platform/kernel/linux-starfive.git
7 years agoMerge remote-tracking branch 'rutland/kvm/common-sysreg' into next-fix
Christoffer Dall [Sun, 9 Apr 2017 14:50:34 +0000 (07:50 -0700)]
Merge remote-tracking branch 'rutland/kvm/common-sysreg' into next-fix

7 years agoKVM: arm/arm64: Advertise support for KVM_CAP_ARM_USER_IRQ
Christoffer Dall [Wed, 1 Feb 2017 11:54:11 +0000 (12:54 +0100)]
KVM: arm/arm64: Advertise support for KVM_CAP_ARM_USER_IRQ

Now that we support both timers and PMU reporting interrupts
to userspace, we can advertise this support.

Reviewed-by: Alexander Graf <agraf@suse.de>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
7 years agoKVM: arm/arm64: Report PMU overflow interrupts to userspace irqchip
Christoffer Dall [Wed, 1 Feb 2017 11:51:52 +0000 (12:51 +0100)]
KVM: arm/arm64: Report PMU overflow interrupts to userspace irqchip

When not using an in-kernel VGIC, but instead emulating an interrupt
controller in userspace, we should report the PMU overflow status to
that userspace interrupt controller using the KVM_CAP_ARM_USER_IRQ
feature.

Reviewed-by: Alexander Graf <agraf@suse.de>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
7 years agoKVM: arm/arm64: Support arch timers with a userspace gic
Alexander Graf [Tue, 27 Sep 2016 19:08:06 +0000 (21:08 +0200)]
KVM: arm/arm64: Support arch timers with a userspace gic

If you're running with a userspace gic or other interrupt controller
(that is no vgic in the kernel), then you have so far not been able to
use the architected timers, because the output of the architected
timers, which are driven inside the kernel, was a kernel-only construct
between the arch timer code and the vgic.

This patch implements the new KVM_CAP_ARM_USER_IRQ feature, where we use a
side channel on the kvm_run structure, run->s.regs.device_irq_level, to
always notify userspace of the timer output levels when using a userspace
irqchip.

This works by ensuring that before we enter the guest, if the timer
output level has changed compared to what we last told userspace, we
don't enter the guest, but instead return to userspace to notify it of
the new level.  If we are exiting, because of an MMIO for example, and
the level changed at the same time, the value is also updated and
userspace can sample the line as it needs.  This is nicely achieved
simply always updating the timer_irq_level field after the main run
loop.

Note that the kvm_timer_update_irq trace event is changed to show the
host IRQ number for the timer instead of the guest IRQ number, because
the kernel no longer know which IRQ userspace wires up the timer signal
to.

Also note that this patch implements all required functionality but does
not yet advertise the capability.

Reviewed-by: Alexander Graf <agraf@suse.de>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
7 years agoKVM: arm/arm64: Add ARM user space interrupt signaling ABI
Alexander Graf [Tue, 27 Sep 2016 19:08:05 +0000 (21:08 +0200)]
KVM: arm/arm64: Add ARM user space interrupt signaling ABI

We have 2 modes for dealing with interrupts in the ARM world. We can
either handle them all using hardware acceleration through the vgic or
we can emulate a gic in user space and only drive CPU IRQ pins from
there.

Unfortunately, when driving IRQs from user space, we never tell user
space about events from devices emulated inside the kernel, which may
result in interrupt line state changes, so we lose out on for example
timer and PMU events if we run with user space gic emulation.

Define an ABI to publish such device output levels to userspace.

Reviewed-by: Alexander Graf <agraf@suse.de>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
7 years agoKVM: arm/arm64: Cleanup the arch timer code's irqchip checking
Christoffer Dall [Tue, 27 Sep 2016 19:08:04 +0000 (21:08 +0200)]
KVM: arm/arm64: Cleanup the arch timer code's irqchip checking

Currently we check if we have an in-kernel irqchip and if the vgic was
properly implemented several places in the arch timer code.  But, we
already predicate our enablement of the arm timers on having a valid
and initialized gic, so we can simply check if the timers are enabled or
not.

This also gets rid of the ugly "error that's not an error but used to
signal that the timer shouldn't poke the gic" construct we have.

Reviewed-by: Alexander Graf <agraf@suse.de>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
7 years agoarm/arm64: Add hyp-stub API documentation
Marc Zyngier [Mon, 3 Apr 2017 18:38:07 +0000 (19:38 +0100)]
arm/arm64: Add hyp-stub API documentation

In order to help people understanding the hyp-stub API that exists
between the host kernel and the hypervisor mode (whether a hypervisor
has been installed or not), let's document said API.

As with any form of documentation, I expect it to become obsolete
and completely misleading within 20 minutes after having being merged.

Acked-by: Russell King <rmk+kernel@armlinux.org.uk>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoARM: hyp-stub: Zero r0 on successful stub handling
Marc Zyngier [Mon, 3 Apr 2017 18:38:06 +0000 (19:38 +0100)]
ARM: hyp-stub: Zero r0 on successful stub handling

We now return HVC_STUB_ERR when a stub hypercall fails, but we
leave whatever was in r0 on success. Zeroing it on return seems
like a good idea.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoarm64: hyp-stub: Zero x0 on successful stub handling
Marc Zyngier [Mon, 3 Apr 2017 18:38:05 +0000 (19:38 +0100)]
arm64: hyp-stub: Zero x0 on successful stub handling

We now return HVC_STUB_ERR when a stub hypercall fails, but we
leave whatever was in x0 on success. Zeroing it on return seems
like a good idea.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoarm64: hyp-stub/KVM: Kill __hyp_get_vectors
Marc Zyngier [Mon, 3 Apr 2017 18:38:04 +0000 (19:38 +0100)]
arm64: hyp-stub/KVM: Kill __hyp_get_vectors

Nobody is using __hyp_get_vectors anymore, so let's remove both
implementations (hyp-stub and KVM).

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoARM: hyp-stub/KVM: Kill __hyp_get_vectors
Marc Zyngier [Mon, 3 Apr 2017 18:38:03 +0000 (19:38 +0100)]
ARM: hyp-stub/KVM: Kill __hyp_get_vectors

Nobody is using __hyp_get_vectors anymore, so let's remove both
implementations (hyp-stub and KVM).

Acked-by: Russell King <rmk+kernel@armlinux.org.uk>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoARM: decompressor: Remove __hyp_get_vectors usage
Marc Zyngier [Mon, 3 Apr 2017 18:38:02 +0000 (19:38 +0100)]
ARM: decompressor: Remove __hyp_get_vectors usage

When the compressed image needs to be relocated to avoid being
overwritten by the decompression process, we need to relocate
the hyp vectors as well so that we can find them once the
decompression has taken effect.

For that, we perform the following calculation:
u32 v = __hyp_get_vectors();
v += offset;
__hyp_set_vectors(v);

But we're guaranteed that the initial value of v as returned by
__hyp_get_vectors is always __hyp_stub_vectors, because we have
just set it by calling __hyp_stub_install.

So let's remove the use of __hyp_get_vectors, and directly use
__hyp_stub_vectors instead.

