[lc-checkins] CVS: linux/mm oom_kill.c,1.7,1.8 memory.c,1.16,1.17 mmap.c,1.5,1.6

[Rodrigo S. de C. <rc...@us...>]

Update of /cvsroot/linuxcompressed/linux/mm
In directory usw-pr-cvs1:/tmp/cvs-serv13325/mm

Modified Files:
	memory.c mmap.c 
Added Files:
	oom_kill.c 
Log Message:
This batch of changes still includes lots of cleanups and code rewrite
to make it simpler. Perfomance increase has been noticed too.

- number_of_pages in comp_cache_t removed. We can check if there are
no fragments by fragments list.

- vswap: no semaphore is needed. I have no idea why the functions
{lock,unlock}_vswap has once been added. I can't see why they are
needed. So they were removed. The same for real_entry field in struct
vswap_address.

- vswap: a new function has been added, namely add_fragment_vswap(),
analogous to remove_fragment_vswap(). It's called from
get_comp_cache_page() and it's a great hand to make things modular.

- vm_enough_memory(): now we take into account compressed cache space
when allowing an application to allocate memory. That is done calling
a function named comp_cache_free_space() which returns, based upon the
estimated_free_space, the number of pages that still can be
compressed.

- move_and_fix_fragments() deleted. comp_cache_free() has a new police
to not move data to and fro all the time like before. We free the
fragment but leave it there waiting for being merged with the free
space. It's pretty simple, check the code. The new code has two new
functions: merge_right_neighbour() and merge_left_neighbour().

- the fragments list is kept sorted by offset field. So, when freeing,
we don't have to search for the next and previous fragments
everytime. Since most of times it's just a plain list_add_tail() in
get_comp_cache_page(), that makes the code simpler and nicer.

- lookup_comp_cache() was partially rewritten, mainly due to the fact
we won't sleep to get a lock on the comp_page.

- find_and_lock_comp_page() function removed and
find_nolock_comp_page() was renamed to find_comp_page(). All functions
that previously called find_and_lock... now calls the find_comp_page()
and locks the comp_page at once with TryLockPage().

- oom_kill() was fixed and takes into account the free space in
compressed cache by calling comp_cache_available_space(). That avoids
killing an application if we have space left in compressed cache yet.




Index: memory.c
===================================================================
RCS file: /cvsroot/linuxcompressed/linux/mm/memory.c,v
retrieving revision 1.16
retrieving revision 1.17
diff -C2 -r1.16 -r1.17
*** memory.c	2001/12/21 18:33:11	1.16
--- memory.c	2002/01/14 12:05:08	1.17
***************
*** 1081,1087 ****
  	for (i = 0; i < num; offset++, i++) {
  		/* Ok, do the async read-ahead now */
- 		lock_vswap(SWP_ENTRY(SWP_TYPE(entry), offset));
  		new_page = read_swap_cache_async(SWP_ENTRY(SWP_TYPE(entry), offset));
- 		unlock_vswap(SWP_ENTRY(SWP_TYPE(entry), offset));
  
  		if (!new_page)
--- 1081,1085 ----
***************
*** 1106,1116 ****
  
  	spin_unlock(&mm->page_table_lock);
- 	lock_vswap(entry);
- 
- 	if (!pte_same(*page_table, orig_pte)) {
- 		unlock_vswap(entry);
- 		return 1;
- 	}
- 	
  	page = lookup_comp_cache(entry);
  
--- 1104,1107 ----
***************
*** 1147,1151 ****
  		unlock_page(page);
  		page_cache_release(page);
- 		unlock_vswap(entry);
  		return 1;
  	}
--- 1138,1141 ----
***************
*** 1155,1159 ****
  
  	remove_pte_vswap(page_table);
- 	unlock_vswap(entry);
  	
  	/* The page isn't present yet, go ahead with the fault. */
--- 1145,1148 ----

Index: mmap.c
===================================================================
RCS file: /cvsroot/linuxcompressed/linux/mm/mmap.c,v
retrieving revision 1.5
retrieving revision 1.6
diff -C2 -r1.5 -r1.6
*** mmap.c	2001/12/13 19:12:58	1.5
--- mmap.c	2002/01/14 12:05:08	1.6
***************
*** 82,85 ****
--- 82,88 ----
  	free += swapper_space.nrpages;
  
+ 	/* Let's count the free space left in compressed cache */
+ 	free += comp_cache_free_space();
+ 
  	/*
  	 * The code below doesn't account for free space in the inode