Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   "Maciej S. Szmigiero" <maciej.szmigiero@oracle.com>
Subject: Re: [PATCH] selftests: kvm: fix overlapping addresses in
 memslot_perf_test
To:     Paolo Bonzini <pbonzini@redhat.com>
Cc:     linux-kernel@vger.kernel.org, kvm@vger.kernel.org
References: <20210528191134.3740950-1-pbonzini@redhat.com>
 <285623f6-52e4-7f8d-fab6-0476a00af68b@oracle.com>
 <fc41bfc4-949f-03c5-3b20-2c1563ad7f62@redhat.com>
Message-ID: <73511f2e-7b5d-0d29-b8dc-9cb16675afb3@oracle.com>
Date:   Sun, 30 May 2021 01:13:36 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101
 Thunderbird/78.10.0
MIME-Version: 1.0
In-Reply-To: <fc41bfc4-949f-03c5-3b20-2c1563ad7f62@redhat.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 29.05.2021 12:20, Paolo Bonzini wrote:
> On 28/05/21 21:51, Maciej S. Szmigiero wrote:
>> On 28.05.2021 21:11, Paolo Bonzini wrote:
>>> The memory that is allocated in vm_create is already mapped close to
>>> GPA 0, because test_execute passes the requested memory to
>>> prepare_vm.  This causes overlapping memory regions and the
>>> test crashes.  For simplicity just move MEM_GPA higher.
>>>
>>> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
>>
>> I am not sure that I understand the issue correctly, is vm_create_default()
>> already reserving low GPAs (around 0x10000000) on some arches or run
>> environments?
> 
> It maps the number of pages you pass in the second argument, see
> vm_create.
> 
>    if (phy_pages != 0)
>      vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS,
>                                  0, 0, phy_pages, 0);
> 
> In this case:
> 
>    data->vm = vm_create_default(VCPU_ID, mempages, guest_code);
> 
> called here:
> 
>    if (!prepare_vm(data, nslots, maxslots, tdata->guest_code,
>                    mem_size, slot_runtime)) {
> 
> where mempages is mem_size, which is declared as:
> 
>          uint64_t mem_size = tdata->mem_size ? : MEM_SIZE_PAGES;
> 
> but actually a better fix is just to pass a small fixed value (e.g. 1024) to vm_create_default,
> since all other regions are added by hand

Yes, but the argument that is passed to vm_create_default() (mem_size
in the case of the test) is not passed as phy_pages to vm_create().
Rather, vm_create_with_vcpus() calculates some upper bound of extra
memory that is needed to cover that much guest memory (including for
its page tables).

The biggest possible mem_size from memslot_perf_test is 512 MiB + 1 page,
according to my calculations this results in phy_pages of 1029 (~4 MiB)
in the x86-64 case and around 1540 (~6 MiB) in the s390x case (here I am
not sure about the exact number, since s390x has some additional alignment
requirements).

Both values are well below 256 MiB (0x10000000UL), so I was wondering
what kind of circumstances can make these allocations collide
(maybe I am missing something in my analysis).

> 
> Paolo
> 

Thanks,
Maciej