Re: PS3: trouble with SPARSEMEM_VMEMMAP and kexec

[Geert Uytterhoeven <Geert.Uytterhoeven@sonycom.com>]

On Mon, 3 Dec 2007, Milton Miller wrote:
> On Dec 2, 2007, at 9:59 PM, Geoff Levand wrote:
> > I'm finding that recently kexec'ed kernels on PS3 will
> > panic on startup.  It seems the trouble was introduced
> > with the ppc64 SPARSEMEM_VMEMMAP support.  The problem
> > is the same when starting either new or old kernels:
> > 
> > 2.6.24 -> 2.6.23 ok
> > 2.6.24 -> 2.6.23 panic
> > 2.6.24 -> 2.6.24 panic
> 
> I'm not sure I completely follow this.  What is the difference between 1 and 2
> ?   Also, you are talking about starting with kexec, but I don't see how that

I think the first line should be `2.6.23 -> 2.6.23' (i.e. a 2.6.23 kernel
kexec's a 2.6.24(-rc) kernel).

> fits in the failure you have below.  In other words, there may be more than
> one failure.  But I can talk a bit about the scope of the problem in the
> current traceback.

The problem can be triggered in other ways, e.g. playing with the ps3fb=xxxM
parameter, which specifies how much memory is reserved for ps3fb.
With 2.6.23, I can boot with `ps3fb=48M'. Very early in 2.6.24-rc*, this
changed.
2.6.24 seems to be easier fragmented than 2.6.23.

