[PATCH v3] net: stmmac: fix FPE events losing
Jianheng Zhang
Jianheng.Zhang at synopsys.com
Thu Nov 30 18:49:40 PST 2023
> -----Original Message-----
> From: Paolo Abeni <pabeni at redhat.com>
> Sent: Thursday, November 30, 2023 10:14 PM
> To: Serge Semin <fancer.lancer at gmail.com>
> Cc: Jianheng Zhang <jianheng at synopsys.com>; Alexandre Torgue <alexandre.torgue at foss.st.com>; Jose
> Abreu <joabreu at synopsys.com>; David S. Miller <davem at davemloft.net>; Eric Dumazet
> <edumazet at google.com>; Jakub Kicinski <kuba at kernel.org>; Maxime Coquelin
> <mcoquelin.stm32 at gmail.com>; Simon Horman <horms at kernel.org>; Andrew Halaney
> <ahalaney at redhat.com>; Bartosz Golaszewski <bartosz.golaszewski at linaro.org>; Shenwei Wang
> <shenwei.wang at nxp.com>; Johannes Zink <j.zink at pengutronix.de>; Russell King (Oracle
> <rmk+kernel at armlinux.org.uk>; Jochen Henneberg <jh at henneberg-systemdesign.com>; Voon Weifeng
> <weifeng.voon at intel.com>; Mohammad Athari Bin Ismail <mohammad.athari.ismail at intel.com>; Ong
> Boon Leong <boon.leong.ong at intel.com>; Tan Tee Min <tee.min.tan at intel.com>; open list:STMMAC
> ETHERNET DRIVER <netdev at vger.kernel.org>; moderated list:ARM/STM32 ARCHITECTURE
> <linux-stm32 at st-md-mailman.stormreply.com>; moderated list:ARM/STM32 ARCHITECTURE
> <linux-arm-kernel at lists.infradead.org>; open list <linux-kernel at vger.kernel.org>; James Li
> <lijames at synopsys.com>; Martin McKenny <mmckenny at synopsys.com>
> Subject: Re: [PATCH v3] net: stmmac: fix FPE events losing
>
> On Thu, 2023-11-30 at 15:09 +0100, Paolo Abeni wrote:
> > On Thu, 2023-11-30 at 16:09 +0300, Serge Semin wrote:
> > > Hi Paolo
> > >
> > > On Thu, Nov 30, 2023 at 10:55:34AM +0100, Paolo Abeni wrote:
> > > > On Tue, 2023-11-28 at 05:56 +0000, Jianheng Zhang wrote:
> > > > > The status bits of register MAC_FPE_CTRL_STS are clear on read. Using
> > > > > 32-bit read for MAC_FPE_CTRL_STS in dwmac5_fpe_configure() and
> > > > > dwmac5_fpe_send_mpacket() clear the status bits. Then the stmmac interrupt
> > > > > handler missing FPE event status and leads to FPE handshaking failure and
> > > > > retries.
> > > > > To avoid clear status bits of MAC_FPE_CTRL_STS in dwmac5_fpe_configure()
> > > > > and dwmac5_fpe_send_mpacket(), add fpe_csr to stmmac_fpe_cfg structure to
> > > > > cache the control bits of MAC_FPE_CTRL_STS and to avoid reading
> > > > > MAC_FPE_CTRL_STS in those methods.
> > > > >
> > > > > Fixes: 5a5586112b92 ("net: stmmac: support FPE link partner hand-shaking procedure")
> > > > > Reviewed-by: Serge Semin <fancer.lancer at gmail.com>
> > > > > Signed-off-by: Jianheng Zhang <jianheng at synopsys.com>
> > > > > ---
> > > > > drivers/net/ethernet/stmicro/stmmac/dwmac5.c | 45 +++++++++-------------
> > > > > drivers/net/ethernet/stmicro/stmmac/dwmac5.h | 4 +-
> > > > > .../net/ethernet/stmicro/stmmac/dwxgmac2_core.c | 3 +-
> > > > > drivers/net/ethernet/stmicro/stmmac/hwif.h | 4 +-
> > > > > drivers/net/ethernet/stmicro/stmmac/stmmac_main.c | 8 +++-
> > > > > drivers/net/ethernet/stmicro/stmmac/stmmac_tc.c | 1 +
> > > > > include/linux/stmmac.h | 1 +
> > > > > 7 files changed, 36 insertions(+), 30 deletions(-)
> > > > >
> > > > > diff --git a/drivers/net/ethernet/stmicro/stmmac/dwmac5.c
> b/drivers/net/ethernet/stmicro/stmmac/dwmac5.c
> > > > > index e95d35f..8fd1675 100644
> > > > > --- a/drivers/net/ethernet/stmicro/stmmac/dwmac5.c
> > > > > +++ b/drivers/net/ethernet/stmicro/stmmac/dwmac5.c
> > > > > @@ -710,28 +710,22 @@ void dwmac5_est_irq_status(void __iomem *ioaddr, struct net_device
> *dev,
> > > > > }
> > > > > }
> > > > >
> > > > > -void dwmac5_fpe_configure(void __iomem *ioaddr, u32 num_txq, u32 num_rxq,
> > > > > +void dwmac5_fpe_configure(void __iomem *ioaddr, struct stmmac_fpe_cfg *cfg,
> > > > > + u32 num_txq, u32 num_rxq,
> > > > > bool enable)
