This message is from the T13 list server.

Jeff G

> Since the world is moving to DMA and SATA FIS's, is there any chance of 
> un-deprecating IDENTIFY DEVICE DMA ?

Fun question, thank you.

> It seems to be supported even in many of today's configurations, and I 
> prefer common paths that are 100% non-data or DMA protocols (even for PATA).

1) Can you tell us more of this preference?

Specifically which measurable evil do we wish to alleviate by reviving
the host option of never talking PIO?  The latency of interrupts in the
256*600 = 153,600ns interval of a PIO block burst?  The code size of a
host designed to work only with devices that support never talking PIO? 
All that and more?

I struggle in part because I've worked only with hosts that begin always
with the PIO op xEC IDENTIFY.  If that doesn't work, then it seems to
me, perhaps falsely, that nothing works.  Does your history of pain
contain systems which do talk DMA but not PIO?  I do remember at t13.org
I've heard tell of ancient drives which read/ write in PIO but do not
support op xEC.  But I've never heard of DMA working without PIO.

2) Can you easily cross-reference us into the relevant Linux source?

At a glance I found only the likely irrelevant:

--- http://lxr.linux.no/search?string=0xEC
--- http://lxr.linux.no/search?string=0xec
--- http://lxr.linux.no/source/drivers/block/paride/pd.c#L260
#define IDE_IDENTIFY            0xec
---

The comments tell me paride is "parallel port IDE", which I take to mean
PC printer port IDE, not PATA i.e. parallel ATA.

Pat LaVarre

Reply via email to