Acked-by: Russell King <rmk+kernel@armlinux.org.uk>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoarm/arm64: KVM: Use HVC_RESET_VECTORS to reinit HYP mode
Marc Zyngier [Mon, 3 Apr 2017 18:38:01 +0000 (19:38 +0100)]
arm/arm64: KVM: Use HVC_RESET_VECTORS to reinit HYP mode

Instead of trying to compare the value given by __hyp_get_vectors(),
which doesn't offer any real guarantee to be the stub's address, use
HVC_RESET_VECTORS to make sure we're in a sane state to reinstall
KVM across PM events.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoarm/arm64: KVM: Remove kvm_get_idmap_start
Marc Zyngier [Mon, 3 Apr 2017 18:38:00 +0000 (19:38 +0100)]
arm/arm64: KVM: Remove kvm_get_idmap_start

With __cpu_reset_hyp_mode having become fairly dumb, there is no
need for kvm_get_idmap_start anymore.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoarm/arm64: KVM: Use __hyp_reset_vectors() directly
Marc Zyngier [Mon, 3 Apr 2017 18:37:59 +0000 (19:37 +0100)]
arm/arm64: KVM: Use __hyp_reset_vectors() directly

__cpu_reset_hyp_mode doesn't need to be passed any argument now,
as the hyp-stub implementations are self-contained, and is now
reduced to just calling __hyp_reset_vectors(). Let's drop the
wrapper and use the stub hypercall directly.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoARM: KVM: Gracefully handle hyp-stubs being restored from under our feet
Marc Zyngier [Mon, 3 Apr 2017 18:37:58 +0000 (19:37 +0100)]
ARM: KVM: Gracefully handle hyp-stubs being restored from under our feet

Should kvm_reboot() be invoked while guest is running, an IPI
wil be issued, forcing the guest to exit and HYP being reset to
the stubs. We will then try to reenter the guest, only to get
an error (HVC_STUB_ERR).

This patch allows this case to be gracefully handled by exiting
the run loop.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoARM: KVM: Implement HVC_SOFT_RESTART in the init code
Marc Zyngier [Mon, 3 Apr 2017 18:37:57 +0000 (19:37 +0100)]
ARM: KVM: Implement HVC_SOFT_RESTART in the init code

Another missing stub hypercall is HVC_SOFT_RESTART. It turns out
that it is pretty easy to implement in terms of HVC_RESET_VECTORS
(since it needs to turn the MMU off).

Tested-by: Keerthy <j-keerthy@ti.com>
Acked-by: Russell King <rmk+kernel@armlinux.org.uk>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoARM: KVM: Convert __cpu_reset_hyp_mode to using __hyp_reset_vectors
Marc Zyngier [Mon, 3 Apr 2017 18:37:56 +0000 (19:37 +0100)]
ARM: KVM: Convert __cpu_reset_hyp_mode to using __hyp_reset_vectors

We are now able to use the hyp stub to reset HYP mode. Time to
kiss __kvm_hyp_reset goodbye, and use __hyp_reset_vectors.

Tested-by: Keerthy <j-keerthy@ti.com>
Acked-by: Russell King <rmk+kernel@armlinux.org.uk>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoARM: KVM: Allow the main HYP code to use the init hyp stub implementation
Marc Zyngier [Mon, 3 Apr 2017 18:37:55 +0000 (19:37 +0100)]
ARM: KVM: Allow the main HYP code to use the init hyp stub implementation

We now have a full hyp-stub implementation in the KVM init code,
but the main KVM code only supports HVC_GET_VECTORS, which is not
enough.

Instead of reinventing the wheel, let's reuse the init implementation
by branching to the idmap page when called with a hyp-stub hypercall.

Tested-by: Keerthy <j-keerthy@ti.com>
Acked-by: Russell King <rmk+kernel@armlinux.org.uk>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoARM: KVM: Implement HVC_GET_VECTORS in the init code
Marc Zyngier [Mon, 3 Apr 2017 18:37:54 +0000 (19:37 +0100)]
ARM: KVM: Implement HVC_GET_VECTORS in the init code

Now that we have an infrastructure to handle hypercalls in the KVM
init code, let's implement HVC_GET_VECTORS there.

Tested-by: Keerthy <j-keerthy@ti.com>
Acked-by: Russell King <rmk+kernel@armlinux.org.uk>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoARM: KVM: Implement HVC_RESET_VECTORS stub hypercall in the init code
Marc Zyngier [Mon, 3 Apr 2017 18:37:53 +0000 (19:37 +0100)]
ARM: KVM: Implement HVC_RESET_VECTORS stub hypercall in the init code

In order to restore HYP mode to its original condition, KVM currently
implements __kvm_hyp_reset(). As we're moving towards a hyp-stub
defined API, it becomes necessary to implement HVC_RESET_VECTORS.

This patch adds the HVC_RESET_VECTORS hypercall to the KVM init
code, which so far lacked any form of hypercall support.

Tested-by: Keerthy <j-keerthy@ti.com>
Acked-by: Russell King <rmk+kernel@armlinux.org.uk>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoARM: hyp-stub: Implement HVC_RESET_VECTORS stub hypercall
Marc Zyngier [Mon, 3 Apr 2017 18:37:52 +0000 (19:37 +0100)]
ARM: hyp-stub: Implement HVC_RESET_VECTORS stub hypercall

Let's define a new stub hypercall that resets the HYP configuration
to its default: hyp-stub vectors, and MMU disabled.

Of course, for the hyp-stub itself, this is a trivial no-op.
Hypervisors will have a bit more work to do.

Tested-by: Keerthy <j-keerthy@ti.com>
Acked-by: Russell King <rmk+kernel@armlinux.org.uk>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoARM: hyp-stub: Define a return value for failed stub calls
Marc Zyngier [Mon, 3 Apr 2017 18:37:51 +0000 (19:37 +0100)]
ARM: hyp-stub: Define a return value for failed stub calls

Define a standard return value to be returned when a hyp stub
call fails.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoARM: Expose the VA/IDMAP offset
Marc Zyngier [Mon, 3 Apr 2017 18:37:50 +0000 (19:37 +0100)]
ARM: Expose the VA/IDMAP offset

The KVM code needs to be able to compute the address of
symbols in its idmap page (the equivalent of a virt_to_idmap()
call). Unfortunately, virt_to_idmap is slightly complicated,
depending on the use of arch_phys_to_idmap_offset or not, and
none of that is readily available at HYP.

Instead, expose a single kimage_voffset variable which contains the
offset between a kernel VA and its idmap address, enabling the
VA->IDMAP conversion. This allows the KVM code to behave similarily
to its arm64 counterpart.

Tested-by: Keerthy <j-keerthy@ti.com>
Acked-by: Russell King <rmk+kernel@armlinux.org.uk>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoARM: hyp-stub: Use r1 for the soft-restart address
Marc Zyngier [Mon, 3 Apr 2017 18:37:49 +0000 (19:37 +0100)]
ARM: hyp-stub: Use r1 for the soft-restart address

It is not really obvious why the restart address should be in r3
when communicated to the hyp-stub. r1 should be perfectly adequate,
and consistent with the rest of the code.

Tested-by: Keerthy <j-keerthy@ti.com>
Acked-by: Russell King <rmk+kernel@armlinux.org.uk>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoARM: Update cpu_v7_reset documentation
Marc Zyngier [Mon, 3 Apr 2017 18:37:48 +0000 (19:37 +0100)]
ARM: Update cpu_v7_reset documentation

cpu_v7_reset() now takes a second parameter indicating whether
we should reboot in HYP or not. Update the documentation to
reflect this.

Tested-by: Keerthy <j-keerthy@ti.com>
Acked-by: Russell King <rmk+kernel@armlinux.org.uk>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoARM: KVM: Convert KVM to use HVC_GET_VECTORS
Marc Zyngier [Mon, 3 Apr 2017 18:37:47 +0000 (19:37 +0100)]
ARM: KVM: Convert KVM to use HVC_GET_VECTORS

The conversion of the HYP stub ABI to something similar to arm64
left the KVM code broken, as it doesn't know about the new
stub numbering. Let's move the various #defines to virt.h, and
let KVM use HVC_GET_VECTORS.

Tested-by: Keerthy <j-keerthy@ti.com>
Acked-by: Russell King <rmk+kernel@armlinux.org.uk>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoARM: soft-reboot into same mode that we entered the kernel
Russell King [Mon, 3 Apr 2017 18:37:46 +0000 (19:37 +0100)]
ARM: soft-reboot into same mode that we entered the kernel

When we soft-reboot (eg, kexec) from one kernel into the next, we need
to ensure that we enter the new kernel in the same processor mode as
when we were entered, so that (eg) the new kernel can install its own
hypervisor - the old kernel's hypervisor will have been overwritten.

In order to do this, we need to pass a flag to cpu_reset() so it knows
what to do, and we need to modify the kernel's own hypervisor stub to
allow it to handle a soft-reboot.

As we are always guaranteed to install our own hypervisor if we're
entered in HYP32 mode, and KVM will have moved itself out of the way
on kexec/normal reboot, we can assume that our hypervisor is in place
when we want to kexec, so changing our hypervisor API should not be a
problem.

Tested-by: Keerthy <j-keerthy@ti.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoARM: hyp-stub: improve ABI
Russell King [Mon, 3 Apr 2017 18:37:45 +0000 (19:37 +0100)]
ARM: hyp-stub: improve ABI

Improve the hyp-stub ABI to allow it to do more than just get/set the
vectors.  We follow the example in ARM64, where r0 is used as an opcode
with the other registers as an argument.

Tested-by: Keerthy <j-keerthy@ti.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoarm64: KVM: Implement HVC_SOFT_RESTART in the init code
Marc Zyngier [Mon, 3 Apr 2017 18:37:44 +0000 (19:37 +0100)]
arm64: KVM: Implement HVC_SOFT_RESTART in the init code

Another missing stub hypercall is HVC_SOFT_RESTART. It turns out
that it is pretty easy to implement in terms of HVC_RESET_VECTORS
(since it needs to turn the MMU off).

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoarm64: KVM: Convert __cpu_reset_hyp_mode to using __hyp_reset_vectors
Marc Zyngier [Mon, 3 Apr 2017 18:37:43 +0000 (19:37 +0100)]
arm64: KVM: Convert __cpu_reset_hyp_mode to using __hyp_reset_vectors

We are now able to use the hyp stub to reset HYP mode. Time to
kiss __kvm_hyp_reset goodbye, and use __hyp_reset_vectors.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoarm64: KVM: Allow the main HYP code to use the init hyp stub implementation
Marc Zyngier [Mon, 3 Apr 2017 18:37:42 +0000 (19:37 +0100)]
arm64: KVM: Allow the main HYP code to use the init hyp stub implementation

We now have a full hyp-stub implementation in the KVM init code,
but the main KVM code only supports HVC_GET_VECTORS, which is not
enough.

Instead of reinventing the wheel, let's reuse the init implementation
by branching to the idmap page when called with a hyp-stub hypercall.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoarm64: KVM: Implement HVC_GET_VECTORS in the init code
Marc Zyngier [Mon, 3 Apr 2017 18:37:41 +0000 (19:37 +0100)]
arm64: KVM: Implement HVC_GET_VECTORS in the init code

Now that we have an infrastructure to handle hypercalls in the KVM
init code, let's implement HVC_GET_VECTORS there.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoarm64: KVM: Implement HVC_RESET_VECTORS stub hypercall in the init code
Marc Zyngier [Mon, 3 Apr 2017 18:37:40 +0000 (19:37 +0100)]
arm64: KVM: Implement HVC_RESET_VECTORS stub hypercall in the init code

In order to restore HYP mode to its original condition, KVM currently
implements __kvm_hyp_reset(). As we're moving towards a hyp-stub
defined API, it becomes necessary to implement HVC_RESET_VECTORS.

This patch adds the HVC_RESET_VECTORS hypercall to the KVM init
code, which so far lacked any form of hypercall support.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoarm64: hyp-stub: Implement HVC_RESET_VECTORS stub hypercall
Marc Zyngier [Mon, 3 Apr 2017 18:37:39 +0000 (19:37 +0100)]
arm64: hyp-stub: Implement HVC_RESET_VECTORS stub hypercall

Let's define a new stub hypercall that resets the HYP configuration
to its default: hyp-stub vectors, and MMU disabled.

Of course, for the hyp-stub itself, this is a trivial no-op.
Hypervisors will have a bit more work to do.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoarm64: hyp-stub: Update documentation in asm/virt.h
Marc Zyngier [Mon, 3 Apr 2017 18:37:38 +0000 (19:37 +0100)]
arm64: hyp-stub: Update documentation in asm/virt.h

Comments in asm/virt.h are slightly out of date, so let's align
them with the new behaviour of the code.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoarm64: hyp-stub: Define a return value for failed stub calls
Marc Zyngier [Mon, 3 Apr 2017 18:37:37 +0000 (19:37 +0100)]
arm64: hyp-stub: Define a return value for failed stub calls

Define a standard return value to be returned when a hyp stub
call fails, and make KVM use it for ARM_EXCEPTION_HYP_GONE
(instead of using a KVM-specific value).

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoarm64: hyp-stub: Don't save lr in the EL1 code
Marc Zyngier [Mon, 3 Apr 2017 18:37:36 +0000 (19:37 +0100)]
arm64: hyp-stub: Don't save lr in the EL1 code

The EL2 code is not corrupting lr anymore, so don't bother preserving
it in the EL1 trampoline code.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoarm64: KVM: Move lr save/restore to do_el2_call
Marc Zyngier [Mon, 3 Apr 2017 18:37:35 +0000 (19:37 +0100)]
arm64: KVM: Move lr save/restore to do_el2_call

At the moment, we only save/restore lr if on VHE, as we rely only
the EL1 code to have preserved it in the non-VHE case.

As we're about to get rid of the latter, let's move the save/restore
code to the do_el2_call macro, unifying both code paths.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoarm64: hyp-stub: Stop pointlessly clobbering lr
Marc Zyngier [Mon, 3 Apr 2017 18:37:34 +0000 (19:37 +0100)]
arm64: hyp-stub: Stop pointlessly clobbering lr

When entering the kernel hyp stub, we check whether or not we've
made it here through an HVC instruction, clobbering lr (aka x30)
in the process.

This is completely pointless, as HVC is the only way to get here
(all traps to EL2 are disabled, no interrupt override is applied).

So let's remove this bit of code whose only point is to corrupt
a valuable register.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoarm: KVM: Treat CP15 accessors returning false as successful
Marc Zyngier [Mon, 27 Mar 2017 16:03:45 +0000 (17:03 +0100)]
arm: KVM: Treat CP15 accessors returning false as successful

Instead of considering that a CP15 accessor has failed when
returning false, let's consider that it is *always* successful
(after all, we won't stand for an incomplete emulation).

