100644
--- a/drivers/staging/qlge/qlge_mpi.c
+++ b/drivers/staging/qlge/qlge_mpi.c
@@ -806,6 +806,7 @@ int qlge_mb_get_port_cfg(struct qlge_adapter *qdev)
{
struct mbox_params mbc;
struct mbox_params *mbcp = &mbc;
+ u32 saved_pause_link_config = 0;
int status = 0;
memset(mbcp, 0, sizeof(struct mbox_params));
@@ -826,7 +827,15 @@ int qlge_mb_get_port_cfg(struct qlge_adapter *qdev)
} else {
netif_printk(qdev, drv, KERN_DEBUG, qdev->ndev,
"Passed Get Port Configuration.\n");
- qdev->link_config = mbcp->mbox_out[1];
+ /*
+ * Don't let the pause parameter be overwritten by
+ *
+ * In this way, follow control can be disabled by default
+ * and the setting could also survive the MPI reset
+ */
+ saved_pause_link_config = qdev->link_config & CFG_PAUSE_STD;
+ qdev->link_config = ~CFG_PAUSE_STD & mbcp->mbox_out[1];
+ qdev->link_config |= saved_pause_link_config;
qdev->max_frame_size = mbcp->mbox_out[2];
}
return status;
--
2.32.0
On 2021-06-21 21:48 +0800, Coiby Xu wrote:
> According to the TODO item,
> > * the flow control implementation in firmware is buggy (sends a flood of pause
> > frames, resets the link, device and driver buffer queues become
> > desynchronized), disable it by default
>
> Currently, qlge_mpi_port_cfg_work calls qlge_mb_get_port_cfg which gets
> the link config from the firmware and saves it to qdev->link_config. By
> default, flow control is enabled. This commit writes the
> save the pause parameter of qdev->link_config and don't let it
> overwritten by link settings of current port. Since qdev->link_config=0
> when qdev is initialized, this could disable flow control by default and
> the pause parameter value could also survive MPI resetting,
> $ ethtool -a enp94s0f0
> Pause parameters for enp94s0f0:
> Autonegotiate: off
> RX: off
> TX: off
>
> The follow control can be enabled manually,
>
> $ ethtool -A enp94s0f0 rx on tx on
> $ ethtool -a enp94s0f0
> Pause parameters for enp94s0f0:
> Autonegotiate: off
> RX: on
> TX: on
>
> Signed-off-by: Coiby Xu <[email protected]>
> ---
> drivers/staging/qlge/TODO | 3 ---
> drivers/staging/qlge/qlge_mpi.c | 11 ++++++++++-
> 2 files changed, 10 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/staging/qlge/TODO b/drivers/staging/qlge/TODO
> index b7a60425fcd2..8c84160b5993 100644
> --- a/drivers/staging/qlge/TODO
> +++ b/drivers/staging/qlge/TODO
> @@ -4,9 +4,6 @@
> ql_build_rx_skb(). That function is now used exclusively to handle packets
> that underwent header splitting but it still contains code to handle non
> split cases.
> -* the flow control implementation in firmware is buggy (sends a flood of pause
> - frames, resets the link, device and driver buffer queues become
> - desynchronized), disable it by default
> * some structures are initialized redundantly (ex. memset 0 after
> alloc_etherdev())
> * the driver has a habit of using runtime checks where compile time checks are
> diff --git a/drivers/staging/qlge/qlge_mpi.c b/drivers/staging/qlge/qlge_mpi.c
> index 2630ebf50341..0f1c7da80413 100644
> --- a/drivers/staging/qlge/qlge_mpi.c
> +++ b/drivers/staging/qlge/qlge_mpi.c
> @@ -806,6 +806,7 @@ int qlge_mb_get_port_cfg(struct qlge_adapter *qdev)
> {
> struct mbox_params mbc;
> struct mbox_params *mbcp = &mbc;
> + u32 saved_pause_link_config = 0;
Initialization is not needed given the code below, in fact the
declaration can be moved to the block below.
> int status = 0;
>
> memset(mbcp, 0, sizeof(struct mbox_params));
> @@ -826,7 +827,15 @@ int qlge_mb_get_port_cfg(struct qlge_adapter *qdev)
> } else {
> netif_printk(qdev, drv, KERN_DEBUG, qdev->ndev,
> "Passed Get Port Configuration.\n");
> - qdev->link_config = mbcp->mbox_out[1];
> + /*
> + * Don't let the pause parameter be overwritten by
> + *
> + * In this way, follow control can be disabled by default
> + * and the setting could also survive the MPI reset
> + */
It seems this comment is incomplete. Also, it's "flow control", not
"follow control".
