[9fans] OT: scsi sense error code hardware error 40 83?

[9fans] OT: scsi sense error code hardware error 40 83?

[Axel Belinfante]

after moving my fake worm kenfs there appears to be
a problem with the main disk.
(the fs kernel doesn't even get to probing the other disks)

the last succesful boot told me the disk is a
Thu Dec  8 15:29:58: scsi#0.0: QUANTUM XP34550S        LXQ1184630312388

as far a I could find the '40 83' means:
    DIAGNOSTIC FAILURE ON COMPONENT NN (80h-FFh)
but I could not find more details.

I still have to try to run some diagnostic program (scsimax(?))
just curious if anybody instantly recognizes the code.



related question (that is here under the wrong heading, but ok)
assuming a kenfs file server with config

filsys main cp(w0)0.50f(w5w1)

now assuming I lost only the w0, but the w5w1 making
up the fake worm are still there, am I right in assuming
that the correct procedure to resurrect the fs would be:
 - get a new disk to replace w0
   (no problem if it differs from the original?)
 - boot the fs kernel
 - enter config mode
 - in config mode enter the usual stuff, as when initializing
   the fs as explained in the fsconfig man page, EXCEPT
 - where the examples say 'ream main' do 'recover main'
right?


These are the boot messages I get (thanks to consolefs)
Tue Oct  3 15:19:37: sysinit
Tue Oct  3 15:19:37: config w0
Tue Oct  3 15:19:37: 	devinit w0
Tue Oct  3 15:19:37: scsi#0.0: unavailable, status 2
Tue Oct  3 15:19:37: 	drive w0:
Tue Oct  3 15:19:37: 		8890759 blocks at 512 bytes each
Tue Oct  3 15:19:37: 		1111345 logical blocks at 4096 bytes each
Tue Oct  3 15:19:37: 		8 multiplier
Tue Oct  3 15:19:37: scsi#0.0: reqsense: 'not ready' code #40 #83
Tue Oct  3 15:19:37: scsi#0.0: byte 2: #02, bytes 15-17: #00 #00 #00
Tue Oct  3 15:19:37: lastcmd (6):  #00 #00 #00 #00 #00 #00
Tue Oct  3 15:19:37: scsi#0.0: unavailable, status 2
Tue Oct  3 15:19:37: scsi#0.0: reqsense: 'hardware error' code #40 #83
Tue Oct  3 15:19:37: scsi#0.0: byte 2: #04, bytes 15-17: #00 #00 #00
Tue Oct  3 15:19:37: lastcmd (10):  #28 #00 #00 #00 #00 #00 #00 #00 #08 #00
[this repeats for a while until it gives up:]
Tue Oct  3 15:19:39: wrenread: w0(0) bad status 0002
Tue Oct  3 15:19:39: stack trace of 2
[stack trace omitted]
Tue Oct  3 15:19:40: 1220 stack used out of 16000
Tue Oct  3 15:19:40: 
Tue Oct  3 15:19:40: panic: config io


Axel.

Re: [9fans] OT: scsi sense error code hardware error 40 83?

[geoff]

I can't help with the sense code; I suspect that the component number
(NN) is specific to the manufacturer and device.

As for replacing your cache disk, yes, your plan sounds right.  Since
your entire cache is being replaced, there is no memory anywhere else
of how big it was, so you can replace it with a disk or disks of any
size.  "recover main" rebuilds your cache disk from the (fake) worm.

Re: [9fans] kenfs config? (was: OT: scsi sense error code hardware error 40 83?)

[Axel Belinfante]

> I can't help with the sense code; I suspect that the component number
> (NN) is specific to the manufacturer and device.

I suspected that too. unfortunately, even though the manufactor site
still lists the disk and has info like jumper settings, it doesn't
have those details. maybe their 'scsimax' thing will tell me something.

