Re: [PATCH v2 4/4] net: macb: Add support for suspend/resume with full power down

[Harini Katakam <harinik@xilinx.com>]

Hi Claudiu,

On Mon, Nov 26, 2018 at 8:16 PM <Claudiu.Beznea@microchip.com> wrote:
>
>
>
> On 26.11.2018 09:07, Harini Katakam wrote:
<snip>
>
> In the previous version you said you encountered some crashes while
> stressing this part if macb_open()/macb_close() was used in here. Could you
> share the tests so that I can debug it on my side?

Sure, my tests include doing a series of suspend/resume -- it is done
as part of a
random regression script. On ZynqMP, it includes the following to do an
FPD off suspend and then use either UART or WOL as a wake source and
repeat the same, sometimes pinging in between:

echo pm_request_wakeup 8 1 0 1 > /sys/kernel/debug/zynqmp-firmware/pm
echo pm_force_powerdown 6 > /sys/kernel/debug/zynqmp-firmware/pm
echo enabled > /sys/devices/platform/amba/ff000000.serial/tty/ttyPS0/power/wakeup
echo enabled > /sys/devices/platform/amba/ff0e0000.ethernet/net/eth0/power/wakeup
echo mem > /sys/power/state

ping does not work sometimes and after a few iterations (sometimes
20), a crash can be
observed tracing from "kfree_skb_list" to "macb_free_consistent". From
the flow, the
only explanation I could come up with was that there was an attempt to
free buffers that
may not have been allocated. Also, this was time consuming in the
suspend/resume path.

<snip>
> > +     } else {
> > +             macb_writel(bp, NCR, MACB_BIT(MPE));
>
> Just asking... shouldn't other registers be restored here after SoC power
> is cut off?
Ideally yes. But in this series, I'm just attempting to re-initialize
to a working
state. I can try context save/restore later.

Regards,
Harini