Re: [PATCH] iommu/arm-smmu-v3: Fix L1 stream table index calculation for AmpereOne

From: Robin Murphy
Date: Wed Oct 02 2024 - 05:59:16 EST


On 2024-10-01 8:48 pm, Yang Shi wrote:


On 10/1/24 12:29 PM, Nicolin Chen wrote:
On Tue, Oct 01, 2024 at 12:09:03PM -0700, Yang Shi wrote:
On 10/1/24 11:27 AM, Nicolin Chen wrote:
On Tue, Oct 01, 2024 at 11:03:46AM -0700, Yang Shi wrote:
Using 64 bit immediate when doing shift can solve the problem.  The
disssembly after the fix looks like:
[...]

          unsigned int last_sid_idx =
-               arm_smmu_strtab_l1_idx((1 << smmu->sid_bits) - 1);
+               arm_smmu_strtab_l1_idx((1UL << smmu->sid_bits) - 1);
Could a 32-bit build be a corner case where UL is no longer a
"64 bit" stated in the commit message?
It shouldn't. Because smmu v3 depends on ARM64.

config ARM_SMMU_V3
         tristate "ARM Ltd. System MMU Version 3 (SMMUv3) Support"
         depends on ARM64
ARM64 can have aarch32 support. I am not sure if ARM64 running a
32-bit OS can be a case though, (and not confined to AmpereOne).

I don't think ARM64 runs 32-bit kernel, at least for newer kernel.

Just use ULL - if the point is that it must be a 64-bit shift for correctness, then being clear about that intent is far more valuable than saving one character of source code.

Thanks,
Robin.



Then, can ssid_bits/s1cdmax be a concern similarly?
IIUC, ssid_bits is determined by IDR1_SSIDSIZE. It is GENMASK(10, 6). So
it shouldn't be 32. IDR1_SIDSIZE is GENMASK(5, 0).
Rechecked the RM. Yea, max sid can be 32 but max ssid is 20 at
this moment, so we should be safe.

Thanks
Nicolin