[PSA] Sandisk SD card causing BSoD

[blueskull]

My 4 year old Sandisk Ultra 256GB has finally bit the dust, despite having seen less than 20 total write cycles (5TBW).
Granted, some part of it was written quite a few times, particularly partition table, maybe 200 times.

Today, it was failing in a funny way. It retains data written to it, partition table, but not drive letter assignment.
It threw PFN_LIST_CORRUPT and KERNEL_MODE_HEAP_CURRUPTION BSoDs when I attempt to reassign drive letter.

I initially thought it was Windows update doing funny things, but I tried the same operation on a 256GB Samsung thumb drive, and it worked.
I then raised suspicion on the SD card reader, and tried another 16GB Sandisk, and it worked.

So, either Windows 10 has a bug handling high capacity cards (in this case, 256GB), or this card is faulty.

Considering my positive experience with the 256GB thumb drive and large SSDs, I bet it's the fault of the card.
Failure of 4 year old card with less than 20 P/E cycles is not acceptable from Sandisk, and I'd like to know if you have similar oops moments.
It was never subject to ESD or other stress situations. It pretty much spent its entire life in a phone (Sony, then OnePlus) or a Surface (Go and Pro 7).

PS. I'm not that shocked either. I had Toshiba card early failures before, so Toshiba (shares fab with Sandisk) NAND failure is expected.

Just ordered a Samsung card. Let's see how long does this one last.

[Ampera]

I'd say it not working in one machine (a Windows machine at that) does not definitively prove that it's gone bad. I'd suggest doing some tests on Linux, if you know how, to see if it's actually gone bad.

[magic]

Today, it was failing in a funny way. It retains data written to it, partition table, but not drive letter assignment.

I'm not entirely sure, I haven't used Windows in ages, but I think drive letter assignments aren't stored on the card but they are specific to your machine and stored locally on your machine. At any rate, it would be a weird card which manages to store all data correctly except for some tiny bit of Windows metadata.

It threw PFN_LIST_CORRUPT and KERNEL_MODE_HEAP_CURRUPTION BSoDs when I attempt to reassign drive letter.

These are memory corruption errors, wrong data being written to a wrong place in RAM.
This is absolutely a software bug, probably a buffer overrun or wrong pointer being used, usually resulting from a failure to correctly handle some corner case the developer didn't anticipate.
This unexpected case may be hardware fault, or it may be some unlucky combination of unusual but valid hardware and user behavior.

I would try "deep" format. Erase/trim the entire card, recreate partitions (be sure to preserve the original alignment) and format. It should restore all software-visible state (except for write latency ) to some sane defaults.

[Ampera]

I don't mean to be pedantic, but don't use the term low level format, as it means something completely different than just wiping the card. It's supposed to refer to the creation of magnetic track layouts (and any other modulation or such) on mechanical hard drives. Just something to avoid confusion.

My suggestion is to start simple and work your way up. It's possible there's something in the partition table that's making Windows bug out. Using a tool like fdisk (linux), rewrite the partition table (GPT or MBR) and then repartition and reformat. Since you're using Darwin Unix, there's likely other manners of going about what you might be trying to accomplish. You could use dd to nuke the first, oh let's say 50MB (to be sure) of the SD card and then give it back to Windows to figure something out. (note partition tables tend to keep redundant copies, so you might want to nuke the last 50MB of the card too, or just rewrite the partition table using a tool)

The fact that it mounts to a folder in Windows is actually quite weird, and while I may, in my spare time, like to pick on Windows, it usually doesn't break like /this/. That being said, there has never been any shortage of trouble people have had with drive letter assignments, so what do I know.

[Ampera]

Huh? I've never heard of that. I mean I can't say I've used large SD cards a whole lot, but that is properly strange.

May I ask what partition table the SD card(s) is(are) using?

[Ampera]

If it wouldn't be too much of a burden (just out of curiosity), could you attempt to replicate this with GPTs?

[magic]

I'll try some DD trick on my mac with the "faulty" card to emulate a low level format and report back.

You can and probably should use trim to erase flash memories.
If your SD reader and/or OS doesn't support that, any digital camera has "full format" option which will discard and reformat the main partition. The remaining few megs could be dd'd if you must.

Hard to do. I've already dumped 200GB+ of data to it.

Flash is not a techonolgy to trust with data existing only in one copy. One borked sector in the flash translation layer can render all your data messed up and scattered and there is not even any standard means of recovery.

