On Wed, 2012-11-07 at 16:29 -0700, Toshi Kani wrote: > On Wed, 2012-11-07 at 12:47 +0100, Karel Zak wrote: > > On Tue, Oct 23, 2012 at 01:49:46PM -0600, Toshi Kani wrote: > > > 1) The system has 4 CPUs > > > $ lscpu -a -e > > > CPU NODE SOCKET CORE L1d:L1i:L2 ONLINE > > > 0 0 0 0 0:0:0 yes > > > 1 0 1 1 1:1:1 yes > > > 2 0 2 2 2:2:2 yes > > > 3 0 3 3 3:3:3 yes > > > > > > 2) Eject cpu2 > > > # echo 1 > /sys/bus/acpi/devices/LNXCPU:02/eject > > > > > > 3) lscpu no longer recognizes cpu3 after cpu2 is ejected > > > $ lscpu -a -e > > > CPU NODE SOCKET CORE L1d:L1i:L2 ONLINE > > > 0 0 0 0 0:0:0 yes > > > 1 0 1 1 1:1:1 yes > > > > > > The following changes are made to address this issue. > > > - Use maxcpus to allocate and parse bitmaps. > > > - Set desc->ncpu from cpu/present, which includes both on-line > > > and off-line CPUs. > > > > In the directory tests we have a dump from some strange machine > > > > tests/ts/lscpu/dumps/sparc64-UltraSparc-T1.tar.gz > > > > where is 32 /sys/devices/system/cpu/cpu[0-9]* subdirectories, but the > > present mask is 0-23, all the masks from the dump: > > > > present: 0-23 > > possible: 0-31 > > online: 0-23 > > offline: 24-31 > > count: 32 (means ls -d sys/devices/system/cpu/cpu[0-9]* | wc -l) > > > > strange, right? But I guess we can ignore this dump, according to > > kernel Documentation/cpu-hotplug.txt the cpu[0-9]* subdirectories > > have to follow the present mask. Maybe the dump is broken or so.. > > Hi Karel, > > Thanks for the updated patch. It looks good. I have tested it on my > setup and it worked fine as well. > > I agree that the present mask and cpu directories need to be consistent. > I looked at the kernel code and found the following change, which might > be the cause of this issue. > > ==== > commit 9f13a1fd452f11c18004ba2422a6384b424ec8a9 > Author: Ben Hutchings <ben@xxxxxxxxxxxxxxx> > Date: Tue Jan 10 03:04:32 2012 +0000 > > cpu: Register a generic CPU device on architectures that currently > do not > > frv, h8300, m68k, microblaze, openrisc, score, um and xtensa > currently > do not register a CPU device. Add the config option > GENERIC_CPU_DEVICES > which causes a generic CPU device to be registered for each present > CPU, > and make all these architectures select it. > > Richard Weinberger <richard@xxxxxx> covered UML and suggested using > per_cpu. > > Signed-off-by: Ben Hutchings <ben@xxxxxxxxxxxxxxx> > Signed-off-by: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> > ==== > > > Hi Ben, > > In the change log above, you described that a generic CPU device to be > registered for each "present" CPU. However, in the code below, it used > for_each_possible_cpu(), instead of for_each_present_cpu(). Is this > what you intended? [...] I think the distinction doesn't matter - the GENERIC_CPU_DEVICES code does not support hotplug and therefore can only be used if present == possible. Ben. -- Ben Hutchings The program is absolutely right; therefore, the computer must be wrong.
Attachment:
signature.asc
Description: This is a digitally signed message part
