(no subject)

[Konrad Rzeszutek Wilk]

This had been posted way way back:  https://lkml.org/lkml/2013/1/16/696

and the discussion about it was:
 - If there are no uses of v2, then why not rip out the support for it.
   [b/c the work for it would be in the networking code and I am still
    hoping that will be done?]
 - Why not do all of this in the toolstack. Enforce that the guest
   can only use v1 ".. if the backend isn't going to make use of them."
   (Matt).
   I am not sure how one would set up the plumbing to properly figure out
   whether the backend (say a driver domain) would communicate to hypervisor
   or XenBus that it is going to use v1 only if it has not yet started.
   (It would not have started b/c the guest hasn't started yet).

   My thinking is that once the frontend and backend start using the
   esoteric features of v2 we can rip this patch out. And also implement
   the proper code for PVHVM/PVH to actually use v2 grants.

[PATCH] xen/grant-table: Force to use v1 of grants.

[Konrad Rzeszutek Wilk]

We have the framework to use v2, but there are no backends that
actually use it. The end result is that on PV we use v2 grants
and on PVHVM v1. The v1 has a capacity of 512 grants per page while
the v2 has 256 grants per page. This means we lose about 50%
capacity - and if we want more than 16 VIFs (each VIF takes
512 grants), then we are hitting the max per guest of 32.

Oracle-bug: 16039922
CC: annie.li@oracle.com
CC: msw@amazon.com
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
---
 drivers/xen/grant-table.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/drivers/xen/grant-table.c b/drivers/xen/grant-table.c
index aa846a4..3005c17 100644
--- a/drivers/xen/grant-table.c
+++ b/drivers/xen/grant-table.c
@@ -1135,10 +1135,8 @@ static void gnttab_request_version(void)
 	int rc;
 	struct gnttab_set_version gsv;
 
-	if (xen_hvm_domain())
-		gsv.version = 1;
-	else
-		gsv.version = 2;
+	gsv.version = 1;
+
 	rc = HYPERVISOR_grant_table_op(GNTTABOP_set_version, &gsv, 1);
 	if (rc == 0 && gsv.version == 2) {
 		grant_table_version = 2;
-- 
1.8.3.1

Re: [Xen-devel] [PATCH] xen/grant-table: Force to use v1 of grants.

[David Vrabel]

On 31/12/13 15:15, Konrad Rzeszutek Wilk wrote:
> We have the framework to use v2, but there are no backends that
> actually use it. The end result is that on PV we use v2 grants
> and on PVHVM v1. The v1 has a capacity of 512 grants per page while
> the v2 has 256 grants per page. This means we lose about 50%
> capacity - and if we want more than 16 VIFs (each VIF takes
> 512 grants), then we are hitting the max per guest of 32.
> 
> Oracle-bug: 16039922
> CC: annie.li@oracle.com
> CC: msw@amazon.com
> Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>

Reviewed-by: David Vrabel <david.vrabel@citrix.com>

What does v2 add anyway? Do we want to remove all v2 related code or are
we expecting to make use of it in the near future?

David

Re: [Xen-devel] [PATCH] xen/grant-table: Force to use v1 of grants.

[Konrad Rzeszutek Wilk]

On Thu, Jan 02, 2014 at 02:54:18PM +0000, David Vrabel wrote:
> On 31/12/13 15:15, Konrad Rzeszutek Wilk wrote:
> > We have the framework to use v2, but there are no backends that
> > actually use it. The end result is that on PV we use v2 grants
> > and on PVHVM v1. The v1 has a capacity of 512 grants per page while
> > the v2 has 256 grants per page. This means we lose about 50%
> > capacity - and if we want more than 16 VIFs (each VIF takes
> > 512 grants), then we are hitting the max per guest of 32.
> > 
> > Oracle-bug: 16039922
> > CC: annie.li@oracle.com
> > CC: msw@amazon.com
> > Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
> 
> Reviewed-by: David Vrabel <david.vrabel@citrix.com>
> 
> What does v2 add anyway? Do we want to remove all v2 related code or are

Better status reporting page and two extra types of grants.

