The standard GITS_TRANSLATER register in ITS is only 4 bytes, but
Hisilicon expands the next 4 bytes to carry some IMPDEF information. That
means, total 8 bytes data will be written to MSIAddress each time.
MSIAddr: |----4bytes----|----4bytes----|
| MSIData | IMPDEF |
There is no problem for ITS, because the next 4 bytes space is reserved
in ITS. But it will overwrite the 4 bytes memory following "sync_count".
It's very fortunately that the previous and the next neighbour of the
"sync_count" are both aligned by 8 bytes, so no problem is met now.
It's good to explicitly add a workaround. Let's enclose the "sync_count"
into a union and companion with a new member "padding" of type u64.
There is no functional change.
Signed-off-by: Zhen Lei <thunder.leizhen@xxxxxxxxxx>
---
drivers/iommu/arm-smmu-v3.c | 18 +++++++++++++++++-
1 file changed, 17 insertions(+), 1 deletion(-)
diff --git a/drivers/iommu/arm-smmu-v3.c b/drivers/iommu/arm-smmu-v3.c
index 6947ccf..4e94730 100644
--- a/drivers/iommu/arm-smmu-v3.c
+++ b/drivers/iommu/arm-smmu-v3.c
@@ -576,7 +576,23 @@ struct arm_smmu_device {
struct arm_smmu_strtab_cfg strtab_cfg;
- u32 sync_count;
+ /*
+ * The member "padding" is used to make sure the member "sync_count" to
+ * be aligned at 8 bytes boundary, and 4 bytes padding memory followed.
+ *
+ * These are required by hi1620 and earlier of Hisilicon. Because the
+ * ITS hardware on hi1620 and earlier will truncate the MSIAddress(Here
+ * it's the address of "sync_count") to 8 bytes boundary first, then
+ * write 32 bits MSIdata at offset 0, and 32 bits IMPDEF data at offset
+ * 4. Without this workaround, the adjacent member maybe overwritten.
+ *
+ * |---4bytes---|---4bytes---|
+ * MSIAddress & (~0x7): MSIdata | IMPDEF data|
+ */
+ union {
+ u32 sync_count;
+ u64 padding;
+ };
/* IOMMU core code handle */
struct iommu_device iommu;
--
1.8.3