[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-scsi
Subject:    Re: [PATCH 22/71] ncr5380: Eliminate selecting state
From:       Finn Thain <fthain () telegraphics ! com ! au>
Date:       2015-11-30 2:12:45
Message-ID: alpine.LNX.2.00.1511301310210.10574 () nippy ! intranet
[Download RAW message or body]


On Sun, 29 Nov 2015, Geert Uytterhoeven wrote:

> If an ISA access takes 8 us, while the CPU runs at 1 GHz, i.e. 500M 
> loops/s, the difference will be huge.

Ondrej showed that an ISA access can take about 1.6 us. I don't know what 
to make of the "8 uS" comment in the mainline driver. Maybe it was an even 
slower ISA card.

Anyway, I made a measurement on my hardware and confirmed that lpj is a 
very bad proxy for device register access throughput. The "loops per 
access" gap is several orders of magnitude:

                                      lpj   HZ    access time (us)  lpa
------------------------------------------------------------------------
Ondrej's 5380 ISA card (PC):      4797252   250   1.6               1.9k
My DMX3191D PCI card (PowerMac):   167079   100   0.42              7.0

> 
> Perhaps you can calibrate an NCR5380_read() loop at driver init time, 
> and use the calibration value later?

I had the same idea but I didn't think that the complexity was justified 
by the low precision requirement. But now that I have some timings I have 
to agree.

-- 
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
[prev in list] [next in list] [prev in thread] [next in thread]