> we expecting to make use of it in the near future?

Yes, that is my understanding.
> 
> David

[PATCH] xen/grant-table: Force to use v1 of grants.

[Konrad Rzeszutek Wilk]

We have the framework to use v2, but there are no backends that
actually use it. The end result is that on PV we use v2 grants
and on PVHVM v1. The v1 has a capacity of 512 grants per page while
the v2 has 256 grants per page. This means we lose about 50%
capacity - and if we want more than 16 VIFs (each VIF takes
512 grants), then we are hitting the max per guest of 32.

Note: The hypervisor provides an over-ride for the default
of 32 frames (pages) per guest: gnttab_max_nr_frames=X

Oracle-bug: 16039922
CC: annie.li@oracle.com
CC: msw@amazon.com
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
---
 drivers/xen/grant-table.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/drivers/xen/grant-table.c b/drivers/xen/grant-table.c
index 95ce9d0..568fd8f 100644
--- a/drivers/xen/grant-table.c
+++ b/drivers/xen/grant-table.c
@@ -1108,10 +1108,8 @@ static void gnttab_request_version(void)
 	int rc;
 	struct gnttab_set_version gsv;
 
-	if (xen_hvm_domain())
-		gsv.version = 1;
-	else
-		gsv.version = 2;
+	gsv.version = 1;
+
 	rc = HYPERVISOR_grant_table_op(GNTTABOP_set_version, &gsv, 1);
 	if (rc == 0 && gsv.version == 2) {
 		grant_table_version = 2;
-- 
1.8.0.2

Re: [PATCH] xen/grant-table: Force to use v1 of grants.

[Matt Wilson]

On Wed, Jan 16, 2013 at 04:22:49PM -0500, Konrad Rzeszutek Wilk wrote:
> We have the framework to use v2, but there are no backends that
> actually use it. The end result is that on PV we use v2 grants
> and on PVHVM v1. The v1 has a capacity of 512 grants per page while
> the v2 has 256 grants per page. This means we lose about 50%
> capacity - and if we want more than 16 VIFs (each VIF takes
> 512 grants), then we are hitting the max per guest of 32.
> 
> Note: The hypervisor provides an over-ride for the default
> of 32 frames (pages) per guest: gnttab_max_nr_frames=X

I think it would be better if the dom0/service domain and hypervisor
prevent the guest from setting the version to 2 if the backend isn't
going to make use of them. In the short term, we could add a Xen
option to limit the supported version to 1.

Matt

> Oracle-bug: 16039922
> CC: annie.li@oracle.com
> CC: msw@amazon.com
> Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
> ---
>  drivers/xen/grant-table.c | 6 ++----
>  1 file changed, 2 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/xen/grant-table.c b/drivers/xen/grant-table.c
> index 95ce9d0..568fd8f 100644
> --- a/drivers/xen/grant-table.c
> +++ b/drivers/xen/grant-table.c
> @@ -1108,10 +1108,8 @@ static void gnttab_request_version(void)
>  	int rc;
>  	struct gnttab_set_version gsv;
>  
> -	if (xen_hvm_domain())
> -		gsv.version = 1;
> -	else
> -		gsv.version = 2;
> +	gsv.version = 1;
> +
>  	rc = HYPERVISOR_grant_table_op(GNTTABOP_set_version, &gsv, 1);
>  	if (rc == 0 && gsv.version == 2) {
>  		grant_table_version = 2;
> -- 
> 1.8.0.2

Re: [Xen-devel] [PATCH] xen/grant-table: Force to use v1 of grants.