[wraper]

Have you tried using diskpart clean or something like that to actually delete everything? Simply deleting partitions and creating new is not equal to doing so on new and unused drive with no data.

[blacksheeplogic]

I've had the occasional issue when using a MSDC between linux & windows where Windows won't recognize the card when inserted, rewrite the image and it's fine....

However, I stopped using SanDisk SD cards altogether now due to there actual usable capacity being less than other MSDC.

[NiHaoMike]

So, maybe Icelake+LPDDR4X 3733 is not really robust enough, or MSFT's memory diagnostic tools is not thorough enough.

The definitive test is memtest86. Might have to let it run overnight for really intermittent faults.

[SparkyFX]

So, maybe Icelake+LPDDR4X 3733 is not really robust enough, or MSFT's memory diagnostic tools is not thorough enough.

RAM fails can happen anytime, had to throw out a couple to solve crashes as well. The self-test at boot is too unreliable to find that.

I am surprised faulty RAM actually showed up as possible cause when googling for your BSOD errors.

[Jeroen3]

BSOD because of an SD card is solid evidence the card interface driver is shitty. Get a better card interface.

A hot-plug device like an SD card may never cause a kernel panic.

[magic]

Jeroen3 is right. A faulty RAM wouldn't know that it is supposed to crap itself out exactly when you try to change the drive letter of a particular SD card. That would be crashes all over the place or at least crashes when you run some computationally intensive workload.

I am surprised faulty RAM actually showed up as possible cause when googling for your BSOD errors.

Nothing surprising, because it is a "memory corruption detected" bluescreen.
If you get those randomly when running Prime95, you can suspect RAM. If you get them 100% reliably by clicking some button in Windows settings, you should probably suspect Windows of shitting its own memory up when you click that button.

[iMo]

Sandisk SDcards are the most faked cards in the world.

[SparkyFX]

Jeroen3 is right. A faulty RAM wouldn't know that it is supposed to crap itself out exactly when you try to change the drive letter of a particular SD card. That would be crashes all over the place or at least crashes when you run some computationally intensive workload.

Just guessing, but the problem is about memory in the heap, in a section tightly watched by the kernel, not some application memory. After booting the memory layout is pretty much the same every time, install new drivers and an area which was reserved space before is now actually in use - maybe even the cards space leads to more reserved space for buffers. Freezes and crashes can also happen after a certain runtime, if the affected memory is actually used.

It depends on the address and whatever Win10 does with its memory. With ECC memory the problem might not even be one.

But i'd also approach this problem by booting into Knoppix and seeing what happens there. The memory test that comes with it is also quite thorough.

[Jeroen3]

A hot-plug device like an SD card may never cause a kernel panic.

A faulty NTFS volume or NTFS driver can. And a faulty RAM can cause all sorts of unexplained bugs.

Only when kernel or driver resources are on it, such as the swap file, which leads to memory corruption. An external or data NTFS volume will not cause a kernel panic. Except when the driver is shit.

[Monkeh]

A hot-plug device like an SD card may never cause a kernel panic.

A faulty NTFS volume or NTFS driver can. And a faulty RAM can cause all sorts of unexplained bugs.

Only when kernel or driver resources are on it, such as the swap file, which leads to memory corruption. An external or data NTFS volume will not cause a kernel panic. Except when the driver is shit.

They're all shit, then. Including AHCI.

Windows has never gracefully handled any drive failures in my experience.

[Haenk]

For formatting you should use "SD Formatter" from sdcard.org *only*. Don't use Windows built in format.
(There are mirrors for the software, as sdcard.org seems to be down for me. Funny, as we are even a paying member of SDA...)

To check integrity, use h2wtest from heise. Another suggested tool is Hard Disk Sentinel, which lets you check all blocks etc. (though better suited for HDDs).

[Monkeh]

For formatting you should use "SD Formatter" from sdcard.org *only*. Don't use Windows built in format.

Which, so far as I can tell, does nothing except use standard Windows APIs to partition and create a filesystem. Nothing you can't do with a dozen other tools, and, frankly, with the built in dialogues.

[magic]

SD spec mandates particular alignment of partitions, internal filesystem data structures and even recommended cluster size for each card capacity. Cards are designed to deliver best performance in those specific conditions.

Panasonic cameras whine on power-up that the card needs reformatting if the alignment is wrong, although they still work.

I have no idea if Windows tools get it right. I know Linux doesn't without some babysitting.