Skip to contentSkip to conversation

Documentation: Include cpupool example in install

PATCH Applied George Dunlap 3 participants 4 messages Feb 14, 2011 – Feb 15, 2011

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Documentation: Include cpupool example in install

xl cpupool-create at the moment requires a config file. Make sure to include the example config file in the install.

tools/examples/Makefile +1 -0
diff -r 4ea36cce2519 -r e66f078e198d tools/examples/Makefile
@@ -22,6 +22,7 @@
 XEN_CONFIGS += xend-pci-quirks.sxp
 XEN_CONFIGS += xend-pci-permissive.sxp
 XEN_CONFIGS += xl.conf
+XEN_CONFIGS += cpupool
 
 .PHONY: all
 all:
#1 lore raw

Am 14.02.2011 16:59, schrieb George Dunlap:

Quoted 2 lines

xl cpupool-create at the moment requires a config file. Make sure to include the example config file in the install.

Good point. Btw: When I did some experiments with cpupools, I couldn't create the cpupool config file on the fly and pipe it to the xl cpupool-create command, as it:
a) explicitly checks for a regular file to be named (and doesn't handle '-' or can cope with /dev/stdin)
b) stats the file to learn the file size (requiring this to be a seekable file, that's why check a)

These are restrictions which apply to guest config files, too, as they use the same functionality.
Is this just a limit in the implementation or is the requirement for a regular on-disk file there for a reason? I could easily create a patch to do away with these restrictions (after the release), are there any arguments against this?

Regards,
Andre.

--
Andre Przywara
AMD-Operating System Research Center (OSRC), Dresden, Germany

#2 lore raw

André Przywara writes ("Re: [Xen-devel] [PATCH] Documentation: Include cpupool example in install"):

Quoted 2 lines

Is this just a limit in the implementation or is the requirement for a regular on-disk file there for a reason?

It's a limit in the implementation. We try to read the config file into a char array in memory, and that's most easily done by stat/malloc/fread.

For a domain, the config file is used both immediately when creating the domain (being parsed by being fed to the flex bytearray scanner entrypoint) but also stored for use when the domain is migrated/saved.

Quoted 2 lines

I could easily create a patch to do away with these restrictions (after the release), are there any arguments against this?

No, that would be fine. The right answer is probably somewhere around libxl_read_file_contents. Perhaps we need libxl_read_fd_contents which reallocs.

Ian.