How do you comprehend the saying that the kernel's memory is not pageable whereas get_free_page use a page-oriented technique?
As per the documentation( https://www.linuxjournal.com/article/6930 ), which says[emphasis mine]: Unfortunately for kernel developers, allocating memory in the kernel is not as simple as allocating memory in userspace. A number of factors contribute to the complication, among them: The kernel is limited to about 1GB of virtual and physical memory. **The kernel's memory is not pageable.** As per the documentation(https://www.oreilly.com <https://www.oreilly.com/library/view/linux-device-drivers/0596005903/ch08.ht...> /library/view/linux-device-drivers/0596005903/ch08.html <https://www.oreilly.com/library/view/linux-device-drivers/0596005903/ch08.ht...> ), which says: get_free_page and Friends If a module needs to allocate big chunks of memory, it is usually better to use a page-oriented technique. I am confused after I have seen these two sayings. I think they have the opposite meaning: the former one clams that the kernel's memory is not pageable whereas the latter one implicitly states there is a page-oriented technique used by the kernel(i.e. function get_free_page depends a page-oriented technique). I have thought and thought about it for a long time, but I still don't comprehend them.I would be grateful to have some help with this question. Thank you for your attention to this matter.
On Sat, 20 Jun 2020 14:18:21 +0800, 孙世龙 sunshilong said:
Unfortunately for kernel developers, allocating memory in the kernel is not as simple as allocating memory in userspace. A number of factors contribute to the complication, among them: The kernel is limited to about 1GB of virtual and physical memory. **The kernel's memory is not pageable.**
If a module needs to allocate big chunks of memory, it is usually better to use a page-oriented technique.
Due to memory fragmentation, if a module needs (say) 2M of memory for an I/O buffer, it's more likely to be able to allocate 512 4K pages scattered through the 1GB of memory than it is to get 1 contiguous chunk of memory. The fact it's not pageable doesn't mean that pages aren't relevant as the unit of allocation.
Unfortunately for kernel developers, allocating memory in the kernel is not as simple as allocating memory in userspace. A number of factors contribute to the complication, among them: The kernel is limited to about 1GB of virtual and physical memory. **The kernel's memory is not pageable.**
If a module needs to allocate big chunks of memory, it is usually better to use a page-oriented technique.
Due to memory fragmentation, if a module needs (say) 2M of memory for an I/O buffer, it's more likely to be able to allocate 512 4K pages scattered through the 1GB of memory than it is to get 1 contiguous chunk of memory.
The fact it's not pageable doesn't mean that pages aren't relevant as the unit of allocation.
Thank you for your help. For the saying that the kernel's memory is not pageable, I think the word "pageable" should be understood in this way: Linux keeps the whole kernel in physical memory at all the times and no such memory would be temporarily moved to swap even if it is not currently in use. Am I right? If my understanding is right, one more question raises, what about the memory related to the user processes, e.g. process control block? Can it be swapped out? Thank you for your attention to this matter. Look forward to hearing from you. Valdis Klētnieks <valdis.kletnieks@vt.edu> 于2020年6月20日周六 下午5:11写道：
On Sat, 20 Jun 2020 14:18:21 +0800, 孙世龙 sunshilong said:
Unfortunately for kernel developers, allocating memory in the kernel is not as simple as allocating memory in userspace. A number of factors contribute to the complication, among them: The kernel is limited to about 1GB of virtual and physical memory. **The kernel's memory is not pageable.**
If a module needs to allocate big chunks of memory, it is usually better to use a page-oriented technique.
Due to memory fragmentation, if a module needs (say) 2M of memory for an I/O buffer, it's more likely to be able to allocate 512 4K pages scattered through the 1GB of memory than it is to get 1 contiguous chunk of memory.
The fact it's not pageable doesn't mean that pages aren't relevant as the unit of allocation.
participants (2)
-
Valdis Klētnieks -
孙世龙 sunshilong