With radix, the guest can do TLB invalidations itself using the tlbie
(global) and tlbiel (local) TLB invalidation instructions.  Linux guests
use local TLB invalidations for translations that have only ever been
accessed on one vcpu.  However, that doesn't mean that the translations
have only been accessed on one physical cpu (pcpu) since vcpus can move
around from one pcpu to another.  Thus a tlbiel might leave behind stale
TLB entries on a pcpu where the vcpu previously ran, and if that task
then moves back to that previous pcpu, it could see those stale TLB
entries and thus access memory incorrectly.  The usual symptom of this
is random segfaults in userspace programs in the guest.

To cope with this, we detect when a vcpu is about to start executing on
a thread in a core that is a different core from the last time it
executed.  If that is the case, then we mark the core as needing a
TLB flush and then send an interrupt to any thread in the core that is
currently running a vcpu from the same guest.  This will get those vcpus
out of the guest, and the first one to re-enter the guest will do the
TLB flush.  The reason for interrupting the vcpus executing on the old
core is to cope with the following scenario:

	CPU 0			CPU 1			CPU 4
	(core 0)			(core 0)			(core 1)

	VCPU 0 runs task X      VCPU 1 runs
	core 0 TLB gets
	entries from task X
	VCPU 0 moves to CPU 4
							VCPU 0 runs task X
							Unmap pages of task X
							tlbiel

				(still VCPU 1)			task X moves to VCPU 1
				task X runs
				task X sees stale TLB
				entries

That is, as soon as the VCPU starts executing on the new core, it
could unmap and tlbiel some page table entries, and then the task
could migrate to one of the VCPUs running on the old core and
potentially see stale TLB entries.

Since the TLB is shared between all the threads in a core, we only
use the bit of kvm->arch.need_tlb_flush corresponding to the first
thread in the core.  To ensure that we don't have a window where we
can miss a flush, this moves the clearing of the bit from before the
actual flush to after it.  This way, two threads might both do the
flush, but we prevent the situation where one thread can enter the
guest before the flush is finished.
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This adds code to  branch around the parts that radix guests don't
need - clearing and loading the SLB with the guest SLB contents,
saving the guest SLB contents on exit, and restoring the host SLB
contents.

Since the host is now using radix, we need to save and restore the
host value for the PID register.

On hypervisor data/instruction storage interrupts, we don't do the
guest HPT lookup on radix, but just save the guest physical address
for the fault (from the ASDR register) in the vcpu struct.
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

POWER9 adds a register called ASDR (Access Segment Descriptor
Register), which is set by hypervisor data/instruction storage
interrupts to contain the segment descriptor for the address
being accessed, assuming the guest is using HPT translation.
(For radix guests, it contains the guest real address of the
access.)

Thus, for HPT guests on POWER9, we can use this register rather
than looking up the SLB with the slbfee. instruction.
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

64-bit Book3S exception handlers must find the dynamic kernel base
to add to the target address when branching beyond __end_interrupts,
in order to support kernel running at non-0 physical address.

Support this in KVM by branching with CTR, similarly to regular
interrupt handlers. The guest CTR saved in HSTATE_SCRATCH1 and
restored after the branch.

Without this, the host kernel hangs and crashes randomly when it is
running at a non-0 address and a KVM guest is started.
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Acked-by: NPaul Mackerras <paulus@ozlabs.org>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Change the calling convention to put the trap number together with
CR in two halves of r12, which frees up HSTATE_SCRATCH2 in the HV
handler.

The 64-bit PR handler entry translates the calling convention back
to match the previous call convention (i.e., shared with 32-bit), for
simplicity.
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Acked-by: NPaul Mackerras <paulus@ozlabs.org>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

POWER9 replaces the various power-saving mode instructions on POWER8
(doze, nap, sleep and rvwinkle) with a single "stop" instruction, plus
a register, PSSCR, which controls the depth of the power-saving mode.
This replaces the use of the nap instruction when threads are idle
during guest execution with the stop instruction, and adds code to
set PSSCR to a value which will allow an SMT mode switch while the
thread is idle (given that the core as a whole won't be idle in these
cases).
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

POWER9 adds new capabilities to the tlbie (TLB invalidate entry)
and tlbiel (local tlbie) instructions.  Both instructions get a
set of new parameters (RIC, PRS and R) which appear as bits in the
instruction word.  The tlbiel instruction now has a second register
operand, which contains a PID and/or LPID value if needed, and
should otherwise contain 0.

