Re: [PATCH] kvm tools: Make the whole guest memory mergeable

[Zang Hongyong <zanghongyong@huawei.com>]

于 2011/12/16,星期五 16:45, Sasha Levin 写道:
> On Fri, 2011-12-16 at 16:36 +0800, Zang Hongyong wrote:
>> 于 2011/12/16,星期五 15:23, Sasha Levin 写道:
>>> On Fri, 2011-12-16 at 15:02 +0800, Zang Hongyong wrote:
>>>> 于 2011/12/16,星期五 13:50, Sasha Levin 写道:
>>>>> On Fri, 2011-12-16 at 09:01 +0800, zanghongyong@huawei.com wrote:
>>>>>> If a guest's ram_size exceeds KVM_32BIT_GAP_START, the corresponding kvm tool's
>>>>>> virtual address size should be (ram_size + KVM_32BIT_GAP_SIZE), rather than ram_size.
>>>>> You're right.
>>>>>
>>>>> There are more places than just the madvise() code which make the same
>>>>> error you've spotted (for example, the memslot allocation code), so
>>>>> instead of trying to fix all of them I'd suggest to just update ram_size
>>>>> in kvm__arch_init() before allocating everything - that should fix all
>>>>> of them at once.
>>>>>
>>>> Yes. There are other scenarios with the same error.
>>>> However ram_size sometimes means real guest ram size, and sometimes 
>>>> means virtual address
>>>> size of kvm tool's user space. Shall we define a new variable?
>>> Let's keep it simple. If the user requests more than RAM than
>>> KVM_32BIT_GAP_START just increase it by KVM_32BIT_GAP_SIZE, this way
>>> mapped size == guest size always (we can madvise(MADV_DONTNEED) the gap
>>> in the mmapped ram).
>>>
>>> Since a user which requests more than KVM_32BIT_GAP_START will have to
>>> be on 64bit host anyway, there shouldn't be any issue with that.
>>>
>> Do you mean increase *kvm->ram_size* by KVM_32BIT_GAP_SIZE?
>> but sometimes kvm->ram_size stands for guest physical ram size (for
>> example in kvm__init_ram() code).
> Yup, kvm->ram_size.
>
> If the user requested more than KVM_32BIT_GAP_START, we pretty much have
> to create the gap, so instead of playing around with different
> interpretations of ram_size, lets add the gap size - this will let us
> have just one ram_size.
>
> mmap()ing extra space for the gap is free, and that was the plan in the
> first place (we just got the math wrong :) ).
>
> Do you see an issue with increasing kvm->ram_size?
>
Yes, it will cause some problems after simply increase the kvm->ram_size.
For examples:
In kvm__init_ram() code we use kvm->ram_size to calculate the size of
the second
RAM range from 4GB to the end of RAM (phys_size = kvm->ram_size -
phys_size;),
so after increase the kvm->ram_size, it will goes wrong.
This problem also happens in e820_setup() code and load_bzimage() code.