Libertas/SDIO deep sleep mode

Libertas/SDIO deep sleep mode

[Daniel Mack]

Hi,

I read about the recently added features in the libertas driver to send
the chip to its deep-sleep mode when the system suspends, so the wakeup
phase is shortened.

I tried this with a recent 2.6.35-rc1 kernel on a PXA300 based design,
and this is what I get:


[    0.721643] Freezing user space processes ... (elapsed 0.01 seconds) done.
[    0.740588] Freezing remaining freezable tasks ... (elapsed 0.01 seconds) done.
[    0.760687] Suspending console(s) (use no_console_suspend to debug)
[    0.879657] libertas: 00:19:88:3a:cb:ac, fw 9.70.3p36, cap 0x00000303
[    0.880567] ------------[ cut here ]------------
[    0.880622] WARNING: at drivers/base/power/main.c:101 device_pm_add+0x94/0xd8()
[    0.880639] Device: ieee80211
[    0.880646] Parentless device registered during a PM transaction (phy1)
[    0.880660] Modules linked in: eeti_ts libertas_sdio libertas pxamci ds2760_battery w1_ds2760 wire
[    0.880754] [<c00493a0>] (unwind_backtrace+0x0/0xec) from [<c0056db0>] (warn_slowpath_common+0x4c/0x7c)
[    0.880793] [<c0056db0>] (warn_slowpath_common+0x4c/0x7c) from [<c0056e74>] (warn_slowpath_fmt+0x30/0x40)
[    0.880829] [<c0056e74>] (warn_slowpath_fmt+0x30/0x40) from [<c01d4cc8>] (device_pm_add+0x94/0xd8)
[    0.880875] [<c01d4cc8>] (device_pm_add+0x94/0xd8) from [<c01ce4dc>] (device_add+0x350/0x510)
[    0.880910] [<c01ce4dc>] (device_add+0x350/0x510) from [<c032d500>] (wiphy_register+0x1ac/0x298)
[    0.881033] [<c032d500>] (wiphy_register+0x1ac/0x298) from [<bf0285dc>] (lbs_cfg_register+0x54/0x9c [libertas])
[    0.881142] [<bf0285dc>] (lbs_cfg_register+0x54/0x9c [libertas]) from [<bf02c228>] (lbs_start_card+0xa0/0x130 [libertas])
[    0.881226] [<bf02c228>] (lbs_start_card+0xa0/0x130 [libertas]) from [<bf040b94>] (if_sdio_probe+0x84c/0x954 [libertas_sdio])
[    0.881276] [<bf040b94>] (if_sdio_probe+0x84c/0x954 [libertas_sdio]) from [<c024fd98>] (sdio_bus_probe+0xcc/0xe8)
[    0.881322] [<c024fd98>] (sdio_bus_probe+0xcc/0xe8) from [<c01d069c>] (driver_probe_device+0xb4/0x198)
[    0.881359] [<c01d069c>] (driver_probe_device+0xb4/0x198) from [<c01cfc08>] (bus_for_each_drv+0x4c/0x8c)
[    0.881396] [<c01cfc08>] (bus_for_each_drv+0x4c/0x8c) from [<c01d08d4>] (device_attach+0x54/0x70)
[    0.881430] [<c01d08d4>] (device_attach+0x54/0x70) from [<c01cfa2c>] (bus_probe_device+0x28/0x50)
[    0.881464] [<c01cfa2c>] (bus_probe_device+0x28/0x50) from [<c01ce510>] (device_add+0x384/0x510)
[    0.881496] [<c01ce510>] (device_add+0x384/0x510) from [<c024fbd4>] (sdio_add_func+0x38/0x54)
[    0.881527] [<c024fbd4>] (sdio_add_func+0x38/0x54) from [<c024f28c>] (mmc_attach_sdio+0x21c/0x2a8)
[    0.881581] [<c024f28c>] (mmc_attach_sdio+0x21c/0x2a8) from [<c024b97c>] (mmc_rescan+0x1ec/0x28c)
[    0.881624] [<c024b97c>] (mmc_rescan+0x1ec/0x28c) from [<c0067a60>] (worker_thread+0x168/0x1f0)
[    0.881667] [<c0067a60>] (worker_thread+0x168/0x1f0) from [<c006abd8>] (kthread+0x78/0x80)
[    0.881720] [<c006abd8>] (kthread+0x78/0x80) from [<c00458bc>] (kernel_thread_exit+0x0/0x8)
[    0.881742] ---[ end trace 919a8d4b06013c07 ]---
[    0.884319] libertas: wlan0: Marvell WLAN 802.11 adapter
[    1.138756] mmc0: card 0001 removed
[    1.139448] PM: suspend of devices complete after 255.035 msecs
[    1.140076] PM: late suspend of devices complete after 0.598 msecs
[    0.000943] PM: early resume of devices complete after 0.634 msecs
[    0.060955] usb usb1: root hub lost power or was reset
[    0.136987] mmc0: new SDIO card at address 0001
[    0.137101] mmc mmc0:0001: parent mmc0 should not be sleeping
[    0.298199] PM: resume of devices complete after 297.118 msecs
[    0.571350] Restarting tasks ... 
[    0.768454] done.
[    1.179764] libertas: 00:19:88:3a:cb:ac, fw 9.70.3p36, cap 0x00000303
[    1.198354] libertas: wlan0: Marvell WLAN 802.11 adapter