This adapts KVM-HV's usage of tlbie and tlbiel to work on POWER9
as well as older processors.  Since we only handle HPT guests so
far, we need RIC=0 PRS=0 R=0, which ends up with the same instruction
word as on previous processors, so we don't need to conditionally
execute different instructions depending on the processor.

The local flush on first entry to a guest in book3s_hv_rmhandlers.S
is a loop which depends on the number of TLB sets.  Rather than
using feature sections to set the number of iterations based on
which CPU we're on, we now work out this number at VM creation time
and store it in the kvm_arch struct.  That will make it possible to
get the number from the device tree in future, which will help with
compatibility with future processors.

Since mmu_partition_table_set_entry() does a global flush of the
whole LPID, we don't need to do the TLB flush on first entry to the
guest on each processor.  Therefore we don't set all bits in the
tlb_need_flush bitmap on VM startup on POWER9.
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

This adds code to handle two new guest-accessible special-purpose
registers on POWER9: TIDR (thread ID register) and PSSCR (processor
stop status and control register).  They are context-switched
between host and guest, and the guest values can be read and set
via the one_reg interface.

The PSSCR contains some fields which are guest-accessible and some
which are only accessible in hypervisor mode.  We only allow the
guest-accessible fields to be read or set by userspace.
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

Some special-purpose registers that were present and accessible
by guests on POWER8 no longer exist on POWER9, so this adds
feature sections to ensure that we don't try to context-switch
them when going into or out of a guest on POWER9.  These are
all relatively obscure, rarely-used registers, but we had to
context-switch them on POWER8 to avoid creating a covert channel.
They are: SPMC1, SPMC2, MMCRS, CSIGR, TACR, TCSCR, and ACOP.
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

On POWER9, the SDR1 register (hashed page table base address) is no
longer used, and instead the hardware reads the HPT base address
and size from the partition table.  The partition table entry also
contains the bits that specify the page size for the VRMA mapping,
which were previously in the LPCR.  The VPM0 bit of the LPCR is
now reserved; the processor now always uses the VRMA (virtual
real-mode area) mechanism for guest real-mode accesses in HPT mode,
and the RMO (real-mode offset) mechanism has been dropped.

When entering or exiting the guest, we now only have to set the
LPIDR (logical partition ID register), not the SDR1 register.
There is also no requirement now to transition via a reserved
LPID value.
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

When switching from/to a guest that has a transaction in progress,
we need to save/restore the checkpointed register state.  Although
XER is part of the CPU state that gets checkpointed, the code that
does this saving and restoring doesn't save/restore XER.

This fixes it by saving and restoring the XER.  To allow userspace
to read/write the checkpointed XER value, we also add a new ONE_REG
specifier.

The visible effect of this bug is that the guest may see its XER
value being corrupted when it uses transactions.

Fixes: e4e38121 ("KVM: PPC: Book3S HV: Add transactional memory support")
Fixes: 0a8eccef ("KVM: PPC: Book3S HV: Add missing code for transaction reclaim on guest exit")
Cc: stable@vger.kernel.org # v3.15+
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>
Reviewed-by: NThomas Huth <thuth@redhat.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

POWER8 has one virtual timebase (VTB) register per subcore, not one
per CPU thread.  The HV KVM code currently treats VTB as a per-thread
register, which can lead to spurious soft lockup messages from guests
which use the VTB as the time source for the soft lockup detector.
(CPUs before POWER8 did not have the VTB register.)

For HV KVM, this fixes the problem by making only the primary thread
in each virtual core save and restore the VTB value.  With this,
the VTB state becomes part of the kvmppc_vcore structure.  This
also means that "piggybacking" of multiple virtual cores onto one
subcore is not possible on POWER8, because then the virtual cores
would share a single VTB register.

PR KVM emulates a VTB register, which is per-vcpu because PR KVM
has no notion of CPU threads or SMT.  For PR KVM we move the VTB
state into the kvmppc_vcpu_book3s struct.

Cc: stable@vger.kernel.org # v3.14+
Reported-by: NThomas Huth <thuth@redhat.com>
Tested-by: NThomas Huth <thuth@redhat.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

In existing real mode ICP code, when updating the virtual ICP
state, if there is a required action that cannot be completely
handled in real mode, as for instance, a VCPU needs to be woken
up, flags are set in the ICP to indicate the required action.
This is checked when returning from hypercalls to decide whether
the call needs switch back to the host where the action can be
performed in virtual mode. Note that if h_ipi_redirect is enabled,
real mode code will first try to message a free host CPU to
complete this job instead of returning the host to do it ourselves.

Currently, the real mode PCI passthrough interrupt handling code
checks if any of these flags are set and simply returns to the host.
This is not good enough as the trap value (0x500) is treated as an
external interrupt by the host code. It is only when the trap value
is a hypercall that the host code searches for and acts on unfinished
work by calling kvmppc_xics_rm_complete.

This patch introduces a special trap BOOK3S_INTERRUPT_HV_RM_HARD
which is returned by KVM if there is unfinished business to be
completed in host virtual mode after handling a PCI passthrough
interrupt. The host checks for this special interrupt condition
and calls into the kvmppc_xics_rm_complete, which is made an
exported function for this reason.

[paulus@ozlabs.org - moved logic to set r12 to BOOK3S_INTERRUPT_HV_RM_HARD
 in book3s_hv_rmhandlers.S into the end of kvmppc_check_wake_reason.]
Signed-off-by: NSuresh Warrier <warrier@linux.vnet.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

Currently, KVM switches back to the host to handle any external
interrupt (when the interrupt is received while running in the
guest). This patch updates real-mode KVM to check if an interrupt
is generated by a passthrough adapter that is owned by this guest.
If so, the real mode KVM will directly inject the corresponding
virtual interrupt to the guest VCPU's ICS and also EOI the interrupt
in hardware. In short, the interrupt is handled entirely in real
mode in the guest context without switching back to the host.

In some rare cases, the interrupt cannot be completely handled in
real mode, for instance, a VCPU that is sleeping needs to be woken
up. In this case, KVM simply switches back to the host with trap
reason set to 0x500. This works, but it is clearly not very efficient.
A following patch will distinguish this case and handle it
correctly in the host. Note that we can use the existing
check_too_hard() routine even though we are not in a hypercall to
determine if there is unfinished business that needs to be
completed in host virtual mode.

The patch assumes that the mapping between hardware interrupt IRQ
and virtual IRQ to be injected to the guest already exists for the
PCI passthrough interrupts that need to be handled in real mode.
If the mapping does not exist, KVM falls back to the default
existing behavior.

The KVM real mode code reads mappings from the mapped array in the
passthrough IRQ map without taking any lock.  We carefully order the
loads and stores of the fields in the kvmppc_irq_map data structure
using memory barriers to avoid an inconsistent mapping being seen by
the reader. Thus, although it is possible to miss a map entry, it is
not possible to read a stale value.

[paulus@ozlabs.org - get irq_chip from irq_map rather than pimap,
 pulled out powernv eoi change into a separate patch, made
 kvmppc_read_intr get the vcpu from the paca rather than being
 passed in, rewrote the logic at the end of kvmppc_read_intr to
 avoid deep indentation, simplified logic in book3s_hv_rmhandlers.S
 since we were always restoring SRR0/1 anyway, get rid of the cached
 array (just use the mapped array), removed the kick_all_cpus_sync()
 call, clear saved_xirr PACA field when we handle the interrupt in
 real mode, fix compilation with CONFIG_KVM_XICS=n.]
Signed-off-by: NSuresh Warrier <warrier@linux.vnet.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

Modify kvmppc_read_intr to make it a C function.  Because it is called
from kvmppc_check_wake_reason, any of the assembler code that calls
either kvmppc_read_intr or kvmppc_check_wake_reason now has to assume
that the volatile registers might have been modified.

