SSD power loss report updates

[Alan_F_Intel]

Intel is aware of the customer sightings on Intel SSD 320 Series. If you experience any issue with your Intel SSD, please contact your Intel representative or Intel customer support (via web: http://www.intel.com/ www.intel.com or phone: http://www.intel.com/p/en_US/support/contact/phone www.intel.com/p/en_US/support/contact/phone) . We will provide an update when we have more information.

Alan

Intel's NVM Solutions Group

[idata]

Wow, I had no idea Intel made a 600GB SSD. The last time I looked at SSDs they topped out at about 320GB. That's entirely my fault. Sorry for the confusion!

Okay, so hdparm showed the right capacity information -- as such, can someone please explain to me where the "8MB limit" is coming from? Where do people see this if the drive capacity returned via ATA is the correct size (actual size of the drive)? Is it that only 8MBytes of data can be written to the drive (starting at LBA 0 and ending at LBA 16383)? I'm sorry if it sounds like I'm asking something stupid or that's already been answered, but again I don't have a broken drive to tinker with so I have to ask a series of questions.

Secondly, regarding HPA -- I'm fully aware Intel SSDs support said ATA command (SET MAX ADDRESS, etc.). The HPA region (I'm not talking about the ATA command set, I'm talking about where the HPA data actually gets stored on the drive) on mechanical hard disks has historically been stored on the platters in a physical area which the drive will not/does not seek to under normal circumstances. Other data is stored in the HPA than just capacity -- for example, SMART attributes are stored there, in addition to the SCT, and much internal drive-specific data. It's my understanding that the physical defect list is also stored within that same normally-inaccessible area of the disk. This is why "messing about" with the HPA is very dangerous (there is no low-level format on mechanical HDDs these days). On mechanical HDDs, if the HPA region goes bad (just like a standard LBA can go bad), you're downright screwed. RMA is your only choice.

Now about SSDs -- there are no platters, and to my knowledge there is no "dedicated section" of NAND cell intended for HPA-like use. Again, an Intel engineer will need to come in here and correct me if I'm wrong. I simply do not know where drive capacity, model, SCT, SMART, etc. data is stored on an SSD. It may be on a separate flash chip (not NAND) or an EEPROM, or some battery-backed RAM. If you examine a http://images.anandtech.com/reviews/storage/Intel/320/DSC_2617.jpg photo of a 320-series SSD there's what appears to be a small flash chip in the upper left (probably drive firmware, but could also be used for HPA-like data -- again, we don't know!). The http://images.anandtech.com/reviews/storage/Intel/320/DSC_2616.jpg back side of the PCB also has some other chip that looks like it could be used for flash (not NAND).

I know there's the ATA command under the Security section labelled "Secure Erase", which on an SSD should cause the FTL to basically be reset (LBA-to-NAND cell page/block mappings thus lost), which has been known to fix/solve some problems. I don't believe that's the case here (pretty sure you would have tried it by now 🙂 ). That's the closest thing to a "low-level format" there is on an SSD anyway.

[idata]

An SSD still has to store page remappings, the FTL, SMART data and what not in the flash. This has to be stored in an area that is protected from the host, just as on a hard drive. I would assume this area, along with every thing else stored in the flash, is virtualized via the FTL. Isn't it still a HPA? Sure, unlike a hard drive it isn't a physical area, but I would argue that the HPA beign in a physical area is in no way essential to it being a HPA; that the HPA is defined by being used to store internal drive data, not that it is a physical place.

[idata]

koitsu,

Forgive me if I'm wrong, but I think you have a wrong idea about what an HPA is and what it's used for.

Software that is HPA-aware can instruct a disk to report a size that is smaller than its true size to the operating system, effectively hiding part of the drive. The hidden part can then be used to store e.g. an OS preload for recovery purposes, or other software that for whatever reason you wish to prevent the OS (or the user) from accessing. Think of it as a hidden partition, albeit at a lower level, that requires specialized software to access.

I don't know how Intel SSDs were designed, but if they are anything like the embedded systems I've had the opportunity to work on, bad block information, ECC checksums, and logical-physical page translation tables are stored along with each individual page (typically 512 or 2048 bytes) in a small flash area dedicated for this sort of miscellaneous information (typically 16 or 64 bytes to go along with each page).

Data such as the serial number, model etc. is most likely stored in a small EEPROM, and I would imagine the drive capacity is calculated based on the size and number of flash chips detected at power-up. Again, I don't know how Intel chose to implement this, but I would say it's unlikely that any of this data is stored in an HPA.

[idata]

Thanks for the comments, mralpha and carsten. I do understand what the HPA is for, but I guess my logic/thought process wasn't really coming across quite right in my previous post. What I was getting at: if SSDs have an HPA, and the HPA is what ultimately goes bad, and the HPA is stored somewhere other than on NAND cells, then that may help narrow down where the actual problem lies. If an SSD has an HPA and the region is kept within a dedicated region of NAND cell, and that cell happens to go bad, then I guess I'm not too surprised at the outcome; I didn't consider the possibility that the FTL might section off/segregate a portion of a NAND cell for the equivalent of an HPA. But if it's stored separately (EEPROM, flash (not NAND), or even battery-backed SRAM), and *that* is what's going bad, then we've at least narrowed it down a bit.

Sorry for any confusion I've caused.

[idata]

koitsu:

It is definitely a bug, firmware or controller related probably.

It is a lot more than 10 days. The earliest 320 report of the bug to Intel that I am aware of is May 17, from "JW". See the "Is Intel 320 firmware buggy" thread here for a link.

If you really want to see a 320 drive failed with the 8MB bug, your best chance is probably to repeatedly power-cycle the drive, many, many times, like JW mentioned in his May 17 report. I'd suggest the easiest way to do it would be to wire in a toggle switch to a 4 pin Molex to SATA-power adapter, so it toggles between normal connection for the 12V and 5V lines, and no connection. Then you can repeatedly power cycle your SSD without rebooting your computer.

BTW, Intel does NOT make a 320GB SSD. The 320 comes in 40, 80, 120, 160, 300, and 600GB sizes.