Re: ata-piix.c (and libata.ko) in RHEL5.1 (build 78 currently) - this kernel is 2.6.18 baqsed

Re: ata-piix.c (and libata.ko) in RHEL5.1 (build 78 currently) - this kernel is 2.6.18 baqsed

[Alan Cox]

On Fri, Feb 22, 2008 at 09:52:45AM -0500, Richardson, Charlotte wrote:
> There is a hack in its probe routine that disables this. I've disabled the hack and ran a long test of
> hotplugs yesterday (added and removed a DVD drive connected to the ESB2 IDE - this is Intel
> device 0x269E) for 26 cycles, and it all appears to function properly. There are other things that do

It ought to work fine for ICH4 or so and later (older ICH you can only warm
plug). Be sure to test the released RHEL as I know the dev tree has completely
broken all hot plug disk support and hangs a lot if you do. Jeff I assume
knows the status of fixing that.

Re: ata-piix.c (and libata.ko) in RHEL5.1 (build 78 currently) - this kernel is 2.6.18 baqsed

[Mark Lord]

Alan Cox wrote:
> On Fri, Feb 22, 2008 at 09:52:45AM -0500, Richardson, Charlotte wrote:
>> There is a hack in its probe routine that disables this. I've disabled the hack and ran a long test of
>> hotplugs yesterday (added and removed a DVD drive connected to the ESB2 IDE - this is Intel
>> device 0x269E) for 26 cycles, and it all appears to function properly. There are other things that do
> 
> It ought to work fine for ICH4 or so and later (older ICH you can only warm
> plug). Be sure to test the released RHEL as I know the dev tree has completely
> broken all hot plug disk support and hangs a lot if you do. Jeff I assume
> knows the status of fixing that.
..

I've missed the front part of this thread, but from the above
message it sounds like somebody is trying to do hotplug on ICH4/5/6/.. 

Correct?

I have previously worked on ICH5 hotplug for a client.
The pieces that were needed there were:

1) Emulate SCR_STATUS reads by reading ICH5 PCS.
If it indicates drive presence, return 0x113 for SCR_STATUS,
otherwise return zero.

2) PCS seems to work fine for detecting *inserted* drives,
but does not find *removed* drives very well.  To get it to
see unplug events, We had to first clear the port bit in the PCS,
wait a usec, then write 0x0003 back into PCS.
Then poll PCS for drive present, typically 5-6usec, sometimes longer.
If not present, we're done.  Otherwise wait for drive !BUSY (~60msec).
Ugly as all can be, but it worked.

3) It is critical to ensure that the ATA "ctl" register is never
written to when no drive is attached.  This means bracketing the SRST
sequence to first do a PCS detection before permitting the SRST.
If "ctl" is accessed with no drive attached, the machine locks up hard.

Or at least that's what we were able to figure out in the time available.

What are you guys doing?

-ml

Re: ata-piix.c (and libata.ko) in RHEL5.1 (build 78 currently) - this kernel is 2.6.18 baqsed

[Alan Cox]

> 3) It is critical to ensure that the ATA "ctl" register is never
> written to when no drive is attached.  This means bracketing the SRST
> sequence to first do a PCS detection before permitting the SRST.
> If "ctl" is accessed with no drive attached, the machine locks up hard.

At least for PATA you probably need to disable IORDY handshaking before
the reset and probe.

Alan

Re: ata-piix.c (and libata.ko) in RHEL5.1 (build 78 currently) - this kernel is 2.6.18 baqsed

[Mark Lord]

Alan Cox wrote:
>> 3) It is critical to ensure that the ATA "ctl" register is never
>> written to when no drive is attached.  This means bracketing the SRST
>> sequence to first do a PCS detection before permitting the SRST.
>> If "ctl" is accessed with no drive attached, the machine locks up hard.
> 
> At least for PATA you probably need to disable IORDY handshaking before
> the reset and probe.
..

Perhaps, but this was for SATA.
I don't even own any PATA disks nowadays.. other than a couple
that are buried inside USB/Firewire enclosures.   My last bare PATA
drive died this morning, while testing a SATA bridge on it.

That drive was one of the infamous IBM DeathStar series from early
in the decade.  R.I.P.  Or pieces, actually.  The platters make
great unbreakable "camping mirrors".  :)

Cheers