i40evf_execute_vf_cmd() uses _atomic_set_cmd() to execute virtual channel commands safely in multi-process mode. However, it returns -1 when one process is pending. Add a spinlock to wait for the virtual channel will handle this issue in concurrent scenarios. Fixes: 4861cde46116 ("i40e: new poll mode driver") Cc: stable@dpdk.org Signed-off-by: Yuying Zhang <yuying.zhang@intel.com> --- drivers/net/i40e/i40e_ethdev.h | 1 + drivers/net/i40e/i40e_ethdev_vf.c | 8 +++++++- 2 files changed, 8 insertions(+), 1 deletion(-) diff --git a/drivers/net/i40e/i40e_ethdev.h b/drivers/net/i40e/i40e_ethdev.h index 19f821829..514c0988b 100644 --- a/drivers/net/i40e/i40e_ethdev.h +++ b/drivers/net/i40e/i40e_ethdev.h @@ -1199,6 +1199,7 @@ struct i40e_vf { uint16_t max_pkt_len; /* Maximum packet length */ bool promisc_unicast_enabled; bool promisc_multicast_enabled; + rte_spinlock_t cmd_send_lock; uint32_t version_major; /* Major version number */ uint32_t version_minor; /* Minor version number */ diff --git a/drivers/net/i40e/i40e_ethdev_vf.c b/drivers/net/i40e/i40e_ethdev_vf.c index 69cab8e73..7fdc58649 100644 --- a/drivers/net/i40e/i40e_ethdev_vf.c +++ b/drivers/net/i40e/i40e_ethdev_vf.c @@ -326,8 +326,11 @@ i40evf_execute_vf_cmd(struct rte_eth_dev *dev, struct vf_cmd_info *args) enum i40evf_aq_result ret; int err, i = 0; - if (_atomic_set_cmd(vf, args->ops)) + rte_spinlock_lock(&vf->cmd_send_lock); + if (_atomic_set_cmd(vf, args->ops)) { + rte_spinlock_unlock(&vf->cmd_send_lock); return -1; + } info.msg = args->out_buffer; info.buf_len = args->out_size; @@ -339,6 +342,7 @@ i40evf_execute_vf_cmd(struct rte_eth_dev *dev, struct vf_cmd_info *args) if (err) { PMD_DRV_LOG(ERR, "fail to send cmd %d", args->ops); _clear_cmd(vf); + rte_spinlock_unlock(&vf->cmd_send_lock); return err; } @@ -406,6 +410,7 @@ i40evf_execute_vf_cmd(struct rte_eth_dev *dev, struct vf_cmd_info *args) break; } + rte_spinlock_unlock(&vf->cmd_send_lock); return err | vf->cmd_retval; } @@ -1249,6 +1254,7 @@ i40evf_init_vf(struct rte_eth_dev *dev) vf->adapter = I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private); vf->dev_data = dev->data; + rte_spinlock_init(&vf->cmd_send_lock); err = i40e_set_mac_type(hw); if (err) { PMD_INIT_LOG(ERR, "set_mac_type failed: %d", err); -- 2.25.1
i40evf_execute_vf_cmd() uses _atomic_set_cmd() to execute virtual channel commands safely in multi-process mode and multi-thread mode. However, it returns -1 when one process or thread is pending. Add rte_spinlock_trylock() to handle this issue in concurrent scenarios. Fixes: 4861cde46116 ("i40e: new poll mode driver") Cc: stable@dpdk.org Signed-off-by: Yuying Zhang <yuying.zhang@intel.com> --- drivers/net/i40e/i40e_ethdev.h | 1 + drivers/net/i40e/i40e_ethdev_vf.c | 16 +++++++++++++++- 2 files changed, 16 insertions(+), 1 deletion(-) diff --git a/drivers/net/i40e/i40e_ethdev.h b/drivers/net/i40e/i40e_ethdev.h index 1466998aa..508a940dc 100644 --- a/drivers/net/i40e/i40e_ethdev.h +++ b/drivers/net/i40e/i40e_ethdev.h @@ -1209,6 +1209,7 @@ struct i40e_vf { bool promisc_unicast_enabled; bool promisc_multicast_enabled; + rte_spinlock_t cmd_send_lock; uint32_t version_major; /* Major version number */ uint32_t version_minor; /* Minor version number */ uint16_t promisc_flags; /* Promiscuous setting */ diff --git a/drivers/net/i40e/i40e_ethdev_vf.c b/drivers/net/i40e/i40e_ethdev_vf.c index 4d6510d1f..282db4721 100644 --- a/drivers/net/i40e/i40e_ethdev_vf.c +++ b/drivers/net/i40e/i40e_ethdev_vf.c @@ -314,7 +314,7 @@ _atomic_set_cmd(struct i40e_vf *vf, enum virtchnl_ops ops) #define ASQ_DELAY_MS 10 static int -i40evf_execute_vf_cmd(struct rte_eth_dev *dev, struct vf_cmd_info *args) +_i40evf_execute_vf_cmd(struct rte_eth_dev *dev, struct vf_cmd_info *args) { struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private); struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private); @@ -405,6 +405,19 @@ i40evf_execute_vf_cmd(struct rte_eth_dev *dev, struct vf_cmd_info *args) return err | vf->cmd_retval; } +static int +i40evf_execute_vf_cmd(struct rte_eth_dev *dev, struct vf_cmd_info *args) +{ + struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private); + int err; + + while (!rte_spinlock_trylock(&vf->cmd_send_lock)) + rte_delay_us_sleep(50); + err = _i40evf_execute_vf_cmd(dev, args); + rte_spinlock_unlock(&vf->cmd_send_lock); + return err; +} + /* * Check API version with sync wait until version read or fail from admin queue */ @@ -1246,6 +1259,7 @@ i40evf_init_vf(struct rte_eth_dev *dev) vf->adapter = I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private); vf->dev_data = dev->data; + rte_spinlock_init(&vf->cmd_send_lock); err = i40e_set_mac_type(hw); if (err) { PMD_INIT_LOG(ERR, "set_mac_type failed: %d", err); -- 2.25.1
> -----Original Message-----
> From: Zhang, Yuying <yuying.zhang@intel.com>
> Sent: Monday, October 19, 2020 10:20 AM
> To: dev@dpdk.org; Zhang, Qi Z <qi.z.zhang@intel.com>; Xing, Beilei
> <beilei.xing@intel.com>
> Cc: Zhang, Yuying <yuying.zhang@intel.com>; stable@dpdk.org
> Subject: [PATCH v2] net/i40e: fix virtual channel confiliction issue
>
> i40evf_execute_vf_cmd() uses _atomic_set_cmd() to execute virtual channel
> commands safely in multi-process mode and multi-thread mode.
