home *** CD-ROM | disk | FTP | other *** search
- Date: Thu, 16 Dec 93 20:01:49 MST
- From: shenson@nyx10.cs.du.edu (Stephen Henson)
- Message-Id: <9312170301.AA21349@nyx10.cs.du.edu>
- X-Disclaimer: Nyx is a public access Unix system run by the University
- of Denver. The University has neither control over nor
- responsibility for the opinions or correct identity of users.
- To: mint@terminator.rs.itd.umich.edu
- Subject: Updates and things ...
-
- Greetings,
- the Minixfs updater does one single Dcntl for MFS_SYNC. This
- function is analagous to Unix sync() in that all drives are handled at
- once. The reason for the path is because there is no way to send info
- to a fileystem without a path of an 'owned' drive. However the update
- demon needs access to Minixfs internals to be sure of finding the
- drive and avoiding diskette whirring (which is caused by the calling of
- routines which determine the filesystem and by the brain damaged hardware
- which kludges disk changes of write protected drives). This sync operation
- also cannot be interrupted by a context switch, so the update daemon is
- run as a user process which makes a syscall.
-
- [side note, from the above, you only need call MFS_SYNC once to sync all
- filesystems (on the first filesystem MFS_VERIFY returns the right magic
- number). The preferred method though is sending anything other than
- SIGSTOP and SIGKILL to the update process, this avoids whirring]
-
- This is not ideal, but the performance increase is so great that
- I felt that some method compatible with the various older MiNT's out there
- was more approprate than something which relied on a (possibly unofficial)
- kernel patch. Besides, I leave the kernel alone unless I have no choice.
- If we want a single update process (or none at all) then I would
- suggest either a new entry in the FILESYS structure (sync?) which is called
- periodically, or some way of specifying an internal function to be called
- at periodic intervals. The latter is perhaps preferable because it allows
- more flexibility (you may not want all update processes to be called at
- the same interval). I'd certainly support this if it ever became 'official'.
- Ah, one more thing, I can make the cache mode easily work on a
- 'per filesystem' basis, however I don't think it wise to make floppies
- permanently not use turbo mode; the performance loss is considerable.
-
- Regards and Seasons Greetings, Steve.
-
-
-