This also adds in the optimization of clearing saved_xirr in the case
where we completely handle and EOI an IPI.  Without this, the next
device interrupt will require two trips through the host interrupt
handling code.

[paulus@ozlabs.org - made kvmppc_check_wake_reason create a stack frame
 when it is calling kvmppc_read_intr, which means we can set r12 to
 the trap number (0x500) after the call to kvmppc_read_intr, instead
 of using r31.  Also moved the deliver_guest_interrupt label so as to
 restore XER and CTR, plus other minor tweaks.]
Signed-off-by: NSuresh Warrier <warrier@linux.vnet.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

It turns out that if the guest does a H_CEDE while the CPU is in
a transactional state, and the H_CEDE does a nap, and the nap
loses the architected state of the CPU (which is is allowed to do),
then we lose the checkpointed state of the virtual CPU.  In addition,
the transactional-memory state recorded in the MSR gets reset back
to non-transactional, and when we try to return to the guest, we take
a TM bad thing type of program interrupt because we are trying to
transition from non-transactional to transactional with a hrfid
instruction, which is not permitted.

The result of the program interrupt occurring at that point is that
the host CPU will hang in an infinite loop with interrupts disabled.
Thus this is a denial of service vulnerability in the host which can
be triggered by any guest (and depending on the guest kernel, it can
potentially triggered by unprivileged userspace in the guest).

This vulnerability has been assigned the ID CVE-2016-5412.

To fix this, we save the TM state before napping and restore it
on exit from the nap, when handling a H_CEDE in real mode.  The
case where H_CEDE exits to host virtual mode is already OK (as are
other hcalls which exit to host virtual mode) because the exit
path saves the TM state.

Cc: stable@vger.kernel.org # v3.15+
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

This moves the transactional memory state save and restore sequences
out of the guest entry/exit paths into separate procedures.  This is
so that these sequences can be used in going into and out of nap
in a subsequent patch.

The only code changes here are (a) saving and restore LR on the
stack, since these new procedures get called with a bl instruction,
(b) explicitly saving r1 into the PACA instead of assuming that
HSTATE_HOST_R1(r13) is already set, and (c) removing an unnecessary
and redundant setting of MSR[TM] that should have been removed by
commit 9d4d0bdd9e0a ("KVM: PPC: Book3S HV: Add transactional memory
support", 2013-09-24) but wasn't.

Cc: stable@vger.kernel.org # v3.15+
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

Functions like power7_wakeup_loss, power7_wakeup_noloss,
power7_wakeup_tb_loss are used by POWER7 and POWER8 hardware. They can
also be used by POWER9. Hence rename these functions hardware agnostic
names.
Suggested-by: NGautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: NShreyas B. Prabhu <shreyas@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

When a guest is assigned to a core it converts the host Timebase (TB)
into guest TB by adding guest timebase offset before entering into
guest. During guest exit it restores the guest TB to host TB. This means
under certain conditions (Guest migration) host TB and guest TB can differ.

When we get an HMI for TB related issues the opal HMI handler would
try fixing errors and restore the correct host TB value. With no guest
running, we don't have any issues. But with guest running on the core
we run into TB corruption issues.

If we get an HMI while in the guest, the current HMI handler invokes opal
hmi handler before forcing guest to exit. The guest exit path subtracts
the guest TB offset from the current TB value which may have already
been restored with host value by opal hmi handler. This leads to incorrect
host and guest TB values.

With split-core, things become more complex. With split-core, TB also gets
split and each subcore gets its own TB register. When a hmi handler fixes
a TB error and restores the TB value, it affects all the TB values of
sibling subcores on the same core. On TB errors all the thread in the core
gets HMI. With existing code, the individual threads call opal hmi handle
independently which can easily throw TB out of sync if we have guest
running on subcores. Hence we will need to co-ordinate with all the
threads before making opal hmi handler call followed by TB resync.

