https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=285681
Mark Johnston <ma...@freebsd.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Attachment #259079|0 |1
is obsolete| |
--- Comment #7 from Mark Johnston <ma...@freebsd.org> ---
Created attachment 259144
--> https://bugs.freebsd.org/bugzilla/attachment.cgi?id=259144&action=edit
possible patch
(In reply to Dimitry Andric from comment #6)
> If I increase to 4G, it panics with "storvsc recvbuf is not large enough".
Weird, I have no idea about that. I guess we should look at the packet length
in vmbus_chan_recv(). A stack trace would also be useful. But, does it happen
at all with the patch in comment 3?
I think that patch is the correct solution. The attached patch should also
eliminate unnecessary bouncing on i386 systems with physical RAM above 4GB. I
think what's happening is that we are getting a 4KB I/O across two pages above
4GB, and busdma is bouncing it into the first 2KB of each of two pages below
4GB. The storvsc driver isn't able to cope with that.
Again, it would be really nice to understand why the assertion fails on arm64,
and whether the attached patch fixes that as well. A storage driver like this
really shouldn't be bouncing on 64-bit systems.
--
You are receiving this mail because:
You are the assignee for the bug.