The return value now simply indicates whether we should skip
the instruction (because it has now been emulated), or if we
should leave the PC alone if the emulation has injected an
exception.

Reviewed-by: Christoffer Dall <cdall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
7 years agoarm: KVM: Make unexpected register accesses inject an undef
Marc Zyngier [Mon, 27 Mar 2017 16:03:44 +0000 (17:03 +0100)]
arm: KVM: Make unexpected register accesses inject an undef

Reads from write-only system registers are generally confined to
EL1 and not propagated to EL2 (that's what the architecture
mantates). In order to be sure that we have a sane behaviour
even in the unlikely event that we have a broken system, we still
handle it in KVM. Same goes for write to RO registers.

In that case, let's inject an undef into the guest.

Reviewed-by: Christoffer Dall <cdall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
7 years agoarm64: KVM: Do not corrupt registers on failed 64bit CP read
Marc Zyngier [Mon, 27 Mar 2017 16:03:43 +0000 (17:03 +0100)]
arm64: KVM: Do not corrupt registers on failed 64bit CP read

If we fail to emulate a mrrc instruction, we:
1) deliver an exception,
2) spit a nastygram on the console,
3) write back some garbage to Rt/Rt2

While 1) and 2) are perfectly acceptable, 3) is out of the scope of
the architecture... Let's mimick the code in kvm_handle_cp_32 and
be more cautious.

Reviewed-by: Christoffer Dall <cdall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoarm64: KVM: Treat sysreg accessors returning false as successful
Marc Zyngier [Mon, 27 Mar 2017 16:03:42 +0000 (17:03 +0100)]
arm64: KVM: Treat sysreg accessors returning false as successful

Instead of considering that a sysreg accessor has failed when
returning false, let's consider that it is *always* successful
(after all, we won't stand for an incomplete emulation).

The return value now simply indicates whether we should skip
the instruction (because it has now been emulated), or if we
should leave the PC alone if the emulation has injected an
exception.

Reviewed-by: Christoffer Dall <cdall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
7 years agoarm64: KVM: PMU: Inject UNDEF on read access to PMSWINC_EL0
Marc Zyngier [Mon, 27 Mar 2017 16:03:41 +0000 (17:03 +0100)]
arm64: KVM: PMU: Inject UNDEF on read access to PMSWINC_EL0

PMSWINC_EL0 is a WO register, so let's UNDEF when reading from it
(in the highly hypothetical case where this doesn't UNDEF at EL1).

Reviewed-by: Christoffer Dall <cdall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
7 years agoarm64: KVM: Make unexpected reads from WO registers inject an undef
Marc Zyngier [Mon, 27 Mar 2017 16:03:40 +0000 (17:03 +0100)]
arm64: KVM: Make unexpected reads from WO registers inject an undef

Reads from write-only system registers are generally confined to
EL1 and not propagated to EL2 (that's what the architecture
mantates). In order to be sure that we have a sane behaviour
even in the unlikely event that we have a broken system, we still
handle it in KVM.

In that case, let's inject an undef into the guest.

Let's also remove write_to_read_only which isn't used anywhere.

Reviewed-by: Christoffer Dall <cdall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
7 years agoarm64: KVM: PMU: Inject UNDEF on non-privileged accesses
Marc Zyngier [Mon, 27 Mar 2017 16:03:39 +0000 (17:03 +0100)]
arm64: KVM: PMU: Inject UNDEF on non-privileged accesses

access_pminten() and access_pmuserenr() can only be accessed when
the CPU is in a priviledged mode. If it is not, let's inject an
UNDEF exception.

Reviewed-by: Christoffer Dall <cdall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
7 years agoarm64: KVM: PMU: Inject UNDEF exception on illegal register access
Marc Zyngier [Mon, 27 Mar 2017 16:03:38 +0000 (17:03 +0100)]
arm64: KVM: PMU: Inject UNDEF exception on illegal register access

Both pmu_*_el0_disabled() and pmu_counter_idx_valid() perform checks
on the validity of an access, but only return a boolean indicating
if the access is valid or not.

Let's allow these functions to also inject an UNDEF exception if
the access was illegal.

Reviewed-by: Christoffer Dall <cdall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
7 years agoarm64: KVM: PMU: Refactor pmu_*_el0_disabled
Marc Zyngier [Mon, 27 Mar 2017 16:03:37 +0000 (17:03 +0100)]
arm64: KVM: PMU: Refactor pmu_*_el0_disabled

There is a lot of duplication in the pmu_*_el0_disabled helpers,
and as we're going to modify them shortly, let's move all the
common stuff in a single function.

No functional change.

Reviewed-by: Christoffer Dall <cdall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoKVM: arm/arm64: vgic: Improve sync_hwstate performance
Christoffer Dall [Sat, 18 Mar 2017 12:48:42 +0000 (13:48 +0100)]
KVM: arm/arm64: vgic: Improve sync_hwstate performance

There is no need to call any functions to fold LRs when we don't use any
LRs and we don't need to mess with overflow flags, take spinlocks, or
prune the AP list if the AP list is empty.

Note: list_empty is a single atomic read (uses READ_ONCE) and can
therefore check if a list is empty or not without the need to take the
spinlock protecting the list.

Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoKVM: arm/arm64: vgic: Don't check vgic_initialized in sync/flush
Christoffer Dall [Sat, 18 Mar 2017 12:41:54 +0000 (13:41 +0100)]
KVM: arm/arm64: vgic: Don't check vgic_initialized in sync/flush

Now when we do an early init of the static parts of the VGIC data
structures, we can do things like checking if the AP lists are empty
directly without having to explicitly check if the vgic is initialized
and reduce a bit of work in our critical path.

Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoKVM: arm/arm64: vgic: Implement early VGIC init functionality
Christoffer Dall [Sat, 18 Mar 2017 12:40:37 +0000 (13:40 +0100)]
KVM: arm/arm64: vgic: Implement early VGIC init functionality

Implement early initialization for both the distributor and the CPU
interfaces.  The basic idea is that even though the VGIC is not
functional or not requested from user space, the critical path of the
run loop can still call VGIC functions that just won't do anything,
without them having to check additional initialization flags to ensure
they don't look at uninitialized data structures.

Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoKVM: arm/arm64: vgic: Get rid of MISR and EISR fields
Christoffer Dall [Thu, 29 Dec 2016 14:57:31 +0000 (15:57 +0100)]
KVM: arm/arm64: vgic: Get rid of MISR and EISR fields

We don't use these fields anymore so let's nuke them completely.

Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
7 years agoKVM: arm/arm64: vgic: Get rid of unnecessary save_maint_int_state
Christoffer Dall [Thu, 29 Dec 2016 14:48:57 +0000 (15:48 +0100)]
KVM: arm/arm64: vgic: Get rid of unnecessary save_maint_int_state

Now when we don't look at the MISR and EISR values anymore, we can get
rid of the logic to save them in the GIC save/restore code.

Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
7 years agoKVM: arm/arm64: vgic: Get rid of unnecessary process_maintenance operation
Christoffer Dall [Thu, 29 Dec 2016 14:44:27 +0000 (15:44 +0100)]
KVM: arm/arm64: vgic: Get rid of unnecessary process_maintenance operation

Since we always read back the LRs that we wrote to the guest and the
MISR and EISR registers simply provide a summary of the configuration of
the bits in the LRs, there is really no need to read back those status
registers and process them.  We might as well just signal the
notifyfd when folding the LR state and save some cycles in the process.
We now clear the underflow bit in the fold_lr_state functions as we only
need to clear this bit if we had used all the LRs, so this is as good a
place as any to do that work.

Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
7 years agoKVM: arm/arm64: vgic: Only set underflow when actually out of LRs
Christoffer Dall [Tue, 21 Mar 2017 20:16:12 +0000 (21:16 +0100)]
KVM: arm/arm64: vgic: Only set underflow when actually out of LRs

We currently assume that all the interrupts in our AP list will be
queued to LRs, but that's not necessarily the case, because some of them
could have been migrated away to different VCPUs and only the VCPU
thread itself can remove interrupts from its AP list.

Therefore, slightly change the logic to only setting the underflow
interrupt when we actually run out of LRs.

As it turns out, this allows us to further simplify the handling in
vgic_sync_hwstate in later patches.

Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoKVM: arm/arm64: vgic: Get rid of live_lrs
Christoffer Dall [Thu, 22 Dec 2016 23:04:59 +0000 (00:04 +0100)]
KVM: arm/arm64: vgic: Get rid of live_lrs

There is no need to calculate and maintain live_lrs when we always
populate the lowest numbered LRs first on every entry and clear all LRs
on every exit.

Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
7 years agoKVM: arm/arm64: vgic: Avoid flushing vgic state when there's no pending IRQ
Shih-Wei Li [Wed, 19 Oct 2016 18:12:34 +0000 (18:12 +0000)]
KVM: arm/arm64: vgic: Avoid flushing vgic state when there's no pending IRQ

We do not need to flush vgic states in each world switch unless
there is pending IRQ queued to the vgic's ap list. We can thus reduce
the overhead by not grabbing the spinlock and not making the extra
function call to vgic_flush_lr_state.

Note: list_empty is a single atomic read (uses READ_ONCE) and can
therefore check if a list is empty or not without the need to take the
spinlock protecting the list.

Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Shih-Wei Li <shihwei@cs.columbia.edu>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoKVM: arm/arm64: vgic: Defer touching GICH_VMCR to vcpu_load/put
Christoffer Dall [Thu, 24 Mar 2016 10:21:04 +0000 (11:21 +0100)]
KVM: arm/arm64: vgic: Defer touching GICH_VMCR to vcpu_load/put

We don't have to save/restore the VMCR on every entry to/from the guest,
since on GICv2 we can access the control interface from EL1 and on VHE
systems with GICv3 we can access the control interface from KVM running
in EL2.

GICv3 systems without VHE becomes the rare case, which has to
save/restore the register on each round trip.

Note that userspace accesses may see out-of-date values if the VCPU is
running while accessing the VGIC state via the KVM device API, but this
is already the case and it is up to userspace to quiesce the CPUs before
reading the CPU registers from the GIC for an up-to-date view.

Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@cs.columbia.edu>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agokvm: arm/arm64: Rework gpa callback handlers
Suzuki K Poulose [Mon, 20 Mar 2017 18:26:42 +0000 (18:26 +0000)]
kvm: arm/arm64: Rework gpa callback handlers

In order to perform an operation on a gpa range, we currently iterate
over each page in a user memory slot for the given range. This is
inefficient while dealing with a big range (e.g, a VMA), especially
while unmaping a range. At present, with stage2 unmap on a range with
a hugepage backed region, we clear the PMD when we unmap the first
page in the loop. The remaining iterations simply traverse the page table
down to the PMD level only to see that nothing is in there.

This patch reworks the code to invoke the callback handlers on the
biggest range possible within the memory slot to to reduce the number of
times the handler is called.

Cc: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Christoffer Dall <cdall@linaro.org>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
7 years agoKVM: arm64: Use common Set/Way sys definitions
Mark Rutland [Fri, 13 Jan 2017 17:51:27 +0000 (17:51 +0000)]
KVM: arm64: Use common Set/Way sys definitions

Now that we have common definitions for the encoding of Set/Way cache
maintenance operations, make the KVM code use these, simplifying the
sys_reg_descs table.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: kvmarm@lists.cs.columbia.edu
7 years agoKVM: arm64: Use common sysreg definitions
Mark Rutland [Thu, 19 Jan 2017 18:39:39 +0000 (18:39 +0000)]
KVM: arm64: Use common sysreg definitions

Now that we have common definitions for the remaining register encodings
required by KVM, make the KVM code use these, simplifying the
sys_reg_descs table and the genericv8_sys_regs table.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: kvmarm@lists.cs.columbia.edu
7 years agoKVM: arm64: use common invariant sysreg definitions
Mark Rutland [Fri, 13 Jan 2017 18:36:26 +0000 (18:36 +0000)]
KVM: arm64: use common invariant sysreg definitions

Now that we have common definitions for the register encodings used by
KVM, make the KVM code uses thse for invariant sysreg definitions. This
makes said definitions a reasonable amount shorter, especially as many
comments are rendered redundant and can be removed.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: kvmarm@lists.cs.columbia.edu
7 years agoKVM: arm64: Use common physical timer sysreg definitions
Mark Rutland [Thu, 9 Mar 2017 16:50:51 +0000 (16:50 +0000)]
KVM: arm64: Use common physical timer sysreg definitions

Now that we have common definitions for the physical timer control
registers, make the KVM code use these, simplifying the sys_reg_descs
table.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: kvmarm@lists.cs.columbia.edu
7 years agoKVM: arm64: Use common GICv3 sysreg definitions
Mark Rutland [Fri, 20 Jan 2017 18:02:39 +0000 (18:02 +0000)]
KVM: arm64: Use common GICv3 sysreg definitions

Now that we have common definitions for the GICv3 register encodings,
make the KVM code use these, simplifying the sys_reg_descs table.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: kvmarm@lists.cs.columbia.edu
7 years agoKVM: arm64: Use common performance monitor sysreg definitions
Mark Rutland [Fri, 20 Jan 2017 17:43:20 +0000 (17:43 +0000)]
KVM: arm64: Use common performance monitor sysreg definitions

Now that we have common definitions for the performance monitor register
encodings, make the KVM code use these, simplifying the sys_reg_descs
table.

The comments for PMUSERENR_EL0 and PMCCFILTR_EL0 are kept, as these
describe non-obvious details regarding the registers. However, a slight
fixup is applied to bring these into line with the usual comment style.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: kvmarm@lists.cs.columbia.edu
7 years agoKVM: arm64: Use common debug sysreg definitions
Mark Rutland [Fri, 13 Jan 2017 17:19:12 +0000 (17:19 +0000)]
KVM: arm64: Use common debug sysreg definitions

Now that we have common definitions for the debug register encodings,
make the KVM code use these, simplifying the sys_reg_descs table.

The table previously erroneously referred to MDCCSR_EL0 as MDCCSR_EL1.
This is corrected (as is necessary in order to use the common sysreg
definition).

