Re[3]: PS3 platform is broken on Linux 3.7.0

[Phileas Fogg]

 And another note.
I took a look at the MMU chapter in the Cell Architecture handbook and indeed the first 15 bits in VA are treated as 0 by the hardware.

Quote:

1. High-order bits above 65 bits in the 80-bit virtual address (VA[0:14]) are not implemented. The hardware always
   treats these bits as `0'. Software must not set these bits to any other value than `0' or the results are undefined in
   the PPE.


regards


Воскресенье, 10 февраля 2013, 12:59 +04:00 от Phileas Fogg <phileas-fogg at mail.ru>:
>Hi,
>
>i found where the problem lies.
>I also printed some values in ps3_hpte_insert with and without 64TB support, i used OpenWRT with Linux 3.7.6 for testing.
>
>Some values without 64TB support:
>-------------------------------------------------
>
>[    0.060487] RPC: Registered named UNIX socket transport module.
>[    0.060511] RPC: Registered udp transport module.
>[    0.060672] RPC: Registered tcp transport module.
>[    0.060873] RPC: Registered tcp NFSv4.1 backchannel transport module.
>[    0.061080] initcall .init_sunrpc+0x0/0xbc returned 0 after 784 usecs
>[    0.061280] calling  .populate_rootfs+0x0/0x120 @ 1
>[    0.061683] ps3_hpte_insert:result=0 vpn=f09b89af50101 pa=d4e0000 ix=dfa0 v=f09b89af5001 r=6c005d4e0194 psize=0 ssize=0 lpar=6c005d4e0000
>[    0.061733] ps3_hpte_insert:result=0 vpn=f09b89af50102 pa=d4e1000 ix=dfb8 v=f09b89af5001 r=6c005d4e1194 psize=0 ssize=0 lpar=6c005d4e1000
>[    0.061895] ps3_hpte_insert:result=0 vpn=f09b89af50103 pa=d4e2000 ix=dfb0 v=f09b89af5001 r=6c005d4e2194 psize=0 ssize=0 lpar=6c005d4e2000
>
>
>Some values with 64TB support:
>-------------------------------------------------
>
>[    0.076477] calling  .init_sunrpc+0x0/0xbc @ 1
>[    0.076992] RPC: Registered named UNIX socket transport module.
>[    0.077017] RPC: Registered udp transport module.
>[    0.077076] RPC: Registered tcp transport module.
>[    0.077277] RPC: Registered tcp NFSv4.1 backchannel transport module.
>[    0.077484] initcall .init_sunrpc+0x0/0xbc returned 0 after 784 usecs
>[    0.077684] calling  .populate_rootfs+0x0/0x120 @ 1
>[    0.078126] ps3_hpte_insert:result=-17 vpn=25008684d80101 pa=d567000 ix=2ec8 v=25008684d8001 r=6c005d567194 psize=0 ssize=0 lpar=6c005d567000
>[    0.078164] ps3_hpte_insert:result=-17 vpn=25008684d80101 pa=d567000 ix=2ec8 v=25008684d8001 r=6c005d567194 psize=0 ssize=0 lpar=6c005d567000
>[    0.078287] ------------[ cut here ]------------
>[    0.078482] Kernel BUG at c00000000002cb3c [verbose debug info unavailable]
>[    0.078686] Oops: Exception in kernel mode, sig: 5 [#1]
>[    0.078883] SMP NR_CPUS=2 PS3
>[    0.079084] Modules linked in:
>[    0.079287] NIP: c00000000002cb3c LR: c00000000002cb38 CTR: 00000000002ffc38
>[    0.079489] REGS: c00000000d04f0e0 TRAP: 0700   Not tainted  (3.7.6)
>[    0.079687] MSR: 8000000000020032 <SF,IR,DR,RI>  CR: 22000022  XER: 00000000
>[    0.079888] SOFTE: 0
>[    0.080090] TASK = c00000000d049060[1] 'swapper/1' THREAD: c00000000d04c000 CPU: 1
>GPR00: c00000000002cb38 c00000000d04f360 c0000000012ec8d0 0000000000000081 
>GPR04: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 
>GPR08: 0000000000000000 c00000000124ce10 0000000000000000 c00000000002bcf0 
>GPR12: 0000000022000022 c000000007ffe280 c000000000008c94 c0000000005cba00 
>
>
>
>And now take a look at 'v' values in both cases.
>
>Without 64TB support:   v=f09b89af5001
>With 64TB support: v=25008684d8001
>
>Number of leading zeros in f09b89af5001 is 16.
>Number of leading zeros in 25008684d8001 is 14.
>
>And that's why lv1_insert_htab_entry fails with -17 which means LV1_ILLEGAL_PARAMETER_VALUE because
>the Hypervisor of PS3 checks 'AVPN' values for number of leading zeros and allows at least 15 bits which in case
>of 'v' value 25008684d8001 is too small of course.
>
>Not sure how to fix it in current Linux kernel. You guys know it better than me.
>
>Regards
>
>
>
>
>Понедельник, 14 января 2013, 15:37 -08:00 от Geoff Levand < geoff at infradead.org >:
>>Hi,
>>
>>On Fri, 2013-01-11 at 18:12 -0800, Geoff Levand wrote:
>>> I checked these, and Michael's 407821a34fce89b4f0b031dbab5cec7d059f46bc
>>> does indeed cause the LV1 hypervisor to panic early, and if that is
>>> reverted, Aneesh's 048ee0993ec8360abb0b51bdf8f8721e9ed62ec4 hits a BUG.
>>
>>Just to give an update, I did a little more work on it and found that
>>the call to lv1_insert_htab_entry() inside ps3_hpte_insert() is
>>failing.
>>
>>     http://git.kernel.org/?p=linux/kernel/git/geoff/ps3-linux.git;a=blob;f=arch/powerpc/platforms/ps3/htab.c;hb=HEAD#l70
>>
>>The values of the variables printed all look strange compared with
>>commit 048ee0993 reverted.  I'll try do some more work on it this
>>week.
>>
>>-Geoff
>>
>>
>>
>>
>>
>>_______________________________________________
>>Linuxppc-dev mailing list
>>Linuxppc-dev at lists.ozlabs.org
>> https://lists.ozlabs.org/listinfo/linuxppc-dev
>
>_______________________________________________
>Linuxppc-dev mailing list
>Linuxppc-dev at lists.ozlabs.org
>https://lists.ozlabs.org/listinfo/linuxppc-dev

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ozlabs.org/pipermail/linuxppc-dev/attachments/20130210/d5a9c567/attachment.html>