9fans - fans of the OS Plan 9 from Bell Labs
 help / color / mirror / Atom feed
From: Geoff Collyer <geoff@collyer.net>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] file server page faulting
Date: Sat, 14 Sep 2002 19:22:18 -0700	[thread overview]
Message-ID: <07e1c8ac664d33b92935754d57baf26a@collyer.net> (raw)

> 1. the S3 is sharing the memory; there is no allowance for this, and
>    when the buffer cache meets the frame buffer horrible things happen

My main file server has the SiS 630E chipset in it, which includes a
built-in VGA chipset that shares host memory.  The motherboard manual
says it's the SiS 300 AGP VGA.  I added a trivial "pcihinv" command
that just calls the otherwise-unused pcihinv function
(/sys/src/fs/pc/pci.c:/^pcihinv).  Here's what pcihinv had to say:

 bus dev type vid  did intl memory
 [ ... ]
 1   0/0 0300 1039 6300 11  0:f0000008 134217728 1:eb000000 131072 2:0000a801 128

Aperture size and shared memory size may be embedded there somewhere.
According to vendor and device ids, this is an SiS 630 integrated AGP
VGA at 0xa800.  Who knows?  The SiS 300 has a different device id
(0x300) but some software (e.g., XFree86) treats the 300 and 630 (and
540) similarly.  SiS's numbering seems to be intended to obfuscate.

I'd been worried about the shared memory too and when I first set up
the file server, had odd crashes shortly after boot, so I inserted a
PCI VGA card into one of the whopping two (002) PCI slots, and told
the BIOS to use it, not the built-in VGA.  Eventually, the odd crashes
turned out to be an uninitialised FPU, but even after fixing that, I
was uneasy about the built-in VGA and so left the PCI VGA card in.
However, I have gradually come to the conclusion that, at least for my
VGA chipset, it's not a problem for these reasons:

- the file server never enables VGA mode but runs the console in CGA
mode, which doesn't need much memory anyway (it can only display 2,000
characters),

- my BIOS reports that the system has 510MB of normal memory + 2MB of
shared video memory (it has 512MB of RAM), now that I've cut the
shared memory size down to 2MB in the BIOS (and the aperture down to
4MB), and

- prints that I added report that memory bank1 runs up to 0x1fe00000
(534773760 = 510MB exactly), so I assume that the BIOS is claiming
only 510MB of RAM or the memory sizing code detects that the memory at
510MB through 512MB behaves strangely and so excludes it.

I then built an fs kernel with devwrite() stubbed out, removed the PCI
VGA card, booted the no-writes fs kernel, read well over 512MB of
distinct data, checked all the file systems (they are roughly 20GB or
40GB each), and everything was still behaving.  The checks are
significant only because they consume RAM.  After running for a while
longer, I then booted my normal fs kernel and things are still okay.

	last boot Thu Sep 12 02:18:44 2002

It's now been over two days, and it's still behaving.


             reply	other threads:[~2002-09-15  2:22 UTC|newest]

Thread overview: 21+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2002-09-15  2:22 Geoff Collyer [this message]
  -- strict thread matches above, loose matches on Subject: below --
2002-09-16 15:36 nigel
2002-09-15  7:21 nigel
2002-09-13 15:57 Fco.J.Ballesteros
2002-09-13 15:37 nigel
2002-09-13 16:37 ` Andrew
2002-09-13 15:14 nigel
2002-09-13 15:23 ` Andrew
2002-09-14  2:09 ` Andrew
2002-09-16  9:47   ` Douglas A. Gwyn
2002-09-16 15:13     ` Jack Johnson
2002-09-16 15:30       ` Andrew
2002-09-13 15:06 Fco.J.Ballesteros
2002-09-13 15:28 ` Andrew
2002-09-12  6:57 nigel
2002-09-13 15:01 ` Andrew
2002-09-12  4:50 Andrew
2002-09-12 14:04 ` FJ Ballesteros
2002-09-14 14:27 ` Andrew
2002-09-14 23:48 ` Andrew
2002-09-15 23:51 ` Andrew

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=07e1c8ac664d33b92935754d57baf26a@collyer.net \
    --to=geoff@collyer.net \
    --cc=9fans@cse.psu.edu \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).