4. DMA驱动初始化：PCIe设备探测（probe）、BAR空间映射、MSI/MSI-X中断申请、DMA引擎初始化

static int my_pcie_probe(struct pci_dev *pdev, const struct pci_device_id *id)
{
    struct my_dev *dev;
    int ret;

    // 1. 分配私有数据结构
    dev = devm_kzalloc(&pdev->dev, sizeof(*dev), GFP_KERNEL);
    if (!dev)
        return -ENOMEM;

    // 2. 使能PCIe设备
    ret = pci_enable_device(pdev);
    if (ret) {
        dev_err(&pdev->dev, "pci_enable_device failed: %d\n", ret);
        return ret;
    }

    // 3. 设置DMA掩码（64位优先）
    ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
    if (ret) {
        ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
        if (ret) {
            dev_err(&pdev->dev, "DMA mask setup failed\n");
            goto err_disable;
        }
    }

    // 4. 映射BAR空间
    dev->bar_phys = pci_resource_start(pdev, 0);
    dev->bar_len  = pci_resource_len(pdev, 0);
    dev->bar_virt = pcim_iomap(pdev, 0, dev->bar_len);
    if (!dev->bar_virt) {
        dev_err(&pdev->dev, "BAR0 iomap failed\n");
        ret = -ENOMEM;
        goto err_disable;
    }

    // 5. 申请MSI-X中断（见4.3节）
    // 6. 初始化DMA引擎（见4.4节）

    pci_set_drvdata(pdev, dev);
    return 0;

err_disable:
    pci_disable_device(pdev);
    return ret;
}

// 查询最大MSI-X向量数
int nvec = pci_msix_vec_count(pdev);
if (nvec < 0) {
    dev_err(&pdev->dev, "MSI-X not supported\n");
    return nvec;
}

// 分配4个MSI-X向量
int ret = pci_alloc_irq_vectors(pdev, 1, 4, PCI_IRQ_MSIX);
if (ret < 0) {
    dev_err(&pdev->dev, "MSI-X allocation failed: %d\n", ret);
    return ret;
}

// 注册每个向量的中断处理函数
for (int i = 0; i < ret; i++) {
    int irq = pci_irq_vector(pdev, i);
    snprintf(dev->irq_name[i], 32, "mydev-%d", i);
    ret = request_irq(irq, my_interrupt_handler, 0,
                      dev->irq_name[i], &dev->dma_chan[i]);
    if (ret) {
        dev_err(&pdev->dev, "request_irq failed for vector %d\n", i);
        goto err_free_vectors;
    }
}

#define RING_SIZE  256
#define DESC_SIZE  32  // 每个描述符32字节

struct dma_desc {
    __le64 src_addr;
    __le64 dst_addr;
    __le32 length;
    __le32 flags;
} __attribute__((packed));

// 分配描述符环（物理连续，64字节对齐）
dev->ring_virt = dma_alloc_coherent(&pdev->dev,
                    RING_SIZE * DESC_SIZE,
                    &dev->ring_phys,
                    GFP_KERNEL);
if (!dev->ring_virt) {
    dev_err(&pdev->dev, "Failed to allocate descriptor ring\n");
    return -ENOMEM;
}

// 初始化所有描述符为无效状态
memset(dev->ring_virt, 0, RING_SIZE * DESC_SIZE);

// 配置DMA控制寄存器（MMIO）
writel(lower_32_bits(dev->ring_phys), dev->bar_virt + DMA_RING_BASE_LOW);
writel(upper_32_bits(dev->ring_phys), dev->bar_virt + DMA_RING_BASE_HIGH);
writel(RING_SIZE, dev->bar_virt + DMA_RING_SIZE);
writel(DMA_ENGINE_ENABLE, dev->bar_virt + DMA_CONTROL);