mptsas hangs caused by ATA pass-through explained

Kashyap wrote:

I have also verified this patch + also done code review with our developers including Eric Moore. Please consider this patch as an ACKed patch and schedule it for next upstream release.

----------------------

Hi,

In comment #20 of this bug entry:

https://bugzilla.kernel.org/show_bug.cgi?id=14831

you state:

"Patch for setting dma boundary is mere avoiding condition which is causing this issue. LSI Gen-1 controller does not have 512byte dma boundary limitation. I have started internal chat with our Firmware engineer. I will update you findings as and when some imp stuffs are found."

Is this still the case, or is the patch the final fix?

Also, comment #25 in the same thread notes a significant negative write performance impact.

Are you seeing this?

I have several machines in production I would like to update, but not if there is a better solution to this on it's way.

Regards,

Richard

Richard,

I remember that default dma boundary was 512 byte earlier sometime 2.6.30 kernel version. Later kernel has introduced 4 byte dma boundary in scsi_lib.c....

As I have mentioned we have limitation of single sgl for ATA PIO commands. So making 512 byte dma boundary will solve all functional issue, but As mentioned in bugzilla #25, people are seeing performance degradation (may due to bounce buffer usage). I need to debug that part.

In my opinion if setting "dma boundary to 512 byte is a solution only if there is no performance degradation" Can I have data how user at bugzilla is testing performance?

Thanks,
Kashyap