SDIO broken in 2.6.29?

SDIO broken in 2.6.29?

[Steve Sakoman]

Overo has a wifi chip on mmc2 that uses the libertas_sdio driver.
I've had my head down working in other areas and failed to noticed
that at some point in the 2.6.29 rc cycle SDIO stopped working.

I'll start bisecting tomorrow morning, but thought I would ask first
to see if anyone else has noticed issues in this area.

Steve

Re: SDIO broken in 2.6.29?

[Tony Lindgren]

* Steve Sakoman <sakoman@gmail.com> [090219 22:38]:
> Overo has a wifi chip on mmc2 that uses the libertas_sdio driver.
> I've had my head down working in other areas and failed to noticed
> that at some point in the 2.6.29 rc cycle SDIO stopped working.
> 
> I'll start bisecting tomorrow morning, but thought I would ask first
> to see if anyone else has noticed issues in this area.

I wonder if it happened when we switched to the mainline version of
omap_hsmmc.c?

Anyways at least now most of the code is in the mainline :) So
hopefully we can get that fix in too during the -rc cycle.

Tony

Re: SDIO broken in 2.6.29?

[David Brownell]

> 
> > I'll start bisecting tomorrow morning, but thought I would ask first
> > to see if anyone else has noticed issues in this area.

I had a couple kernels where it didn't find the device, but
the last builds I installed found it just fine... 

Re: SDIO broken in 2.6.29?

[Steve Sakoman]

On Fri, Feb 20, 2009 at 7:49 AM, Tony Lindgren <tony@atomide.com> wrote:
> * Steve Sakoman <sakoman@gmail.com> [090219 22:38]:
>> Overo has a wifi chip on mmc2 that uses the libertas_sdio driver.
>> I've had my head down working in other areas and failed to noticed
>> that at some point in the 2.6.29 rc cycle SDIO stopped working.
>>
>> I'll start bisecting tomorrow morning, but thought I would ask first
>> to see if anyone else has noticed issues in this area.
>
> I wonder if it happened when we switched to the mainline version of
> omap_hsmmc.c?
>
> Anyways at least now most of the code is in the mainline :) So
> hopefully we can get that fix in too during the -rc cycle.

I've tried to reproduce Dave's reported success with recent kernels,
but have been unable to do so.

I haven't narrowed it down to a specific commit yet, but have verified
that the problem surfaces for me at the rc2 -> rc3 transition.

This is suspiciously near the time of the switch to the mainline
version of omap_hsmmc.c . . .

I'll keep digging.

Steve

Re: SDIO broken in 2.6.29?

[Steve Sakoman]

On Fri, Feb 20, 2009 at 7:49 AM, Tony Lindgren <tony@atomide.com> wrote:
> * Steve Sakoman <sakoman@gmail.com> [090219 22:38]:
>> Overo has a wifi chip on mmc2 that uses the libertas_sdio driver.
>> I've had my head down working in other areas and failed to noticed
>> that at some point in the 2.6.29 rc cycle SDIO stopped working.
>>
>> I'll start bisecting tomorrow morning, but thought I would ask first
>> to see if anyone else has noticed issues in this area.
>
> I wonder if it happened when we switched to the mainline version of
> omap_hsmmc.c?
>
> Anyways at least now most of the code is in the mainline :) So
> hopefully we can get that fix in too during the -rc cycle.

A quick update . . .

I started adding debug printf's and discovered that the issue seems to
be related to the power supply negotiation code:

mmc_select_voltage: card ocr = 0x00000000
mmc_select_voltage: host ocr_avail = 0x000fe080
mmc1: host doesn't support card's voltages
mmc1: error -22 whilst initialising SDIO card

Seemed strange that the card was returning 0 for ocr, so I added a
printf at the point of the read, and suddenly the card was giving a
reasonable ocr value!

diff --git a/drivers/mmc/core/core.c b/drivers/mmc/core/core.c
index df6ce4a..ecde069 100644
--- a/drivers/mmc/core/core.c
+++ b/drivers/mmc/core/core.c

@@ -761,6 +764,7 @@ void mmc_rescan(struct work_struct *work)
                * First we search for SDIO...
                */
               err = mmc_send_io_op_cond(host, 0, &ocr);
+               pr_warning("mmc_rescan: card ocr from io_op=0x%08x,
err = %d\n", ocr, err);
               if (!err) {
                       if (mmc_attach_sdio(host, ocr))
                               mmc_power_off(host);

With that printf I now get:

mmc_rescan: card ocr from io_op=0x90ff8000, err = 0
<snip>
mmc1: new SDIO card at address 0001
<snip>
libertas_sdio: Libertas SDIO driver
libertas_sdio: Copyright Pierre Ossman
libertas_sdio mmc1:0001:1: firmware: requesting sd8686_helper.bin
<snip>
libertas_sdio mmc1:0001:1: firmware: requesting sd8686.bin
<snip>
libertas: 00:19:88:05:b5:23, fw 9.70.3p24, cap 0x00000303
eth0 (libertas_sdio): not using net_device_ops yet
<snip>
libertas: PREP_CMD: command 0x00a3 failed: 2
libertas: PREP_CMD: command 0x00a3 failed: 2
libertas: eth0: Marvell WLAN 802.11 adapter

Nice, a completely normal boot!  Not quite time to celebrate though --
4 reboots demonstrated a failure to load the libertas firmware 50% of
the time.

To collect more data I tried this code on 4 boards.  Two of the 4 act
as described above, the other 2 return 0 for ocr even with the printf.

I rolled back the code to 2.6.28 and again tested all 4 boards.
Result: 100% success with all boards.

These symptoms scream "timing" to me.

As far as I can tell this issue cropped up somewhere around the 2.6.29
rc2->rc3 transition.

Any ideas or words or wisdom?  I am far from an expert on the mmc/sdio code.

Steve