Re: linux-next: Tree for Nov 27 (scsi/aha1542)

Re: linux-next: Tree for Nov 27 (scsi/aha1542)

[Randy Dunlap]

On 11/26/18 8:25 PM, Stephen Rothwell wrote:
> Hi all,
> 
> Changes since 20181126:
> 

on i386:

ERROR: "__udivdi3" [drivers/scsi/aha1542.ko] undefined!

somewhere in aha1542_interrupt() according to objdump.


-- 
~Randy

Re: linux-next: Tree for Nov 27 (scsi/aha1542)

[Stephen Rothwell]

[-- Attachment #1: Type: text/plain, Size: 468 bytes --]

Hi all,

On Tue, 27 Nov 2018 20:14:58 -0800 Randy Dunlap <rdunlap@infradead.org> wrote:
>
> On 11/26/18 8:25 PM, Stephen Rothwell wrote:
> > Hi all,
> > 
> > Changes since 20181126:
> >   
> 
> on i386:
> 
> ERROR: "__udivdi3" [drivers/scsi/aha1542.ko] undefined!
> 
> somewhere in aha1542_interrupt() according to objdump.

Presumably caused by commit

  1794ef2b150d ("scsi: aha1542: convert to DMA mapping API")

-- 
Cheers,
Stephen Rothwell

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: linux-next: Tree for Nov 27 (scsi/aha1542)

[James Bottomley]

[-- Attachment #1: Type: text/plain, Size: 1387 bytes --]

On Wed, 2018-11-28 at 15:38 +1100, Stephen Rothwell wrote:
> Hi all,
> 
> On Tue, 27 Nov 2018 20:14:58 -0800 Randy Dunlap <rdunlap@infradead.or
> g> wrote:
> > 
> > On 11/26/18 8:25 PM, Stephen Rothwell wrote:
> > > Hi all,
> > > 
> > > Changes since 20181126:
> > >   
> > 
> > on i386:
> > 
> > ERROR: "__udivdi3" [drivers/scsi/aha1542.ko] undefined!
> > 
> > somewhere in aha1542_interrupt() according to objdump.
> 
> Presumably caused by commit
> 
>   1794ef2b150d ("scsi: aha1542: convert to DMA mapping API")

Yes, it's because dma_addr_t can be u64 on pae systems but
isa_virt_to_bus only ever returns unsigned long (because an ISA
physical address can only be 24 bits).

I think this is the fix; there doesn't seem to be much point converting
to do_div given all the limitations.

James

---

diff --git a/drivers/scsi/aha1542.c b/drivers/scsi/aha1542.c
index a9c29757172f..afb693d7b44f 100644
--- a/drivers/scsi/aha1542.c
+++ b/drivers/scsi/aha1542.c
@@ -325,7 +325,7 @@ static irqreturn_t aha1542_interrupt(int irq, void *dev_id)
 			return IRQ_HANDLED;
 		};
 
-		mbo = (scsi2int(mb[mbi].ccbptr) - aha1542->ccb_handle) / sizeof(struct ccb);
+		mbo = (scsi2int(mb[mbi].ccbptr) - (unsigned long)aha1542->ccb_handle) / sizeof(struct ccb);
 		mbistatus = mb[mbi].status;
 		mb[mbi].status = 0;
 		aha1542->aha1542_last_mbi_used = mbi;

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 228 bytes --]

Re: linux-next: Tree for Nov 27 (scsi/aha1542)

[Christoph Hellwig]

On Tue, Nov 27, 2018 at 09:41:24PM -0800, James Bottomley wrote:
> Yes, it's because dma_addr_t can be u64 on pae systems but
> isa_virt_to_bus only ever returns unsigned long (because an ISA
> physical address can only be 24 bits).
> 
> I think this is the fix; there doesn't seem to be much point converting
> to do_div given all the limitations.

Yes, something along these lines should fix it.  I don't have the code
in front of me, but I vaguely remember there is a second division like
this, though.  It might be worth to factor the calculation into a helper
with a comment like the message above explaining it.

Re: linux-next: Tree for Nov 27 (scsi/aha1542)

[Randy Dunlap]

On 11/27/18 9:41 PM, James Bottomley wrote:
> On Wed, 2018-11-28 at 15:38 +1100, Stephen Rothwell wrote:
>> Hi all,
>>
>> On Tue, 27 Nov 2018 20:14:58 -0800 Randy Dunlap <rdunlap@infradead.or
>> g> wrote:
>>>
>>> On 11/26/18 8:25 PM, Stephen Rothwell wrote:
>>>> Hi all,
>>>>
>>>> Changes since 20181126:
>>>>   
>>>
>>> on i386:
>>>
>>> ERROR: "__udivdi3" [drivers/scsi/aha1542.ko] undefined!
>>>
>>> somewhere in aha1542_interrupt() according to objdump.
>>
>> Presumably caused by commit
>>
>>   1794ef2b150d ("scsi: aha1542: convert to DMA mapping API")
> 
> Yes, it's because dma_addr_t can be u64 on pae systems but
> isa_virt_to_bus only ever returns unsigned long (because an ISA
> physical address can only be 24 bits).
> 
> I think this is the fix; there doesn't seem to be much point converting
> to do_div given all the limitations.
> 
> James

Yes, that works.  Thanks.

Acked-by: Randy Dunlap <rdunlap@infradead.org>

> ---
> 
> diff --git a/drivers/scsi/aha1542.c b/drivers/scsi/aha1542.c
> index a9c29757172f..afb693d7b44f 100644
> --- a/drivers/scsi/aha1542.c
> +++ b/drivers/scsi/aha1542.c
> @@ -325,7 +325,7 @@ static irqreturn_t aha1542_interrupt(int irq, void *dev_id)
>  			return IRQ_HANDLED;
>  		};
>  
> -		mbo = (scsi2int(mb[mbi].ccbptr) - aha1542->ccb_handle) / sizeof(struct ccb);
> +		mbo = (scsi2int(mb[mbi].ccbptr) - (unsigned long)aha1542->ccb_handle) / sizeof(struct ccb);
>  		mbistatus = mb[mbi].status;
>  		mb[mbi].status = 0;
>  		aha1542->aha1542_last_mbi_used = mbi;
> 


-- 
~Randy