[PATCH v2] KVM: arm64: Remove size-order align in the nVHE hyp private VA range

[Marek Szyprowski]

On 28.08.2023 15:19, Marc Zyngier wrote:
> On Mon, 28 Aug 2023 13:16:51 +0100,
> Marek Szyprowski <m.szyprowski at samsung.com> wrote:
>> On 11.08.2023 13:20, Vincent Donnefort wrote:
>>> commit f922c13e778d ("KVM: arm64: Introduce
>>> pkvm_alloc_private_va_range()") and commit 92abe0f81e13 ("KVM: arm64:
>>> Introduce hyp_alloc_private_va_range()") added an alignment for the
>>> start address of any allocation into the nVHE hypervisor private VA
>>> range.
>>>
>>> This alignment (order of the size of the allocation) intends to enable
>>> efficient stack verification (if the PAGE_SHIFT bit is zero, the stack
>>> pointer is on the guard page and a stack overflow occurred).
>>>
>>> But this is only necessary for stack allocation and can waste a lot of
>>> VA space. So instead make stack-specific functions, handling the guard
>>> page requirements, while other users (e.g.  fixmap) will only get page
>>> alignment.
>>>
>>> Signed-off-by: Vincent Donnefort <vdonnefort at google.com>
>> This change, merged to linux-next as f156a7d13fc3 ("KVM: arm64: Remove
>> size-order align in the nVHE hyp private VA range"), introduced the
>> following regression on Raspberry Pi4b board. Here is a log observed on
>> linux next-20230828 release:
>>
>> --->8---
>>
>> kvm [1]: IPA Size Limit: 44 bits
>> ------------[ cut here ]------------
>> WARNING: CPU: 0 PID: 1 at arch/arm64/kvm/hyp/pgtable.c:453
>> hyp_map_walker+0xb0/0x120
>> Modules linked in:
>> CPU: 0 PID: 1 Comm: swapper/0 Not tainted 6.5.0-next-20230828 #13889
>> Hardware name: Raspberry Pi 4 Model B (DT)
>> pstate: 00000005 (nzcv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
>> pc : hyp_map_walker+0xb0/0x120
>> lr : hyp_map_walker+0x30/0x120
>> ...
>> Call trace:
>>    hyp_map_walker+0xb0/0x120
>>    kvm_pgtable_visitor_cb.isra.0+0x38/0x5c
>>    __kvm_pgtable_walk+0x1dc/0x32c
>>    __kvm_pgtable_walk+0xd0/0x32c
>>    __kvm_pgtable_walk+0xd0/0x32c
>>    __kvm_pgtable_walk+0xd0/0x32c
>>    kvm_pgtable_walk+0xd0/0x224
>>    kvm_pgtable_hyp_map+0x8c/0xf0
>>    __create_hyp_mappings+0x98/0xc0
>>    __create_hyp_private_mapping+0xf0/0x14c
>>    create_hyp_io_mappings+0x8c/0x118
>>    vgic_v2_probe+0xbc/0x1ec
>>    kvm_vgic_hyp_init+0xcc/0x210
>>    kvm_arm_init+0x6a4/0x1040
> Ah, wonderful. Can you give the snippet below a go?
>
> diff --git a/arch/arm64/kvm/mmu.c b/arch/arm64/kvm/mmu.c
> index 11c1d786c506..50be51cc40cc 100644
> --- a/arch/arm64/kvm/mmu.c
> +++ b/arch/arm64/kvm/mmu.c
> @@ -652,6 +652,9 @@ int hyp_alloc_private_va_range(size_t size, unsigned long *haddr)
>   
>   	mutex_unlock(&kvm_hyp_pgd_mutex);
>   
> +	if (!ret)
> +		*haddr = base;
> +
>   	return ret;
>   }
>   

This fixed the regression. Feel free to add:

Reported-by: Marek Szyprowski <m.szyprowski at samsung.com>

Tested-by: Marek Szyprowski <m.szyprowski at samsung.com>


Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland