Re: [PATCH v7 0/5] clk: add driver for the SiFive FU740

[Zong Li]

On Thu, May 20, 2021 at 2:17 AM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
>
> Hi Zong,
>
> On Wed, May 19, 2021 at 5:55 PM Zong Li <zong.li@xxxxxxxxxx> wrote:
> > On Tue, May 11, 2021 at 4:57 PM Yixun Lan <yixun.lan@xxxxxxxxx> wrote:
> > > On Wed, Apr 14, 2021 at 2:25 PM Zong Li <zong.li@xxxxxxxxxx> wrote:
> > > > On Mon, Apr 12, 2021 at 7:31 PM Andreas Schwab <schwab@xxxxxxxxxxxxxx> wrote:
> > > > > On Mär 31 2021, Zong Li wrote:
> > > > > > I found that the gemgxlpll was disabled immediately by power
> > > > > > management after macb driver install. The mainline's defconfig doesn't
> > > > > > enable CONFIG_PM, so the network is fine on it. The opensuse defconfig
> > > > > > enables CONFIG_PM, and the patch
> > > > > > 732374a0b440d9a79c8412f318a25cd37ba6f4e2 added the enable/disable
> > > > > > callback functions, so the gemgxlpll PLL, I have no idea why power
> > > > > > management disable it, I would keep trace it.
> > > > >
> > > > > Does that mean that CONFIG_PM also affects the FU740?
> > > >
> > > > Yes, we got the same problem on the FU740. We are checking the issue.
> > > >
> > > Just a mild ping, any progress regarding this issue?
> >
> > Currently, if runtime power management is enabled, macb driver would
> > go to sleep at the end of macb_probe, then the gigabit ethernet PLL
> > would be disabled.  During this period of time, the system would hang
> > up if we try to access GEMGXL control registers, it means that we
> > can't access GEMGXL control registers before the gigabit ethernet PLL
> > is resumed again. There are some cases, for example, if we execute the
>
> Sounds familiar.
>
> > 'ifconfig' command, it would eventually go to the macb_get_status to
>
> Do you mean mac_get_stats()? macb_get_status() does not exist.

Sorry for the typo, it should be macb_get_stats.

>
> > access GEMGXL control registers and cause the system to hang up. Give
> > more example here, if we execute 'ip link set lo up & ip addr add
> > 127.0.0.1/8 dev lo', it would cause the system to hang up, because
> > these commands would try to query the interfaces and eventually go to
> > macb_get_status as well. However, if we can resume the gigabit
> > ethernet PLL first, such as 'ip link set eth0 up' or 'udhcpc', then
> > everything goes well. I'm trying to figure out if there are some hooks
> > that we can check the PLL status in the macb driver before it actually
> > touches the control registers. If anyone has an idea about that,
> > please feel free to point it out to me, thanks.
>
> And you cannot call pm_runtime_get_sync(), as this is called from
> atomic contect. Other drivers avoid accessing the registers while
> the device is not up, cfr. e.g. commit 7fa2955ff70ce453 ("sh_eth:
> Fix sleeping function called from invalid context").
>

Thanks for your help. I have done the similar modification by
following the patch you provided. I also verified the bug that we use
pm_runtime_get_sync there. I'm going to post the fix patch by adding
the is_opened flag.

Thanks again.

> Gr{oetje,eeting}s,
>
>                         Geert
>
> --
> Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
>
> In personal conversations with technical people, I call myself a hacker. But
> when I'm talking to journalists I just say "programmer" or something like that.
>                                 -- Linus Torvalds