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: kvmarm@lists.cs.columbia.edu
7 years agoKVM: arm64: add SYS_DESC()
Mark Rutland [Fri, 13 Jan 2017 17:25:11 +0000 (17:25 +0000)]
KVM: arm64: add SYS_DESC()

This patch adds a macro enabling us to initialise sys_reg_desc
structures based on common sysreg encoding definitions in
<asm/sysreg.h>. Subsequent patches will use this to simplify the KVM
code.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: kvmarm@lists.cs.columbia.edu
7 years agoLinux 4.11-rc3
Linus Torvalds [Mon, 20 Mar 2017 02:09:39 +0000 (19:09 -0700)]
Linux 4.11-rc3

7 years agomm/swap: don't BUG_ON() due to uninitialized swap slot cache
Linus Torvalds [Mon, 20 Mar 2017 02:00:47 +0000 (19:00 -0700)]
mm/swap: don't BUG_ON() due to uninitialized swap slot cache

This BUG_ON() triggered for me once at shutdown, and I don't see a
reason for the check.  The code correctly checks whether the swap slot
cache is usable or not, so an uninitialized swap slot cache is not
actually problematic afaik.

I've temporarily just switched the BUG_ON() to a WARN_ON_ONCE(), since
I'm not sure why that seemingly pointless check was there.  I suspect
the real fix is to just remove it entirely, but for now we'll warn about
it but not bring the machine down.

Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Tim Chen <tim.c.chen@linux.intel.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoMerge tag 'powerpc-4.11-5' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
Linus Torvalds [Mon, 20 Mar 2017 01:49:28 +0000 (18:49 -0700)]
Merge tag 'powerpc-4.11-5' of git://git./linux/kernel/git/powerpc/linux

Pull more powerpc fixes from Michael Ellerman:
 "A couple of minor powerpc fixes for 4.11:

   - wire up statx() syscall

   - don't print a warning on memory hotplug when HPT resizing isn't
     available

  Thanks to: David Gibson, Chandan Rajendra"

* tag 'powerpc-4.11-5' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux:
  powerpc/pseries: Don't give a warning when HPT resizing isn't available
  powerpc: Wire up statx() syscall

7 years agoMerge branch 'parisc-4.11-2' of git://git.kernel.org/pub/scm/linux/kernel/git/deller...
Linus Torvalds [Mon, 20 Mar 2017 01:11:13 +0000 (18:11 -0700)]
Merge branch 'parisc-4.11-2' of git://git./linux/kernel/git/deller/parisc-linux

Pull parisc fixes from Helge Deller:

 - Mikulas Patocka added support for R_PARISC_SECREL32 relocations in
   modules with CONFIG_MODVERSIONS.

 - Dave Anglin optimized the cache flushing for vmap ranges.

 - Arvind Yadav provided a fix for a potential NULL pointer dereference
   in the parisc perf code (and some code cleanups).

 - I wired up the new statx system call, fixed some compiler warnings
   with the access_ok() macro and fixed shutdown code to really halt a
   system at shutdown instead of crashing & rebooting.

* 'parisc-4.11-2' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: Fix system shutdown halt
  parisc: perf: Fix potential NULL pointer dereference
  parisc: Avoid compiler warnings with access_ok()
  parisc: Wire up statx system call
  parisc: Optimize flush_kernel_vmap_range and invalidate_kernel_vmap_range
  parisc: support R_PARISC_SECREL32 relocation in modules

7 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending
Linus Torvalds [Mon, 20 Mar 2017 01:06:31 +0000 (18:06 -0700)]
Merge git://git./linux/kernel/git/nab/target-pending

Pull SCSI target fixes from Nicholas Bellinger:
 "The bulk of the changes are in qla2xxx target driver code to address
  various issues found during Cavium/QLogic's internal testing (stable
  CC's included), along with a few other stability and smaller
  miscellaneous improvements.

  There are also a couple of different patch sets from Mike Christie,
  which have been a result of his work to use target-core ALUA logic
  together with tcm-user backend driver.

  Finally, a patch to address some long standing issues with
  pass-through SCSI export of TYPE_TAPE + TYPE_MEDIUM_CHANGER devices,
  which will make folks using physical (or virtual) magnetic tape happy"

* git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending: (28 commits)
  qla2xxx: Update driver version to 9.00.00.00-k
  qla2xxx: Fix delayed response to command for loop mode/direct connect.
  qla2xxx: Change scsi host lookup method.
  qla2xxx: Add DebugFS node to display Port Database
  qla2xxx: Use IOCB interface to submit non-critical MBX.
  qla2xxx: Add async new target notification
  qla2xxx: Export DIF stats via debugfs
  qla2xxx: Improve T10-DIF/PI handling in driver.
  qla2xxx: Allow relogin to proceed if remote login did not finish
  qla2xxx: Fix sess_lock & hardware_lock lock order problem.
  qla2xxx: Fix inadequate lock protection for ABTS.
  qla2xxx: Fix request queue corruption.
  qla2xxx: Fix memory leak for abts processing
  qla2xxx: Allow vref count to timeout on vport delete.
  tcmu: Convert cmd_time_out into backend device attribute
  tcmu: make cmd timeout configurable
  tcmu: add helper to check if dev was configured
  target: fix race during implicit transition work flushes
  target: allow userspace to set state to transitioning
  target: fix ALUA transition timeout handling
  ...

7 years agoMerge branch 'libnvdimm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdim...
Linus Torvalds [Sun, 19 Mar 2017 22:45:02 +0000 (15:45 -0700)]
Merge branch 'libnvdimm-fixes' of git://git./linux/kernel/git/nvdimm/nvdimm

Pull device-dax fixes from Dan Williams:
 "The device-dax driver was not being careful to handle falling back to
  smaller fault-granularity sizes.

  The driver already fails fault attempts that are smaller than the
  device's alignment, but it also needs to handle the cases where a
  larger page mapping could be established. For simplicity of the
  immediate fix the implementation just signals VM_FAULT_FALLBACK until
  fault-size == device-alignment.

  One fix is for -stable to address pmd-to-pte fallback from the
  original implementation, another fix is for the new (introduced in
  4.11-rc1) pud-to-pmd regression, and a typo fix comes along for the
  ride.

  These have received a build success notification from the kbuild
  robot"

* 'libnvdimm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm:
  device-dax: fix debug output typo
  device-dax: fix pud fault fallback handling
  device-dax: fix pmd/pte fault fallback handling

7 years agoqla2xxx: Update driver version to 9.00.00.00-k
Himanshu Madhani [Wed, 15 Mar 2017 16:48:56 +0000 (09:48 -0700)]
qla2xxx: Update driver version to 9.00.00.00-k

Signed-off-by: Himanshu Madhani <himanshu.madhani@cavium.com>
signed-off-by: Giridhar Malavali <giridhar.malavali@cavium.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agoqla2xxx: Fix delayed response to command for loop mode/direct connect.
Quinn Tran [Wed, 15 Mar 2017 16:48:55 +0000 (09:48 -0700)]
qla2xxx: Fix delayed response to command for loop mode/direct connect.

Current driver wait for FW to be in the ready state before
processing in-coming commands. For Arbitrated Loop or
Point-to- Point (not switch), FW Ready state can take a while.
FW will transition to ready state after all Nports have been
logged in. In the mean time, certain initiators have completed
the login and starts IO. Driver needs to start processing all
queues if FW is already started.

Signed-off-by: Quinn Tran <quinn.tran@cavium.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@cavium.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agoqla2xxx: Change scsi host lookup method.
Quinn Tran [Wed, 15 Mar 2017 16:48:54 +0000 (09:48 -0700)]
qla2xxx: Change scsi host lookup method.

For target mode, when new scsi command arrive, driver first performs
a look up of the SCSI Host. The current look up method is based on
the ALPA portion of the NPort ID. For Cisco switch, the ALPA can
not be used as the index. Instead, the new search method is based
on the full value of the Nport_ID via btree lib.

Signed-off-by: Quinn Tran <quinn.tran@cavium.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@cavium.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agoqla2xxx: Add DebugFS node to display Port Database
Himanshu Madhani [Wed, 15 Mar 2017 16:48:53 +0000 (09:48 -0700)]
qla2xxx: Add DebugFS node to display Port Database

Signed-off-by: Himanshu Madhani <himanshu.madhani@cavium.com>
Signed-off-by: Giridhar Malavali <giridhar.malavali@cavium.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agoqla2xxx: Use IOCB interface to submit non-critical MBX.
Quinn Tran [Wed, 15 Mar 2017 16:48:52 +0000 (09:48 -0700)]
qla2xxx: Use IOCB interface to submit non-critical MBX.

The Mailbox interface is currently over subscribed. We like
to reserve the Mailbox interface for the chip managment and
link initialization. Any non essential Mailbox command will
be routed through the IOCB interface. The IOCB interface is
able to absorb more commands.

Following commands are being routed through IOCB interface

- Get ID List (007Ch)
- Get Port DB (0064h)
- Get Link Priv Stats (006Dh)

Signed-off-by: Quinn Tran <quinn.tran@cavium.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@cavium.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agoqla2xxx: Add async new target notification
Quinn Tran [Wed, 15 Mar 2017 16:48:51 +0000 (09:48 -0700)]
qla2xxx: Add async new target notification

Signed-off-by: Quinn Tran <quinn.tran@cavium.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@cavium.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agoqla2xxx: Export DIF stats via debugfs
Anil Gurumurthy [Wed, 15 Mar 2017 16:48:50 +0000 (09:48 -0700)]
qla2xxx: Export DIF stats via debugfs

Signed-off-by: Anil Gurumurthy <anil.gurumurthy@cavium.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@cavium.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agoqla2xxx: Improve T10-DIF/PI handling in driver.
Quinn Tran [Wed, 15 Mar 2017 16:48:49 +0000 (09:48 -0700)]
qla2xxx: Improve T10-DIF/PI handling in driver.

Add routines to support T10 DIF tag.

Signed-off-by: Quinn Tran <quinn.tran@cavium.com>
Signed-off-by: Anil Gurumurthy <anil.gurumurthy@cavium.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@cavium.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agoqla2xxx: Allow relogin to proceed if remote login did not finish
Quinn Tran [Wed, 15 Mar 2017 16:48:48 +0000 (09:48 -0700)]
qla2xxx: Allow relogin to proceed if remote login did not finish

If the remote port have started the login process, then the
PLOGI and PRLI should be back to back. Driver will allow
the remote port to complete the process. For the case where
the remote port decide to back off from sending PRLI, this
local port sets an expiration timer for the PRLI. Once the
expiration time passes, the relogin retry logic is allowed
to go through and perform login with the remote port.

Signed-off-by: Quinn Tran <quinn.tran@qlogic.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@cavium.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agoqla2xxx: Fix sess_lock & hardware_lock lock order problem.
Quinn Tran [Wed, 15 Mar 2017 16:48:47 +0000 (09:48 -0700)]
qla2xxx: Fix sess_lock & hardware_lock lock order problem.

The main lock that needs to be held for CMD or TMR submission
to upper layer is the sess_lock. The sess_lock is used to
serialize cmd submission and session deletion. The addition
of hardware_lock being held is not necessary. This patch removes
hardware_lock dependency from CMD/TMR submission.

Use hardware_lock only for error response in this case.

Path1
       CPU0                    CPU1
       ----                    ----
  lock(&(&ha->tgt.sess_lock)->rlock);
                               lock(&(&ha->hardware_lock)->rlock);
                               lock(&(&ha->tgt.sess_lock)->rlock);
  lock(&(&ha->hardware_lock)->rlock);

Path2/deadlock
*** DEADLOCK ***
Call Trace:
dump_stack+0x85/0xc2
print_circular_bug+0x1e3/0x250
__lock_acquire+0x1425/0x1620
lock_acquire+0xbf/0x210
_raw_spin_lock_irqsave+0x53/0x70
qlt_sess_work_fn+0x21d/0x480 [qla2xxx]
process_one_work+0x1f4/0x6e0

Cc: <stable@vger.kernel.org>
Cc: Bart Van Assche <Bart.VanAssche@sandisk.com>
Reported-by: Bart Van Assche <Bart.VanAssche@sandisk.com>
Signed-off-by: Quinn Tran <quinn.tran@cavium.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@cavium.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agoqla2xxx: Fix inadequate lock protection for ABTS.
Quinn Tran [Wed, 15 Mar 2017 16:48:46 +0000 (09:48 -0700)]
qla2xxx: Fix inadequate lock protection for ABTS.

Normally, ABTS is sent to Target Core as Task MGMT command.
In the case of error, qla2xxx needs to send response, hardware_lock
is required to prevent request queue corruption.

Cc: <stable@vger.kernel.org>
Signed-off-by: Quinn Tran <quinn.tran@cavium.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@cavium.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agoqla2xxx: Fix request queue corruption.
Quinn Tran [Wed, 15 Mar 2017 16:48:45 +0000 (09:48 -0700)]
qla2xxx: Fix request queue corruption.

When FW notify driver or driver detects low FW resource,
driver tries to send out Busy SCSI Status to tell Initiator
side to back off. During the send process, the lock was not held.

Cc: <stable@vger.kernel.org>
Signed-off-by: Quinn Tran <quinn.tran@qlogic.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@cavium.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agoqla2xxx: Fix memory leak for abts processing
Quinn Tran [Wed, 15 Mar 2017 16:48:44 +0000 (09:48 -0700)]
qla2xxx: Fix memory leak for abts processing

Cc: <stable@vger.kernel.org>
Signed-off-by: Quinn Tran <quinn.tran@cavium.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@cavium.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agoqla2xxx: Allow vref count to timeout on vport delete.
Joe Carnuccio [Wed, 15 Mar 2017 16:48:43 +0000 (09:48 -0700)]
qla2xxx: Allow vref count to timeout on vport delete.

Cc: <stable@vger.kernel.org>
Signed-off-by: Joe Carnuccio <joe.carnuccio@cavium.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@cavium.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agotcmu: Convert cmd_time_out into backend device attribute
Nicholas Bellinger [Sat, 18 Mar 2017 22:04:13 +0000 (15:04 -0700)]
tcmu: Convert cmd_time_out into backend device attribute

Instead of putting cmd_time_out under ../target/core/user_0/foo/control,
which has historically been used by parameters needed for initial
backend device configuration, go ahead and move cmd_time_out into
a backend device attribute.

In order to do this, tcmu_module_init() has been updated to create
a local struct configfs_attribute **tcmu_attrs, that is based upon
the existing passthrough_attrib_attrs along with the new cmd_time_out
attribute.  Once **tcm_attrs has been setup, go ahead and point
it at tcmu_ops->tb_dev_attrib_attrs so it's picked up by target-core.

Also following MNC's previous change, ->cmd_time_out is stored in
milliseconds but exposed via configfs in seconds.  Also, note this
patch restricts the modification of ->cmd_time_out to before +
after the TCMU device has been configured, but not while it has
active fabric exports.

