Re: [PATCH 1/9] cmd64x: implement clear_irq() method

[Sergei Shtylyov <sshtylyov@ru.mvista.com>]

Hello.

Bartlomiej Zolnierkiewicz wrote:

>>>> Convert the driver's two dma_end() methods into clear_irq() methods -- the
>>>> driver will now use the standard dma_end() method implementation, ide_dma_end().
>>>>
>>>> Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>
>>>>
>>>> ---
>>>> The patch is atop of ide-2.6.git 'for-next' branch.
>>>>
>>>>  drivers/ide/cmd64x.c |   31 +++++++++++++++++--------------
>>>>  1 files changed, 17 insertions(+), 14 deletions(-)
>>>>     
>>>>         
>>> [...]
>>>   
>>>       
>>>> @@ -226,11 +226,10 @@ static void cmd64x_set_dma_mode(ide_driv
>>>>  		(void) pci_write_config_byte(dev, pciU, regU);
>>>>  }
>>>>  
>>>> -static int cmd648_dma_end(ide_drive_t *drive)
>>>> +static void cmd648_clear_irq(ide_drive_t *drive)
>>>>  {
>>>>  	ide_hwif_t *hwif	= drive->hwif;
>>>>  	unsigned long base	= hwif->dma_base - (hwif->channel * 8);
>>>>     
>>>>         
>>> Don't we need to check whether hwif->dma_base is valid now?
>>>   
>>>       
>>    You're right, I have managed to overlook this. I'll change this to 
>> pci_resource_start() call instead...
>>     
>
> Currently this driver should operate fine without BAR4 set so even if
> this is pci_resource_start(), the return value still needs to be checked
> against 0 -- it is the reliability/maintainability issue.
>   

   Nay, the PCI device just must not be enabled in this case: all BARs 
must be allocated, period -- that's what the PCI specs say, IIRC. 
Besides, 0 doesn't generally mean "unassigned" in the PCI world.
I know, I know, we consider it unallocated but that's not correct -- 
only PCI 2.2 (IIRC) used to have words about 0 meaning "unassigned". 
Note that we don't check pci_resource_start() result for 0 in the 
drivers that do call it, like hpt366.c... I'm seeing the checks in 
pdc202xx_*.c but they seem pretty useless -- this just must not happen. 
So, I don't agree here...

>>> Also shouldn't this patch be done after fixing core code first?  
>>>       
>>    Fixing what exactly, ide_timer_expiry()?
>>     
>
> Yes.  Currently IRQ is cleared in ide_timer_expiry()'s code-path:
>
>  -> hwif->handler [ ide_dma_intr() ]
>      -> dma_ops->dma_end [ cmd64{x,8}_dma_end() ]
>
> After this patch it won't be.
>   

   Ah, that's a correct spotting...

MBR, Sergei