On 27/09/10 10:50, Mark Morgan Lloyd wrote:
Henry Vermaak wrote:
On 27/09/10 08:50, Michael Schnell wrote:
On 09/25/2010 09:07 AM, Bo Berglund wrote:
After writing to the card, same performs the storage procedure
internally which takes a undefined amount of time and can't be watched
from outside. So you don't know how long you need to apply power to the
card after switching off the system. If you remove the power to the card
too early, _all_ (even unrelated) data on the card might be lost. This
is why using a special file system (a Flash File system and/or a
journaling file system) does not help.
See sync, fsync, fdatasync, etc. You can also mount some filesystems
with sync, but that's usually not a good idea for performance and
flash cycle issues. Best is to remove the power only after your kernel
has powered down, obviously, so you know that all the buffers have
been flushed.
I think I'd like to see a clarification from Michael here: did he mean
that the kernel/filesystem/drivers take an indefinite time committing
the data or that the card takes an indefinite time after receiving data
from the operating system?
The first of those can at least be initiated using a sync() call, and
it's reasonable to assume that a filesystem designed for the purpose
won't get screwed if the driver finds it can no longer talk to the
storage device. However if the storage device itself is susceptible to
errors caused by power removal that is far more sinister.
The only delay I can see (after fsync) would be due to the hardware
cache of the disk in use (or a defective drive). On linux you can call
hdparm to flush on-disk caches.
Henry
--
_______________________________________________
Lazarus mailing list
[email protected]
http://lists.lazarus.freepascal.org/mailman/listinfo/lazarus
