[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <1506346478-1631-1-git-send-email-eric.auger@redhat.com>
Date: Mon, 25 Sep 2017 15:34:29 +0200
From: Eric Auger <eric.auger@...hat.com>
To: eric.auger.pro@...il.com, eric.auger@...hat.com,
linux-kernel@...r.kernel.org, kvm@...r.kernel.org,
marc.zyngier@....com, cdall@...aro.org, peter.maydell@...aro.org,
andre.przywara@....com, wanghaibin.wang@...wei.com
Cc: wu.wubin@...wei.com
Subject: [PATCH 0/9] vITS Migration fixes and reset
This series fixes various bugs observed when saving/restoring the
ITS state before the guest writes the ITS registers (on first boot or
after reset/reboot).
This is a follow up of Wanghaibin's series [1] plus additional
patches following additional code review. It also proposes one
ITS reset implementation. I dared to include 2 Wanghaibin patches,
taking into account the discussions held on the ML but obviously
they can live separately.
On guest reset or when shutdown -r is initiated from guest, the ITS
caches are not saved into guest RAM. However currently the ITS is not
resettable, meaning the ITS will contain previously set values in
its registers. If we initiate a state backup before the guest
re-writes the ITS registers, we use the old register values to restore
the ITS tables. And since they may have never been written they
may contain invalid entries. This leads to inconsistencies detected
by the save/restore code and causes save/restore failure.
Patches [1-5] should be cc'ed stable I think:
- KVM: arm/arm64: vgic-its: fix return value for restore (Wanghaibin)
happens on restore whenever the device or ITT tables contain
only invalid data
- Always allow clearing GITS_CREADR/CWRITER
On first boot, if GITS_CBASER queue size is 0 and if we save at
this moment, restore fails on CREAD write.
- if GITS_BASER<n> point to invalid tables we currently return -EINVAL
on state save. QEMU aborts on this error whereas it forgives -EFAULT
as we considered we should not abort on guest bad programming.
- Then we should never attempt to use GITS_BASER<n> and GITS_CBASER
if they are not valid. Patches 4 and 5 fix that.
Patches [6-9] implement a new ITS reset IOCTL
- maybe patch "free caches when GITS_BASER Valid bit is cleared" can
be omitted if we consider the caches should not be cleared when
GITS_BASER<n>.Valid is cleared.
Best Regards
Eric
Git: complete series available at
https://github.com/eauger/linux/tree/v4.14-rc2-its-reset-v1
* Testing:
- on Cavium using a virtio-net-pci guest and various sequences of
guest shutdown -r now, virsh reset, virsh suspend/resume,
virsh reboot, virsh save.restore, virsh shutdown
References:
[1] [RFC PATCH 0/3] fix migrate failed when vm is in booting
https://www.spinics.net/lists/kvm-arm/msg27121.html
History:
PATCH v1
- series including 2 modified patches of Wanghaibin
Eric Auger (7):
KVM: arm/arm64: vgic-its: Always allow clearing GITS_CREADR/CWRITER
KVM: arm/arm64: vgic-its: Improve error reporting on device table save
KVM: arm/arm64: vgic-its: Check GITS_BASER Valid bit before saving
tables
KVM: arm/arm64: vgic-its: Check GITS_CBASER validity before processing
commands
KVM: arm/arm64: vgic-its: free caches when GITS_BASER Valid bit is
cleared
KVM: arm/arm64: Document KVM_DEV_ARM_ITS_CTRL_RESET
KVM: arm/arm64: vgic-its: Implement KVM_DEV_ARM_ITS_CTRL_RESET
wanghaibin (2):
KVM: arm/arm64: vgic-its: fix return value for restore
KVM: arm/arm64: vgic-its: new helper functions to free the caches
Documentation/virtual/kvm/devices/arm-vgic-its.txt | 16 ++
arch/arm/include/uapi/asm/kvm.h | 1 +
arch/arm64/include/uapi/asm/kvm.h | 1 +
virt/kvm/arm/vgic/vgic-its.c | 211 +++++++++++++--------
4 files changed, 155 insertions(+), 74 deletions(-)
--
2.5.5
Powered by blists - more mailing lists