This patch introduces a sibling subcore state structure (shared by all
threads in the core) in paca which holds information about whether sibling
subcores are in Guest mode or host mode. An array in_guest[] of size
MAX_SUBCORE_PER_CORE=4 is used to maintain the state of each subcore.
The subcore id is used as index into in_guest[] array. Only primary
thread entering/exiting the guest is responsible to set/unset its
designated array element.

On TB error, we get HMI interrupt on every thread on the core. Upon HMI,
this patch will now force guest to vacate the core/subcore. Primary
thread from each subcore will then turn off its respective bit
from the above bitmap during the guest exit path just after the
guest->host partition switch is complete.

All other threads that have just exited the guest OR were already in host
will wait until all other subcores clears their respective bit.
Once all the subcores turn off their respective bit, all threads will
will make call to opal hmi handler.

It is not necessary that opal hmi handler would resync the TB value for
every HMI interrupts. It would do so only for the HMI caused due to
TB errors. For rest, it would not touch TB value. Hence to make things
simpler, primary thread would call TB resync explicitly once for each
core immediately after opal hmi handler instead of subtracting guest
offset from TB. TB resync call will restore the TB with host value.
Thus we can be sure about the TB state.

One of the primary threads exiting the guest will take up the
responsibility of calling TB resync. It will use one of the top bits
(bit 63) from subcore state flags bitmap to make the decision. The first
primary thread (among the subcores) that is able to set the bit will
have to call the TB resync. Rest all other threads will wait until TB
resync is complete.  Once TB resync is complete all threads will then
proceed.
Signed-off-by: NMahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

Upcoming in-kernel VFIO acceleration needs different handling in real
and virtual modes which makes it hard to support both modes in
the same handler.

This creates a copy of kvmppc_rm_h_stuff_tce and kvmppc_rm_h_put_tce
in addition to the existing kvmppc_rm_h_put_tce_indirect.

This also fixes linker breakage when only PR KVM was selected (leaving
HV KVM off): the kvmppc_h_put_tce/kvmppc_h_stuff_tce functions
would not compile at all and the linked would fail.
Signed-off-by: NAlexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Thomas Huth discovered that a guest could cause a hard hang of a
host CPU by setting the Instruction Authority Mask Register (IAMR)
to a suitable value.  It turns out that this is because when the
code was added to context-switch the new special-purpose registers
(SPRs) that were added in POWER8, we forgot to add code to ensure
that they were restored to a sane value on guest exit.

This adds code to set those registers where a bad value could
compromise the execution of the host kernel to a suitable neutral
value on guest exit.

Cc: stable@vger.kernel.org # v3.14+
Fixes: b005255eReported-by: NThomas Huth <thuth@redhat.com>
Reviewed-by: NDavid Gibson <david@gibson.dropbear.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

No code changes.
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This adds real and virtual mode handlers for the H_PUT_TCE_INDIRECT and
H_STUFF_TCE hypercalls for user space emulated devices such as IBMVIO
devices or emulated PCI. These calls allow adding multiple entries
(up to 512) into the TCE table in one call which saves time on
transition between kernel and user space.

The current implementation of kvmppc_h_stuff_tce() allows it to be
executed in both real and virtual modes so there is one helper.
The kvmppc_rm_h_put_tce_indirect() needs to translate the guest address
to the host address and since the translation is different, there are
2 helpers - one for each mode.

This implements the KVM_CAP_PPC_MULTITCE capability. When present,
the kernel will try handling H_PUT_TCE_INDIRECT and H_STUFF_TCE if these
are enabled by the userspace via KVM_CAP_PPC_ENABLE_HCALL.
If they can not be handled by the kernel, they are passed on to
the user space. The user space still has to have an implementation
for these.

Both HV and PR-syle KVM are supported.
Signed-off-by: NAlexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: NDavid Gibson <david@gibson.dropbear.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

In the old DABR register, the BT (Breakpoint Translation) bit
is bit number 61. In the new DAWRX register, the WT (Watchpoint
Translation) bit is bit number 59. So to move the DABR-BT bit
into the position of the DAWRX-WT bit, it has to be shifted by
two, not only by one. This fixes hardware watchpoints in gdb of
older guests that only use the H_SET_DABR/X interface instead
of the new H_SET_MODE interface.

