PciHostBridge driver calls CpuIo to perform IO operation, which is different from the previous implementation in PcAtChipsetPkg.
I suggest we put the IO optimization in CpuIo driver. It can also benefit all other modules. Jeff, Do you agree to sync below commit to UefiCpuPkg/CpuIo driver? commit 1fd376d979221e9765dbec215916f5a6ab92879f PcAtChipsetPkg/PciHostBridgeDxe: Improve KVM FIFO I/O read/write performance Regards, Ray >-----Original Message----- >From: Justen, Jordan L >Sent: Friday, April 8, 2016 4:50 AM >To: Laszlo Ersek <ler...@redhat.com>; Ni, Ruiyu <ruiyu...@intel.com>; Mark ><kram...@gmail.com> >Cc: edk2-devel-01 <edk2-devel@lists.01.org>; Gao, Liming <liming....@intel.com> >Subject: Re: [edk2] [PATCH 0/2] OvmfPkg: drop PciHostBridgeDxe fork > >On 2016-04-07 12:36:36, Laszlo Ersek wrote: >> On 04/07/16 20:24, Jordan Justen wrote: >> > On 2016-04-07 08:07:35, Laszlo Ersek wrote: >> >> I like the diffstat on this one. >> >> >> > >> > :) >> > >> > Series Reviewed-by: Jordan Justen <jordan.l.jus...@intel.com> >> >> Thanks. Committed as aa47e5297851..4a392a451d4d. >> >> > Ray, >> > >> > I noticed one possible issue. With the new driver, did we lose this >> > commit? >> > >> > commit 1fd376d979221e9765dbec215916f5a6ab92879f >> > >> > PcAtChipsetPkg/PciHostBridgeDxe: Improve KVM FIFO I/O read/write >> > performance >> >> I think so. Apologies for not noticing it. :( >> >> I hope that Ray can resurrect / reapply the C code directly from the >> commit you reference above, and the assembly (in NASM syntax) from >> 6026bf460037 and ace1d0517b65. >> > >We may have to hold off on the NASM conversion for now, since the >driver is in MdeModulePkg. We haven't converted this package to NASM >yet. > >-Jordan > >> > Laszlo, >> > >> > Someone was reporting slowness with the Windows 7 ISO on irc a few >> > days back. Maybe this is the cause? >> > >> > I don't think this would impact virtio performance. >> >> I think you are referring to the following discussion on April 4th: >> >> > <Naruni> hello, im having performance issues when using ovmf >> > (OVMF_CODE-pure-efi.fd from gerd hoffman repo) in QEMU >> > emulator version 2.3.0 (Debian 1:2.3+dfsg-5ubuntu9.2) >> > >> > <Naruni> the specific issue im having is slow performance while trying >> > to install win7, the initial screen 'loading windows files' >> > is very slow and im only getting about 1 MB/sec disk IO >> > according to iotop >> > >> > <Naruni> if i use a VM in bios mode it operates at full performance >> > >> > <Naruni> this behavior is the same across multiple distros (ubuntu, >> > debian, arch, and gentoo) >> > >> > <Naruni> kvm is enabled and working according to qemu monitor >> > >> > <lersek> Naruni, what kind of virtual disk are you using? Are you >> > using virtio-blk or virtio-scsi with OVMF as well? >> > >> > <lersek> Naruni, also, I guess the best "advice" I can give at the >> > moment is: "just use the most recent upstream QEMU release". >> >> Unfortunately, "Naruni" didn't respond (I'll admit that I noticed his >> question only 90 minutes later or so). >> >> However, I am now putting this together with an email, on the vfio-users >> list (also April 4th): >> >> https://www.redhat.com/archives/vfio-users/2016-April/msg00029.html >> >> The description is practically identical, so I'm thinking the same user >> reported it. I'm adding the email address from the latter message to the >> Cc list now. Mark, are you Naruni? :) >> >> Unfortunately, the report doesn't name the version of Gerd's OVMF RPM in >> use. It also doesn't say whether the CPU was pegged while experiencing >> the slow performance. >> >> The "multiple distros" comment in the IRC report is not clear -- does >> that mean multiple distros as guests, or hosts? I guess it's the former; >> first, the host is reported as "QEMU emulator version 2.3.0 (Debian >> 1:2.3+dfsg-5ubuntu9.2)", while the domain XML pasted into Mark's email >> lists an ArchLinux installer ISO (i.e., guest). >> >> The domain XML clarifies another thing: the CD-ROM is IDE, not >> virtio-scsi. Also, there is no other disk drive (of any kind). >> >> Let me see if I can reproduce this, and perhaps bisect it. >> >> Mark, if you want to follow up on this list (= edk2-devel), please note >> that first you will have to subscribe to the list (and await the >> confirmation), at <https://lists.01.org/mailman/listinfo/edk2-devel>. >> >> Thanks >> Laszlo _______________________________________________ edk2-devel mailing list edk2-devel@lists.01.org https://lists.01.org/mailman/listinfo/edk2-devel