dma_mask on scsi_calculate_bounce_limit

dma_mask on scsi_calculate_bounce_limit

[王神]

Hi

I have enabled DMA for my host device and when i connect a scsi device
to the host, it crashes in the following code..

Function: scsi_calculate_bounce_limit

 if (host_dev && host_dev->dma_mask)
                bounce_limit = *host_dev->dma_mask;

"*dma_mask" is illegal pointer operation. I would like to understand
is this known and is there reason behind such a code ?

TIA
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: dma_mask on scsi_calculate_bounce_limit

[James Bottomley]

On Thu, 2010-11-18 at 14:20 +0530, 王神 wrote:
> Hi
> 
> I have enabled DMA for my host device and when i connect a scsi device
> to the host, it crashes in the following code..
> 
> Function: scsi_calculate_bounce_limit
> 
>  if (host_dev && host_dev->dma_mask)
>                 bounce_limit = *host_dev->dma_mask;
> 
> "*dma_mask" is illegal pointer operation. I would like to understand
> is this known and is there reason behind such a code ?

It likely means you're operating on a new architecture and its generic
devices haven't been set up correctly.

James


--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: dma_mask on scsi_calculate_bounce_limit

[王神]

On Thu, Nov 18, 2010 at 11:21 PM, James Bottomley
<James.Bottomley@suse.de> wrote:
> On Thu, 2010-11-18 at 14:20 +0530, 王神 wrote:
>> Hi
>>
>> I have enabled DMA for my host device and when i connect a scsi device
>> to the host, it crashes in the following code..
>>
>> Function: scsi_calculate_bounce_limit
>>
>>  if (host_dev && host_dev->dma_mask)
>>                 bounce_limit = *host_dev->dma_mask;
>>
>> "*dma_mask" is illegal pointer operation. I would like to understand
>> is this known and is there reason behind such a code ?
>
> It likely means you're operating on a new architecture and its generic
> devices haven't been set up correctly.

Could you please more specific on what is generic devices haven't been
setup properly...

>
> James
>
>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: dma_mask on scsi_calculate_bounce_limit

[James Bottomley]

On Fri, 2010-11-19 at 01:54 +0800, 王神 wrote:
> On Thu, Nov 18, 2010 at 11:21 PM, James Bottomley
> <James.Bottomley@suse.de> wrote:
> > On Thu, 2010-11-18 at 14:20 +0530, 王神 wrote:
> >> Hi
> >>
> >> I have enabled DMA for my host device and when i connect a scsi device
> >> to the host, it crashes in the following code..
> >>
> >> Function: scsi_calculate_bounce_limit
> >>
> >>  if (host_dev && host_dev->dma_mask)
> >>                 bounce_limit = *host_dev->dma_mask;
> >>
> >> "*dma_mask" is illegal pointer operation. I would like to understand
> >> is this known and is there reason behind such a code ?
> >
> > It likely means you're operating on a new architecture and its generic
> > devices haven't been set up correctly.
> 
> Could you please more specific on what is generic devices haven't been
> setup properly...

It means the architecture setup code left a NULL pointer in the generic
device which shouldn't be there ... without knowing which architecture
and seeing the code, it's pretty impossible to be more specific.

James


--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: dma_mask on scsi_calculate_bounce_limit

[王神]

On Fri, Nov 19, 2010 at 6:22 AM, James Bottomley
<James.Bottomley@suse.de> wrote:
> On Fri, 2010-11-19 at 01:54 +0800, 王神 wrote:
>> On Thu, Nov 18, 2010 at 11:21 PM, James Bottomley
>> <James.Bottomley@suse.de> wrote:
>> > On Thu, 2010-11-18 at 14:20 +0530, 王神 wrote:
>> >> Hi
>> >>
>> >> I have enabled DMA for my host device and when i connect a scsi device
>> >> to the host, it crashes in the following code..
>> >>
>> >> Function: scsi_calculate_bounce_limit
>> >>
>> >>  if (host_dev && host_dev->dma_mask)
>> >>                 bounce_limit = *host_dev->dma_mask;
>> >>
>> >> "*dma_mask" is illegal pointer operation. I would like to understand
>> >> is this known and is there reason behind such a code ?
>> >
>> > It likely means you're operating on a new architecture and its generic
>> > devices haven't been set up correctly.
>>
>> Could you please more specific on what is generic devices haven't been
>> setup properly...
>
> It means the architecture setup code left a NULL pointer in the generic
> device which shouldn't be there ... without knowing which architecture
> and seeing the code, it's pretty impossible to be more specific.