Cc: stable@vger.kernel.org # v3.14+
Signed-off-by: NThomas Huth <thuth@redhat.com>
Reviewed-by: NLaurent Vivier <lvivier@redhat.com>
Reviewed-by: NDavid Gibson <david@gibson.dropbear.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

As we saw with the TM Bad Thing type of program interrupt occurring
on the hrfid that enters the guest, it is not completely impossible
to have a trap occurring in the guest entry/exit code, despite the
fact that the code has been written to avoid taking any traps.

This adds a check in the kvmppc_handle_exit_hv() function to detect
the case when a trap has occurred in the hypervisor-mode code, and
instead of treating it just like a trap in guest code, we now print
a message and return to userspace with a KVM_EXIT_INTERNAL_ERROR
exit reason.

Of the various interrupts that get handled in the assembly code in
the guest exit path and that can return directly to the guest, the
only one that can occur when MSR.HV=1 and MSR.EE=0 is machine check
(other than system call, which we can avoid just by not doing a sc
instruction).  Therefore this adds code to the machine check path to
ensure that if the MCE occurred in hypervisor mode, we exit to the
host rather than trying to continue the guest.
Signed-off-by: NPaul Mackerras <paulus@samba.org>

When handling a hypervisor data or instruction storage interrupt (HDSI
or HISI), we look up the SLB entry for the address being accessed in
order to translate the effective address to a virtual address which can
be looked up in the guest HPT.  This lookup can occasionally fail due
to the guest replacing an SLB entry without invalidating the evicted
SLB entry.  In this situation an ERAT (effective to real address
translation cache) entry can persist and be used by the hardware even
though there is no longer a corresponding SLB entry.

Previously we would just deliver a data or instruction storage interrupt
(DSI or ISI) to the guest in this case.  However, this is not correct
and has been observed to cause guests to crash, typically with a
data storage protection interrupt on a store to the vmemmap area.

Instead, what we do now is to synthesize a data or instruction segment
interrupt.  That should cause the guest to reload an appropriate entry
into the SLB and retry the faulting instruction.  If it still faults,
we should find an appropriate SLB entry next time and be able to handle
the fault.
Tested-by: NThomas Huth <thuth@redhat.com>
Reviewed-by: NDavid Gibson <david@gibson.dropbear.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Currently a CPU running a guest can receive a H_DOORBELL in the
following two cases:
1) When the CPU is napping due to CEDE or there not being a guest
vcpu.
2) The CPU is running the guest vcpu.

Case 1), the doorbell message is not cleared since we were waking up
from nap. Hence when the EE bit gets set on transition from guest to
host, the H_DOORBELL interrupt is delivered to the host and the
corresponding handler is invoked.

However in Case 2), the message gets cleared by the action of taking
the H_DOORBELL interrupt. Since the CPU was running a guest, instead
of invoking the doorbell handler, the code invokes the second-level
interrupt handler to switch the context from the guest to the host. At
this point the setting of the EE bit doesn't result in the CPU getting
the doorbell interrupt since it has already been delivered once. So,
the handler for this doorbell is never invoked!

This causes softlockups if the missed DOORBELL was an IPI sent from a
sibling subcore on the same CPU.

This patch fixes it by explitly invoking the doorbell handler on the
exit path if the exit reason is H_DOORBELL similar to the way an
EXTERNAL interrupt is handled. Since this will also handle Case 1), we
can unconditionally clear the doorbell message in
kvmppc_check_wake_reason.
Signed-off-by: NGautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

For the machine check interrupt that happens while we are in the guest,
kvm layer attempts the recovery, and then delivers the machine check interrupt
directly to the guest if recovery fails. On successful recovery we go back to
normal functioning of the guest. But there can be cases where a machine check
interrupt can happen with MSR(RI=0) while we are in the guest. This means
MC interrupt is unrecoverable and we have to deliver a machine check to the
guest since the machine check interrupt might have trashed valid values in
SRR0/1. The current implementation do not handle this case, causing guest
to crash with Bad kernel stack pointer instead of machine check oops message.

