svn commit: r205487 - head/sys/vm

[K. Macy]

On Tue, Mar 23, 2010 at 6:54 AM, John Baldwin <jhb at freebsd.org> wrote:
> On Tuesday 23 March 2010 6:33:51 am K. Macy wrote:
>> The size change causes the initialization path to change in a way that
>> currently causes crashes.
>
> Are you planning on debugging it further?  Does UMA_BOOTPAGES or NKPT need to
> be larger?
>
>> On Mar 23, 2010 2:57 AM, "Ivan Voras" <ivoras at freebsd.org> wrote:
>>
>> On 22 March 2010 23:39, Kip Macy <kmacy at freebsd.org> wrote:
>> > Author: kmacy
>> > Date: Mon Mar 22 22:39...
>> Does this mean you have determined that aligning these structures is
>> not as beneficial on i386 or is there something else going on?
>>
>
> --
> John Baldwin


>From a mail I sent to Jeff:

This pushes things through keg_large_init (whereas previously it went
through keg_small_init) which breaks things on i386.
I can workaround the fact that hash_alloc gets called before hashzone
is initialized but trying to allocate a slab while during
initialization of the slab zone is problematic. Suggestions are
welcome.


(gdb) bt
#0  keg_alloc_slab (keg=0xc0ab3000, zone=0xc0547c80, wait=2)
   at /usr/home/kmacy/svn_checkouts/head_flowtable/sys/vm/uma_core.c:823
#1  0xc02dbfaf in keg_fetch_slab (keg=0xc0ab3000, zone=0xc0547c80, flags=2)
   at /usr/home/kmacy/svn_checkouts/head_flowtable/sys/vm/uma_core.c:2159
#2  0xc02dc2cc in zone_fetch_slab (zone=0xc0547c80, keg=0xc0ab3000, flags=2)
   at /usr/home/kmacy/svn_checkouts/head_flowtable/sys/vm/uma_core.c:2219
#3  0xc02db3ab in zone_alloc_item (zone=0xc0547c80, udata=0xc073cc48, flags=2)
   at /usr/home/kmacy/svn_checkouts/head_flowtable/sys/vm/uma_core.c:2475
#4  0xc02db74f in uma_zcreate (name=0xc037ef33 "UMA Slabs", size=284,
ctor=0, dtor=0, uminit=0,
   fini=0, align=3, flags=536870912)
   at /usr/home/kmacy/svn_checkouts/head_flowtable/sys/vm/uma_core.c:1827
#5  0xc02dcc6b in uma_startup (bootmem=0xc0a64000, boot_pages=80)