Cc: Mike Christie <mchristi@redhat.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agotcmu: make cmd timeout configurable
Mike Christie [Thu, 9 Mar 2017 08:42:09 +0000 (02:42 -0600)]
tcmu: make cmd timeout configurable

A single daemon could implement multiple types of devices
using multuple types of real devices that may not support
restarting from crashes and/or handling tcmu timeouts. This
makes the cmd timeout configurable, so handlers that do not
support it can turn if off for now.

Signed-off-by: Mike Christie <mchristi@redhat.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agotcmu: add helper to check if dev was configured
Mike Christie [Thu, 9 Mar 2017 08:42:08 +0000 (02:42 -0600)]
tcmu: add helper to check if dev was configured

This adds a helper to check if the dev was configured. It
will be used in the next patch to prevent updates to some
config settings after the device has been setup.

Signed-off-by: Mike Christie <mchristi@redhat.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agoMerge tag 'openrisc-for-linus' of git://github.com/openrisc/linux
Linus Torvalds [Sat, 18 Mar 2017 22:50:39 +0000 (15:50 -0700)]
Merge tag 'openrisc-for-linus' of git://github.com/openrisc/linux

Pull OpenRISC fixes from Stafford Horne:
 "OpenRISC fixes for build issues that were exposed by kbuild robots
  after 4.11 merge. All from allmodconfig builds. This includes:

   - bug in the handling of 8-byte get_user() calls

   - module build failure due to multile missing symbol exports"

* tag 'openrisc-for-linus' of git://github.com/openrisc/linux:
  openrisc: Export symbols needed by modules
  openrisc: fix issue handling 8 byte get_user calls
  openrisc: xchg: fix `computed is not used` warning

7 years agotarget: fix race during implicit transition work flushes
Mike Christie [Thu, 2 Mar 2017 10:59:50 +0000 (04:59 -0600)]
target: fix race during implicit transition work flushes

This fixes the following races:

1. core_alua_do_transition_tg_pt could have read
tg_pt_gp_alua_access_state and gone into this if chunk:

if (!explicit &&
        atomic_read(&tg_pt_gp->tg_pt_gp_alua_access_state) ==
           ALUA_ACCESS_STATE_TRANSITION) {

and then core_alua_do_transition_tg_pt_work could update the
state. core_alua_do_transition_tg_pt would then only set
tg_pt_gp_alua_pending_state and the tg_pt_gp_alua_access_state would
not get updated with the second calls state.

2. core_alua_do_transition_tg_pt could be setting
tg_pt_gp_transition_complete while the tg_pt_gp_transition_work
is already completing. core_alua_do_transition_tg_pt then waits on the
completion that will never be called.

To handle these issues, we just call flush_work which will return when
core_alua_do_transition_tg_pt_work has completed so there is no need
to do the complete/wait. And, if core_alua_do_transition_tg_pt_work
was running, instead of trying to sneak in the state change, we just
schedule up another core_alua_do_transition_tg_pt_work call.

Note that this does not handle a possible race where there are multiple
threads call core_alua_do_transition_tg_pt at the same time. I think
we need a mutex in target_tg_pt_gp_alua_access_state_store.

Signed-off-by: Mike Christie <mchristi@redhat.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agotarget: allow userspace to set state to transitioning
Mike Christie [Thu, 2 Mar 2017 10:59:49 +0000 (04:59 -0600)]
target: allow userspace to set state to transitioning

Userspace target_core_user handlers like tcmu-runner may want to set the
ALUA state to transitioning while it does implicit transitions. This
patch allows that state when set from configfs.

Signed-off-by: Mike Christie <mchristi@redhat.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agotarget: fix ALUA transition timeout handling
Mike Christie [Thu, 2 Mar 2017 10:59:48 +0000 (04:59 -0600)]
target: fix ALUA transition timeout handling

The implicit transition time tells initiators the min time
to wait before timing out a transition. We currently schedule
the transition to occur in tg_pt_gp_implicit_trans_secs
seconds so there is no room for delays. If
core_alua_do_transition_tg_pt_work->core_alua_update_tpg_primary_metadata
needs to write out info to a remote file, then the initiator can
easily time out the operation.

Signed-off-by: Mike Christie <mchristi@redhat.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agotarget: Use system workqueue for ALUA transitions
Mike Christie [Thu, 2 Mar 2017 05:13:26 +0000 (23:13 -0600)]
target: Use system workqueue for ALUA transitions

If tcmu-runner is processing a STPG and needs to change the kernel's
ALUA state then we cannot use the same work queue for task management
requests and ALUA transitions, because we could deadlock. The problem
occurs when a STPG times out before tcmu-runner is able to
call into target_tg_pt_gp_alua_access_state_store->
core_alua_do_port_transition -> core_alua_do_transition_tg_pt ->
queue_work. In this case, the tmr is on the work queue waiting for
the STPG to complete, but the STPG transition is now queued behind
the waiting tmr.

Note:
This bug will also be fixed by this patch:
http://www.spinics.net/lists/target-devel/msg14560.html
which switches the tmr code to use the system workqueues.

For both, I am not sure if we need a dedicated workqueue since
it is not a performance path and I do not think we need WQ_MEM_RECLAIM
to make forward progress to free up memory like the block layer does.

Signed-off-by: Mike Christie <mchristi@redhat.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agotarget: fail ALUA transitions for pscsi
Mike Christie [Thu, 2 Mar 2017 05:13:25 +0000 (23:13 -0600)]
target: fail ALUA transitions for pscsi

We do not setup the LU group for pscsi devices, so if you write
a state to alua_access_state that will cause a transition you will
get a NULL pointer dereference.

This patch will fail attempts to try and transition the path
for backend devices that set the TRANSPORT_FLAG_PASSTHROUGH_ALUA
flag.

Signed-off-by: Mike Christie <mchristi@redhat.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agotarget: allow ALUA setup for some passthrough backends
Mike Christie [Thu, 2 Mar 2017 05:13:24 +0000 (23:13 -0600)]
target: allow ALUA setup for some passthrough backends

This patch allows passthrough backends to use the core/base LIO
ALUA setup and state checks, but still handle the execution of
commands.

This will allow the target_core_user module to execute STPG and RTPG
in userspace, and not have to duplicate the ALUA state checks, path
information (needed so we can check if command is executable on
specific paths) and setup (rtslib sets/updates the configfs ALUA
interface like it does for iblock or file).

For STPG, the target_core_user userspace daemon, tcmu-runner will
still execute the STPG, and to update the core/base LIO state it
will use the existing configfs interface. For RTPG, tcmu-runner
will loop over configfs and/or cache the state.

Signed-off-by: Mike Christie <mchristi@redhat.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agotcmu: return on first Opt parse failure
Mike Christie [Thu, 2 Mar 2017 05:14:40 +0000 (23:14 -0600)]
tcmu: return on first Opt parse failure

We only were returing failure if the last opt to be parsed failed.
This has a return failure when we first detect a failure.

Signed-off-by: Mike Christie <mchristi@redhat.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
7 years agotcmu: allow hw_max_sectors greater than 128
Mike Christie [Thu, 2 Mar 2017 05:14:39 +0000 (23:14 -0600)]
tcmu: allow hw_max_sectors greater than 128

tcmu hard codes the hw_max_sectors to 128 which is a litle small.
Userspace uses the max_sectors to report the optimal IO size and
some initiators perform better with larger IOs (open-iscsi seems
to do better with 256 to 512 depending on the test).

(Fix do not display hw max sectors twice - MNC)

Signed-off-by: Mike Christie <mchristi@redhat.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>