Hi Andrew,
On 2/21/19 12:22 PM, Andrew Cooper wrote:
pci_release_devices() takes the global PCI lock. Once pci_release_devices()
has completed, it will be called redundantly each time paging_teardown() and
vcpu_destroy_pagetables() continue.
This is liable to be millions of times for a reasonably sized guest, and is a
serialising bottleneck now that domain_kill() can be run concurrently on
different domains.
Instead of propagating the opencoding of the relinquish state machine, take
the opportunity to clean it up.
Leave a proper set of comments explaining that domain_relinquish_resources()
implements a co-routine. Introduce a documented PROGRESS() macro to avoid
latent bugs such as the RELMEM_xen case, and make the new PROG_* states
private to domain_relinquish_resources().
Signed-off-by: Andrew Cooper <andrew.coop...@citrix.com>
---
CC: Jan Beulich <jbeul...@suse.com>
CC: Wei Liu <wei.l...@citrix.com>
CC: Roger Pau Monné <roger....@citrix.com>
CC: Juergen Gross <jgr...@suse.com>
CC: Stefano Stabellini <sstabell...@kernel.org>
CC: Julien Grall <julien.gr...@arm.com>
So I know Xen 4.12 isn't going to crash and burn without this change, but I
also can't un-see the unnecessary global PCI lock contention. In terms of
risk, this is extremely low - this function has complete coverage in testing,
and its behaviour isn't changing dramatically.
ARM: There are no problems, latent or otherwise, with your version of
domain_relinquish_resources(), but I'd recommend the same cleanup in due
course.
I will add in my todo list of cleanup!
Cheers,
--
Julien Grall
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel
