Re: [RFC][PATCH 4/5] tools:firmware:hvmloader: reserve RMRR mappings in e820

[Andrew Cooper <andrew.cooper3@citrix.com>]

On 08/08/14 16:53, Jan Beulich wrote:
>>>> On 07.08.14 at 13:02, <tiejun.chen@intel.com> wrote:
>> We need to reserve all RMRR mappings in e820 to avoid any
>> potential guest memory conflict.
>>
>> Signed-off-by: Tiejun Chen <tiejun.chen@intel.com>
>> ---
>>  tools/firmware/hvmloader/e820.c | 14 ++++++++++++++
>>  tools/firmware/hvmloader/e820.h |  6 ++++++
>>  tools/firmware/hvmloader/util.c | 13 +++++++++++++
>>  tools/firmware/hvmloader/util.h |  1 +
>>  4 files changed, 34 insertions(+)
>> @@ -124,6 +126,18 @@ int build_e820_table(struct e820entry *e820,
>>      e820[nr].type = E820_RAM;
>>      nr++;
>>  
>> +    /* We'd better reserve RMRR mapping for each VM to avoid potential
>> +     * memory conflict.
>> +     */
>> +    e820_rmrr_map = get_rmrr_map_info();
>> +    for ( i = 0; i <= e820_rmrr_map->nr_map; i++ )
>> +    {
>> +        e820[nr].addr = e820_rmrr_map->map[i].addr;
>> +        e820[nr].size = e820_rmrr_map->map[i].size + 1;
>> +        e820[nr].type = E820_RESERVED;
>> +        nr++;
>> +    }
> You can't just put this in an arbitrary place, without caring for overlaps
> (and ordering perhaps - I don't recall offhand whether E820 is
> required to be sorted, or whether that's just common practice). In fact
> I'm not certain the code block following your insertion (interestingly
> another Intel special) doesn't violate this too.
>
> Jan

Whether it is required to be sorted or not, it is certainly the case
that there exist many BIOSes which present all manor of garbage in the
E820, including entries not aligned on page boundaries.

Any decent OS should be able to cope, but it is very bad form for
HVMLoader to present an E820 table like this.  It should be sorted and
lacking any overlapping regions.

~Andrew