[26281.490060] Bad kernel stack pointer 3fff9ccce5b0 at c00000000000490c
[26281.490434] Oops: Bad kernel stack pointer, sig: 6 [#1]
[26281.490472] SMP NR_CPUS=2048 NUMA pSeries

This patch fixes this issue by checking MSR(RI=0) in KVM layer and forwarding
unrecoverable interrupt to guest which then panics with proper machine check
Oops message.
Signed-off-by: NMahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Acked-by: NPaul Mackerras <paulus@samba.org>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

In guest_exit_cont we call kvmhv_commence_exit which expects the trap
number as the argument. However r3 doesn't contain the trap number at
this point and as a result we would be calling the function with a
spurious trap number.

Fix this by copying r12 into r3 before calling kvmhv_commence_exit as
r12 contains the trap number.

Cc: stable@vger.kernel.org # v4.1+
Fixes: eddb60fbSigned-off-by: NGautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

The code that handles the case when we receive a H_DOORBELL interrupt
has a comment which says "Hypervisor doorbell - exit only if host IPI
flag set".  However, the current code does not actually check if the
host IPI flag is set.  This is due to a comparison instruction that
got missed.

As a result, the current code performs the exit to host only
if some sibling thread or a sibling sub-core is exiting to the
host.  This implies that, an IPI sent to a sibling core in
(subcores-per-core != 1) mode will be missed by the host unless the
sibling core is on the exit path to the host.

This patch adds the missing comparison operation which will ensure
that when HOST_IPI flag is set, we unconditionally exit to the host.

Fixes: 66feed61
Cc: stable@vger.kernel.org # v4.1+
Signed-off-by: NGautham R. Shenoy <ego@linux.vnet.ibm.com>
Reviewed-by: NDavid Gibson <david@gibson.dropbear.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

The current dynamic micro-threading code has a race due to which a
secondary thread naps when it is supposed to be running a vcpu. As a
side effect of this, on a guest exit, the primary thread in
kvmppc_wait_for_nap() finds that this secondary thread hasn't cleared
its vcore pointer. This results in "CPU X seems to be stuck!"
warnings.

The race is possible since the primary thread on exiting the guests
only waits for all the secondaries to clear its vcore pointer. It
subsequently expects the secondary threads to enter nap while it
unsplits the core. A secondary thread which hasn't yet entered the nap
will loop in kvm_no_guest until its vcore pointer and the do_nap flag
are unset. Once the core has been unsplit, a new vcpu thread can grab
the core and set the do_nap flag *before* setting the vcore pointers
of the secondary. As a result, the secondary thread will now enter nap
via kvm_unsplit_nap instead of running the guest vcpu.

Fix this by setting the do_nap flag after setting the vcore pointer in
the PACA of the secondary in kvmppc_run_core. Also, ensure that a
secondary thread doesn't nap in kvm_unsplit_nap when the vcore pointer
in its PACA struct is set.

Fixes: b4deba5cSigned-off-by: NGautham R. Shenoy <ego@linux.vnet.ibm.com>
Reviewed-by: NDavid Gibson <david@gibson.dropbear.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

In 64 bit kernels, the Fixed Point Exception Register (XER) is a 64
bit field (e.g. in kvm_regs and kvm_vcpu_arch) and in most places it is
accessed as such.

This patch corrects places where it is accessed as a 32 bit field by a
64 bit kernel.  In some cases this is via a 32 bit load or store
instruction which, depending on endianness, will cause either the
lower or upper 32 bits to be missed.  In another case it is cast as a
u32, causing the upper 32 bits to be cleared.

This patch corrects those places by extending the access methods to
64 bits.
Signed-off-by: NSam Bobroff <sam.bobroff@au1.ibm.com>
Reviewed-by: NLaurent Vivier <lvivier@redhat.com>
Reviewed-by: NThomas Huth <thuth@redhat.com>
Tested-by: NThomas Huth <thuth@redhat.com>
Signed-off-by: NAlexander Graf <agraf@suse.de>

This adds implementations for the H_CLEAR_REF (test and clear reference
bit) and H_CLEAR_MOD (test and clear changed bit) hypercalls.

