[dm-devel] dm-cache-mq policy can't allocate memory

[Joe Thornber]

On Fri, Apr 05, 2013 at 04:37:57PM +0200, Étienne BERSAC wrote:
> Hi,
> 
> I'm testing the experimental dm-cache target and think i encountered a
> bug.
> 
> I tested version d08d528dc1848fb369a0b27cdb0749d8f6f38063 of linus tree.
> Compiled with make -j4 deb-pkg. Config file attached.
> 
> Here is the current dmsetup configuration :
> 
> ------------------->8-----------------
> root at st-brice:~/etudes# dmsetup table
> st--brice-root: 0 151486464 linear 8:5 2048
> st--brice-swap_1: 0 4169728 linear 8:5 151488512
> -------------------8<-----------------
> 
> 8:5 is /dev/sda, a disk reserved for the system. I'm using another
> physical disk for playing. I directly use physical partitions as blocks
> backing the cache target, i'm just playing with the same physical device
> yet :
> 
> ------------>8---------------
> root at st-brice:~# fdisk -l /dev/sdb
> Disk /dev/sdb: 80.0 GB, 80000000000 bytes
> 
>    Device Boot      Start         End      Blocks   Id  System
> /dev/sdb1              63     3903794     1951866   8e  Linux LVM
> /dev/sdb2         3903795    82027889    39062047+  8e  Linux LVM
> /dev/sdb3        82027890   156249999    37111055   8e  Linux LVM
> -------------8<---------------
> 
> /dev/sdb1 is zeroed before issuing this command. I setup the target
> using the following table :
> 
>         /sbin/dmsetup create dmc0 --table '0 74222110 cache /dev/sdb1 /dev/sdb2 /dev/sdb3 64 0 default 0'

The mq policy uses a fair bit of memory per cache entry (I'm working
on reducing this).  I suggest you bump the block size right up.  eg,
2048 rather than 64.  Tell me if this doesn't help, the only unusual
thing you're doing is using a cache device that's bigger than the
origin, it should work though.

- Joe