Re: zeroes read back more often than appended

[Pat LaVarre <p.lavarre@ieee.org>]

> From: Ben Fennema
>
> I'll see if I can figure out
> whats going on in 2.6.0-test6-smp.
>
> Somehow unrecorded block(s) are getting
> inserted in the data.

Intriguing, thank you.  Two questions immediately occur to me:

1) Do you mean to say you already have seen 2.6.0-test6-smp fail?

I ask because, if you say yes, then I should stop investing more time
into finding a second pc where I can reproduce my observations.

2) Do you mean to say you have seen the metadata of a failure, and you
saw that metadata say some of the data was allocated but not written?

I suspect yes because I know with a real device I saw all the write data
reach the disk.  In a soft trace of usb cdb's, I saw that reading the
file back skipped over part of the disk.

With a real device I tried a pattern more complex than the byte xAA
repeated.  My more complex pattern let me see that that the data
supposedly not written had been written to the disk.  Only the read back
skipped over that portion of the disk.

If that theory holds true, then we will fix only metadata, not data.

> > > ... 2.6.0-test6 ...
> > > CPU0: Intel(R) Pentium(R) 4 CPU 2.40GHz stepping 09
> > > CPU1: Intel(R) Pentium(R) 4 CPU 2.40GHz stepping 09
> > ... read more zeroes than appended ...
>
> > ... 2.4.22-xfs ...
> > ... 866.394 MHz ... CPU0: Intel Pentium III ...
> > correct '\xAA' read back, no troublesome zeroes, ...
>
> can't repeat ... 2.4 or 2.6.0-test6 non-smp ...
>
> whats going on in 2.6.0-test6-smp ...

I also saw correct '\xAA' read back, no troublesome zeroes, in Red Hat
2.6.0-test3 on a (fifth?) pc, in 20..30 s/test, with a loop on ext3:

318MB LOWMEM available.
Detected 697.733 MHz processor.
CPU0: Intel Celeron (Coppermine) stepping 06
..... CPU clock speed is 697.0620 MHz.
..... host bus clock speed is 66.0440 MHz.

Pat LaVarre