> > > > > {
> > > > > u32 value;
> > > > >
> > > > > - if (!enable) {
> > > > > - value = readl(ioaddr + MAC_FPE_CTRL_STS);
> > > > > -
> > > > > - value &= ~EFPE;
> > > > > -
> > > > > - writel(value, ioaddr + MAC_FPE_CTRL_STS);
> > > > > - return;
> > > > > + if (enable) {
> > > > > + cfg->fpe_csr = EFPE;
> > > > > + value = readl(ioaddr + GMAC_RXQ_CTRL1);
> > > > > + value &= ~GMAC_RXQCTRL_FPRQ;
> > > > > + value |= (num_rxq - 1) << GMAC_RXQCTRL_FPRQ_SHIFT;
> > > > > + writel(value, ioaddr + GMAC_RXQ_CTRL1);
> > > > > + } else {
> > > > > + cfg->fpe_csr = 0;
> > > > > }
> > > > > -
> > > > > - value = readl(ioaddr + GMAC_RXQ_CTRL1);
> > > > > - value &= ~GMAC_RXQCTRL_FPRQ;
> > > > > - value |= (num_rxq - 1) << GMAC_RXQCTRL_FPRQ_SHIFT;
> > > > > - writel(value, ioaddr + GMAC_RXQ_CTRL1);
> > > > > -
> > > > > - value = readl(ioaddr + MAC_FPE_CTRL_STS);
> > > > > - value |= EFPE;
> > > > > - writel(value, ioaddr + MAC_FPE_CTRL_STS);
> > > > > + writel(cfg->fpe_csr, ioaddr + MAC_FPE_CTRL_STS);
> > > > > }
> > > > >
> > > > > int dwmac5_fpe_irq_status(void __iomem *ioaddr, struct net_device *dev)
> > > > > @@ -741,6 +735,9 @@ int dwmac5_fpe_irq_status(void __iomem *ioaddr, struct net_device
> *dev)
> > > > >
> > > > > status = FPE_EVENT_UNKNOWN;
> > > > >
> > > > > + /* Reads from the MAC_FPE_CTRL_STS register should only be performed
> > > > > + * here, since the status flags of MAC_FPE_CTRL_STS are "clear on read"
> > > > > + */
> > > > > value = readl(ioaddr + MAC_FPE_CTRL_STS);
> > > > >
> > > > > if (value & TRSP) {
> > > > > @@ -766,19 +763,15 @@ int dwmac5_fpe_irq_status(void __iomem *ioaddr, struct net_device
> *dev)
> > > > > return status;
> > > > > }
> > > > >
> > > > > -void dwmac5_fpe_send_mpacket(void __iomem *ioaddr, enum stmmac_mpacket_type type)
> > > > > +void dwmac5_fpe_send_mpacket(void __iomem *ioaddr, struct stmmac_fpe_cfg *cfg,
> > > > > + enum stmmac_mpacket_type type)
> > > > > {
> > > > > - u32 value;
> > > > > + u32 value = cfg->fpe_csr;
> > > > >
> > > > > - value = readl(ioaddr + MAC_FPE_CTRL_STS);
> > > > > -
> > > > > - if (type == MPACKET_VERIFY) {
> > > > > - value &= ~SRSP;
> > > > > + if (type == MPACKET_VERIFY)
> > > > > value |= SVER;
> > > > > - } else {
> > > > > - value &= ~SVER;
> > > > > + else if (type == MPACKET_RESPONSE)
> > > > > value |= SRSP;
> > > > > - }
> > > > >
> > > > > writel(value, ioaddr + MAC_FPE_CTRL_STS);
> > > > > }
> > > >
> > >
> > > > It's unclear to me why it's not necessary to preserve the SVER/SRSP
> > > > bits across MAC_FPE_CTRL_STS writes. I guess they are not part of the
> > > > status bits? perhaps an explicit comment somewhere will help?
> > >
> > > The SRSP and SVER are self-cleared flags with no effect on zero
> > > writing. Their responsibility is to emit the Respond and Verify
> > > mPackets respectively. As soon as the packets are sent, the flags will
> > > be reset by hardware automatically. So no, they aren't a part of the
> > > status bits.
> > >
> > > Note since 'value' now isn't read from the MAC_FPE_CTRL_STS register,
> > > there is no point in clearing up these flags in the local variable
> > > because 'value' has now them cleared by default.
> > >
> > > Not sure whether a comment about that is required, since the described
> > > behavior is well documented in the Synopsys HW-manual.
Thanks for Serge’s explanation.
> >
> > Thanks for the explanation, it clarifies the things to me. I agree
> > there is no need for a patch change.
>
> I'm sorry, I have to take back the last sentence: the submitter and SoB
> email address still don't match. @Jianheng: please fix it for good at
> the next iteration.
OH, sorry~ I mixed up my internet email address with the internal one. I will fix it and resend the patch.
Best Regards,
Jianheng
>
> Cheers,
>
> Paolo
More information about the linux-arm-kernel
mailing list