Re: [Qemu-devel] [PATCH 15/28] ide: stop PIO transfer on errors

[Paolo Bonzini]

Il 31/07/2014 14:23, Stefan Hajnoczi ha scritto:
> On Mon, Jul 07, 2014 at 02:17:56PM -0400, John Snow wrote:
>> From: Paolo Bonzini <address@hidden>
>>
>> This will provide a hook for sending the result of the command via the
>> FIS receive area.
>>
>> Signed-off-by: Paolo Bonzini <address@hidden>
>> Signed-off-by: John Snow <address@hidden>
>> ---
>>  hw/ide/core.c | 7 ++++---
>>  1 file changed, 4 insertions(+), 3 deletions(-)
>>
>> diff --git a/hw/ide/core.c b/hw/ide/core.c
>> index bd3bd34..d900ba0 100644
>> --- a/hw/ide/core.c
>> +++ b/hw/ide/core.c
>> @@ -422,6 +422,9 @@ static inline void ide_abort_command(IDEState *s)
>>  {
>>      s->status = READY_STAT | ERR_STAT;
>>      s->error = ABRT_ERR;
>> +    if (s->end_transfer_func != ide_transfer_stop) {
>> +        ide_transfer_stop(s);
>> +    }
>>  }
> 
> I don't understand this.
> 
> ide_transfer_stop() sets s->send_transfer_func = ide_transfer_stop.
> This has the side-effect of making ide_is_pio_out() return true (not
> sure if that poses a problem).

ide_is_pio_out is only called if DRQ_STAT is set, and it is never set
after ide_transfer_stop returns, so that is not a problem.

> Why can't we call ide_transfer_stop() when s->end_transfer_func ==
> ide_transfer_stop?

The idea was to only call it for PIO commands (if s->end_transfer_func
== ide_transfer_stop, the PIO command cannot fail anymore), but we might
as well call it unconditionally like ide_dma_error does, i.e.

 {
+    ide_transfer_stop(s);
     s->status = READY_STAT | ERR_STAT;
     s->error = ABRT_ERR;
 }

It's simpler indeed.

Paolo