NVFS devices are still crashing in Plucker occasionally. About half a
year ago I've put a lot of effort into making Plucker NVFS-safe by
removing all calls to DmQueryRecord() except those where the handle is
used close to immediately, and replacing them with
DmGetRecord()/DmReleaseRecord() pairs. There is code in ramfile.c that
keeps track of handles, taking care of nesting, etc.
But now I have a worry. I was assuming that one could count on an
unlocked handle fetched by DmGetRecord() remaining valid until the
DmReleaseRecord() call because I was assuming that NVFS would know about
this. But I saw something somewhere about how one can't count on
DmGetRecord(), either. Does anybody know if this is true?
Fortunately, if it IS true, it might be fixable by simply locking
records. And since all the long-term record use goes through ramfile.c,
this is not a very difficult task. But I don't want to do this unless I
have to.
Frankly, I don't understand why this problem exists at all. Surely Palm
could have implemented NVFS to keep track of a ton of fake handles, and
if the data that the handle was pointing to got kicked out of dbcache,
it could get loaded back from flash/HD as soon as MemHandleLock() was
called on it. Maybe the handles could be like the fake handles to ROM
stuff that don't persist between apps, but it would still help a lot.
Alex
p.s. I'm working on an NVFS backup program that backs up data direct
from the NVFS internal storage, without looking at the RAM copy at all.
This should make the backup much more complete and reliable. For a
beta, see:
http://www.1src.com/forums/showthread.php?t=109500
_______________________________________________
plucker-dev mailing list
plucker-dev@rubberchicken.org
http://lists.rubberchicken.org/mailman/listinfo/plucker-dev
