Alpha: bad unaligned access handling

Alpha: bad unaligned access handling

[Jan-Benedict Glaw]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset=us-ascii, Size: 1882 bytes --]

Hi!

With my currently installed ping (netkit-ping 0.10-6 from Debian Woody)
I get unaligned accesses:

ping(15953): unaligned trap at 00000001200030e4: 0000000120026b34 29 1
ping(15953): unaligned trap at 0000000120003110: 0000000120026b2c 29 2

The worse part is: they seem to be handled The Wrong Way:

[jbglaw@air:/home/jbglaw] $> ping -c 1 localhost
PING localhost (127.0.0.1): 56 data bytes
64 bytes from 127.0.0.1: icmp_seq=0 ttl=255 time=13.8 ms
wrong data byte #8 should be 0x8 but was 0xdc
        c d e f 10 11 12 13 14 15 16 17 18 19 1a 1b 1c 1d 1e 1f 20 21 22 23 24 25 26 27 28 29 2a 2b 
        2c 2d 2e 2f 0 0 0 0 0 0 0 0 0 0 0 0 

--- localhost ping statistics ---
1 packets transmitted, 1 packets received, 0% packet loss
round-trip min/avg/max = 13.8/13.8/13.8 ms


This is on a NoName Alpha box, running 2.4.0-test8-pre1 (with very good
uptimes), but I think 2.4.2-pre2 would do the same (wrong) things as
arch/alpha/kernel/traps.c wasn't really changed since ages... I don't
know much about alpha assemlbly language so somebody else should have
a look at that;( Btw., when such an unaligned trap has happened, I also
get transmit timeouts on my network card (which is:
00:0c.0 Ethernet controller: 3Com Corporation 3c905 100BaseTX [Boomerang]


NETDEV WATCHDOG: eth0: transmit timed out
eth0: transmit timed out, tx_status 00 status e000.
  Flags; bus-master 1, full 1; dirty 1470429(13) current 1470445(13).
  Transmit list 477a42d0 vs. fffffc00077a42d0.
  0: @fffffc00077a4200  length 80000042 status 00000042
...

MfG, JBG

-- 
Fehler eingestehen, Größe zeigen: Nehmt die Rechtschreibreform zurück!!!
/* Jan-Benedict Glaw <jbglaw@lug-owl.de> -- +49-177-5601720 */
keyID=0x8399E1BB fingerprint=250D 3BCF 7127 0D8C A444 A961 1DBD 5E75 8399 E1BB
     "insmod vi.o and there we go..." (Alexander Viro on linux-kernel)

[-- Attachment #2: Type: application/pgp-signature, Size: 240 bytes --]

Re: Alpha: bad unaligned access handling

[Carlos Carvalho]

Jan-Benedict Glaw (jbglaw@lug-owl.de) wrote on 14 February 2001 15:48:
 >With my currently installed ping (netkit-ping 0.10-6 from Debian Woody)
 >I get unaligned accesses:
 >
 >ping(15953): unaligned trap at 00000001200030e4: 0000000120026b34 29 1
 >ping(15953): unaligned trap at 0000000120003110: 0000000120026b2c 29 2
 >
 >The worse part is: they seem to be handled The Wrong Way:
 >
 >[jbglaw@air:/home/jbglaw] $> ping -c 1 localhost
 >PING localhost (127.0.0.1): 56 data bytes
 >64 bytes from 127.0.0.1: icmp_seq=0 ttl=255 time=13.8 ms
 >wrong data byte #8 should be 0x8 but was 0xdc
 >        c d e f 10 11 12 13 14 15 16 17 18 19 1a 1b 1c 1d 1e 1f 20 21 22 23 24 25 26 27 28 29 2a 2b 
 >        2c 2d 2e 2f 0 0 0 0 0 0 0 0 0 0 0 0 
 >
 >--- localhost ping statistics ---
 >1 packets transmitted, 1 packets received, 0% packet loss
 >round-trip min/avg/max = 13.8/13.8/13.8 ms
 >
 >
 >This is on a NoName Alpha box, running 2.4.0-test8-pre1 (with very good
 >uptimes), but I think 2.4.2-pre2 would do the same (wrong) things as
 >arch/alpha/kernel/traps.c wasn't really changed since ages...

I also get these, with 2.2.18pre5 (plus some Andrea patches) and
vanilla 2.2.19pre10 on a SMP UP2000.

Re: Alpha: bad unaligned access handling

[Sean Hunter]

On Wed, Feb 14, 2001 at 03:11:17PM -0200, Carlos Carvalho wrote:
> Jan-Benedict Glaw (jbglaw@lug-owl.de) wrote on 14 February 2001 15:48:
>  >With my currently installed ping (netkit-ping 0.10-6 from Debian Woody)
>  >I get unaligned accesses:
>  >
>  >ping(15953): unaligned trap at 00000001200030e4: 0000000120026b34 29 1
>  >ping(15953): unaligned trap at 0000000120003110: 0000000120026b2c 29 2
>  >
>  >The worse part is: they seem to be handled The Wrong Way:
>  >
>  >[jbglaw@air:/home/jbglaw] $> ping -c 1 localhost
>  >PING localhost (127.0.0.1): 56 data bytes
>  >64 bytes from 127.0.0.1: icmp_seq=0 ttl=255 time=13.8 ms
>  >wrong data byte #8 should be 0x8 but was 0xdc
>  >        c d e f 10 11 12 13 14 15 16 17 18 19 1a 1b 1c 1d 1e 1f 20 21 22 23 24 25 26 27 28 29 2a 2b 
>  >        2c 2d 2e 2f 0 0 0 0 0 0 0 0 0 0 0 0 
>  >
>  >--- localhost ping statistics ---
>  >1 packets transmitted, 1 packets received, 0% packet loss
>  >round-trip min/avg/max = 13.8/13.8/13.8 ms
>  >
>  >
>  >This is on a NoName Alpha box, running 2.4.0-test8-pre1 (with very good
>  >uptimes), but I think 2.4.2-pre2 would do the same (wrong) things as
>  >arch/alpha/kernel/traps.c wasn't really changed since ages...
> 
> I also get these, with 2.2.18pre5 (plus some Andrea patches) and
> vanilla 2.2.19pre10 on a SMP UP2000.

This is an application problem, not a kernel one.  You need to upgrade your
netkit.

Sean

P.S.  I wrote a small wrapper to aid in the debugging of unaligned traps, which
I'll send to anyone who's interested.

> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

Re: Alpha: bad unaligned access handling

[Carlos Carvalho]

Sean Hunter (sean@dev.sportingbet.com) wrote on 14 February 2001 17:26:
 >This is an application problem, not a kernel one.  You need to upgrade your
 >netkit.

Yes, I was quite confident of this. However, unaligned traps are a
frequent problem with alphas. For a looong time we had zsh produce
lots of it, to the point of making it unusable. Strangely, the problem
disappeared without changing anything in zsh. It was either a library
or kernel problem.

 >P.S. I wrote a small wrapper to aid in the debugging of unaligned
 >traps, which I'll send to anyone who's interested.

I'd like it!

Re: Alpha: bad unaligned access handling

[Sean Hunter]

On Wed, Feb 14, 2001 at 03:38:33PM -0200, Carlos Carvalho wrote:
> Sean Hunter (sean@dev.sportingbet.com) wrote on 14 February 2001 17:26:
>  >This is an application problem, not a kernel one.  You need to upgrade your
>  >netkit.
> 
> Yes, I was quite confident of this. However, unaligned traps are a
> frequent problem with alphas. For a looong time we had zsh produce
> lots of it, to the point of making it unusable. Strangely, the problem
> disappeared without changing anything in zsh. It was either a library
> or kernel problem.

Definitely library, I'd think.

> 
>  >P.S. I wrote a small wrapper to aid in the debugging of unaligned
>  >traps, which I'll send to anyone who's interested.
> 
> I'd like it!
> 

OK, my alpha is a sick bunny at the moment, so I'll have to wait until I get
home (so I can see why I can't ssh to it).  What the wrapper does is set some
settings so your program gets sigbus when it generates an unaligned trap, and
then runs your program in gdb so gdb helpfully stops at the line which
generated the trap.  It goes without saying you need to build the program in
question with debugging symbols so that you see the code.

You then need to fix the unaligned access.  This sometimes requires real alpha
guruhood (Which I do not possess, but Richard Henderson or Michal Jagerman do,
if you need advice), but sometimes simply requires adding __attribute__
((__unaligned__)) to a struct member in a c file.

Sean