kernel-packages team mailing list archive

[Launchpad Bug Tracker <1338666@xxxxxxxxxxxxxxxxxx>]

You have been subscribed to a public bug:

== Comment: #0 - David A. Harnett <doberman@xxxxxxxxxx> - 2014-03-12 20:43:09 ==
---Problem Description---
On Ubuntu 14.04 ppc64le in a Power7 LPAR , if kernel module nx_crypto is already loaded ,
attempts to load additional kernel modules fail with this message :

modprobe: ERROR: could not insert 'xxx': Invalid argument

where xxx is the name of module we tried to load .
dmesg shows:
[   33.012917] nx: scatter/gather list overflow, pid: 806

---uname output---
Linux c277f3u1crash2 3.13.0-17-generic #37-Ubuntu SMP Mon Mar 10 21:47:19 UTC 2014 ppc64le ppc64le ppc64le GNU/Linux
 
---Additional Hardware Info---
special PowerVM Little Endian firmware level b0108l_1446.780 
 
Machine Type = Power 780+ 9179-MHD LPAR 
 
---Debugger---
A debugger is not configured
 
---Steps to Reproduce---
 "lsmod" , if nx_crypto module not loaded , then "modprobe nx_crypto" ;
try to load any kernel module , "modprobe nfs"

 
Stack trace output:
 no
 
Oops output:
 no
 
System Dump Info:
  The system is not configured to capture a system dump.
 
*Additional Instructions for Dave Harnett/doberman@xxxxxxxxxx , no backup: 
-Post a private note with access information to the machine that the bug is occuring on. 
-Attach sysctl -a output output to the bug.

== Comment: #3 - David A. Harnett <doberman@xxxxxxxxxx> - 2014-03-12 21:19:42 ==
Circumvention : add directive "blacklist nx_crypto" to a file in directory /etc/modprobe.d

Problem does not occur on Ubuntu 14.04 powerpc64 Big Endian image
running in LPAR on same hardware .

== Comment: #5 - Leonidas Da Silva Barbosa <leosilva@xxxxxxxxxx> - 2014-03-13 14:48:17 ==
David, 

can you provide us with a

cat /etc/issue 
cat /proc/cpuinfo
uname -a 

from the system where the problem does not happen?

== Comment: #6 - David A. Harnett <doberman@xxxxxxxxxx> - 2014-03-13 15:24:55 ==
This is the requested output from the BE image in another LPAR on the same physical machine .

$ cat /etc/issue
Ubuntu Trusty Tahr (development branch) \n \l

$ cat /proc/cpuinfo
processor       : 0
cpu             : POWER7 (architected), altivec supported
clock           : 3724.000000MHz
revision        : 2.0 (pvr 004a 0200)

processor       : 1
cpu             : POWER7 (architected), altivec supported
clock           : 3724.000000MHz
revision        : 2.0 (pvr 004a 0200)

processor       : 2
cpu             : POWER7 (architected), altivec supported
clock           : 3724.000000MHz
revision        : 2.0 (pvr 004a 0200)

processor       : 3
cpu             : POWER7 (architected), altivec supported
clock           : 3724.000000MHz
revision        : 2.0 (pvr 004a 0200)

timebase        : 512000000
platform        : pSeries
model           : IBM,9179-MHD
machine         : CHRP IBM,9179-MHD

$ uname -a
Linux c277f3u1admin2 3.13.0-17-powerpc64-smp #37-Ubuntu SMP Mon Mar 10 21:58:15 UTC 2014 ppc64 ppc64 ppc64 GNU/Linux

== Comment: #10 - Leonidas Da Silva Barbosa <leosilva@xxxxxxxxxx> - 2014-03-14 13:19:42 ==
David,

The hardware that are you running both image is over a PowerVM LE ?

== Comment: #11 - David A. Harnett <doberman@xxxxxxxxxx> - 2014-03-14 13:32:42 ==
This model Power 780 (Saturn MR class) Power7 machine has special IBM internal use only firmware  that permits both LE and BE OS images to run in separate PowerVM LPARs simultaneously .  Our contact for the firmware is David Larson , larson1@xxxxxxxxxx

This is the third problem I observed on this machine where LE OS has
problems but the same function works correctly on a nearly equivalent
level of BE OS on the same physical machine .

1) nvram command fails with segfault - to be fixed in powerpc-utils package
2) ibmveth kernel driver for PowerVM virtual ethernet device - fixed in an Ubuntu 14.04 kernel
3) this problem
Common thread seems to be drivers that interface with firmware/hardware platform features .

== Comment: #17 - Cody P. Schafer <cpschafe@xxxxxxxxxx> - 2014-05-14 14:02:52 ==
I've seen issues on POWER8-PowerVM-ppc64le with nx-crypto, dmesg attached.

note: this is still LE, but is also on P8 (not P7 as this bug was
originally reported).

== Comment: #18 - Cody P. Schafer <cpschafe@xxxxxxxxxx> - 2014-05-14 14:04:55 ==
Also, in my case further modules load without issue. The only problem is the log spam.

== Comment: #19 - Cody P. Schafer <cpschafe@xxxxxxxxxx> - 2014-05-14 14:05:17 ==
kernel is v3.15-rc4-260-g38583f0 (from linus's tree)
uname -a:
Linux host 3.15.0-rc4-testing+ #8 SMP Wed May 14 02:21:38 PDT 2014 ppc64le ppc64le ppc64le GNU/Linux

== Comment: #20 - George C. Wilson <gcwilson@xxxxxxxxxx> - 2014-05-14 14:13:32 ==
Hi Cody. NX crypto isn't supported on little endian kernels. We don't intend to support it because the P8 VMX crypto instructions should provide faster AES acceleration and also userspace support for it. The nx_crypto should be blacklisted or disabled in the kernel config for little endian.

== Comment: #21 - Cody P. Schafer <cpschafe@xxxxxxxxxx> - 2014-05-14 14:15:28 ==
Ok, in that case someone needs to submit a patch to add a dependency on BE in the kernel's KConfig.

== Comment: #22 - George C. Wilson <gcwilson@xxxxxxxxxx> - 2014-05-14 14:23:46 ==
Leonidas and Marcelo, could you please submit such a patch? It should probably go upstream into the defconfig as well, I'd imagine.

== Comment: #26 -  Leonidas Da Silva Barbosa 2014-06-26 09:35:22 EDT ==

Kernel upstream already has this dependency:

here is the upstream commit
8b9f9269bcfb4fe9b44bb52aac1ce796834109fe

** Affects: linux (Ubuntu)
     Importance: Undecided
         Status: Fix Committed


** Tags: architecture-ppc64le bot-comment bugnameltc-105629 severity-critical
-- 
ppc64le kernel module nx_crypto blocks loading other kernel modules
https://bugs.launchpad.net/bugs/1338666
You received this bug notification because you are a member of Kernel Packages, which is subscribed to linux in Ubuntu.