> As for replacing your cache disk, yes, your plan sounds right.  Since
> your entire cache is being replaced, there is no memory anywhere else
> of how big it was, so you can replace it with a disk or disks of any
> size.  "recover main" rebuilds your cache disk from the (fake) worm.

thanks for confirming.


different question:
what is the safe procedure to add one (or more?) mirror disks to the
fake cached worm of an existing kenfs? like one mirror disk for the
fake worm, and a different one to the cache?

I gues your answer above implies that for the cache it would be ok to
just configure the new filesys with mirror-ed cache and then recover.

The fsconconfig man page seems to suggest that for the mirror
of the fake worm it would also be sufficient to just specify
the mirror, with the existing fake worm device at the left.

	{device...}
		A pseudo-device formed from the mirroring of the first
		device in the list onto all the others. [...]

Is this right? or is an explicit copydev of the fake worm needed
(or advisable?)


one more:
when extending an existing config say on w0, one does
  config w0    (identify config location + erase current)
  mention all old config items
  add new config items
  end
right? or is there a way to just extend an existing config?


one last question, mere curiosity:
fsconfig(8) also says:
    Copydev will copy the device from-dev to the device to-dev.
    block by block, and panic.

why the panic? to be allowed (forced?) to reboot and config?


sorry for all these questions, but configuring a kenfs is not
something I (any of us?) daily do, and experience some years
ago taught me to be a bit careful (yeah, it's in the archives) 
(not that I learned enough to set up mirroring before I need it :-(

Thanks, a lot!
Axel.

Re: [9fans] kenfs config? (was: OT: scsi sense error code hardware

[geoff]

Right, you could just configure the cache as mirrored disks and
recover main.  The mirror device does not copy new devices when they
are added to a mirror; you'd need to copydev first (or boot as a cpu
server and use dd) to copy the original disk to its mirror disks).

You should only use "config w0" when you initialise the configuration.
Thereafter, the file server kernel reads that configuration (from w0
in this case, which name it gets from plan9.nvr) at start-up and
merges any changes due to config commands that you then type.  Typing
a "filsys" command for an existing file system will replace the old
definition; "filsys scratch" would delete any definition for a file
system named "scratch"; and declaring a file system with a name as yet
unknown will add the file system.

The panic after copydev is indeed to force a reboot and give an
opportunity for reconfiguration.

I understand your caution.  There was a period when I was configuring
kenfs daily or pretty close to it, but that's certainly unusual.

Re: [9fans] kenfs config? (was: OT: scsi sense error code hardware

[Axel Belinfante]

> Right, you could just configure the cache as mirrored disks and
> recover main.  The mirror device does not copy new devices when they
> are added to a mirror; you'd need to copydev first (or boot as a cpu
> server and use dd) to copy the original disk to its mirror disks).

finally I had time to look at my fs.
seems I'm lucky: the fake worm is ok.

I've replaced the cache disk and recovered; I encountered
'panic: fworm: rbounds 13219302' (my fs kernel is a bit old (2003))
but the fix mentioned here a while ago (fix conf.firstsb) worked.

since I wanted a backup (and if possible: mirror) 'right now' even
though my /sys/src/fs is out of date and does not yet have copydev,
and I don't want to pull a lot of changes from sources right now,
I tried copyworm
from:  filsys main cw0f(w5w1)
to:    filsys output f(w2)

this seems to have worked, but I have not yet accessed the result
(just switched off the machine while it looped after copyworm).

now my real question is:
would it be ok to use the copy f(w2) as mirror,
as in    filsys main cw0{f(w5w1)f(w2)}    ?

Axel.

Re: [9fans] kenfs config? (was: OT: scsi sense error code hardware

[geoff]

The disks in a mirror need to start out with identical
contents.  If (w5w1) is exactly the same size as w2,
I think you'll be okay.  If they aren't exactly the
same size, the fake-worm blocks-written bitmaps should
be in different places, thus it won't work right.

If you can boot your file server as a CPU server,
you can use dd to copy disks.

