Re: [Xen-devel] sources of freebsd with xen-4.2 kernel updates

[Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx>]

On Mon, Dec 30, 2013 at 01:53:01PM +0100, Roger Pau Monné wrote:
> On 30/12/13 13:42, Wei Liu wrote:
> > On Mon, Dec 30, 2013 at 04:18:28PM +0400, Igor Kozhukhov wrote:
> >>
> >> On Dec 30, 2013, at 3:25 PM, Roger Pau Monné wrote:
> >>
> >>> On 30/12/13 11:57, Igor Kozhukhov wrote:
> >>>> Hi Roger,
> >>>>
> >>>> thanks for the link.
> >>>> I have the same/(similar the same) implementation on 
> >>>> illumos(OpenSolaris).
> >>>>
> >>>> do you have plans for grant table v1 & v2 implementation on FreeBSD ?
> >>>
> >>> Grant table v1 is the current implementation, and I don't have plans
> >>> currently to add support for v2 (cannot speak of others though).
> >>>
> >>> Linux has v2 implemented, but AFAIK there's no user of the new features
> >>> introduced by v2. Right now with v2 you just end up with less grants,
> >>> because v2 grants metadata use twice as much memory as v1 grants, and
> >>> the number of grant frames is limited by the hypervisor.
> >>>
> >> Thanks for your info.
> >>
> >> Could you please let me know - can we use v1 for PV and HVM guests ?
> >> I looked v2 implementation on Linux and found that v1 is using for HVM, 
> >> and v2 for PV - but maybe i'm wrong.
> >>
> > 
> > You can use V1 for all guests for sure.
> > 
> > My understanding is that the code in Linux's grant-table.c requests V2
> > form the hypervisor, but it will fall back to using V1 if V2 is not
> > avaiable.
> 
> Linux explicitly requests v1 when running as an HVM guest (I have no
> idea why), for the rest it tries v2 first and falls back to v1 if

B/c it did not have the code to support mapping of the status page:

commit 69e8f430e243d657c2053f097efebc2e2cd559f0
Author: Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx>
Date:   Wed Jan 25 00:13:20 2012 -0500

    xen/granttable: Disable grant v2 for HVM domains.
    
    As proper scaffolding for supporting error status is not yet
    implemented.
    
    BUG: unable to handle kernel NULL pointer dereference at 0000000000000400
    IP: [<ffffffff81375ae9>] gnttab_end_foreign_access_ref_v2+0x29/0x40
    PGD 32aa3067 PUD 32a87067 PMD 0
    Oops: 0000 [#1] PREEMPT SMP
    CPU 0
    Modules linked in: sg sr_mod cdrom ata_generic ata_piix libata scsi_mod 
xen_blkfront xen_netfront fb_sys_fops sysimgblt sysfillr
    cmd
    
    Pid: 2307, comm: ip Not tainted 3.3.0-rc1 #1 Xen HVM domU
    RIP: 0010:[<ffffffff81375ae9>]  [<ffffffff81375ae9>] 
gnttab_end_foreign_access_ref_v2+0x29/0x40
    RSP: 0018:ffff88003be03d38  EFLAGS: 00010206
    RAX: 0000000000000000 RBX: ffff880033210640 RCX: 0000000000000040
    RDX: 0000000000002000 RSI: 0000000000000000 RDI: 0000000000000200
    RBP: ffff88003be03d38 R08: 0000000000000101 R09: 0000000000000000
    R10: dead000000100100 R11: 0000000000000000 R12: ffff88003be03e48
    R13: 0000000000000001 R14: ffff880039461c00 R15: 0000000000000200
    FS:  00007fb1f84ec700(0000) GS:ffff88003be00000(0000) knlGS:0000000000000000
    ...

It should be implemented at some point.. but since there are no real
users of V2 it is a the bottom of the pile of 'todo'.

> there's no v2 support:
> 
> http://lxr.free-electrons.com/source/drivers/xen/grant-table.c#L1116
> 
> As Wei points out, you can safely use v1 for all kind of guests, at
> least on x86.
> 
> Roger.
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@xxxxxxxxxxxxx
> http://lists.xen.org/xen-devel

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel