svn commit: r267071 - stable/10/usr.sbin/bhyve

[Willem Jan Withagen]

On 4-6-2014 20:08, John Baldwin wrote:
> Author: jhb
> Date: Wed Jun  4 18:08:08 2014
> New Revision: 267071
> URL: http://svnweb.freebsd.org/changeset/base/267071
> 
> Log:
>   MFC 260999:
>   Increase the block-layer backend maximum number of requests to match
>   the AHCI command queue depth.
> 
> Modified:
>   stable/10/usr.sbin/bhyve/block_if.c
> Directory Properties:
>   stable/10/   (props changed)
> 
> Modified: stable/10/usr.sbin/bhyve/block_if.c
> ==============================================================================
> --- stable/10/usr.sbin/bhyve/block_if.c	Wed Jun  4 17:57:48 2014	(r267070)
> +++ stable/10/usr.sbin/bhyve/block_if.c	Wed Jun  4 18:08:08 2014	(r267071)
> @@ -50,7 +50,7 @@ __FBSDID("$FreeBSD$");
>  
>  #define BLOCKIF_SIG	0xb109b109
>  
> -#define BLOCKIF_MAXREQ	16
> +#define BLOCKIF_MAXREQ	32
>  
>  enum blockop {
>  	BOP_READ,

I once asked Peter about this, because of the trouble I had when running
the bhyve_svm branch.

This is not perse a good enhancement....
I've had a combination of HEAD and bhyve where running Linux and AMD
would regularly generate timeouts for FPDMA for slots > 16....

After building the most recent bhyve_svm with patch set to get to a more
recent almost-head-snapshot I ran some test, and did not run into these
timeouts.
So can very well be that fixes in HEAD got ride off the timeout. But
then again I would say: thread with caution.

--WjW