[Jan Beulich]

>>> On 17.01.13 at 13:22, Matt Wilson <msw@amazon.com> wrote:
> On Wed, Jan 16, 2013 at 04:22:49PM -0500, Konrad Rzeszutek Wilk wrote:
>> We have the framework to use v2, but there are no backends that
>> actually use it. The end result is that on PV we use v2 grants
>> and on PVHVM v1. The v1 has a capacity of 512 grants per page while
>> the v2 has 256 grants per page. This means we lose about 50%
>> capacity - and if we want more than 16 VIFs (each VIF takes
>> 512 grants), then we are hitting the max per guest of 32.
>> 
>> Note: The hypervisor provides an over-ride for the default
>> of 32 frames (pages) per guest: gnttab_max_nr_frames=X
> 
> I think it would be better if the dom0/service domain and hypervisor
> prevent the guest from setting the version to 2 if the backend isn't
> going to make use of them. In the short term, we could add a Xen
> option to limit the supported version to 1.

But that's something each guest has to decide for itself.

Anyway - if no-one's using v2, what did it get introduced for
(not only in the hypervisor, but also in the kernel)? I thought
the networking drivers were intended to make use of it...

Jan

Re: [Xen-devel] [PATCH] xen/grant-table: Force to use v1 of grants.

[konrad wilk]

On 1/17/2013 8:05 AM, Jan Beulich wrote:
>>>> On 17.01.13 at 13:22, Matt Wilson<msw@amazon.com>  wrote:
>> On Wed, Jan 16, 2013 at 04:22:49PM -0500, Konrad Rzeszutek Wilk wrote:
>>> We have the framework to use v2, but there are no backends that
>>> actually use it. The end result is that on PV we use v2 grants
>>> and on PVHVM v1. The v1 has a capacity of 512 grants per page while
>>> the v2 has 256 grants per page. This means we lose about 50%
>>> capacity - and if we want more than 16 VIFs (each VIF takes
>>> 512 grants), then we are hitting the max per guest of 32.
>>>
>>> Note: The hypervisor provides an over-ride for the default
>>> of 32 frames (pages) per guest: gnttab_max_nr_frames=X
>> I think it would be better if the dom0/service domain and hypervisor
>> prevent the guest from setting the version to 2 if the backend isn't
>> going to make use of them. In the short term, we could add a Xen
>> option to limit the supported version to 1.
> But that's something each guest has to decide for itself.
>
> Anyway - if no-one's using v2, what did it get introduced for
> (not only in the hypervisor, but also in the kernel)? I thought
> the networking drivers were intended to make use of it...

That was the idea, but the work on that stalled. I think that once the 
netfront or netback has the right
code, then we can revert this patch - or just make this be more 
automatic and change itself to V2 if
it can.
> Jan
>

Re: [Xen-devel] [PATCH] xen/grant-table: Force to use v1 of grants.

[Ian Campbell]

On Wed, 2013-01-16 at 21:22 +0000, Konrad Rzeszutek Wilk wrote:
> Note: The hypervisor provides an over-ride for the default
> of 32 frames (pages) per guest: gnttab_max_nr_frames=X

Should we consider increasing this value?

Ian.

Re: [Xen-devel] [PATCH] xen/grant-table: Force to use v1 of grants.

[Konrad Rzeszutek Wilk]

On Thu, Jan 17, 2013 at 01:12:36PM +0000, Ian Campbell wrote:
> On Wed, 2013-01-16 at 21:22 +0000, Konrad Rzeszutek Wilk wrote:
> > Note: The hypervisor provides an over-ride for the default
> > of 32 frames (pages) per guest: gnttab_max_nr_frames=X
> 
> Should we consider increasing this value?

The fuzzy math is that for a VIF, you end up using 512 grants. For a VBD
it is 352 grants. With V1 you have 512 grants in a page, with V2 you have
256. So roughly the math is that you can at max 31VIFs and 1VBD with V1
(PVHVM guests) and 15VIFs and 1VBD with V2 (so all PV guests).

I would increase the value, but I don't know what a good value is.
> 
> Ian.
>