Re: [PATCH v3 9/9] LoongArch: KVM: INTC: Add address alignment check

From: Bibo Mao
Date: Thu Jun 26 2025 - 22:14:07 EST


That is a little different with x86/arm64 kvm.
With iocsr address space, there is access size with 1/2/4/8. With IO memory address space, there is generic access size with 1/2/4/8, however if irqchip driver of VM is deployed in user space, it will be possible with 16/32 bytes with SIMD instruction.

Regards
Bibo Mao

On 2025/6/21 下午9:04, Huacai Chen wrote:
Hi, David,

On Sat, Jun 21, 2025 at 7:21 PM David Laight
<david.laight.linux@xxxxxxxxx> wrote:

On Thu, 19 Jun 2025 16:47:22 +0800
Huacai Chen <chenhuacai@xxxxxxxxxx> wrote:

Hi, Bibo,

On Wed, Jun 11, 2025 at 9:51 AM Bibo Mao <maobibo@xxxxxxxxxxx> wrote:

IOCSR instruction supports 1/2/4/8 bytes access, the address should
be naturally aligned with its access size. Here address alignment
check is added in eiointc kernel emulation.

At the same time len must be 1/2/4/8 bytes from iocsr exit emulation
function kvm_emu_iocsr(), remove the default case in switch case
statements.
Robust code doesn't depend its callers do things right, so I suggest
keeping the default case, which means we just add the alignment check
here.

kernel code generally relies on callers to DTRT - except for values
that come from userspace.

Otherwise you get unreadable and slow code that continuously checks
for things that can't happen.
Generally you are right - but this patch is not the case.

Adding a "default" case here doesn't make code slower or unreadable,
and the code becomes more robust.

Huacai


David


And I think this patch should also Cc stable and add a Fixes tag.


Huacai


Signed-off-by: Bibo Mao <maobibo@xxxxxxxxxxx>
---
arch/loongarch/kvm/intc/eiointc.c | 21 +++++++++++++--------
1 file changed, 13 insertions(+), 8 deletions(-)

diff --git a/arch/loongarch/kvm/intc/eiointc.c b/arch/loongarch/kvm/intc/eiointc.c
index 8b0d9376eb54..4e9d12300cc4 100644
--- a/arch/loongarch/kvm/intc/eiointc.c
+++ b/arch/loongarch/kvm/intc/eiointc.c
@@ -311,6 +311,12 @@ static int kvm_eiointc_read(struct kvm_vcpu *vcpu,
return -EINVAL;
}

+ /* len must be 1/2/4/8 from function kvm_emu_iocsr() */
+ if (addr & (len - 1)) {
+ kvm_err("%s: eiointc not aligned addr %llx len %d\n", __func__, addr, len);
+ return -EINVAL;
+ }
+
vcpu->stat.eiointc_read_exits++;
spin_lock_irqsave(&eiointc->lock, flags);
switch (len) {
@@ -323,12 +329,9 @@ static int kvm_eiointc_read(struct kvm_vcpu *vcpu,
case 4:
ret = loongarch_eiointc_readl(vcpu, eiointc, addr, val);
break;
- case 8:
+ default:
ret = loongarch_eiointc_readq(vcpu, eiointc, addr, val);
break;
- default:
- WARN_ONCE(1, "%s: Abnormal address access: addr 0x%llx, size %d\n",
- __func__, addr, len);
}
spin_unlock_irqrestore(&eiointc->lock, flags);

@@ -682,6 +685,11 @@ static int kvm_eiointc_write(struct kvm_vcpu *vcpu,
return -EINVAL;
}

+ if (addr & (len - 1)) {
+ kvm_err("%s: eiointc not aligned addr %llx len %d\n", __func__, addr, len);
+ return -EINVAL;
+ }
+
vcpu->stat.eiointc_write_exits++;
spin_lock_irqsave(&eiointc->lock, flags);
switch (len) {
@@ -694,12 +702,9 @@ static int kvm_eiointc_write(struct kvm_vcpu *vcpu,
case 4:
ret = loongarch_eiointc_writel(vcpu, eiointc, addr, val);
break;
- case 8:
+ default:
ret = loongarch_eiointc_writeq(vcpu, eiointc, addr, val);
break;
- default:
- WARN_ONCE(1, "%s: Abnormal address access: addr 0x%llx, size %d\n",
- __func__, addr, len);
}
spin_unlock_irqrestore(&eiointc->lock, flags);

--
2.39.3