1. Optimize DMABUF Mkey Page Size in mlx5
优化 mlx5 中的 DMABUF Mkey 页大小

This patch series enables the mlx5 driver to dynamically select the optimal page size for DMABUF-based memory keys (mkeys), rather than relying on a fixed page size during registration.
该补丁集使 mlx5 驱动能够为基于 DMABUF 的内存密钥（mkey）动态选择最优的页大小，而不再在注册时依赖固定的页大小。

Previously, DMABUF memory registration always used a fixed 4KB page size for mkeys. This could lead to suboptimal performance, especially when the underlying memory layout supports larger page sizes.
此前，DMABUF 内存注册始终为 mkey 使用固定的 4KB 页大小。这种做法在底层内存布局支持更大页大小的情况下，会导致性能不佳。

The previous approach did not leverage the hardware's advertised capabilities for larger page sizes. Moreover, the driver failed to set the appropriate page size mask in the mkey configuration, which could result in invalid registrations when switching to very large pages.
这种旧的实现方式没有利用硬件公开支持的大页能力。此外，驱动程序在配置 mkey 时未设置正确的页大小掩码，在切换到非常大的页时可能导致注册无效。

This series improves DMABUF performance by:
本系列补丁通过以下方式提升了 DMABUF 的性能：

Dynamically selecting the best page size for a given memory region (MR) at creation time and upon page faults.
在内存区域（MR）创建时以及发生页错误时，动态选择最合适的页大小。

Correcting gaps in the previous implementation related to page size mask handling.
修复了此前实现中关于页大小掩码处理的缺陷。

By reducing the number of page table entries (and thus MTT/KSM descriptors) that the Host Channel Adapter (HCA) must walk through, this optimization lowers cache-line fetches and improves overall efficiency.
通过减少主机通道适配器（HCA）需遍历的页表项数量（从而减少 MTT/KSM 描述符），此优化降低了缓存行的访问次数，提高了整体效率。

2. RDMA: Support CQs with User Memory
RDMA：支持使用用户空间内存创建 CQ

This patch series introduces a standardized mechanism for creating Completion Queues (CQs) using preallocated memory supplied by userspace.
本系列补丁引入了一种标准化机制，允许使用用户空间预分配的内存创建完成队列（C