Yet another approach to memory fragmentation

Posted Feb 14, 2005 8:40 UTC (Mon) by Nir (guest, #27440)
Parent article: Yet another approach to memory fragmentation

I have a question. I hope I am not too rude. i am trying to understand whether high memory is visible in interrupt context ? When sending/getting a buffer to disk or NIC from an application that mapped high memory pages ( > 1GB ) are they copied to a normal memory and then copied to the user buffer ? Thank you. Raz Ben Jehuda razb@bitband.com

to post comments

mapping of high-memory pages

Posted Feb 18, 2005 3:57 UTC (Fri) by xoddam (subscriber, #2322) [Link]

Ultimately, the high-memory pages will be mapped temporarily into the
kernel address space.

Driver writers have the option of using a buffer in normal memory and
copy_to_user()/copy_from_user() functions to transfer data to/from
userspace. These functions hide the details of the virtual address
mapping.

Alternatively the driver can use get_user_pages() to 'pin' the userspace
pages into memory and kmap() to get the kernel address of each page.
kmap() will return the fixed address of a normal page, or map a high
memory page into the kernel. The pages must be released afterwards with
kunmap() (a noop if the page is normal) and page_cache_release().