kern/dev/xen: Use the per-blkback-device uma(9) zone for the bio.

[seigo.tanimura_gmail.com]

Summary

The usage is essentially the same as the VM swap pager.

• New Loader Tunable & Sysctl Entry

• hw.xbd.xbb_reserved_new_bios The number of the bios reserved and preallocated per device. This is writable for hotplug.

Signed-off-by: Seigo Tanimura <[e-mail]>

[ehem_freebsd_m5p.com]

The one to consider D45381 is @royger who handles the Xen interface for FreeBSD.

At first glance this looks plausible, though running on Xen might mean some decisions should be different.

[royger]

Some questions. I mostly miss where g_io_new_uz() and the new set of helpers are introduced, as I'm not finding them in main git branch yet?

Since this seems to be a performance improvement, do you have some figures on how much of a performance difference does this change make?

sys/dev/xen/blkback/blkback.c

75

Ideally includes should be sorted alphabetically if possible, so vm/uma.h should be at the top of the vm/ group.

765

static unsigned int, and the max amount of in-flight BIOs blkback can currently handle is set by XBB_MAX_REQUESTS.

2002

Does this change depend on some unmerged stuff? I'm not able to find g_new_bio_uz() on the current main branch, neither g_io_new_uz(). How does it behave when it gets passed NULL as argument.

2003

If the uma pool is allocated to account for the maximum amuont of possible in-flights BIOs, the bio == NULL is no longer possible, and hence a KASSERT should be added inside of the if expression block.

[ehem_freebsd_m5p.com]

@royger under "Revision Contents" select the red tab "Stack (2 Open)". This is how Phabricator handles series and interrelated commits. The bits you were wondering about are there.