[BUG] Kernel error when first driver unbind with empty MMC slot

[BUG] Kernel error when first driver unbind with empty MMC slot

[Petr Cvek]

During testing of these patches

	[PATCH] mmc: pxamci: fix card detect threaded interrupt
	[PATCH 1/3] dmaengine: virt-dma: don't always free descriptor upon completion

I have found unrelated error.	

How to reproduce:

1) Remove any SD card
2) No CPLD initial power for card (in magician.c, probably irrelevant)
3) Boot into an initrd filesystem
4) Run command echo -n "pxa2xx-mci.0" > /sys/class/mmc_host/mmc0/device/driver/unbind
5) Error message will be printed:

[ 97.877519] irq 39: nobody cared (try booting with the "irqpoll" option)
[ 97.884283] CPU: 0 PID: 153 Comm: sh Not tainted 4.2.0-next-20150910-00015-g03de31f-dirty #16
[ 97.892757] Hardware name: HTC Magician
[ 97.896726] [<c000edac>] (unwind_backtrace) from [<c000cb0c>] (show_stack+0x10/0x14)
[ 97.904503] [<c000cb0c>] (show_stack) from [<c00450ac>] (__report_bad_irq+0x24/0xb4)
[ 97.912221] [<c00450ac>] (__report_bad_irq) from [<c00453f4>] (note_interrupt+0x250/0x2b4)
[ 97.920452] [<c00453f4>] (note_interrupt) from [<c0043450>] (handle_irq_event_percpu+0xa8/0x164)
[ 97.929195] [<c0043450>] (handle_irq_event_percpu) from [<c0043538>] (handle_irq_event+0x2c/0x40)
[ 97.938027] [<c0043538>] (handle_irq_event) from [<c0045d0c>] (handle_level_irq+0x94/0x118)
[ 97.946342] [<c0045d0c>] (handle_level_irq) from [<c0042dc4>] (generic_handle_irq+0x20/0x30)
[ 97.954739] [<c0042dc4>] (generic_handle_irq) from [<c0042ebc>] (__handle_domain_irq+0x4c/0xa4)
[ 97.963395] [<c0042ebc>] (__handle_domain_irq) from [<c0009414>] (ichp_handle_irq+0x28/0x38)
[ 97.971794] [<c0009414>] (ichp_handle_irq) from [<c000d5b0>] (__irq_svc+0x50/0x64)
[ 97.979321] Exception stack(0xc23f1de0 to 0xc23f1e28)
[ 97.984366] 1de0: c34030e0 00000027 00000000 00000000 c3400220 fffffffa c34002d4 ffffffff
[ 97.992506] 1e00: 000000d0 c15536c0 000000ec 0116f2dc 00000001 c23f1e34 c0043c48 c01a0504
[ 98.000633] 1e20: 20000013 ffffffff
[ 98.004159] [<c000d5b0>] (__irq_svc) from [<c01a0504>] (__radix_tree_lookup+0x60/0xcc)
[ 98.012043] [<c01a0504>] (__radix_tree_lookup) from [<c0043c48>] (free_irq+0x10/0x88)
[ 98.019895] [<c0043c48>] (free_irq) from [<c024e350>] (pxamci_remove+0xa4/0x154)
[ 98.027285] [<c024e350>] (pxamci_remove) from [<c0202940>] (platform_drv_remove+0x18/0x30)
[ 98.035578] [<c0202940>] (platform_drv_remove) from [<c0201574>] (__device_release_driver+0x84/0x104)
[ 98.044769] [<c0201574>] (__device_release_driver) from [<c0201610>] (device_release_driver+0x1c/0x28)
[ 98.054034] [<c0201610>] (device_release_driver) from [<c01ffad4>] (unbind_store+0x58/0x90)
[ 98.062346] [<c01ffad4>] (unbind_store) from [<c01ff358>] (drv_attr_store+0x20/0x2c)
[ 98.070103] [<c01ff358>] (drv_attr_store) from [<c00f14c4>] (sysfs_kf_write+0x44/0x48)
[ 98.077998] [<c00f14c4>] (sysfs_kf_write) from [<c00f0b30>] (kernfs_fop_write+0xec/0x1ac)
[ 98.086171] [<c00f0b30>] (kernfs_fop_write) from [<c009b7b8>] (__vfs_write+0x2c/0xd8)
[ 98.093980] [<c009b7b8>] (__vfs_write) from [<c009c660>] (vfs_write+0x94/0x150)
[ 98.101260] [<c009c660>] (vfs_write) from [<c009c8ec>] (SyS_write+0x40/0x8c)
[ 98.108284] [<c009c8ec>] (SyS_write) from [<c000a460>] (ret_fast_syscall+0x0/0x38)
[ 98.115800] handlers:
[ 98.118073] [<c024f19c>] pxamci_irq
[ 98.121559] Disabling IRQ #39

Error exists in versions before the dmaengine patches was created. Specifically:

	1efdb5f0a9243ca8f3460a5ce1b407b06a021f02

Applying the DMA engine patches (list at the beginning of this mail) does not make any change to this bug behavior.

Petr Cvek

Re: [BUG] Kernel error when first driver unbind with empty MMC slot

[Robert Jarzmik]

Petr Cvek <petr.cvek@tul.cz> writes:

> During testing of these patches
>
> 	[PATCH] mmc: pxamci: fix card detect threaded interrupt
> 	[PATCH 1/3] dmaengine: virt-dma: don't always free descriptor upon completion
>
> I have found unrelated error.	
>
> How to reproduce:
>
> 1) Remove any SD card
> 2) No CPLD initial power for card (in magician.c, probably irrelevant)
> 3) Boot into an initrd filesystem
> 4) Run command echo -n "pxa2xx-mci.0" > /sys/class/mmc_host/mmc0/device/driver/unbind
> 5) Error message will be printed:
>
> [ 97.877519] irq 39: nobody cared (try booting with the "irqpoll" option)

To go forward, I need to either :
 - see the debug messages activated, especially the one pxamci_irq()
 - or apply this patch [1] to see if it fixes the issue :

Cheers.

--
Robert

[1] Patch, totally untested/not compiled
---8<---
diff --git a/drivers/mmc/host/pxamci.c b/drivers/mmc/host/pxamci.c
index 67c9d1443597..3e0a7dd8da84 100644
--- a/drivers/mmc/host/pxamci.c
+++ b/drivers/mmc/host/pxamci.c
@@ -890,9 +890,7 @@ static int pxamci_remove(struct platform_device *pdev)
                        host->pdata->exit(&pdev->dev, mmc);
 
                pxamci_stop_clock(host);
-               writel(TXFIFO_WR_REQ|RXFIFO_RD_REQ|CLK_IS_OFF|STOP_CMD|
-                      END_CMD_RES|PRG_DONE|DATA_TRAN_DONE,
-                      host->base + MMC_I_MASK);
+               pxamci_disable_irq(host,  MMC_I_MASK_ALL);
 
                free_irq(host->irq, host);
                dmaengine_terminate_all(host->dma_chan_rx);

Re: [BUG] Kernel error when first driver unbind with empty MMC slot

[Petr Cvek]

Dne 13.9.2015 v 11:56 Robert Jarzmik napsal(a):
> Petr Cvek <petr.cvek@tul.cz> writes:
> 
>> During testing of these patches
>>
>> 	[PATCH] mmc: pxamci: fix card detect threaded interrupt
>> 	[PATCH 1/3] dmaengine: virt-dma: don't always free descriptor upon completion
>>
>> I have found unrelated error.	
>>
>> How to reproduce:
>>
>> 1) Remove any SD card
>> 2) No CPLD initial power for card (in magician.c, probably irrelevant)
>> 3) Boot into an initrd filesystem
>> 4) Run command echo -n "pxa2xx-mci.0" > /sys/class/mmc_host/mmc0/device/driver/unbind
>> 5) Error message will be printed:
>>
>> [ 97.877519] irq 39: nobody cared (try booting with the "irqpoll" option)
> 
> To go forward, I need to either :
>  - see the debug messages activated, especially the one pxamci_irq()
>  - or apply this patch [1] to see if it fixes the issue :
> 
> Cheers.
> 
> --
> Robert
> 
> [1] Patch, totally untested/not compiled
> ---8<---
> diff --git a/drivers/mmc/host/pxamci.c b/drivers/mmc/host/pxamci.c
> index 67c9d1443597..3e0a7dd8da84 100644
> --- a/drivers/mmc/host/pxamci.c
> +++ b/drivers/mmc/host/pxamci.c
> @@ -890,9 +890,7 @@ static int pxamci_remove(struct platform_device *pdev)
>                         host->pdata->exit(&pdev->dev, mmc);
>  
>                 pxamci_stop_clock(host);
> -               writel(TXFIFO_WR_REQ|RXFIFO_RD_REQ|CLK_IS_OFF|STOP_CMD|
> -                      END_CMD_RES|PRG_DONE|DATA_TRAN_DONE,
> -                      host->base + MMC_I_MASK);
> +               pxamci_disable_irq(host,  MMC_I_MASK_ALL);
>  
>                 free_irq(host->irq, host);
>                 dmaengine_terminate_all(host->dma_chan_rx);
> 

Yes, the patch fixes it. Log on unfixed version with debugging information enabled consists of many of a line:

	[  125.016666] PXAMCI: irq 00000200 stat 00002042

Petr Cvek