netdev - Re: [PATCH 1/2] r8169: Implement dynamic ASPM mechanism

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:   Wed, 4 Aug 2021 12:05:10 +0800
From:   Kai-Heng Feng <kai.heng.feng@...onical.com>
To:     Heiner Kallweit <hkallweit1@...il.com>
Cc:     nic_swsd <nic_swsd@...ltek.com>,
        "David S. Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        "open list:8169 10/100/1000 GIGABIT ETHERNET DRIVER" 
        <netdev@...r.kernel.org>, open list <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH 1/2] r8169: Implement dynamic ASPM mechanism

On Wed, Aug 4, 2021 at 3:57 AM Heiner Kallweit <hkallweit1@...il.com> wrote:
>
> On 03.08.2021 17:28, Kai-Heng Feng wrote:
> > r8169 NICs on some platforms have abysmal speed when ASPM is enabled.
> > Same issue can be observed with older vendor drivers.
> >
> > The issue is however solved by the latest vendor driver. There's a new
>
> Is there any errata document from Realtek recommending this workaround?
> Any prove that it solves the issues in all cases of ASPM issues we've
> seen so far?

Actually I don't know. Let me ask Realtek.

> Also your heuristics logic seems to be different from the one in r8168.
> The vendor driver considers also rx packets.

rx packets are accumulated in rtl_rx().

>
> In addition you use this logic also for chip versions not covered by
> r8168, like RTL8125. Any info from Realtek regarding these chip versions?

Right, maybe 8125 doesn't need dynamic ASPM. Let me ask them...

>
> > mechanism, which disables r8169's internal ASPM when the NIC has
> > substantial network traffic, and vice versa.
> >
> 10 packets per second I wouldn't call substantial traffic.

I'll change the wording in v2.

> I'm afraid we may open a can of worms and may be bothered
> with bug reports and complaints again.

Let's hope this time it works.

>
> > So implement the same mechanism here to resolve the issue.
> >
> For me this risk is too high to re-enable ASPM for a lot of chip
> versions w/o any official errata and workaround information.
> I propose you make this change downstream, and if there are no
> user complaints after some months I may consider to have something
> like that in the mainline driver.

Sure, let's see how it works in downstream kernel first.

>
> > Signed-off-by: Kai-Heng Feng <kai.heng.feng@...onical.com>
> > ---
> >  drivers/net/ethernet/realtek/r8169_main.c | 36 +++++++++++++++++++++++
> >  1 file changed, 36 insertions(+)
> >
> > diff --git a/drivers/net/ethernet/realtek/r8169_main.c b/drivers/net/ethernet/realtek/r8169_main.c
> > index c7af5bc3b8af..e257d3cd885e 100644
> > --- a/drivers/net/ethernet/realtek/r8169_main.c
> > +++ b/drivers/net/ethernet/realtek/r8169_main.c
> > @@ -624,6 +624,10 @@ struct rtl8169_private {
> >
> >       unsigned supports_gmii:1;
> >       unsigned aspm_manageable:1;
> > +     unsigned aspm_enabled:1;
> > +     struct timer_list aspm_timer;
> > +     u32 aspm_packet_count;
> > +
> >       dma_addr_t counters_phys_addr;
> >       struct rtl8169_counters *counters;
> >       struct rtl8169_tc_offsets tc_offset;
> > @@ -2671,6 +2675,8 @@ static void rtl_hw_aspm_clkreq_enable(struct rtl8169_private *tp, bool enable)
> >               RTL_W8(tp, Config5, RTL_R8(tp, Config5) & ~ASPM_en);
> >       }
> >
> > +     tp->aspm_enabled = enable;
> > +
> >       udelay(10);
> >  }
> >
> > @@ -4408,6 +4414,7 @@ static void rtl_tx(struct net_device *dev, struct rtl8169_private *tp,
> >
> >       dirty_tx = tp->dirty_tx;
> >
> > +     tp->aspm_packet_count += tp->cur_tx - dirty_tx;
> >       while (READ_ONCE(tp->cur_tx) != dirty_tx) {
> >               unsigned int entry = dirty_tx % NUM_TX_DESC;
> >               u32 status;
> > @@ -4552,6 +4559,8 @@ static int rtl_rx(struct net_device *dev, struct rtl8169_private *tp, int budget
> >               rtl8169_mark_to_asic(desc);
> >       }
> >
> > +     tp->aspm_packet_count += count;
> > +
> >       return count;
> >  }
> >
> > @@ -4659,8 +4668,31 @@ static int r8169_phy_connect(struct rtl8169_private *tp)
> >       return 0;
> >  }
> >
> > +#define ASPM_PACKET_THRESHOLD 10
> > +#define ASPM_TIMER_INTERVAL 1000
> > +
> > +static void rtl8169_aspm_timer(struct timer_list *timer)
> > +{
> > +     struct rtl8169_private *tp = from_timer(tp, timer, aspm_timer);
> > +     bool enable;
> > +
> > +     enable = tp->aspm_packet_count <= ASPM_PACKET_THRESHOLD;
> > +
> > +     if (tp->aspm_enabled != enable) {
> > +             rtl_unlock_config_regs(tp);
> > +             rtl_hw_aspm_clkreq_enable(tp, enable);
> > +             rtl_lock_config_regs(tp);
>
> All this in interrupt context w/o locking?

Sorry, I forgot the timer is in interrupt context.
Or is it safe to use workqueue for rtl_{,un}lock_config_regs() and
rtl_hw_aspm_clkreq_enable()?

Kai-Heng

>
> > +     }
> > +
> > +     tp->aspm_packet_count = 0;
> > +
> > +     mod_timer(timer, jiffies + msecs_to_jiffies(ASPM_TIMER_INTERVAL));
> > +}
> > +
> >  static void rtl8169_down(struct rtl8169_private *tp)
> >  {
> > +     del_timer_sync(&tp->aspm_timer);
> > +
> >       /* Clear all task flags */
> >       bitmap_zero(tp->wk.flags, RTL_FLAG_MAX);
> >
> > @@ -4687,6 +4719,10 @@ static void rtl8169_up(struct rtl8169_private *tp)
> >       rtl_reset_work(tp);
> >
> >       phy_start(tp->phydev);
> > +
> > +     timer_setup(&tp->aspm_timer, rtl8169_aspm_timer, 0);
> > +     mod_timer(&tp->aspm_timer,
> > +               jiffies + msecs_to_jiffies(ASPM_TIMER_INTERVAL));
> >  }
> >
> >  static int rtl8169_close(struct net_device *dev)
> >
>