Re: [PATCH net-next 2/2] net: stmmac: dwmac-motorcomm: support Wake-on-LAN from S5

From
Yao Zi <me@ziyao.cc>
Date
2026-11-16 17:27:47
Also in
linux-arm-kernel, lkml
Reply to
Magnus Leatti, 2026-11-16
On Mon, Nov 16, 2026 at 01:48:47PM -0300, Magnus Leatti wrote:
Hi Yao,

On Mon, Nov 16, 2026 at 03:17:00PM +0000, Yao Zi wrote:
quoted
quoted
Does the chip automatically enables WoL and loads MAC from eFuse on
power-on?
Yes. I pulled the power cable, kept the driver from loading and dumped
the registers before anything touched the chip: OOB WoL enabled, eFuse
address in filter 0, PMT armed for magic packet, receiver on. The
packet wake bit in WOL_CTRL is clear.
Nice. But with packet wake disabled by default, wouldn't the card fail
to wake the machine if it just recovers from a power failure?

This behavior is confusing to me, but yes, useful.
quoted
Aside from a clean power-on, its behavior on reset also matters.
Same state, about 2 s after SYS_RESET. Probe disables OOB WoL before
that, so a cold and a warm boot look the same to the driver. But the
chip arms with the eFuse address, not the one in use. The vendor
driver writes the address again after enabling OOB WoL, for the same
reason.
Makes sense.
quoted
we might probably call stmmac_dvr_remove() in .shutdown() to detach
the NIC from kernel's infrastructure, then reset it with SYS_RESET to
get a functional WoL state in S5.
YES! Bingo, and I built AND tested now :) .shutdown calls
stmmac_dvr_remove(), then with "wol g":

  reset only                          no wake, 1 of 1
  reset, wait, address                no wake, 1 of 1
  reset, wait, packet wake            wake on the eFuse address, 1 of 1
  reset, wait, address, packet wake   wake on the address in use, 3 of 3
  "wol d", remove only                no wake, 1 of 1
  reboot, remove only                 link back, 1 of 1

"wait" polls PMT until the chip has armed itself, 2.6 to 2.7 s, 7 of 7.
"packet wake" is the bit in WOL_CTRL, the reset clears it. PME was
enabled in every row but the first. It also wakes without PME and in
D3hot, 1 of 1 each. No warning in any shutdown log.

So v2 would be, glue only:

 - every shutdown: stmmac_dvr_remove()
 - power-off with WoL set: motorcomm_reset(), poll PMT, write the
   address in use, set the packet wake bit, pci_wake_from_d3(), D3hot
IMHO, this would be the right way to go, but I'd like to drop the part
of writing the MAC address, since AFAIK WoL from S5 is usually achieved
through the real hardware address by convention, and
Two questions:

 1) The glue writes MAC_ADDR0 directly after the remove, two writel().
    Is it reasonable?
 2) Power-off takes about 2.7 s longer with WoL set. I can't write the
    address earlier: in a test that left the registers returning
    garbage until the driver was bound again.
if we keep using the real hardware address, at least problem 1
disappears, and I do think it's a little cursed (but not much) to write
the core MAC in a glue driver. And at least now we don't need to write
the address on our own, so if packet WOL bit could be set before PMT
polling succeeds, the second problem goes away, too.
Maxime, this needs no helper in the core, please ignore my question.

So my plan for this week: I'll write a v2 that addresses everything
you've mentioned (once again, thanks) and send it your way. If
Then feel free to add a Suggested-by tag :)
something else comes up there will be a v3, and a v4, until WoL works
on every YT6801, NUC, laptop or add-in card. Deal?

Magnus
-- 
Thanks,
Yao Zi

Meow-mrrp-mrow, I'm a cat.
Thread (15 messages) 15 messages, 5 authors, last reply 1d ago
Keyboard shortcuts
hback out one level
jnext message in thread / next series-diff position
kprevious message in thread / previous series-diff position
ldrill in
Escclose help / fold thread tree
?toggle this help