It is for ARM and the error is not due to NULL pointer. The code just
checks if it is host mode + dma enabled and it access a normal
interger variable with indirection causing the crash;

int dma_mask;

and accessing dma_mask as "*dma_mask" ??

Anything I am missing :O

>
> James
>
>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: dma_mask on scsi_calculate_bounce_limit

[Douglas Gilbert]

On 10-11-19 01:41 AM, 王神 wrote:
> On Fri, Nov 19, 2010 at 6:22 AM, James Bottomley
> <James.Bottomley@suse.de>  wrote:
>> On Fri, 2010-11-19 at 01:54 +0800, 王神 wrote:
>>> On Thu, Nov 18, 2010 at 11:21 PM, James Bottomley
>>> <James.Bottomley@suse.de>  wrote:
>>>> On Thu, 2010-11-18 at 14:20 +0530, 王神 wrote:
>>>>> Hi
>>>>>
>>>>> I have enabled DMA for my host device and when i connect a scsi device
>>>>> to the host, it crashes in the following code..
>>>>>
>>>>> Function: scsi_calculate_bounce_limit
>>>>>
>>>>>   if (host_dev&&  host_dev->dma_mask)
>>>>>                  bounce_limit = *host_dev->dma_mask;
>>>>>
>>>>> "*dma_mask" is illegal pointer operation. I would like to understand
>>>>> is this known and is there reason behind such a code ?
>>>>
>>>> It likely means you're operating on a new architecture and its generic
>>>> devices haven't been set up correctly.
>>>
>>> Could you please more specific on what is generic devices haven't been
>>> setup properly...
>>
>> It means the architecture setup code left a NULL pointer in the generic
>> device which shouldn't be there ... without knowing which architecture
>> and seeing the code, it's pretty impossible to be more specific.
> 
> It is for ARM and the error is not due to NULL pointer. The code just
> checks if it is host mode + dma enabled and it access a normal
> interger variable with indirection causing the crash;
> 
> int dma_mask;
> 
> and accessing dma_mask as "*dma_mask" ??
> 
> Anything I am missing :O

Yes, supplying more information.

I'm familiar with AT91SAM9G20 which is an ARM based
microcontroller. The directory of interest in the kernel
source is arch/arm/mach-at91 and for the G20 the
at91sam9260_devices.c file defines its DMA masks for
its devices. In all cases that I can see the dma_mask is
the address of a u64 holding DMA_BIT_MASK(32) .

There are no real SCSI devices on the G20 but USB
might be regarded as SCSI "once removed". And the
USB devices on the G20 have dma_mask values as
described above.

Doug Gilbert

--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: dma_mask on scsi_calculate_bounce_limit

[James Bottomley]

On Fri, 2010-11-19 at 12:11 +0530, 王神 wrote:
> On Fri, Nov 19, 2010 at 6:22 AM, James Bottomley
> <James.Bottomley@suse.de> wrote:
> > On Fri, 2010-11-19 at 01:54 +0800, 王神 wrote:
> >> On Thu, Nov 18, 2010 at 11:21 PM, James Bottomley
> >> <James.Bottomley@suse.de> wrote:
> >> > On Thu, 2010-11-18 at 14:20 +0530, 王神 wrote:
> >> >> Hi
> >> >>
> >> >> I have enabled DMA for my host device and when i connect a scsi device
> >> >> to the host, it crashes in the following code..
> >> >>
> >> >> Function: scsi_calculate_bounce_limit
> >> >>
> >> >>  if (host_dev && host_dev->dma_mask)
> >> >>                 bounce_limit = *host_dev->dma_mask;
> >> >>
> >> >> "*dma_mask" is illegal pointer operation. I would like to understand
> >> >> is this known and is there reason behind such a code ?
> >> >
> >> > It likely means you're operating on a new architecture and its generic
> >> > devices haven't been set up correctly.
> >>
> >> Could you please more specific on what is generic devices haven't been
> >> setup properly...
> >
> > It means the architecture setup code left a NULL pointer in the generic
> > device which shouldn't be there ... without knowing which architecture
> > and seeing the code, it's pretty impossible to be more specific.
> 
> It is for ARM and the error is not due to NULL pointer. The code just
> checks if it is host mode + dma enabled and it access a normal
> interger variable with indirection causing the crash;

OK, it's crashing because the pointer isn't set up correctly ... is that
more semantically accurate?

The pointer setup is done in the arch code, so that's where the problem
is.

> int dma_mask;

That's not the definition in struct device; this is:

        u64             *dma_mask;      /* dma mask (if dma'able device) */


> and accessing dma_mask as "*dma_mask" ??
> 
> Anything I am missing :O

Yes, but you supply too little information to know what.  It's
definitely a problem somewhere in the arch setup code.

James


--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html