[DTrace-devel] [PATCH] Implement the io provider

[Eugene Loh]

I also tried running the tests.  OL7 OL8 OL9 UEK6 UEK7 x86 and ARM. 8 
platforms.  3 passes of the io tests.  All tests pass except on OL9 ARM, 
where local2 and nfs2 failed consistently.  I don't really understand 
this stuff (I devised these tests just by looking for conditions to 
check that held up empirically and could be imagined to be reasonable), 
but I guess I can look more.

On test/unittest/io/tst.nfs2.sh, the output file log.write expects a 
single start of 1048576 bytes in column 7;  I see

run1:
io: vmlinux: :       start 0 520    1002859   1002859   0 0 0 52958581 
52958581    0   0 0 0   0 0 0 nfs nfs <nfs>
io: vmlinux: :       start 0 520      49152     49152   0 0 0 52958581 
52958581    0   0 0 0   0 0 0 nfs nfs <nfs>

run2:
io: vmlinux: :       start 0 520     992032    992032   0 0 0 1324185 
1324185    0   0 0 0   0 0 0 nfs nfs <nfs>
io: vmlinux: :       start 0 520      57344     57344   0 0 0 1324185 
1324185    0   0 0 0   0 0 0 nfs nfs <nfs>

run3:
io: vmlinux: :       start 0 520     964185    964185   0 0 0 19900749 
19900749    0   0 0 0   0 0 0 nfs nfs <nfs>
io: vmlinux: :       start 0 520      86016     86016   0 0 0 19900749 
19900749    0   0 0 0   0 0 0 nfs nfs <nfs>

On tst.local2.sh, log.write shows no "start" records with nonzero byte 
counts.

At the very least, I can add some more diagnostics with the tests to 
make it easier to see what's going on.