Re: [v2 PATCH] iommu/arm-smmu-v3: Fix L1 stream table index calculation for 32-bit sid size

[Yang Shi]

On 10/4/24 9:29 AM, Yang Shi wrote:
> On 10/4/24 9:14 AM, Jason Gunthorpe wrote:
> > On Fri, Oct 04, 2024 at 09:05:46AM -0700, Yang Shi wrote:
> > > On 10/4/24 5:43 AM, Jason Gunthorpe wrote:
> > > > On Thu, Oct 03, 2024 at 08:31:23AM -0700, Yang Shi wrote:
> > > > > If I understand correctly, the check is mainly used to avoid the 
> > > > > u64 -> u32
> > > > > overflow. This check guarantee no overflow. If some crazy hardware 
> > > > > really
> > > > > requests that large memory, the allocation will fail.
> > > > Sure, the kalloc will print a warn on anyhow if it is too big
> > > Thank you. Will spin a new revision.
> > Oh wait a sec, it is not so simple, the 31 is too big because the
> > multiply will overflow or truncate to size_t too. This is why I picked
> > something lower.
>
> How about define the size as u64?
>
>  static int arm_smmu_init_strtab_linear(struct arm_smmu_device *smmu)
>  {
> -       u32 size;
> +       u64 size;
>         struct arm_smmu_strtab_cfg *cfg = &smmu->strtab_cfg;
>         u32 num_sids;
>
> It won't overflow and the large allocation will fail anyway.

The size parameter passed to dmam_alloc_coherent() is size_t type. We 
may do:

+#define SIZE_4G (4 * 1024 * 1024 * 1024ULL)
+
 static int arm_smmu_init_strtab_linear(struct arm_smmu_device *smmu)
 {
-       u32 size;
+       u64 size;
        struct arm_smmu_strtab_cfg *cfg = &smmu->strtab_cfg;
-       u32 num_sids;
+       u64 num_sids;

        /* It is not practical to have too large stream id size for 
linear */
        if (smmu->sid_bits > 31)
@@ -3667,6 +3669,9 @@ static int arm_smmu_init_strtab_linear(struct 
arm_smmu_device *smmu)
        num_sids = arm_smmu_strtab_num_sids(smmu);

        size = num_sids * sizeof(struct arm_smmu_ste);
+       if (size > SIZE_4G)
+               return -EINVAL;
+
        cfg->linear.table = dmam_alloc_coherent(smmu->dev, size,
&cfg->linear.ste_dma,
                                                GFP_KERNEL);

> > Jason