Re: [9fans] kenfs config? (was: OT: scsi sense error code hardware

[Axel Belinfante]

> The disks in a mirror need to start out with identical
> contents.  If (w5w1) is exactly the same size as w2,
> I think you'll be okay.  If they aren't exactly the
> same size, the fake-worm blocks-written bitmaps should
> be in different places, thus it won't work right.

They are not the same size.

Just checking (and increasing, with your help :-) my understanding:

It isn't the case that in a config of cw0{f(w5w1)f(w2)}
there are two fake-worm pseudo devices that each take care of
their own bitmaps, which thus need not be at the same place?
Or is that a rather weird config, and should in a
c<dev1><dev2> config the <dev2> always directly be a worm
or fake-worm device?

Whereas in a config of cw0f{(w5w1)w2} there is a single
fake-worm pseudo device and thus single bitmap, stored
(I guess) at the same place on resp. (w5w1) and w2?
(position will depend on computed size of {(w5w1)w2}
 I guess, which would be the minimum of sizes of (w5w1) and w2)


> If you can boot your file server as a CPU server,
> you can use dd to copy disks.

I did not yet explore that option - have to fiddle a bit to
take root filesystem from somewhere (my 'selfcontained'
cpu/auth server, or laptop, or maybe sources). Will work out ok.
I've also considered booting a live cd as a 'rescue disk' -
but then I'd have to connect a cd drive first... :-)


Thanks (a lot!) for your help so far - it's much appreciated!
Axel.

Re: [9fans] kenfs config? (was: OT: scsi sense error code hardware

[geoff]

Good point; I was thinking of cw0f{(w5w1)w2}.  cw0{f(w5w1)f(w2)}, or
equivalently cw0{f(w5w1)fw2}, should work if (w5w1) is smaller than
w2.  The mirror device will limit both devices to the size of the
smaller one, so if w2 is smaller than (w5w1), f(w5w1) will be
truncated in the mirror {f(w5w1)fw2}.

There's still the question of how to initialise the second device in
the mirror-to-be.  I would have copyworm-ed f(w5w1) to f(w2).  I'm
nervous that copying from cw0f(w5w1) may not yield exactly the same
bytes as copying from f(w5w1).

Re: [9fans] kenfs config? (was: OT: scsi sense error code hardware

[Axel Belinfante]

> There's still the question of how to initialise the second device in
> the mirror-to-be.  I would have copyworm-ed f(w5w1) to f(w2).  I'm
> nervous that copying from cw0f(w5w1) may not yield exactly the same
> bytes as copying from f(w5w1).

Ok, thanks! I've tried looking at the code but (at this time of day)
initially only got more confused - /sys/src/fs/port/config.c:^wormof
seems not get the worm part of a cw, which is what I would expect it
to do (instead, it seems to take are of a cw within an fworm?).


I then realized that the copyworm code I run today was older:
  /n/sourcesdump/2003/0314/plan9/sys/src/fs/port/config.c 
it did copy the  f(w5w1) part of cw0f(w5w1) (the written blocks).

  copying worm from f(w5w1) to fw2, starting in 8 seconds
  limit 1505835
  copying worm
  fworm: read 0
  0 not written
  fworm: read 1
  1 not written
  block 20000 Tue Oct 17 15:57:26 2006
  [...]
  copied 1505835 blocks from f(w5w1) to fw2


Even earlier today I had tried the same with the slightly newer kernel
(code from march 14 2003) which I had patched to allow recover to work,
and then it not only did not give exactly the same bytes, but it failed:

  copying worm from cw0f(w5w1) (worm cw0f(w5w1)) to fw2, starting in 8 seconds
  devsize(cw0f(w5w1)) = 1505936
  fworm: read 0
  i/o error reading cw0f(w5w1) block 0
  panic: no blocks to copy on cw0f(w5w1)

I did not compare in detail, but the code seems to be essentially
the same as in the current version on sources.
my first superblock seems to be block 2 - that explains fworm: read 0?

related to the 'fworm of mirror' vs. 'mirror of fworms':
I guess that copyworm would not be able to deal with the case of
'cached worm where worm is mirror of fake worms' (or would it?)

Axel.
(turns out I have to redo the copy anyway because confused by
the big numbers of half-k and 4k blocks I chose a too small w2 disk
that won't allow much growth - but is was big enough to hold the
written part of f(w5w1) so I guess at least I have a backup)

[9fans] gs

[Charles Forsyth]

as tedious as waiting for gs to do something is waiting
for db to load all its symbols to debug it not doing something.
if they ever find another candidate for a planet to replace pluto,
they should call it `ghostscript'.  it will be big enough to qualify.

out of idle interest, why has there never been a replacement?
is it really that hard?

Re: [9fans] gs

[geoff]

In first edition, there was a home-grown postscript interpreter, psi,
but it was replaced by gs in second edition.  The degree of difficulty
depends upon what `it' is.  gs can interpret and emit Postscript or
PDF of varying versions, and includes a substantial set of Postscript
fonts.  As I recall, just getting a Postscript interpreter to render
reasonably quickly (er, less than pessimally slowly) is quite a job.
PDF seems to be an attempt to rein in the full generality of the
Postscript language.

Perhaps the people who make Forth interpreters could make a fast
Postscript interpreter, though I suspect that graphics is the
bottleneck.

Re: [9fans] gs

[Francisco J Ballesteros]

I'm curious. any other interpreter that could be ported?

On 10/3/06, Charles Forsyth <forsyth@terzarima.net> wrote:
> as tedious as waiting for gs to do something is waiting
> for db to load all its symbols to debug it not doing something.
> if they ever find another candidate for a planet to replace pluto,
> they should call it `ghostscript'.  it will be big enough to qualify.
>
> out of idle interest, why has there never been a replacement?
> is it really that hard?
>
>
>

Re: [9fans] gs

[Jack Johnson]

On 10/3/06, Charles Forsyth <forsyth@terzarima.net> wrote:
> out of idle interest, why has there never been a replacement?
> is it really that hard?

That's a great question.  I Googled for some leads:

"Saffron includes an all-Java PostScript(r) interpreter"
http://www.dynalivery.com/customersupport/Saffron_FAQs/SaffronInfo.html

"The RoPS interpreter is a Microsoft Windows implementation of the
PostScript programming language interpreter"
http://www.rops.org/

There is some interesting detail stuff here that I'd want to re-read
if I were going to write an interpreter from scratch:
http://www.anastigmatix.net/postscript/direct.html

This might also make for a good jumping-off point:
http://serl.cs.colorado.edu/~arcadia/Software/tps.html

...but from the looks of it, maybe it would be easier to "fix"
ghostscript by making it faster/slimmer/etc.  I wonder if it runs
better on some architectures than others?  If so, sounds like a good
job for a cpu server.

-Jack

Re: [9fans] gs

[LiteStar numnums]

[-- Attachment #1: Type: text/plain, Size: 2031 bytes --]

You can scratch RoPS of that list: anything that it does do, it not only
does it slowly, but improperly. To make matters worse, 'paid' version is no
better.
I had a discussion with Manfred von Thun (you know, the gentleman who
created Joy) about this while I was working on a language that was similar
to postscript & joy. He was interested in the design aspects we were
discussing, but was hesitant about the PS aspects I was using. I should
think that the real reason that PostScript is so slow is that the graphics
rely upon someone who should have really optimised some the workings out of
their source file...
Just my $0.02 (or 0.01 GBP)

