Re: kern/56291: XEN3_DOM0 nvme abysmal performance - very slow boot

[Frank Kardel <kardel%netbsd.org@localhost>]

The following reply was made to PR kern/56291; it has been noted by GNATS.

From: Frank Kardel <kardel%netbsd.org@localhost>
To: Joerg Sonnenberger <joerg%bec.de@localhost>
Cc: gnats-bugs%netbsd.org@localhost, kern-bug-people%netbsd.org@localhost,
 gnats-admin%netbsd.org@localhost, netbsd-bugs%netbsd.org@localhost
Subject: Re: kern/56291: XEN3_DOM0 nvme abysmal performance - very slow boot
Date: Fri, 2 Jul 2021 17:29:33 +0200

 Yes, I have the same suspicion PR/55667 started out like that until it 
 got stuck in a hypervisor panic issue.
 
  From the initial PR 55667 report:
 
 	XEN3_DOM0 used to boot flawlessly on this machine. Probably the
 	MSI/MSIX interrupt rework introduced the regression.
 	The boot stalls after attaching ums0 to wsmouse0. Basic
 	timing seems to work as the IPMI driver logs the version
 	16 seconds after boot and while the system seems hung.
 	This is a regression as 9.99.55 used to work fine.
 
 So this is the second platform with DOM0 boot issues seen on both Intel 
 and AMD cpus.
 
 Frank
 
 On 07/02/21 13:30, Joerg Sonnenberger wrote:
 > On Fri, Jul 02, 2021 at 10:53:23AM +0200, Frank Kardel wrote:
 >> Yes, there seems to be trouble with interrupts overall.
 >>
 >> Even without nvme the system is slow. (I had to boot with -a as the
 >>
 >> root device was not recognized correctly - different issue):
 >>
 >> Interrupts in a non Xen environment are at a rate of ~120. On Xen
 >>
 >> without nvme they are around ~2100.
 >>
 >> Seems that high interrupt rates occur with the bge driver (20 on an idle
 >> interface (bge0), 800 on
 >>
 >> an enabled interface (bge1))
 >>
 >> The two files contain two sets of statistics 10 seconds apart.
 > I wonder if the MSIX use is the problem.
 >
 > Joerg