Linux 4.1 To Bring Support For NCQ Autosense

[phoronix]

Phoronix: Linux 4.1 To Bring Support For NCQ Autosense

While usually not presenting any major features each release cycle, the libata feature pull request for Linux 4.1 is a bit more interesting this time around...

http://www.phoronix.com/scan.php?page=news_item&px=Linux-4.1-NCQ-Autosense

[uid313]

But this data needs to be relayed to the user too.
Not just outputted to dmesg or syslog.

[Milan Kerslager]

Phoronix: Linux 4.1 To Bring Support For NCQ Autosense

While usually not presenting any major features each release cycle, the libata feature pull request for Linux 4.1 is a bit more interesting this time around...

The admin is in the first place to take care of it. In any case, there is kdbus which will be in the mainline so this could be easy to deliver this sort of messages to the user even into the GUI.

[BradN]

But this data needs to be relayed to the user too.
Not just outputted to dmesg or syslog.

What about reporting it to a cache layer to lock some lost writes in the cache instead of just totally losing data? Like that will ever happen

[gilboa]

What about reporting it to a cache layer to lock some lost writes in the cache instead of just totally losing data? Like that will ever happen

I doubt that you understand the complexity of what you are asking.
Lock the memory and do -what-?

- Gilboa

[uid313]

I doubt that you understand the complexity of what you are asking.
Lock the memory and do -what-?

- Gilboa

Well you stream/write the memory a virtual file (not to disk) on the file system.
Such as /sys/fs/failed_writes, then you have a software reads the virtual file and;
A) Writes it to a external backup drive.
B) Sends it over the network to a remote machine.
C) Informs the user, emails someone, or writes something to a log.

[gilboa]

Well you stream/write the memory a virtual file (not to disk) on the file system.
Such as /sys/fs/failed_writes, then you have a software reads the virtual file and;
A) Writes it to a external backup drive.
B) Sends it over the network to a remote machine.
C) Informs the user, emails someone, or writes something to a log.

... A disk failure - I mean full disk failure - will be useless without an application layer that understands what exactly was updated and where.
Writing a couple of random pages into a couple of random files sitting in some random remote location is beyond useless.
Far better to inform the application layer that the write has failed, and let it handle it cleanly.

The mere fact that some applications are stupid enough to assume that every write succeeds without implementing proper error handling is not the kernel's fault.

- Gilboa

[BradN]

Sorry for the late reply, I should have been more specific. I was referring to like bcache or things like that which have a permanent memory backed cache. Like, for utilizing the flash based cache some spinning drives come with, or using an SSD to boost a mechanical drive.