On 9/16/2026 9:44 AM, David Marchand wrote:
On Thu, 10 Sept 2026 at 14:56, Anatoly Burakov <[email protected]> wrote:@@ -500,7 +507,8 @@ rte_dpaa2_get_iommu_class(void) return RTE_IOVA_DC;/* check if all devices on the bus support Virtual addressing or not */ - if (fslmc_all_device_support_iova() != 0 && dev_vfio_noiommu_is_enabled() == 0) + if (fslmc_all_device_support_iova() != 0 && + dev_vfio_get_iommu_mode() == DEV_VFIO_IOMMU_MODE_UNSAFE) return RTE_IOVA_VA; return RTE_IOVA_PA;IA flagged this part: In `fslmc_bus.c`, the condition changed from: ```c /* Old */ if (fslmc_all_device_support_iova() != 0 && dev_vfio_noiommu_is_enabled() == 0) return RTE_IOVA_VA; /* New */ if (fslmc_all_device_support_iova() != 0 && dev_vfio_get_iommu_mode() == DEV_VFIO_IOMMU_MODE_UNSAFE) return RTE_IOVA_VA; ``` The old code returned `RTE_IOVA_VA` when NOIOMMU was **disabled** (== 0), but the new code returns `RTE_IOVA_VA` when IOMMU mode is **UNSAFE** (i.e., NOIOMMU is enabled). Looking at `dev_vfio_noiommu_is_enabled()`: ```c int dev_vfio_noiommu_is_enabled(void) { return vfio_global_cfg.iommu_mode == DEV_VFIO_IOMMU_MODE_UNSAFE; } ``` So: - Old: `dev_vfio_noiommu_is_enabled() == 0` → NOIOMMU is NOT enabled → IOMMU is SAFE - New: `dev_vfio_get_iommu_mode() == DEV_VFIO_IOMMU_MODE_UNSAFE` → NOIOMMU IS enabled **This is an inverted condition.** The old code returned `RTE_IOVA_VA` when IOMMU was safe (NOIOMMU disabled), but the new code returns `RTE_IOVA_VA` when IOMMU is unsafe (NOIOMMU enabled). This appears to be a correctness bug that will change the IOVA mode selection behavior. **Suggested fix:** The new condition should likely be: ```c if (fslmc_all_device_support_iova() != 0 && dev_vfio_get_iommu_mode() != DEV_VFIO_IOMMU_MODE_UNSAFE) return RTE_IOVA_VA; ``` Or if the intent was to check for safe mode: ```c if (fslmc_all_device_support_iova() != 0 && dev_vfio_get_iommu_mode() == DEV_VFIO_IOMMU_MODE_SAFE) return RTE_IOVA_VA; ```
Yep, a typo. -- Thanks, Anatoly