Also, the 8686's supply voltage (which is controlled via a regulator in
the pxamci code) drops to zero during the suspend phase.

Here are some questions that I have about this:

* The "Parentless device registered during a PM transaction" looks like
  a bug to me. Note that I added kobject_name(&dev->kobj) to the output
  to get a clue about what's going on.

* The firmware survives the suspend/resume transition and works fine
  after that. I just wonder why, as the voltage was switched off?

* I believe that if we need to keep the power switched on, there should
  be a call to sdio_set_host_pm_flags(MMC_PM_KEEP_POWER) somewhere in
  the libertas_sdio layer, right? How should the pxamci core get to
  know it shouldn't switch off its regulator?

* Should the message about mmc0 which should not be sleeping go away
  once the other issues are solved?


Thanks,
Daniel

Re: Libertas/SDIO deep sleep mode

[Chris Ball]

Hi,

   > Hi, I read about the recently added features in the libertas
   > driver to send the chip to its deep-sleep mode when the system
   > suspends, so the wakeup phase is shortened.

I think there might be a bit of terminology confusion here -- as
I understand it, "deep sleep" refers to putting the 8686 into a
power-saving mode after a certain amount of idle time, while the
rest of the system is running.

If you want to preserve power to the 8686 over a system suspend,
that's a largely different set of code, which isn't complete in
mainline yet; as you've seen, the MMC and SDHCI side of it is
upstream, but not libertas or pxamci code yet.

   > * The "Parentless device registered during a PM transaction"
   > looks like a bug to me. Note that I added kobject_name(&dev->kobj)
   > to the output to get a clue about what's going on.

Not sure what's going on there.

   > * The firmware survives the suspend/resume transition and works
   > fine after that. I just wonder why, as the voltage was switched
   > off?

That's odd.  Perhaps the 8686 is being alternatively powered somehow
on your hardware?

   > * I believe that if we need to keep the power switched on, there
   > should be a call to sdio_set_host_pm_flags(MMC_PM_KEEP_POWER)
   > somewhere in the libertas_sdio layer, right? How should the
   > pxamci core get to know it shouldn't switch off its regulator?

That's right.  We're using this code at OLPC, to go into system
suspend with the 8686 powered and then be woken when a new packet
arrives.  The libertas side of that code hasn't been submitted yet,
but you're welcome to check it out at our git repo:

http://dev.laptop.org/git/olpc-2.6/commit/?h=olpc-2.6.31&id=db09aa1f537f7d5eae6e52ecf59cdf3571150ee6
http://dev.laptop.org/git/olpc-2.6/log/drivers/net/wireless/libertas?h=olpc-2.6.31

- Chris.
-- 
Chris Ball   <cjb@laptop.org>
One Laptop Per Child