#9702 BLOC 1.5-har: Linux (OS42) crashes on resume on an XO-1.5 B3

[Zarro Boogs per Child]

#9702: Linux (OS42) crashes on resume on an XO-1.5 B3
--------------------------------+-------------------------------------------
           Reporter:  wad       |       Owner:  dsaxena                          
               Type:  defect    |      Status:  new                              
           Priority:  blocker   |   Milestone:  1.5-hardware-C                   
          Component:  kernel    |     Version:  Development build as of this date
         Resolution:            |    Keywords:                                   
        Next_action:  diagnose  |    Verified:  1                                
Deployment_affected:            |   Blockedby:                                   
           Blocking:            |  
--------------------------------+-------------------------------------------

Comment(by cjb):

 The mdelays posted work to get resume working in the presence of a powered
 SD card on a B3.

 They don't work, even though "echo 9 > /proc/sys/kernel/printk" does, on
 resume of the wifi card on a B3.  Resume on a wifi card on a B2 requires
 no mdelays, so this is odd.

 The error seen is:

 {{{
 [  192.205993] sdhci-pci 0000:00:0c.0: setting latency timer to 64
 [  192.206049] sdhci-pci 0000:00:0c.0: setting latency timer to 64
 [  192.206137] mmc1:0001: error -123 reading SDIO_CCCR_INTx
 [  192.210029] sdhci_set_power: new power value = 14
 [  192.256085] mmc_sdio_resume: init card failed error -123
 [  192.297558] mmc1: error -123 during resume (card was removed?)
 }}}

 It's odd that the attempt to read from CCCR comes *before* the set_power
 call.  The call is made by the pending irq thread, which seems to run
 independently of the mmc resume sequence.  So, perhaps we have a race
 where the kernel thread looks for pending IRQs before the card has been
 resumed properly?  But that's still odd, because we don't hit the same
 race on a B2 for some reason..

-- 
Ticket URL: <http://dev.laptop.org/ticket/9702#comment:21>
One Laptop Per Child <http://laptop.org/>
OLPC bug tracking system