It is a configuration policy. The policy here is to adjust the cost of current page. The cost means the overhead for kernel to operate the page if the page is swapped out. Kernel adopts a decay policy. Specifically, if current cost is greater than lrusize/4, its cost will be divided by 2. If kernel has no this policy, after a long-term running, the cost of each page is very high. Kernel might mistakenly reserve pages which are frequently visited long time ago but are inactive currently. It might cause performance degradation. The value (lrusize/4) here has a trade-off between performance and sensitivity. For example, if the value is too small, kernel will frequently adjust the priority of each page, resulting in process's performance degradation. If the value is too large, kernel might be misleaded by historical data, causing wrongly swapping currently popular pages and further performance degradation.

It is a configuration policy. The policy here is to determine the size of page cluster. "2" and "3" is determined externally to the kernel. Page cluster is the actual execution unit when swapping. If the machine is small-memory, kernel executes swap operation on 4 (2^2) pages for each time. Otherwise, kernel operats 8 (2^3) for each time. The rational here is to avoid much memory pressure for small-memory system and to improve performance for large-memory system.

It is a configuration policy. It is used to count the physical page for memory control group. The policy here is one physical page corresponds to one count to the memory control group. If using huge page, the value here might not be 1,

It is a configuration policy. If kernel cannot allocate an enough virtual space with alignment, while nofail is specified to disallow failure, the kernel will let the process to sleep for 1 time slot. In this period, the process cannot be interrupted and quit the schedule queue of CPU. The "1" here is a configuration parameter, made externally to the kernel. It is not large enough for latency-sensitive process and is not small enough to retry frequently.

It is an algorithmic policy and also a configuration policy. The algorithmic policy here is to maxmize the data locality of virtual memory address, by letting the space be in continuous virtual pages. If the demanded size of virtual memory is larger than one page, the kernel will allocate multiple continuous pages for it by using __vmalloc_node(). Otherwise, the kernel will call kmalloc_node() to place it in a single page. On the other hand, to allocate continuous pages, we should invoke __vmalloc_node() by declaring align = 1, which is a configuration policy.

It is an algorithmic policy. The show_numa_info() is used for printing how the virtual memory of v (vm_struct) distributes its pages across numa nodes. The two if conditions are used for filtering the invalid v. The first if means the virtual memory has not been associated with physical page. The second if means the virtual memory has not been initialized yet.

It is an algorithmic policy. The policy here is to avoid lock/unlock overhead by using copy_page_to_iter_nofault() function plus if-else branch. Because copy_page_to_iter_nofault() is based on no page fault, we need to check the physical page is still alive. So the kernel uses an if condition to guarantee copy_page_to_iter_nofault() without page fault. On the other hand, if the page is not valid, kernel chooses to fill the iter using zero value instead of returning an error. I think it is because it can make invocation more convenient. The caller does not need to do a verification of the return value and does not need further actions to handle it.

It is an algorithmic policy. The two if conditions are used for checking the feasibility of the allocation. The first if means the adress of the allocation is beyond the allowable address, causing overflow. In this case, since the base address is decided from higher one to smaller one, reducing base address cannot work. Note that the base address is dependent on the va. The second if means we don't find a valid va which has appropriate base address. So the policy here is going to overflow branch to re-allocate va list.

It is an algorithmic policy. It is used for checking whether the start address of current va is valid. The policy here is to change a new va, different from the solution of "end address check" (explained in my other annotation). The reason is that we find the base address from high address to low address. If we continue to reduce the base address, the if condition will be always wrong. Changing a new va with a lower start address is the only solution.

It is an algorithmic policy. The outer loop is an endless loop until finding a vaild virtual memory space satisfying the requirements. The if condition here is used to guarantee the space in va is large enough. If not, we continue to scan the memory space from high address to low address while satisfying the alignment requirement. Specifically, tuning the base address to a lower address. And then, it will retry for verification.

It is an algorithmic policy. The for loop is used for transferring the virtual memory managment by using from vm_struct (link list) to vm_area (tree?). Inside the for loop, if we cannot allocate a new vmap_area from vmap_area_cachep (va = NULL), the program will simply print a warning and skip current tmp, instread of retry or anything to guarantee the integrity.

It is an algorithmic policy. The vmalloc_dump_obj() function is used for printing the information of specified virtual memory. Due to concurrency, before manipulating the critical state, we must obtain the lock. The policy here is spin_trylock. It only tries once. If fails, the function will simply returns false. I think the policy here is to avoid long-term waiting to improve the performance.