From mboxrd@z Thu Jan  1 00:00:00 1970
To: 9fans@cse.psu.edu
From: nemo@gsyc.escet.urjc.es
MIME-Version: 1.0
Content-Type: multipart/mixed;
	boundary="upas-xpcnjxxsvsafmlufobudibijwt"
Message-Id: <20001124140220.2550119A07@mail.cse.psu.edu>
Subject: [9fans] [Q] are the kernel notes decent enough?
Date: Fri, 24 Nov 2000 15:04:32 +0000
Topicbox-Message-UUID: 31991c9c-eac9-11e9-9e20-41e7f4b1d025

This is a multi-part message in MIME format.
--upas-xpcnjxxsvsafmlufobudibijwt
Content-Disposition: inline
Content-Type: text/plain; charset="US-ASCII"
Content-Transfer-Encoding: 7bit

This is a question for the authors of the system.

I would love to distribute the notes both through the web and by
getting it in /usr/doc within the distribution.

I wouldn't like to distribute the document if authors of the code do
not consider the notes as good enough.  That's because I wouldn't like
people that didn't read the kernel code before reading the notes to
think that the kernel is not so good because of my faults while
describing it.

Do you think that this is good enough to be distributed?  If that's
the case, what do you think about putting this thing into /sys/doc/?
If that's not the case, just drop me a line about how to improve the
document.

Of course, distribution will have to wait in any case until I iterate
through the last chapters fixing everything I can notice as wrong.
But probably I'll do this in  2 or 3 of weeks.

thanks, and sorry to bother you.


--upas-xpcnjxxsvsafmlufobudibijwt
Content-Type: message/rfc822
Content-Disposition: inline

                         Fixed since the first version
                                       
   Since the last version of this document, these things have been fixed.
     * Introduction chapter:
          + It now says which kernel tree the notes refer to (3rd
            edition; june release).
          + Figures added for Intel stuff
          + Brief description of assembler.
     * Starting up chapter: I consider this chapter as fixed now.
          + Note about why using 9load.
          + 9load jumps to tokzero as I was saying, but that does not set
            the PC to KZERO.
          + fixes in 9load probe routine.
          + fixes about the mapping for the Mach structure.
          + fixes in ram scanning.
          + Small section about PC hardware added.
          + Horrendous bug about how do QIDs work fixed (I said that vers
            was used to detect removed-and-recreated files, which is not
            the case; In fact I think I understood that was not the case
            long ago before starting this document. Too sleepy that day?
            Too few caffeine? Who knows?).
          + pageinit has nothing to do with software for paging, it is
            just initializing the page allocator. The starting up chapter
            was saying it had to do with software MMU, although on the
            next paragraph it was described and the reader could know
            what it really does.
          + psstate only holds the process state for ps.
          + process groups have name spaces, not processes. The
            description was confussing.
          + Some figures added.
          + Note added about the check for pidalloc wrap.
          + Note about why to reserve room in the bottom of the kernel
            stack.
          + Fix about eflags in touser.
     * Process chapter: still fixing it.
          + Note added about why to use a scheduler stack.
          + Note added about why to have two scheduling classes.
          + Note added about other synch. means.
          + postnote fix.
          + Note for wakeup in pexit.
          + procstopwait fix.
          + Some figures added.
     * Files chapter: still fixing it.
          + Note added about why to have a limit on file descriptors.
          + Non-empty directories can be removed.
          + Hint about saveregisters.
          + The example for ``..'' now uses walk and not cd, which was
            unfortunate because cd does not crosses the mount point.
          + Note about why a read from a union must clone the mounted
            channels anyway. The comment could be suggesting that the
            clone could be avoided, that is not the case.
          + Note about how to avoid the CHDIR trick for exportfs.
          + Nested calls to the mount driver are not a problem; msg sizes
            are.
          + Note about coherence of distributed caches.
          + Note about using just pages to do caching.
     * Memory chapter: still fixing it.
          + Note about why to use just physical memory for caching remote
            files.
          + Note about why the pager avoids waiting for locks.
          + Note about the lock+unlock when paging-in.
          + Note fix about the use of share while cloning segments.
          + Note fix about the check for ESEG while attaching segments.
       
--upas-xpcnjxxsvsafmlufobudibijwt--