Date: Mon, 18 Oct 2010 23:01:33 +0200 From: Giovanni Trematerra <gianni@freebsd.org> To: Andriy Gapon <avg@freebsd.org> Cc: alc@freebsd.org, freebsd-current@freebsd.org Subject: Re: panic in uma_startup for many-core amd64 system Message-ID: <AANLkTim-z-rNvPa%2BFDGAb7oroKB2DWxZSECti=ioH8GD@mail.gmail.com> In-Reply-To: <4CBC5719.1020807@freebsd.org> References: <4C9B9B9C.6000807@freebsd.org> <4CBBEBDF.3060905@freebsd.org> <AANLkTi=O4GtAKDqEr%2BR27E5Xe%2BdGBZc0d2_=KpobtuSW@mail.gmail.com> <4CBC5719.1020807@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, Oct 18, 2010 at 4:18 PM, Andriy Gapon <avg@freebsd.org> wrote: > on 18/10/2010 16:40 Giovanni Trematerra said the following: >> On Mon, Oct 18, 2010 at 8:40 AM, Andriy Gapon <avg@freebsd.org> wrote: >>> on 23/09/2010 21:25 Andriy Gapon said the following: >>>> >>>> Jeff, >>>> >>>> just for the kicks I tried to emulate a machine with 64 logical CPUs u= sing >>>> qemu-devel port: >>>> qemu-system-x86_64 -smp sockets=3D4,cores=3D8,threads=3D2 ... >>>> >>>> It seems that FreeBSD agreed to recognize only first 32 CPUs, but it p= aniced anyway. >>>> >>>> Here's a backtrace: >>>> #34 0xffffffff804fe7f5 in zone_alloc_item (zone=3D0xffffffff80be1554, >>>> udata=3D0xffffffff80be1550, flags=3D1924) at /usr/src/sys/vm/uma_core.= c:2506 >>>> #35 0xffffffff804ff35d in hash_alloc (hash=3D0xffffff001ffdb030) at >>>> /usr/src/sys/vm/uma_core.c:483 >>>> #36 0xffffffff804ff642 in keg_ctor (mem=3DVariable "mem" is not availa= ble. >>>> ) at /usr/src/sys/vm/uma_core.c:1396 >>>> #37 0xffffffff804fe91b in zone_alloc_item (zone=3D0xffffffff80a1f300, >>>> udata=3D0xffffffff80be1b60, flags=3D2) at /usr/src/sys/vm/uma_core.c:2= 544 >>>> #38 0xffffffff804ff92e in zone_ctor (mem=3DVariable "mem" is not avail= able. >>>> ) at /usr/src/sys/vm/uma_core.c:1832 >>>> #39 0xffffffff804ffca4 in uma_startup (bootmem=3D0xffffff001ffac000, b= oot_pages=3D48) >>>> at /usr/src/sys/vm/uma_core.c:1741 >>>> #40 0xffffffff80514822 in vm_page_startup (vaddr=3D1844674407157681766= 4) at >>>> /usr/src/sys/vm/vm_page.c:360 >>>> #41 0xffffffff805060c5 in vm_mem_init (dummy=3DVariable "dummy" is not= available. >>>> ) at /usr/src/sys/vm/vm_init.c:118 >>>> #42 0xffffffff803258b9 in mi_startup () at /usr/src/sys/kern/init_main= .c:253 >>>> #43 0xffffffff8017177c in btext () at /usr/src/sys/amd64/amd64/locore.= S:81 >>>> [[[ >>>> Note: >>>> 1. Frame numbers are high because the backtrace is obtained via gdb re= motely >>>> connected to qemu and also there is bunch of extra frames from DDB, et= c. >>>> 2. Line numbers in uma_core. won't match those in FreeBSD tree, becaus= e I've doing >>>> some unrelated hacking in the file. >>>> ]]] >>>> >>>> The problem seems to be with creation of "UMA Zones" zone and keg. >>>> Because of the large number of processors, size argument in the follow= ing snippet >>>> is set to a value of 4480: >>>> >>>> args.name =3D "UMA Zones"; >>>> args.size =3D sizeof(struct uma_zone) + >>>> =A0 =A0 (sizeof(struct uma_cache) * (mp_maxid + 1)); >>>> >>>> Because of this, keg_ctor() calls keg_large_init(): >>>> >>>> else if ((keg->uk_size+UMA_FRITM_SZ) > >>>> =A0 =A0 (UMA_SLAB_SIZE - sizeof(struct uma_slab))) >>>> =A0 =A0 =A0 =A0 keg_large_init(keg); >>>> else >>>> =A0 =A0 =A0 =A0 keg_small_init(keg); >>>> >>>> keg_large_init sets UMA_ZONE_OFFPAGE and UMA_ZONE_HASH flags for this = keg. >>>> This leads to hash_alloc() being invoked from keg_ctor(): >>>> >>>> if (keg->uk_flags & UMA_ZONE_HASH) >>>> =A0 =A0 =A0 =A0 hash_alloc(&keg->uk_hash); >>>> >>>> But the problem is that "UMA Hash" zone is not created yet and thus th= e call leads >>>> to the panic. =A0"UMA Hash" zone is the last of system zones created. >>>> >>>> Not sure what the proper fix here could/should be. >>>> Would it work to simply not set UMA_ZONE_HASH flag when UMA_ZFLAG_INTE= RNAL is set? >>>> >>>> >>>> And some final calculations. >>>> On the test system sizeof(struct uma_cache) is 128 bytes and (mp_maxid= + 1) is 32, >>>> so it's already UMA_SLAB_SIZE =3D PAGE_SIZE =3D 4096. >>>> >>> >>> Here is a simple solution that seems to work: >>> http://people.freebsd.org/~avg/uma-many-cpus.diff >>> Not sure if it's the best we can do. >>> >> >> I don't know if it makes sense I only want to raise a flag. >> Is it safe to call kmem_malloc() before bucket_init() during >> uma_startup() to reserve room for CPU caches? > > Hmm, not sure what exactly you mean. Sorry, nevermind > >> Reading the top uma_int.h comment, it seems that the best way to >> handle this issue >> would be to implement and allow for dynamic slab sizes. > > Again, not sure if I follow you, I don't see relation between per-cpu cac= hes and > dynamic slab size. Your patch seems just a work around about initial slab size where the keg is backed. Having dynamic slab sizes would allow to have the keg backed on a larger sl= ab without going OFFPAGE. -- Giovanni Trematerra
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?AANLkTim-z-rNvPa%2BFDGAb7oroKB2DWxZSECti=ioH8GD>