kvm.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Marc Zyngier <maz@kernel.org>
To: linux-arm-kernel@lists.infradead.org, kvm@vger.kernel.org,
	kvmarm@lists.cs.columbia.edu
Cc: James Morse <james.morse@arm.com>,
	Julien Thierry <julien.thierry.kdev@gmail.com>,
	Suzuki K Poulose <suzuki.poulose@arm.com>,
	Alexandru Elisei <alexandru.elisei@arm.com>,
	Eric Auger <eric.auger@redhat.com>,
	Hector Martin <marcan@marcan.st>,
	Mark Rutland <mark.rutland@arm.com>,
	kernel-team@android.com
Subject: [PATCH 00/11] KVM: arm64: Initial host support for the Apple M1
Date: Tue, 16 Mar 2021 17:46:05 +0000	[thread overview]
Message-ID: <20210316174617.173033-1-maz@kernel.org> (raw)

I've spent the past few weekends trying to see how I could support the
M1 as a KVM host. It started by being pretty ugly, but the end result
is actually not too horrible.

Just a wee bit horrible.

The M1 has no GIC. And as everybody know, "KVM" stands for "The GIC
Emulator". Yes, Avi got the TLA wrong. Blame him.

It has no GIC, but it has a *partial* implementation of a vGICv3.
Which is not advertised by the CPU, because you can't have one without
the other. And it is partial because it cannot implement the automatic
HW interrupt deactivation trick (no GIC, remember?). But despite its
flaws, this gives us a fighting chance to expose something standard to
the guests.

"But how will the timers work?", I hear someone asking.

Well, that's a very good question. We rely on the GIC active state and
the HW-based deactivation to make the whole thing work, preventing a
pending timer interrupt from kicking us out of the guest, and
transparently re-enabling the interrupt when the guest EOIs it.

None of that can work on the M1, because (/all together now/) *it
doesn't have a GIC*!

What we can do instead is to fallback to regular masking on guest
entry, and rely on an exit on EOI to unmask the timer. Is that free?
No. Does it work? Yes! The trick is to make this as transparent as
possible to the rest of KVM so that the overhead is squarely on the M1
side. And my (very limited) testing indicates that this overhead is
pretty small.

Note that there is another way to implement the virtual timer: it
appears that this timer can be directly injected thanks to an IMPDEF
mechanism (HACR_EL2[20] + S3_5_15_1_2). It works fine, but:

- it is IMPDEF
- it is massively invasive for the GIC state machine
- it bloats the entry/exit fast path
- it forbids the implementation of GICR_ISPENDR0
- it doesn't work the physical timer
- it is IMPDEF

So scratch that, it doesn't work. I've kept the code for posterity,
but I don't think it is worth it. "Cute Embedded Nonsense Hack", to
quote someone.

This series is on top of Hector's latest drop[1], itself on top of the
FIQ enablement code gathered by Mark[2], plus 5.12-rc3 (which is
needed because this machine lacks an architectural PMU).

You will also need a bleeding edge userspace (qemu[3], kvmtool[4])
that configures the VM with a tiny IPA space (as the HW supports at
most 36 bits while KVM defaults to 40...).

[1] https://lore.kernel.org/r/20210304213902.83903-1-marcan@marcan.st
[2] https://lore.kernel.org/r/20210315115629.57191-1-mark.rutland@arm.com
[3] https://lore.kernel.org/r/20210310135218.255205-1-drjones@redhat.com
[4] https://lore.kernel.org/r/20210309163909.822149-1-maz@kernel.org

Marc Zyngier (11):
  irqchip/gic: Split vGIC probing information from the GIC code
  KVM: arm64: Handle physical FIQ as an IRQ while running a guest
  KVM: arm64: vgic: Be tolerant to the lack of maintenance interrupt
  KVM: arm64: vgic: Let an interrupt controller advertise lack of HW
    deactivation
  KVM: arm64: vgic: move irq->get_input_level into an ops structure
  KVM: arm64: vgic: Implement SW-driven deactivation
  KVM: arm64: timer: Refactor IRQ configuration
  KVM: arm64: timer: Add support for SW-based deactivation
  irqchip/apple-aic: Fix [un]masking of guest timers
  irqchip/apple-aic: Initialise SYS_APL_VM_TMR_FIQ_ENA_EL1 at boot time
  irqchip/apple-aic: Advertise some level of vGICv3 compatibility

 arch/arm64/kvm/arch_timer.c            | 153 ++++++++++++++++++++-----
 arch/arm64/kvm/hyp/hyp-entry.S         |   6 +-
 arch/arm64/kvm/vgic/vgic-init.c        |  33 +++++-
 arch/arm64/kvm/vgic/vgic-v2.c          |  19 ++-
 arch/arm64/kvm/vgic/vgic-v3.c          |  19 ++-
 arch/arm64/kvm/vgic/vgic.c             |  14 +--
 drivers/irqchip/irq-apple-aic.c        |  18 ++-
 drivers/irqchip/irq-gic-common.c       |  13 ---
 drivers/irqchip/irq-gic-common.h       |   2 -
 drivers/irqchip/irq-gic-v3.c           |   6 +-
 drivers/irqchip/irq-gic.c              |   6 +-
 include/kvm/arm_vgic.h                 |  41 +++++--
 include/linux/irqchip/arm-gic-common.h |  25 +---
 include/linux/irqchip/arm-vgic-info.h  |  43 +++++++
 14 files changed, 290 insertions(+), 108 deletions(-)
 create mode 100644 include/linux/irqchip/arm-vgic-info.h

-- 
2.29.2


             reply	other threads:[~2021-03-16 17:47 UTC|newest]

Thread overview: 12+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-03-16 17:46 Marc Zyngier [this message]
2021-03-16 17:46 ` [PATCH 01/11] irqchip/gic: Split vGIC probing information from the GIC code Marc Zyngier
2021-03-16 17:46 ` [PATCH 02/11] KVM: arm64: Handle physical FIQ as an IRQ while running a guest Marc Zyngier
2021-03-16 17:46 ` [PATCH 03/11] KVM: arm64: vgic: Be tolerant to the lack of maintenance interrupt Marc Zyngier
2021-03-16 17:46 ` [PATCH 04/11] KVM: arm64: vgic: Let an interrupt controller advertise lack of HW deactivation Marc Zyngier
2021-03-16 17:46 ` [PATCH 05/11] KVM: arm64: vgic: move irq->get_input_level into an ops structure Marc Zyngier
2021-03-16 17:46 ` [PATCH 06/11] KVM: arm64: vgic: Implement SW-driven deactivation Marc Zyngier
2021-03-16 17:46 ` [PATCH 07/11] KVM: arm64: timer: Refactor IRQ configuration Marc Zyngier
2021-03-16 17:46 ` [PATCH 08/11] KVM: arm64: timer: Add support for SW-based deactivation Marc Zyngier
2021-03-16 17:46 ` [PATCH 09/11] irqchip/apple-aic: Fix [un]masking of guest timers Marc Zyngier
2021-03-16 17:46 ` [PATCH 10/11] irqchip/apple-aic: Initialise SYS_APL_VM_TMR_FIQ_ENA_EL1 at boot time Marc Zyngier
2021-03-16 17:46 ` [PATCH 11/11] irqchip/apple-aic: Advertise some level of vGICv3 compatibility Marc Zyngier

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20210316174617.173033-1-maz@kernel.org \
    --to=maz@kernel.org \
    --cc=alexandru.elisei@arm.com \
    --cc=eric.auger@redhat.com \
    --cc=james.morse@arm.com \
    --cc=julien.thierry.kdev@gmail.com \
    --cc=kernel-team@android.com \
    --cc=kvm@vger.kernel.org \
    --cc=kvmarm@lists.cs.columbia.edu \
    --cc=linux-arm-kernel@lists.infradead.org \
    --cc=marcan@marcan.st \
    --cc=mark.rutland@arm.com \
    --cc=suzuki.poulose@arm.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).