Hi Stephe,
The spinlock of rxq_reset_lock, and txq_reset than I need it to make sure
one time just one user can call
queue_stop operate a queue reset, hardware only support reset queue head to
zero only one user operate.
When queue_stop 0 and queue_stop 1 operate at the same time, the hardware
queue reset will be fault.
Even port0 and port1 of one pf, operate queue reset will be also fault.
For rx_burst and tx_burst that I don't used the spinlock
Regards Wenbo
> -----Original Message-----
> From: Stephen Hemminger <step...@networkplumber.org>
> Sent: 2025年2月12日 0:02
> To: Wenbo Cao <caowe...@mucse.com>
> Cc: tho...@monjalon.net; dev@dpdk.org; ferruh.yi...@amd.com;
> andrew.rybche...@oktetlabs.ru; yao...@mucse.com
> Subject: Re: [PATCH v9 08/28] net/rnp: add queue setup and release
operations
>
> On Tue, 11 Feb 2025 16:49:58 +0800
> Wenbo Cao <caowe...@mucse.com> wrote:
>
> > diff --git a/drivers/net/rnp/base/rnp_hw.h
> > b/drivers/net/rnp/base/rnp_hw.h index da2c1d3b0e..ea4e1b8d83 100644
> > --- a/drivers/net/rnp/base/rnp_hw.h
> > +++ b/drivers/net/rnp/base/rnp_hw.h
> > @@ -120,6 +120,9 @@ struct rnp_hw {
> > bool lane_is_sgmii[RNP_MAX_PORT_OF_PF];
> > struct rnp_mbx_info mbx;
> > struct rnp_fw_info fw_info;
> > +
> > + spinlock_t rxq_reset_lock; /* reset op isn't thread safe */
> > + spinlock_t txq_reset_lock; /* reset op isn't thread safe */
> > };
> >
>
> In DPDK control operations do not need to be thread safe.
> Many other drivers do not allow start/stop in one thread while tx/rx are
> happening etc.
> This is because rx/tx should not have overhead of any locks.
> The cost of one spinlock at 10G will drop performance a lot
