MemTest86 10.1 Test 14 Not Working

[zenmota]

I downloaded the new 10.1 version of MemTest86 (Pro Version). I used the ImageUSB program that comes with it to make the USB drive to boot MemTest86 with. USB drive is 8gig. However, when I boot from the USB drive and run MemTest86 and try to run Test 14 DMA Test I get an error message:

Cannot Enable test, DMA test partition not found

Seems the ImageUSB program is not making the DMA test partition like it did in version 10.0. Tried the Extend/Add Partition option and it did make another partition, but Test 14 DMA test still does work.

Thank you for your time

[usko]

The DMA Test Partition isn't in the image file for 10.1 (I use the free version, but I believe it's the same). The DMA Test Partition appears to require to be Partition #3, start at 512MiB on the disk (sector 1048576) and be 1048543 sectors big (ending at 209711 and have the volume id of 50415353-2A40-4503-B2F7-474E00EBFBAA. You can manually create these using gdisk in Ubuntu Linux; I don't know how to make partitions with specific sectors and specific volume ids in Windows or MacOS.

If they ever choose to fix this, it would be really nice to make the transfer partition size just be a factor of 2^n instead of a fixed 2^29, start at any sector on the disk, not hard-code the volume id (but hard code the partition name), and use a partition with a single file in it, consuming the entire partition and probably with the partition hidden flag set.

[David (PassMark)]

Yes, looks like we broke it with the V10.1 release. Opps.
Probably due to the attempt to make the entire disk image smaller (to allowed older USB flash drives to be used).

We'll get it fixed shortly.

[zenmota]

Thanks for looking into it. Looking forward to fixed release.

[David (PassMark)]

We've re-released the V10.1 package and the Test#14 should be OK now.
https://www.memtest86.com/download.htm

(software itself hasn't changed, just the disk image install package)

[zenmota]

New image works fine, DMA test is working. Seems that when the program creates the usb drive it makes a separate partition for the DMA testing. However, this partition seems to be unformatted. so Windows keeps popping wanting to format the partition (at least right after the usb drive is created or every time you plug it in when in Windows). Is it possible to format that partition so Windows don't keep nagging about? Not a big deal, just a question. Thanks for fixing the problem so quickly.

[Station]

Hi


Can you please properly document the requirements for this partition?

Like usko wrote, the default layout has it starting on sector 1048576, ending on sector 2097118 with a total length of 1048543 sectors, with it being the third partition and still having a partition UUID of 50415353-2a40-4503-b2f7-474e00ebfbaa (even in 11.0) with a partition label of "DMA Test Partition". It looks rather empty, so doesn't seem to be formatted with any file system - which makes sense for scratch data.

Which of these parameters are mandatory for MemTest86 to find it?

Is everything hardcoded, as in, the partition number, start and sectors, the label and the UUID? Alternatively, if it's not hardcoded, does MemTest86 scan all partitions until it finds a suitable partition/scratch space? If the latter, what does it search for exactly?


(Tiny side track: I know that the feature/test is marked as experimental and not enabled by default... but even if it weren't experimental, I do believe enabling it by default would be a very bad idea. Enabling it just for 10 test iterations for 32 GB of system memory would amount to 320 GB of writes to flash memory, which is rather excessive wear users probably won't expect. If MemTest86 would try to scan the whole system - not just the device it was booted from! - for a suitable scratch partition, and either use the first one it can find, or even let users select it specifically, users could provide a scratch partition on spinning rust, i.e., old-style hard disk drives, which aren't terribly affected by wear, unlike flash media. Then again, putting the scratch space on HDD would be magnitudes slower than on fast flash media accessed via USB 3.0/3.1 [gen 1/2/whatever]/4.0/… I guess you sadly can't have both acceptable testing speeds and wear protection. )

[David (PassMark)]

I do believe enabling it by default would be a very bad idea

Flash drives are insanely cheap. So we aren't too worried about the wear.

But it also turns out that USB flash drives are VERY unreliable. People buy very cheap USB flash drives and the failure rate is very high. So we saw more errors on the flash drives than the memory. This was the major problem with the test. Years ago we bought a batch of 100 no-name flash drives. 30% failed on the first write.

[Station]

True. I have a habit of avoiding cheap flash drives and always testing them before use to make sure that they at least aren't counterfeit and contain the advertised memory. However, this doesn't mean that they couldn't partially fail after that and figuring that out is extremely difficult due to wear leveling.

This said, I'd be really pissed if MemTest86 would inadvertently cause so much wear to my $50 32 GB or $100 1 TB flash drives.

[David (PassMark)]

You can get name brand 32GB USB drives for $5 and 16GB for $3.
Writing data to a somewhat random disk sector on a random internal hard disk has the high risk of data loss.
Anyway the test is turned off by default.

[Station]

Okay, I've played around a bit and found out the set of parameters that are necessary for the partition to be recognized:

name	irrelevant (tested "blargh")
number	must be 3
type UUID	irrelevant (tested 0FC63DAF-8483-4772-8E79-3D69D8477DE4 mapping to "Linux filesystem")
UUID	must be 50415353-2A40-4503-B2F7-474E00EBFBAA
start sector	must be exactly 1048576
end sector	must be exactly 2097118
attributes	irrelevant (tested hidden, hide from EFI, read-only, do not automount)

Hence, the software (at least up to version 11.0) is currently looking for the partition at exactly the given sector values (even one sector more or less will make the detection fail), it must have partition number 3 and the UUID as specified.


It would be great if the partition number and start and end sectors could be made dynamic. Having a fixed size with a fixed UUID isn't a big deal, but the fixed location and partition number is inconvenient.

Writing data to a somewhat random disk sector on a random internal hard disk has the high risk of data loss.

Oh, no, I don't mean writing to random locations, that would be rather stupid. However, if users provide a partition of exactly 1048543 sectors with a UUID of 50415353-2A40-4503-B2F7-474E00EBFBAA on a HDD, then that would be a clear invitation for MemTest86 to use it.

[David (PassMark)]

Just confirming: The scratch partition for DMA test is fixed at a specific GUID, offset and size to prevent MemTest86 inadvertently finding and destroying data from another partition. So only USB drives created from official MemTest86 images can be used with this optional DMA test.

If enough people wanted something different we would take a look at it. Holy grail would be to use DMA, without any disk being involved at all. We attempted a solution like this, but couldn't make it work on a wide range of machines.