On Mon, May 21, 2018 at 04:22:07PM +0200, Jaromír Dole?ek wrote:
> The very basic test is to do some moral equivalent of "dd if=foo
> of=/dev/null bs=64k" with some reasonably sized foo file, with default
> setting twice (once to test reading off the disk, second off cache),
> then set ubc_direct to 1 via DDB, remount the filesystem and re-run.
> Also would be good to check that the transferred contents are correct,
> e.g. via sha1.
>
> If that doesn't trigger any panic, further test would be to run some
> non-trivial build with ubc_direct set to 1.
Here are timing results:
[/tmp] root@gemini # /usr/bin/time dd if=/netbsd of=/dev/zero bs=64k
213+1 records in
213+1 records out
14005088 bytes transferred in 0.735 secs (19054541 bytes/sec)
0.75 real 0.00 user 0.17 sys
[/tmp] root@gemini # /usr/bin/time dd if=/netbsd of=/dev/zero bs=64k
213+1 records in
213+1 records out
14005088 bytes transferred in 0.120 secs (116709066 bytes/sec)
0.12 real 0.00 user 0.12 sys
and again with ubc_direct=1 (rebooted with -d, changed it in ddb and continued)
[/tmp] root@gemini # /usr/bin/time dd if=/netbsd of=/dev/zero bs=64k
213+1 records in
213+1 records out
14005088 bytes transferred in 0.741 secs (18900253 bytes/sec)
0.76 real 0.00 user 0.13 sys
[/tmp] root@gemini # /usr/bin/time dd if=/netbsd of=/dev/zero bs=64k
213+1 records in
213+1 records out
14005088 bytes transferred in 0.088 secs (159148727 bytes/sec)
0.09 real 0.00 user 0.09 sys
Unfortunatley with ubc_direct enabled, it panics quickly:
[ 399.9154127] System Machine Check (660): Rev 0x1, Code 0x202, Flags 0x0
[ 399.9154127] System Machine Check (660): Rev 0x1, Code 0x202, Flags 0x0
[ 399.9154127] Software Error Summary Flags = 0x0000000000000001
[ 399.9154127] Software Error Summary Flags = 0x0000000000000001
[ 399.9154127] CPU Device Interrupt Requests = 0x4000000000000000
[ 399.9154127] CPU Device Interrupt Requests = 0x4000000000000000
[ 399.9154127] DIR = 0x4000000000000000<Pchip 0 error>
[ 399.9154127] DIR = 0x4000000000000000<Pchip 0 error>
[ 399.9154127] Cchip Miscellaneous Register = 0x0000000120000000
[ 399.9154127] Pchip 0 Error Register = 0x006038d678700041
[ 399.9154127] Cchip Miscellaneous Register = 0x0000000120000001
[ 399.9154127] error = 0x41<Error lost,Target abort>
[ 399.9154127] Pchip 0 Error Register = 0x006038d678700041
[ 399.9154127] address = 0x38d67870, 0x0<No DAC>
[ 399.9154127] error = 0x41<Error lost,Target abort>
[ 399.9154127] command = 0x6<PCI memory read>
[ 399.9154127] address = 0x38d67870, 0x0<No DAC>
[ 399.9154127] Pchip 1 Error Register = 0x0000000000000000
[ 399.9154127] command = 0x6<PCI memory read>
[ 399.9154127] Pchip 1 Error Register = 0x0000000000000000
[ 399.9154127] unexpected machine check:
[ 399.9154127] unexpected machine check:
[ 399.9154127] mces = 0x1
[ 399.9154127] vector = 0x660
[ 399.9154127] mces = 0x1
[ 399.9154127] param = 0xfffffc0000006080
[ 399.9154127] vector = 0x660
[ 399.9154127] param = 0xfffffc0000006480
[ 399.9154127] pc = 0xfffffc0001109fe0
[ 399.9154127] pc = 0xfffffc00010fb0fc
[ 399.9154127] ra = 0xfffffc00010fb120
[ 399.9154127] code = 0x100000202
[ 399.9154127] ra = 0xfffffc00010fb0fc
[ 399.9154127] curlwp = 0xfffffc003fe3f980
[ 399.9154127] code = 0x100000202
[ 399.9154127] pid = 0.2, comm = system
[ 399.9154127] curlwp = 0xfffffc003fda0020
[ 399.9154127] pid = 0.15, comm = system
[ 399.9154127] panic:
399.9154127] machine check
[ 399.9154127] cpu0: Begin traceback...
[ 399.9154127] alpha trace requires known PC =eject=
[ 399.9154127] cpu0: End traceback...
[ 399.9154127] WARNING: Unable to halt secondary CPUs (0x3)
Martin