> However, it returns -1 when one process or thread is pending. Add
> rte_spinlock_trylock() to handle this issue in concurrent scenarios.
>
> Fixes: 4861cde46116 ("i40e: new poll mode driver")
> Cc: stable@dpdk.org
>
> Signed-off-by: Yuying Zhang <yuying.zhang@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
Applied to dpdk-next-net-intel.
Thanks
Qi
On 10/19/2020 3:20 AM, Yuying Zhang wrote: > i40evf_execute_vf_cmd() uses _atomic_set_cmd() to execute virtual > channel commands safely in multi-process mode and multi-thread mode. > However, it returns -1 when one process or thread is pending. Add > rte_spinlock_trylock() to handle this issue in concurrent scenarios. > Should '_atomic_set_cmd()' removed, since spinlock it added? > Fixes: 4861cde46116 ("i40e: new poll mode driver") > Cc: stable@dpdk.org > > Signed-off-by: Yuying Zhang <yuying.zhang@intel.com> <...>
> -----Original Message-----
> From: Ferruh Yigit <ferruh.yigit@intel.com>
> Sent: Friday, October 23, 2020 6:42 PM
> To: Zhang, Yuying <yuying.zhang@intel.com>; dev@dpdk.org; Zhang, Qi Z
> <qi.z.zhang@intel.com>; Xing, Beilei <beilei.xing@intel.com>
> Cc: stable@dpdk.org
> Subject: Re: [dpdk-dev] [PATCH v2] net/i40e: fix virtual channel confiliction issue
>
> On 10/19/2020 3:20 AM, Yuying Zhang wrote:
> > i40evf_execute_vf_cmd() uses _atomic_set_cmd() to execute virtual
> > channel commands safely in multi-process mode and multi-thread mode.
> > However, it returns -1 when one process or thread is pending. Add
> > rte_spinlock_trylock() to handle this issue in concurrent scenarios.
> >
>
> Should '_atomic_set_cmd()' removed, since spinlock it added?
>
> > Fixes: 4861cde46116 ("i40e: new poll mode driver")
> > Cc: stable@dpdk.org
> >
> > Signed-off-by: Yuying Zhang <yuying.zhang@intel.com>
>
> <...>
'_atomic_set_cmd()' is also used for sync with i40evf_handle_aq_msg to handle the request <-> response.
It is not just used for multi-task. I prefer to keep the atomic.
On 10/30/2020 5:32 AM, Zhang, Yuying wrote:
>
>
>> -----Original Message-----
>> From: Ferruh Yigit <ferruh.yigit@intel.com>
>> Sent: Friday, October 23, 2020 6:42 PM
>> To: Zhang, Yuying <yuying.zhang@intel.com>; dev@dpdk.org; Zhang, Qi Z
>> <qi.z.zhang@intel.com>; Xing, Beilei <beilei.xing@intel.com>
>> Cc: stable@dpdk.org
>> Subject: Re: [dpdk-dev] [PATCH v2] net/i40e: fix virtual channel confiliction issue
>>
>> On 10/19/2020 3:20 AM, Yuying Zhang wrote:
>>> i40evf_execute_vf_cmd() uses _atomic_set_cmd() to execute virtual
>>> channel commands safely in multi-process mode and multi-thread mode.
>>> However, it returns -1 when one process or thread is pending. Add
>>> rte_spinlock_trylock() to handle this issue in concurrent scenarios.
>>>
>>
>> Should '_atomic_set_cmd()' removed, since spinlock it added?
>>
>>> Fixes: 4861cde46116 ("i40e: new poll mode driver")
>>> Cc: stable@dpdk.org
>>>
>>> Signed-off-by: Yuying Zhang <yuying.zhang@intel.com>
>>
>> <...>
> '_atomic_set_cmd()' is also used for sync with i40evf_handle_aq_msg to handle the request <-> response.
> It is not just used for multi-task. I prefer to keep the atomic.
>
OK