> > These are the commits that seem to introduce the problem:
> > 
> >   d29eff7bca60c9ee401d691d4562a4abca8de543 ppc64: SPARSEMEM_VMEMMAP suppor
> >   8f6aac419bd590f535fb110875a51f7db2b62b5b Generic Virtual Memmap support
> > for SPARSEMEM
> > 
> > 
> > Below is a startup dump.  Any help in finding the problem
> > would be appreciated.
> > 
> > -Geoff
> > 
> > 
> > 
> > ps3_mm_add_memory:317: start_addr 740320000000h, start_pfn 740320000h,
> > nr_pages 17000h
> > <4>swapper: page allocation failure. order:12, mode:0x80d0
> > Call Trace:
> > [c000000006047820] [c00000000000e700] .show_stack+0x68/0x1b0 (unreliable)
> > [c0000000060478c0] [c000000000089eb4] .__alloc_pages+0x358/0x3ac
> > [c0000000060479b0] [c0000000000a3964] .vmemmap_alloc_block+0x6c/0xf4
> > [c000000006047a40] [c000000000026544] .vmemmap_populate+0x74/0x100
> > [c000000006047ae0] [c0000000000a385c] .sparse_mem_map_populate+0x38/0x5c
> > [c000000006047b70] [c0000000000a36e4] .sparse_add_one_section+0x64/0x128
> > [c000000006047c20] [c0000000000aa74c] .__add_pages+0xac/0x18c
> > [c000000006047cd0] [c000000000025fd4] .arch_add_memory+0x44/0x60
> > [c000000006047d60] [c0000000000aa5b0] .add_memory+0xd4/0x124
> > [c000000006047e00] [c000000000452544] .ps3_mm_add_memory+0x8c/0x108
> > [c000000006047ea0] [c0000000004417c4] .kernel_init+0x1f4/0x3b8
> > [c000000006047f90] [c000000000021d88] .kernel_thread+0x4c/0x68
> > Mem-info:
> > DMA per-cpu:
> > CPU    0: Hot: hi:   42, btch:   7 usd:   0   Cold: hi:   14, btch:   3 usd:
> > 0
> > CPU    1: Hot: hi:   42, btch:   7 usd:   0   Cold: hi:   14, btch:   3 usd:
> > 0
> > Active:0 inactive:0 dirty:0 writeback:0 unstable:0
> >  free:18094 slab:122 mapped:0 pagetables:0 bounce:0
> > DMA free:72376kB min:0kB low:0kB high:0kB active:0kB inactive:0kB
> > present:129280kB pages_scanned:0 all_unreclaimable? no
> > lowmem_reserve[]: 0 0 0
> > DMA: 8*4kB 5*8kB 5*16kB 7*32kB 3*64kB 5*128kB 4*256kB 3*512kB 5*1024kB
> > 3*2048kB 4*4096kB 5*8192kB 0*16384kB = 72376kB
> > Swap cache: add 0, delete 0, find 0/0, race 0+0
> > Free swap  = 0kB
> > Total swap = 0kB
> > Free swap:            0kB
> > 32768 pages of RAM
> > 10403 reserved pages
> > 0 pages shared
> > 0 pages swap cached
> 
> The kernel is using 16MB pages for the linear mapping and, since its in the
> same region, the sparse virtural memmap.  PS3 uses hotplug for all most all of
> its memory.   In this case, its trying to allocate an additional page to cover
> a new region of the memory map.   However, the initial 128 MB is fragmented,
> we have 8 8M chunks but no 16MB ones.
> 
> > <1>Unable to handle kernel paging request for data at address
> > 0xcf0001960b000010
> > <1>Faulting instruction address: 0xc000000000087340
> > Oops: Kernel access of bad area, sig: 11 [#1]
> > SMP NR_CPUS=2 PS3
> > Modules linked in:
> > NIP: c000000000087340 LR: c00000000008733c CTR: 0000000000000000
> > REGS: c000000006047900 TRAP: 0300   Not tainted
> > (2.6.24-rc3-ps3-linux-dev-g91428d55-dirty)
> > MSR: 8000000000008032 <EE,IR,DR>  CR: 22004444  XER: 00000000
> > DAR: cf0001960b000010, DSISR: 0000000042000000
> > TASK = c000000006041080[1] 'swapper' THREAD: c000000006044000 CPU: 1
> > <6>GPR00: 0000000000000000 c000000006047b80 c00000000052b410
> > c000000006001b40
> > <6>GPR04: 0000000000000001 0000000000000003 0000000000000008
> > 0000000000000000
> > <6>GPR08: 0000000000000002 cf0001960b000008 c000000006051240
> > 0000000000000003
> > <6>GPR12: 0000000000000003 c000000000484080 00000000100d0000
> > 0000000000bc5000
> > <6>GPR16: 0000000007fff000 0000000000000001 00000000100a0000
> > 00000000100d0000
> > <6>GPR20: 0000000000000000 00000000100df628 00000000100df458
> > 00000000100df678
> > <6>GPR24: 0000000000740336 c000000000492c00 0000000000000000
> > 0000000000000001
> > <6>GPR28: 0000000740325000 0000000740324924 c0000000004ce9a8
> > cf0001960affffe0
> > NIP [c000000000087340] .memmap_init_zone+0xf0/0x134
> > LR [c00000000008733c] .memmap_init_zone+0xec/0x134
> > Call Trace:
> > [c000000006047b80] [c0000000001da530] .add_memory_block+0xd8/0x108
> > (unreliable)
> > [c000000006047c20] [c0000000000aa7ac] .__add_pages+0x10c/0x18c
> > [c000000006047cd0] [c000000000025fd4] .arch_add_memory+0x44/0x60
> > [c000000006047d60] [c0000000000aa5b0] .add_memory+0xd4/0x124
> > [c000000006047e00] [c000000000452544] .ps3_mm_add_memory+0x8c/0x108
> > [c000000006047ea0] [c0000000004417c4] .kernel_init+0x1f4/0x3b8
> > [c000000006047f90] [c000000000021d88] .kernel_thread+0x4c/0x68
> > Instruction dump:
> > 901f000c 38000400 7d20f8a8 7d290378 7d20f9ad 40a2fff4 7ba00521 7fe3fb78
> > 38800002 41820008 4bffff0d 393f0028 <f9290008> f93f0028 3bbd0001 3bff0038
> > <0>Kernel panic - not syncing: Attempted to kill init!
> 
> Instead of detecting the fail and aborting the add, we proceed to dereference
> the memory map.
> 
> Chris, as you can see, PS3 needs to allocate 1/8th of total initial memory to
> add any more memory.  Geoff, can you predict what linear address the
> additional memory will occupy?  Judging from the attempted address toa add,
> maybe not.   If not, my only thought is to pre-reserve an additional page and
> consume it on the first add.   Additional adds will likely draw from the first
> added region, pinning.

To me it sounds a bit strange that hotplug memory relies on having huge
contiguous blocks of memory available. If this isn't done very early in the
boot process, changes are high it will fail.

Would it be possible to allocate the memory from the newly added block, which
is guaranteed to be unfragmented?

With kind regards,
 
Geert Uytterhoeven
Software Architect

Sony Network and Software Technology Center Europe
The Corporate Village · Da Vincilaan 7-D1 · B-1935 Zaventem · Belgium
 
Phone:    +32 (0)2 700 8453	
Fax:      +32 (0)2 700 8622	
E-mail:   Geert.Uytterhoeven@sonycom.com	
Internet: http://www.sony-europe.com/
 	
Sony Network and Software Technology Center Europe	
A division of Sony Service Centre (Europe) N.V.	
Registered office: Technologielaan 7 · B-1840 Londerzeel · Belgium	
VAT BE 0413.825.160 · RPR Brussels	
Fortis Bank Zaventem · Swift GEBABEBB08A · IBAN BE39001382358619