On Thu, Jan 30, 2003 at 10:08:43AM +0530, MAASK Group wrote: > From what knowledge I have, the dirty bit in the page table entry of a > page is used to indicate whether the page has been modified. I suppose its > usage is for knowing if this page has been written to the buffer cache (to > be written to the disk later), before swapping it out. But this is in case > of pages to be read/written to the I/O. What is the use of the dirty bit in > case of shared memory? I saw that the swapper wont swap out dirty pages and > thus doesnt swap out the shared memory pages which is needed. Is there any > other use pf the dirty buit in case shared memory pages? Correct me if I am > wrong anywhere here!! Generaly speaking, buffer cache is specific to block device access and other filesystems (eq. network) do without it. It does not matter from the perspective of the page cache. A page in page cache either belongs to a file or is anonymous (and thus belongs to swap). If it belongs to a file and is dirty too long, it's writepage method is called to flush it to disk. If some pages are needed, the least recently used page is taken and if it's dirty, it's writepage method is called. Then is cleaned and reused - that is, if it's clean, it's simply forgoten, because it exists somewhere. As for not swaping used pages, another technique is used. Kernel walks the pages, copies dirty bits to page structs (not all pages in page cache have PTE - only those mmaped) and marks the pages unmapped. When such a page is accessed, a minor page fault occurs and it's simply marked mapped again. When it's not accessed for some time, it's swapped out (by the rule above - if it's dirty, it's writen, else simply discarded). Some architectures have an access bit in addition to dirty bit, so unmaping is not needed - kernel just checks if the accessed bit wasn't set within some period of time. ------------------------------------------------------------------------------- Jan 'Bulb' Hudec <bulb@ucw.cz> -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