> + saved_pause_link_config = qdev->link_config & CFG_PAUSE_STD;
> + qdev->link_config = ~CFG_PAUSE_STD & mbcp->mbox_out[1];
> + qdev->link_config |= saved_pause_link_config;
> qdev->max_frame_size = mbcp->mbox_out[2];
> }
> return status;
> --
> 2.32.0
>
On Tue, Jun 22, 2021 at 04:49:51PM +0900, Benjamin Poirier wrote:
>On 2021-06-21 21:48 +0800, Coiby Xu wrote:
>> According to the TODO item,
>> > * the flow control implementation in firmware is buggy (sends a flood of pause
>> > frames, resets the link, device and driver buffer queues become
>> > desynchronized), disable it by default
>>
>> Currently, qlge_mpi_port_cfg_work calls qlge_mb_get_port_cfg which gets
>> the link config from the firmware and saves it to qdev->link_config. By
>> default, flow control is enabled. This commit writes the
>> save the pause parameter of qdev->link_config and don't let it
>> overwritten by link settings of current port. Since qdev->link_config=0
>> when qdev is initialized, this could disable flow control by default and
>> the pause parameter value could also survive MPI resetting,
>> $ ethtool -a enp94s0f0
>> Pause parameters for enp94s0f0:
>> Autonegotiate: off
>> RX: off
>> TX: off
>>
>> The follow control can be enabled manually,
>>
>> $ ethtool -A enp94s0f0 rx on tx on
>> $ ethtool -a enp94s0f0
>> Pause parameters for enp94s0f0:
>> Autonegotiate: off
>> RX: on
>> TX: on
>>
>> Signed-off-by: Coiby Xu <[email protected]>
>> ---
>> drivers/staging/qlge/TODO | 3 ---
>> drivers/staging/qlge/qlge_mpi.c | 11 ++++++++++-
>> 2 files changed, 10 insertions(+), 4 deletions(-)
>>
>> diff --git a/drivers/staging/qlge/TODO b/drivers/staging/qlge/TODO
>> index b7a60425fcd2..8c84160b5993 100644
>> --- a/drivers/staging/qlge/TODO
>> +++ b/drivers/staging/qlge/TODO
>> @@ -4,9 +4,6 @@
>> ql_build_rx_skb(). That function is now used exclusively to handle packets
>> that underwent header splitting but it still contains code to handle non
>> split cases.
>> -* the flow control implementation in firmware is buggy (sends a flood of pause
>> - frames, resets the link, device and driver buffer queues become
>> - desynchronized), disable it by default
>> * some structures are initialized redundantly (ex. memset 0 after
>> alloc_etherdev())
>> * the driver has a habit of using runtime checks where compile time checks are
>> diff --git a/drivers/staging/qlge/qlge_mpi.c b/drivers/staging/qlge/qlge_mpi.c
>> index 2630ebf50341..0f1c7da80413 100644
>> --- a/drivers/staging/qlge/qlge_mpi.c
>> +++ b/drivers/staging/qlge/qlge_mpi.c
>> @@ -806,6 +806,7 @@ int qlge_mb_get_port_cfg(struct qlge_adapter *qdev)
>> {
>> struct mbox_params mbc;
>> struct mbox_params *mbcp = &mbc;
>> + u32 saved_pause_link_config = 0;
>
>Initialization is not needed given the code below,
Thanks for the spotting this issue!
> in fact the
>declaration can be moved to the block below.
I thought I need to put the declaration in the beginning of the
function. But it seems Linux kernel coding style doesn't require it.
I'll move it to the else block below then.
>
>> int status = 0;
>>
>> memset(mbcp, 0, sizeof(struct mbox_params));
>> @@ -826,7 +827,15 @@ int qlge_mb_get_port_cfg(struct qlge_adapter *qdev)
>> } else {
>> netif_printk(qdev, drv, KERN_DEBUG, qdev->ndev,
>> "Passed Get Port Configuration.\n");
>> - qdev->link_config = mbcp->mbox_out[1];
>> + /*
>> + * Don't let the pause parameter be overwritten by
>> + *
>> + * In this way, follow control can be disabled by default
>> + * and the setting could also survive the MPI reset
>> + */
>
>It seems this comment is incomplete. Also, it's "flow control", not
>"follow control".
Ah, yes. I should state it as "Don't let the pause parameter be
overwritten by be overwritten by the firmware.". And thanks for
correcting the typo.
>
>> + saved_pause_link_config = qdev->link_config & CFG_PAUSE_STD;
>> + qdev->link_config = ~CFG_PAUSE_STD & mbcp->mbox_out[1];
>> + qdev->link_config |= saved_pause_link_config;
>> qdev->max_frame_size = mbcp->mbox_out[2];
>> }
>> return status;
>> --
>> 2.32.0
>>
--
Best regards,
Coiby