[Xen-devel] [PATCH v3 0/2] qemu-qdisk: Implementation of grant copy operation.

[Paulina Szubarczyk <paulinaszubarczyk@xxxxxxxxx>]

Hi,

It is a proposition for implementation of grant copy operation in qemu-qdisk 
and 
interface in libxc/libs. 

Changes since v2:
Interface:
- dropped the changes in libxc/include/xenctrl_compat
- changed the MINOR version in Makefile
- replaced 'return -1' -> 'abort()'in libs/gnttab/gnttab_unimp.c
- moved the struct 'xengnttab_copy_grant_segment' to 
  libs/gnttab/include/xengnttab.h
- added explicit assingment to ioctl_gntdev_grant_copy_segment 
  to the linux part

qemu-qdisk:
- to use the xengnttab_* function directly added -lxengnttab to configure
  and include <xengnttab.h> in include/hw/xen/xen_common.h
- in ioreq_copy removed an out path, changed a log level, made explicit 
  assignement to 'xengnttab_copy_grant_segment'
* I did not change the way of testing if grant_copy operation is implemented.
  As far as I understand if the code from gnttab_unimp.c is used then the 
gnttab 
  device is unavailable and the handler to gntdev would be invalid. But 
  if the handler is valid then the ioctl should return operation unimplemented 
  if the gntdev does not implement the operation.


Changes since v1:
Interface:
- changed the interface to call grant copy operation to match ioctl
        int xengnttab_grant_copy(xengnttab_handle *xgt,
                                 uint32_t count,
                                 xengnttab_grant_copy_segment_t* segs)

- added a struct 'xengnttab_copy_grant_segment' definition to tools/libs        
  /gnttab/private.h, tools/libxc/include/xenctrl_compat.h

- changed the function 'osdep_gnttab_grant_copy' which right now just
  call the ioctl

- added a new VER1.1 to tools/libs/gnttab/libxengnttab.map 

qemu-qdisk:
- removed the 'ioreq_write','ioreq_read_init','ioreq_read' functions 
- implemented 'ioreq_init_copy_buffers', 'ioreq_copy' 
- reverted the removal of grant map and introduced conditional invoking
  grant copy or grant map
- resigned from caching the local buffers on behalf of allocating the 
  required amount of pages at once. The cached structure would require 
  to have an lock guard and I suppose that the performance improvement 
  would degraded. 
 

For the functional test I attached the device with a qdisk backend to the 
guest, 
mounted, performed some reads and writes.

I run fio tests[1] with different iodepth and size of the block. The test can 
be 
accessed on my github[2] but mainly after the warm up I run for 60 seconds:
    fio --time_based \
                --clocksource=clock_gettime \
                --rw=randread \
                --random_distribution=pareto:0.9 \
                --size=10g \
            --direct='1' \
            --ioengine=libaio \
                --filename=$DEV \
                --iodepth=$IODEPTH \
                --bs=$BS \
                --name=$NAME \
                --runtime=$RUNTIME >> $FILENAME
The test were repeated at least three times. 

[1] 
https://docs.google.com/spreadsheets/d/1E6AMiB8ceJpExL6jWpH9u2yy6DZxzhmDUyFf-eUuJ0c/edit?usp=sharing

[2] https://github.com/paulina-szubarczyk/xen-benchmark
    - multitest_with_iodepth.sh


Thanks and regards, 
Paulina

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel