Re: include/linux/kvm_host.h:488 suspicious rcu_dereference_check() usage!

[Paolo Bonzini <pbonzini@redhat.com>]

On 21/09/2015 17:10, Paolo Bonzini wrote:
> 
> 
> On 20/09/2015 18:48, Borislav Petkov wrote:
>> [26421.593927] ------ spte 0x3e5a22027 level 4.
>> [26421.598228] ------ spte 0x38a00b027 level 3.
>> [26421.602505] ------ spte 0x387334027 level 2.
>> [26421.602506] ------ spte 0xffff0000000b8f67 level 1.
>> [26421.602506] ------------[ cut here ]------------
>> [26421.602530] WARNING: CPU: 2 PID: 17000 at arch/x86/kvm/mmu.c:3385 handle_mmio_page_fault.part.93+0x1a/0x20 [kvm]()
>> [26421.602550] Modules linked in: tun sha256_ssse3 sha256_generic drbg binfmt_misc ipv6 vfat fat fuse dm_crypt dm_mod kvm_amd kvm crc32_pclmul aesni_intel aes_x86_64 lrw gf128mul glue_helper ablk_helper cryptd fam15h_power amd64_edac_mod k10temp edac_core amdkfd amd_iommu_v2 radeon acpi_cpufreq
>> [26421.602552] CPU: 2 PID: 17000 Comm: qemu-system-i38 Not tainted 4.3.0-rc1+ #1
>> [26421.602553] Hardware name: To be filled by O.E.M. To be filled by O.E.M./M5A97 EVO R2.0, BIOS 1503 01/16/2013
>> [26421.602555]  ffffffffa02fc7d2 ffff880386c0fb80 ffffffff812c8c2a 0000000000000000
>> [26421.602556]  ffff880386c0fbb8 ffffffff81053e55 ffff880429ff8000 000000000000000f
>> [26421.602558]  00000000000b8000 0000000000000000 00000000ffffffff ffff880386c0fbc8
>> [26421.602558] Call Trace:
>> [26421.602564]  [<ffffffff812c8c2a>] dump_stack+0x4e/0x84
>> [26421.602566]  [<ffffffff81053e55>] warn_slowpath_common+0x95/0xe0
>> [26421.602567]  [<ffffffff81053f5a>] warn_slowpath_null+0x1a/0x20
>> [26421.602577]  [<ffffffffa02cf73a>] handle_mmio_page_fault.part.93+0x1a/0x20 [kvm]
> 
> Can you test this patch for this second call trace?
> 
> diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
> index fb16a8ea3dee..10c281ec9a25 100644
> --- a/arch/x86/kvm/mmu.c
> +++ b/arch/x86/kvm/mmu.c
> @@ -3320,6 +3320,9 @@ walk_shadow_page_get_mmio_spte(struct kvm_vcpu *vcpu, u64 addr, u64 *sptep)
>  		if (!is_shadow_present_pte(spte))
>  			break;
>  
> +		if (leaf == 1 && is_mmio_spte(spte))
> +			break;
> +
>  		reserved |= is_shadow_zero_bits_set(&vcpu->arch.mmu, spte,
>  						    leaf);
>  	}

And of course I need a v2 of this one too. :(

First, the leaf test would have to be == 0, because I prepared the
patch on the first 4.3 pull request instead of the latest Linus
tree.  However even this would not be a good change, because

is_shadow_present_pte(spte) == !(pte & PT_PRESENT_MASK) || is_mmio_spte(pte)

and thus is_shadow_present_pte implies the "if" I'm adding above.

So can you instead please add this debugging printk?

diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index fb16a8ea3dee..90e8ef264861 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -3334,6 +3334,7 @@ walk_shadow_page_get_mmio_spte(struct kvm_vcpu *vcpu, u64 addr, u64 *sptep)
 			       sptes[root - 1], root);
 			root--;
 		}
+		pr_err("shadow_mmio_mask: %lx\n", shadow_mmio_mask);
 	}
 exit:
 	*sptep = spte;

Paolo