Micron Working On NVMe SSD Abrupt Shutdown Support For Linux

[Developer12]

Some (and the keyword here is some) SSD/NVMe devices have full power failure protection (sometimes referred to as Super Capacitors) to insure that any final writes are complete, and the device will not corrupt other flash memory lines or system state (sometimes effectively bricking the device or in loss of data on the device) when power fails. Those devices tend to Enterprise qualified devices, with the expected pricing premium. The compromise for consumer pricing in recent generation devices is that when power fails the devices will try to not corrupt the system state, but that says nothing about the state of the flash memory line (and the filesystem on it), although some vendors will market that (incorrectly) as handling power failures.

Having any additional hint to flush the buffers before the power is abruptly pulled can only be a good thing. Of course, not all controllers will properly support such, but if they do, your data may thank you.

Do you know what to look for when buying an SSD that has full power failure protection?

[barti_ddu]

Do you know what to look for when buying an SSD that has full power failure protection?

For the array of tantalum capacitors (or a large one if it is a SATA SSD), I think

[CommunityMember]

Admittedly, I don't have much experience in enterprise environments, but I feel like if your super-important system's emergency backup doesn't last long enough to do a normal shutdown, the solution shouldn't be to add a "mostly harmless" storage shutdown protocol

Under many jurisdictions EPO is required to shut down any/all power instantaneously, including backup power (UPS) systems (at least if you want them to be compliant, and if they are not, you may not get a certificate of occupancy).

And that does not even include the "stuff happens" issues where the entire power distribution system goes "WHHAAAAACK!" (in a very loud and wanting to be somewhere else kind of way)

In any case, the proper way to design a "super important system" to be reliable typically includes geographic diversity.

[Grayswan]

I am currently testing this patch to see if it wor

[mdedetrich]

Under many jurisdictions EPO is required to shut down any/all power instantaneously, including backup power (UPS) systems (at least if you want them to be compliant, and if they are not, you may not get a certificate of occupancy).

And that does not even include the "stuff happens" issues where the entire power distribution system goes "WHHAAAAACK!" (in a very loud and wanting to be somewhere else kind of way)

In any case, the proper way to design a "super important system" to be reliable typically includes geographic diversity.

Ontop of this, UPS's can also fail so you shouldn't 100% rely on them (although obviously you should always have UPS).

[CommunityMember]

Do you know what to look for when buying an SSD that has full power failure protection?

For Intel, you can go to https://ark.intel.com/content/www/us...5&0_EPLDP=True

I would presume other Enterprise vendors have something equivalent. And while I would not swear to the current status, for Samsung it used to be that the SSD and/or NVMe stick had to end in a 3 (so 863, 963, 983, etc.).

[stormcrow]

A laptop should know well in advance when it's going to run out of power. Besides, my laptop running Linux can do a full clean shutdown within four seconds.

"Should" know. Linux power management isn't the greatest thing out there. I've seen older laptops go from 10% to 1% in a few seconds. Bugs happen, software, firmware, and hardware. What do you have against the capability anyway? The safety of data is a major issue. Do you KNOW once your system shuts down all your data is written to your drive? I can guarantee, you don't know for sure on all hardware, and that's dogged filesystem developers for decades. Drive makers have been very sloppy over the years in signalling a flush to disk complete while still having data in onboard write cache, even during power down. There's a reason FreeBSD issues several buffer flushes to storage before it powers off on a normal shutdown sequence. Having a reasonably safe emergency shutdown capability exposed from hardware to OS would be outstanding for any system.

[ed31337]

Excessive capacitance can incur a sharp overcurrent situation during power on as the capacitors initially charge up. That high current can cause corona discharge inside switches and stress power delivery circuitry. In the case of USB attached storage, the USB specification has 10uF max capacitance allowed for bus powered devices. Even if the cost of "adding a capacitor" seems like it should not be much, the required additional circuitry to provide current limited soft-start of a high capacitance device adds up.

If your SSD contains capacitors, capacitors can break down over time. I've had many electronics gadgets wind up as dead bricks due to failing electrolytic capacitors taking out the power supply, which then takes out everything else down the line.

If software can take care of the safe shutdown issue instead, we can save money, board size, and weight upfront using cheaper SSDs that don't have excess capacitors. It should improve long term reliability from lack of failing caps, saving us a lot of time and headaches in the long run too.

[kcrudup]

i am currently testing this patch to see if it wor

~px~~px~~~~~~~p
NO CARRIER