On 10/3/06, Jack Johnson <knapjack@gmail.com> wrote:
>
> On 10/3/06, Charles Forsyth <forsyth@terzarima.net> wrote:
> > out of idle interest, why has there never been a replacement?
> > is it really that hard?
>
> That's a great question.  I Googled for some leads:
>
> "Saffron includes an all-Java PostScript(r) interpreter"
> http://www.dynalivery.com/customersupport/Saffron_FAQs/SaffronInfo.html
>
> "The RoPS interpreter is a Microsoft Windows implementation of the
> PostScript programming language interpreter"
> http://www.rops.org/
>
> There is some interesting detail stuff here that I'd want to re-read
> if I were going to write an interpreter from scratch:
> http://www.anastigmatix.net/postscript/direct.html
>
> This might also make for a good jumping-off point:
> http://serl.cs.colorado.edu/~arcadia/Software/tps.html
>
> ...but from the looks of it, maybe it would be easier to "fix"
> ghostscript by making it faster/slimmer/etc.  I wonder if it runs
> better on some architectures than others?  If so, sounds like a good
> job for a cpu server.
>
> -Jack
>



-- 
If work and leisure are soon to be subordinated to this one utopian
principle -- absolute busyness -- then utopia and melancholy will come to
coincide: an age without conflict will dawn, perpetually busy -- and without
consciousness.

-- Günter Grass

[-- Attachment #2: Type: text/html, Size: 2706 bytes --]

Re: [9fans] gs

[Andy Newman]

LiteStar numnums wrote:
> I should think that the real reason that PostScript is so slow is
> that the graphics rely upon someone who should have really optimised
> some the workings out of their source file...

In a previous life I worked on graphics systems similar in nature
to Postscript (language + graphics lib).  Profiling such things,
including older versions of gs, shows a general pattern of around
50% of the time in an average render (for some value of average)
is spent writing runs of pixels to the output image.  The memory
ops hurt a lot. When accelerating these things (i.e. throwing h/w
at it) typically the first thing you want to do is have the h/w
generate pixel runs from some run-length representation used in
the s/w.  Increased device spatial and color resolutions just make
this more true with the associated increase in memory requirements
for images.  Modern CPU/memory characteristics may have altered
that 50% figure, it's been a while since I measured it, but I don't
see it being vastly different (famous last words :).  The nature
of image rendering, and RIPs, place high demands on memory systems
and it is easy to NOT get good rendering throughput solely due to
memory related issues.  Also the language interpreter itself can
be a bottleneck if not optimized for handling large inputs.  The
programs these interpreters generally process are very large but
relatively simple - few loops, some subroutines but lots and lots
(many millions if not tens of millions) of lexemes to read.

Re: [9fans] gs

[jmk]

On Tue Oct  3 16:49:10 EDT 2006, knapjack@gmail.com wrote:
> ...
> ...but from the looks of it, maybe it would be easier to "fix"
> ghostscript by making it faster/slimmer/etc.  I wonder if it runs
> better on some architectures than others?  If so, sounds like a good
> job for a cpu server.
> 
> -Jack

As we get older, faster/slimmer/etc. gets harder and harder
and eventually we just buy bigger jeans.

Re: [9fans] gs

[Bruce Ellis]

writing a postscript interpreter that kinda works is not a very
hard task.  writing a ps interpreter that gets the graphics right,
particularly colour, is very tedious and takes a lot of man hours.
optimizing the interpreter so it spits out >20 pages/min is not fun.
i've been involved in the writing of three ps interpreters.
i believe there are still printers in the wild, and even being sold,
that have the undocumented %brucee inbuilt file (my 48x48x1,
a one liner can image it full-page).

brucee

On 10/4/06, jmk@plan9.bell-labs.com <jmk@plan9.bell-labs.com> wrote:
> On Tue Oct  3 16:49:10 EDT 2006, knapjack@gmail.com wrote:
> > ...
> > ...but from the looks of it, maybe it would be easier to "fix"
> > ghostscript by making it faster/slimmer/etc.  I wonder if it runs
> > better on some architectures than others?  If so, sounds like a good
> > job for a cpu server.
> >
> > -Jack
>
> As we get older, faster/slimmer/etc. gets harder and harder
> and eventually we just buy bigger jeans.
>