This message is from the T13 list server.
If a cached write fails, there's no asynchronous way to post notification of the failure... that's one of the drawbacks of the huge performance gains with enabled write caches. I think most drives will post a 0x71 status indicating write fault for all subsequent cached write commands, and that should be plenty of indication that something catastrophic has happened. Modern ATA drives don't write fault unless they're having severe problems... error recovery in desktop tends to be very "heroic" (while in large SCSI arrays it is often faster to rebuild the block from your RAID5 data than recover it from the flaky disk) In my opinion, something designed to stay BSY=1 indefinitely is asking for trouble. I also don't see off the top of my head how a drive is capable of aborting a command and staying busy. --eric > -----Original Message----- > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On > Behalf Of Larry Barras > Sent: Tuesday, March 08, 2005 3:18 PM > To: Harlan Andrews; John Masiewicz > Cc: [email protected] > Subject: Re: [t13] Inquiries into SATA material during ATA/8 review > > > This message is from the T13 list server. > > > There is a subtle difference between "freeze" and abort in this case. > > What the drive should actually do in this case is abort the command > and stick busy. > > Freeze would mean literally "do nothing". > > I brought up the same point at the discussion, but I see what they > are getting at with the term freeze meaning "do nothing". What we > want in this case is the drive to abort and stay BSY=1 until power > cycled as a signal that things have gone very wrong. > > > > At 2:06 PM -0800 3/8/05, Harlan Andrews wrote: > >John, > > > >One case where drives currently "Freeze" is when a cashed write > >fails. At that point, the host must be notified that one or more > >of the previously successfully acknowledged writes was unable to be > >written. The current behavior of several drives is to freeze > >until a power cycle occurs. NOTE: a COMRESET does NOT ( and > >should not) reset this condition. Only the power cycle resets this > >condition. > > > >I would MUCH rather see a more identifiable error condition. > >Ideally, the drive should return a specific error to ALL commands > >until it is explicitly reset by the host. The "reset" should NOT > >be any currently defined command. It should be explicitly reserved > >to reset the failed cashed write. The host must not be allowed to > >continue without specifically acknowledging that data on the drive > >HAS BEEN CORRUPTED. > > > >...Harlan > > > > > >On Mar 8, 2005, at 1:44 PM, John Masiewicz wrote: > > > >>Brian, > >> > >>I don't think you asked quite the right questions as posed in the > >>meeting, especially regarding the BIST. > >> > >>BIST was not listed are re-transmittable in the list, but if you > >>look at the Device Transport BIST Activate state diagram, a retry > >>is explicitly directed. This was the inconsistency we were trying > >>correct. I don't have any problem adding a note, but the two > >>section are not in agreement, thus the request for clarification, > >>it there is any. > >> > >> > >> > >>The issue with "Freeze" is that it is recommended. I cannot think > >>of any condition where I would recommend a drive "Freeze" as a > >>policy. (I have no objection to hosts freezing) Freeze means rely > >>on timeouts, which in ATA are as long as 30 seconds. This does not > >>sound like a good recommendation for error handling if you know you > >>are in an unrecoverable situation. Freeze, as discussed in the > >>meeting, implies do nothing, and respond to nothing (except > >>COMRESET). If a device is so perturbed that it unrecoverable and > >>it knows this, certainly a COMINIT from the device or COMRESET > >>from the host might be recommended, but not "freezing". > >> > >> > >> > >>I prefer eliminating the unnecessary recommendation, since it > >>offers nothing in terms of error recovery, but rather says do not > >>perform error recovery. Similarly, it is in a section entitled > >>"Software error recovery" and I would not recommend that the > >>software "freeze". > >> > >> > >> > >>Please, however, review the state machines on the BIST retry, and > >>reconsider your recommendation. > >> > >> > >> > >>John Masiewicz > >> > >>Western Digital > >> > >> > >> > >> > >>From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of > >>Dees, Brian M > >>Sent: Tuesday, March 08, 2005 8:44 AM > >>To: [email protected] > >>Subject: [t13] Inquiries into SATA material during ATA/8 review > >> > >> > >> > >>Hi, > >> During the review of the ATA/8 Serial Transport material > >>in the February T13 Summit a few items were given to me as the > >>liaison into the SATA-IO to research the meaning or intent. Below > >>are the questions and responses I've recorded in finding the > >>details. > >> > >>1. What was the original intent of listing 'Freeze' as a valid > >>category for error responses? (see sections 10.1 & 10.5.2 of > >>d1697r0a-ATA8-ST, where Freeze has been removed) > >> > >><Original reference in sections 11.1 (pg 273) and 11.5.2 (pg 279) > >>of Serial ATA 1.0a.> > >>Freeze is the condition the drive enters where a host action like a > >>reset is required to recover the drive. Whether you admit it or > >>not, it is not unusual for drives to enter this state. Serial ATA > >>might have been the first to explicitly call this behavior out, > >>although it is not an unusual occurrence. In hopelessly messed up > >>situations, it is probably safer for a drive to halt and await the > >>host cleaning up with a reset than it might be for the drive to > >>proceed and potentially exacerbate the situation. > >> > >>2. Why isn't BIST Activate included in the list of frame types that > >>can be retransmitted? (see section 10.4.3.2 of d1697r0a-ATA8-ST) > >> > >><Original reference in section 11.4.2.2 (page 278) of > Serial ATA 1.0a.> > >>BIST is not a function performed during the normal course of > >>operation for a disk drive. Since it's an engineering lab-bench > >>function to be used in controlled conditions by knowledgeable > >>experts, it does not have the kinds of automatic recovery that > >>normal user functions might have. In some lab settings, it could > >>even be detrimental for auto-recovery operations to be performed by > >>the interface, since some lab functions are for the purpose of > >>examining such behaviors. If the BIST Activate function fails in a > >>lab setting, the engineer is presumably sufficiently capable to > >>know what is happening and attempt the operation over again. > >> > >> With the understanding of the above explanations, > I believe > >>that 'Freeze' should be re-inserted back into the ATA/8 material > >>and that BIST Activate should still not exist in the list referred > >>to by question #2. This would require three total edits, although > >>minor, into the next revision of the document. > >> > >>Regards, > >>Brian > > > -- > > --------------------- > I make stuff go. > --------------------- > > Larry Barras > Apple Computer Inc. > 1 Infinite Loop > MS: 306-2TC > Cupertino, CA 95014 > (408) 974-3220 >
