Re: RE: Calling scsi_done from queuecommand()

Re: RE: Calling scsi_done from queuecommand()

[Zhikun Wang]

> On Wednesday, January 11, 2006 9:21 PM, Zhikun Wang wrote: 

> Mr. Moore:
> 	Can we call scsi_done() to complete the I/O from the 
> queuecommand() ? (e.g. to make 
> queuecommand synchronous).   I try this method in my LsiLogic 
> 22320 SCSI card driver 
> inMoore, Eric Redhat9, and i can correctly handle read command. But 
> Write command is stop when i 
> call scsi_done() in the queuecommand. I am very puzzled!   
> Thanks for any answers!
> 

> I see no problem with that as long as your waiting for the firmware

> to complete the command.
My problem is write command is completed for the firmware, but call scsi_done() is stopped!

Calling scsi_done from queuecommand()

[Zhikun Wang]

Hi list, 
	Can we call scsi_done() to complete the I/O from the queuecommand() ? (e.g. to make queuecommand synchronous). 
	I try this method in my LsiLogic 22320 SCSI card driver in Redhat9, and i can correctly handle read command. But Write command is stop when i call scsi_done() in the queuecommand. I am very puzzled! 
	Thanks for any answers!

Regards, 
-zkwang 

Re: Calling scsi_done from queuecommand()

[Arjan van de Ven]

On Thu, 2006-01-12 at 11:57 +0800, Zhikun Wang wrote:
> Hi list, 
> 	Can we call scsi_done() to complete the I/O from the queuecommand() ? 

yes that should be fine *IF THE COMMAND COMPLETED*. (for example that
can be done for battery backed ram cards)

> (e.g. to make queuecommand synchronous). 
> 	I try this method in my LsiLogic 22320 SCSI card driver 

but... that sounds unlikely to be right. That is real scsi, which means
you have to wait a REALLY long time (easily 5 miliseconds) before the
command completes, and you cannot sleep in queuecommand (it's called
from interrupt context). So this sounds really fishy

> in Redhat9,

you do realize you're using a truely antique linux kernel/distribution
there, right? I suggest you go to a far more recent (and an actual
maintained) distribution

Re: Re: Calling scsi_done from queuecommand()

[Zhikun Wang]

>On Thu, 2006-01-12 at 11:57 +0800, Zhikun Wang wrote:
> Hi list, 
> 	Can we call scsi_done() to complete the I/O from the queuecommand() ? 

>>yes that should be fine *IF THE COMMAND COMPLETED*. (for example that
>>can be done for battery backed ram cards)

> (e.g. to make queuecommand synchronous). 
> 	I try this method in my LsiLogic 22320 SCSI card driver 

>>but... that sounds unlikely to be right. That is real scsi, which means
>>you have to wait a REALLY long time (easily 5 miliseconds) before the
>>command completes, and you cannot sleep in queuecommand (it's called
>>from interrupt context). So this sounds really fishy

I intend to use a global variable to make queuecommand synchronously. In the <queuecommand function> i set the global variable to 0.
	......
	global = 0;
	......
	while(!global)
    	barrier();
    SCpnt->scsi_done(SCpnt);
	......
When the command completes, in the <interrupt routine> i set the global variable to 1.
	......
	global = 1;
    ......
this method works for read command, but write command will stop at scsi_done routine and then timeout!
i am very puzzled about this!

> in Redhat9,

>>you do realize you're using a truely antique linux kernel/distribution
>>there, right? I suggest you go to a far more recent (and an actual
>>maintained) distribution

i know Redhat9 is out of time, but i must use this kernel/distribution
in my project!