Re: [RFC RESEND v5 3/4] tests/qtest/libqos: Add SMMUv3 helper library

[Pierrick Bouvier <pierrick.bouvier@linaro.org>]

On 11/26/25 7:45 AM, Tao Tang wrote:
> Introduce qos-smmuv3, a reusable library for SMMUv3-related qtest
> operations. This module encapsulates common tasks like:
> 
> - SMMUv3 initialization (enabling, configuring command/event queues)
> - Stream Table Entry (STE) and Context Descriptor (CD) setup
> - Multi-level page table construction (L0-L3 for 4KB granules)
> - Support for Stage 1, Stage 2, and nested translation modes
> - Could be easily extended to support multi-space testing infrastructure
>      (Non-Secure, Secure, Root, Realm)
> 
> The library provides high-level abstractions that allow test code to
> focus on IOMMU behavior validation rather than low-level register
> manipulation and page table encoding. Key features include:
> 
> - Automatic memory allocation for translation structures with proper
>      alignment
> - Helper functions to build valid STEs/CDs for different translation
>      scenarios
> - Page table walkers that handle address offset calculations per
>      security space
> - Command queue management for SMMU configuration commands
> 
> This infrastructure is designed to be used by iommu-testdev-based tests
> and future SMMUv3 test suites, reducing code duplication and improving
> test maintainability.
> 
> Signed-off-by: Tao Tang <tangtao1634@phytium.com.cn>
> ---
>   tests/qtest/libqos/meson.build  |   3 +
>   tests/qtest/libqos/qos-smmuv3.c | 731 ++++++++++++++++++++++++++++++++
>   tests/qtest/libqos/qos-smmuv3.h | 267 ++++++++++++
>   3 files changed, 1001 insertions(+)
>   create mode 100644 tests/qtest/libqos/qos-smmuv3.c
>   create mode 100644 tests/qtest/libqos/qos-smmuv3.h
> 

...

> +
> +void qsmmu_single_translation(QSMMUTestContext *ctx)
> +{
> +    uint32_t config_result;
> +    uint32_t dma_result;
> +    bool test_passed;
> +
> +    /* Configure SMMU translation */
> +    config_result = qsmmu_setup_and_enable_translation(ctx);
> +    if (config_result != 0) {
> +        g_test_message("Configuration failed: mode=%u status=0x%x",
> +                       ctx->config.trans_mode, config_result);
> +        return;

Is that expected to silently return if we can't configure translation?

> +    }
> +
> +    /* Trigger DMA operation */
> +    dma_result = qsmmu_trigger_dma(ctx);
> +    if (dma_result != 0) {
> +        g_test_message("DMA failed: mode=%u result=0x%x",
> +                       ctx->config.trans_mode, dma_result);
> +    } else {
> +        g_test_message("-> DMA succeeded: mode=%u", ctx->config.trans_mode);
> +    }
> +
> +    /* Validate test result */
> +    test_passed = qsmmu_validate_test_result(ctx);
> +    g_assert_true(test_passed);
> +
> +    /* Clean up translation state to prepare for the next test */
> +    qsmmu_cleanup_translation(ctx);
> +}
> +
> +void qsmmu_translation_batch(const QSMMUTestConfig *configs, size_t count,
> +                             QTestState *qts, QPCIDevice *dev,
> +                             QPCIBar bar, uint64_t smmu_base)
> +{
> +    for (int i = 0; i < count; i++) {
> +        /* Initialize test memory */
> +        qtest_memset(qts, configs[i].dma_iova, 0x00, configs[i].dma_len);
> +        /* Execute each test configuration */
> +        QSMMUTestContext ctx = {
> +            .qts = qts,
> +            .dev = dev,
> +            .bar = bar,
> +            .smmu_base = smmu_base,
> +            .config = configs[i],
> +            .trans_status = 0,
> +            .dma_result = 0,
> +            .sid = dev->devfn,
> +            .tx_space = qsmmu_sec_sid_to_space(configs[i].sec_sid),
> +        };
> +
> +        qsmmu_single_translation(&ctx);
> +        g_test_message("--> Test %d completed: mode=%u sec_sid=%u "
> +                       "status=0x%x result=0x%x", i, configs[i].trans_mode,
> +                       configs[i].sec_sid, ctx.trans_status, ctx.dma_result);
> +    }
> +}

What is the reason for batching operations?
We are not in a performance critical scenario for running this test, so 
it's probably better to have distinct calls to single_translation.

...

For the rest of the patch, which is quite consequent, congrats. It's 
hard to review all the setup phase here, but knowing it works with the 
current smmuv3 implementation, that's a good proof that it's working as 
expected.