Re: [PATCH v2 06/12] xen/gnttab: Pass max_{grant, maptrack}_frames into grant_table_create()

[Julien Grall <julien.grall@arm.com>]

Hi Andrew,

On 08/15/2018 02:08 PM, Andrew Cooper wrote:
> On 15/08/18 14:04, Julien Grall wrote:
>> Hi Andrew,
>>
>> On 08/13/2018 11:01 AM, Andrew Cooper wrote:
>>> ... rather than setting the limits up after domain_create() has
>>> completed.
>>>
>>> This removes all the gnttab infrastructure for calculating the number
>>> of dom0
>>> grant frames, opting instead to require the dom0 construction code to
>>> pass a
>>> sane value in via the configuration.
>>>
>>> In practice, this now means that there is never a partially
>>> constructed grant
>>> table for a reference-able domain.
>>>
>>> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
>>> ---
>>> CC: Jan Beulich <JBeulich@suse.com>
>>> CC: Stefano Stabellini <sstabellini@kernel.org>
>>> CC: Julien Grall <julien.grall@arm.com>
>>> CC: Wei Liu <wei.liu2@citrix.com>
>>>
>>> v2:
>>>    * Split/rearrange to avoid the post-domain-create error path.
>>> ---
>>>    xen/arch/arm/domain_build.c       |  3 ++-
>>>    xen/arch/arm/setup.c              | 12 ++++++++++++
>>>    xen/arch/x86/setup.c              |  3 +++
>>>    xen/common/domain.c               |  3 ++-
>>>    xen/common/grant_table.c          | 16 +++-------------
>>>    xen/include/asm-arm/grant_table.h | 12 ------------
>>>    xen/include/asm-x86/grant_table.h |  5 -----
>>>    xen/include/xen/grant_table.h     |  6 ++----
>>>    8 files changed, 24 insertions(+), 36 deletions(-)
>>>
>>> diff --git a/xen/arch/arm/domain_build.c b/xen/arch/arm/domain_build.c
>>> index 1351572..737e0f3 100644
>>> --- a/xen/arch/arm/domain_build.c
>>> +++ b/xen/arch/arm/domain_build.c
>>> @@ -2079,7 +2079,8 @@ static void __init find_gnttab_region(struct
>>> domain *d,
>>>         * enough space for a large grant table
>>>         */
>>>        kinfo->gnttab_start = __pa(_stext);
>>> -    kinfo->gnttab_size = gnttab_dom0_frames() << PAGE_SHIFT;
>>> +    kinfo->gnttab_size = min_t(unsigned int, opt_max_grant_frames,
>>> +                               PFN_DOWN(_etext - _stext)) <<
>>> PAGE_SHIFT;
>>
>>
>> I agree with Jan's comment on v1 that there is a risk someone will
>> update the size here but ...
>>
>>
>>>      #ifdef CONFIG_ARM_32
>>>        /*
>>> diff --git a/xen/arch/arm/setup.c b/xen/arch/arm/setup.c
>>> index 45f3841..3d3b30c 100644
>>> --- a/xen/arch/arm/setup.c
>>> +++ b/xen/arch/arm/setup.c
>>> @@ -20,6 +20,7 @@
>>>    #include <xen/compile.h>
>>>    #include <xen/device_tree.h>
>>>    #include <xen/domain_page.h>
>>> +#include <xen/grant_table.h>
>>>    #include <xen/types.h>
>>>    #include <xen/string.h>
>>>    #include <xen/serial.h>
>>> @@ -693,6 +694,17 @@ void __init start_xen(unsigned long
>>> boot_phys_offset,
>>>        struct domain *dom0;
>>>        struct xen_domctl_createdomain dom0_cfg = {
>>>            .max_evtchn_port = -1,
>>> +
>>> +        /*
>>> +         * The region used by Xen on the memory will never be mapped
>>> in DOM0
>>> +         * memory layout. Therefore it can be used for the grant table.
>>> +         *
>>> +         * Only use the text section as it's always present and will
>>> contain
>>> +         * enough space for a large grant table
>>> +         */
>>> +        .max_grant_frames = min_t(unsigned int, opt_max_grant_frames,
>>> +                                  PFN_DOWN(_etext - _stext)),
>>
>> ... not here. So I would prefer if we either keep an helper to find
>> the size of pass that size around to domain_build. Do we store the
>> size in the domain information?
> 
> I have to admit that I'm somewhat perplexed by ARM's
> find_gnttab_region(), and I'm not sure why it exists.

Dom0 is using the host memory layout that may differ between platforms. 
So there is not a region address that would fit everyone.

This function is here to find at boot a suitable region in the layout 
where the OS can map the grant-table. The result will be written in the 
firmware table.

> 
> The value is available from d->grant_table.max_grant_frames but ISTR
> finding that the order of construction meant that it wasn't available
> when needed (although this was all from code inspection, so I could very
> easily be wrong).

I think it should be fine for Dom0 as find_gnttab_region is called from 
construct_dom0 and d->grant_table.max_grant_frames would be set before 
via domain_create().

Assuming d->grant_table.max_grant_frames can only be 0 before 
initialization, I would potentially add a 
BUG_ON(!d->grant_table.max_grant_frames) to make sure this always stay 
like that.

Cheers,

-- 
Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel