Yeah - I see lots of BUG_ON(!irqs_disabled()) in zcache code while placing and compressing the page(and also other places). I am not clear what you mean when you say zcache does so for other reasons. My guess was when clean zcache_put_page is called it holds locks (mapping->tree_lock) that conflicts xsmalloc pool allocations. If zsmalloc has no such issues, can these be removed? Thanks, Asim - Adding the correct kernel newbies address with entire conversation intact for reference. On Wed, Jun 27, 2012 at 1:26 PM, Seth Jennings <sjenning@linux.vnet.ibm.com> wrote:
On 06/27/2012 01:04 PM, Asim wrote:
This query relates zcache and the new zsmalloc patches.
I was looking through zcache code and your various zsmalloc patches. I am aware the zcache disables interrupts due to xvmalloc pool allocations.However, does the zsmalloc code also needs to disable interrupts while placing a page in zcache?
(Adding Nitin to Cc)
None of the functions need interrupts/preemption to be disabled before calling them, even though zcache does for other reasons.
zs_map_object() will return with page faults and preemption disabled by kmap_atomic() and get_cpu_var() respectively. This is to insure atomic access to the mapping_area structure. They are re-enabled in zs_unmap_object().
-- Seth