Messages in this thread | | | Date | Wed, 25 Mar 2015 14:05:33 +0900 | From | Heesub Shin <> | Subject | Re: [PATCH 1/2] zsmalloc: do not remap dst page while prepare next src page |
| |
Hello,
On 03/25/2015 12:24 AM, Sergey Senozhatsky wrote: > object may belong to different pages. zs_object_copy() handles > this case and maps a new source page (get_next_page() and > kmap_atomic()) when object crosses boundaries of the current > source page. But it also performs unnecessary kunmap/kmap_atomic > of the destination page (it remains unchanged), which can be > avoided.
No, it's not unnecessary. We should do kunmap_atomic() in the reverse order of kmap_atomic(), so unfortunately it's inevitable to kunmap_atomic() both on d_addr and s_addr.
> > Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com> > --- > mm/zsmalloc.c | 2 -- > 1 file changed, 2 deletions(-) > > diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c > index d920e8b..7af4456 100644 > --- a/mm/zsmalloc.c > +++ b/mm/zsmalloc.c > @@ -1536,12 +1536,10 @@ static void zs_object_copy(unsigned long src, unsigned long dst, > break; > > if (s_off + size >= PAGE_SIZE) { > - kunmap_atomic(d_addr); > kunmap_atomic(s_addr);
Removing kunmap_atomic(d_addr) here may cause BUG_ON() at __kunmap_atomic().
I tried yours to see it really happens: > kernel BUG at arch/arm/mm/highmem.c:113! > Internal error: Oops - BUG: 0 [#1] SMP ARM > Modules linked in: > CPU: 2 PID: 1774 Comm: bash Not tainted 4.0.0-rc2-mm1+ #105 > Hardware name: ARM-Versatile Express > task: ee971300 ti: e8a26000 task.ti: e8a26000 > PC is at __kunmap_atomic+0x144/0x14c > LR is at zs_object_copy+0x19c/0x2dc
regards heesub
| |