When clearing the reference or change bit in the guest view of the HPTE,
we also have to clear it in the real HPTE so that we can detect future
references or changes.  When we do so, we transfer the R or C bit value
to the rmap entry for the underlying host page so that kvm_age_hva_hv(),
kvm_test_age_hva_hv() and kvmppc_hv_get_dirty_log() know that the page
has been referenced and/or changed.

These hypercalls are not used by Linux guests.  These implementations
have been tested using a FreeBSD guest.
Signed-off-by: NPaul Mackerras <paulus@samba.org>
Signed-off-by: NAlexander Graf <agraf@suse.de>

This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip.  Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core.  With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.

Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.

To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread.  In addition we extend the core_info struct to
have information on each subcore.  When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.

Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode.  It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.

This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap.  In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered.  The default is 6.

With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch.  These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.

It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest.  In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path.  In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: NPaul Mackerras <paulus@samba.org>
Signed-off-by: NAlexander Graf <agraf@suse.de>

When running a virtual core of a guest that is configured with fewer
threads per core than the physical cores have, the extra physical
threads are currently unused.  This makes it possible to use them to
run one or more other virtual cores from the same guest when certain
conditions are met.  This applies on POWER7, and on POWER8 to guests
with one thread per virtual core.  (It doesn't apply to POWER8 guests
with multiple threads per vcore because they require a 1-1 virtual to
physical thread mapping in order to be able to use msgsndp and the
TIR.)

The idea is that we maintain a list of preempted vcores for each
physical cpu (i.e. each core, since the host runs single-threaded).
Then, when a vcore is about to run, it checks to see if there are
any vcores on the list for its physical cpu that could be
piggybacked onto this vcore's execution.  If so, those additional
vcores are put into state VCORE_PIGGYBACK and their runnable VCPU
threads are started as well as the original vcore, which is called
the master vcore.

After the vcores have exited the guest, the extra ones are put back
onto the preempted list if any of their VCPUs are still runnable and
not idle.

This means that vcpu->arch.ptid is no longer necessarily the same as
the physical thread that the vcpu runs on.  In order to make it easier
for code that wants to send an IPI to know which CPU to target, we
now store that in a new field in struct vcpu_arch, called thread_cpu.
Reviewed-by: NDavid Gibson <david@gibson.dropbear.id.au>
Tested-by: NLaurent Vivier <lvivier@redhat.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>
Signed-off-by: NAlexander Graf <agraf@suse.de>

PACA_DSCR offset macro tracks dscr_default element in the paca
structure. Better change the name of this macro to match that of the
data element it tracks. Makes the code more readable.
Signed-off-by: NAnshuman Khandual <khandual@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This uses msgsnd where possible for signalling other threads within
the same core on POWER8 systems, rather than IPIs through the XICS
interrupt controller.  This includes waking secondary threads to run
the guest, the interrupts generated by the virtual XICS, and the
interrupts to bring the other threads out of the guest when exiting.

Aggregated statistics from debugfs across vcpus for a guest with 32
vcpus, 8 threads/vcore, running on a POWER8, show this before the
change:

 rm_entry:     3387.6ns (228 - 86600, 1008969 samples)
  rm_exit:     4561.5ns (12 - 3477452, 1009402 samples)
  rm_intr:     1660.0ns (12 - 553050, 3600051 samples)

and this after the change:

 rm_entry:     3060.1ns (212 - 65138, 953873 samples)
  rm_exit:     4244.1ns (12 - 9693408, 954331 samples)
  rm_intr:     1342.3ns (12 - 1104718, 3405326 samples)

for a test of booting Fedora 20 big-endian to the login prompt.

The time taken for a H_PROD hcall (which is handled in the host
kernel) went down from about 35 microseconds to about 16 microseconds
with this change.

The noinline added to kvmppc_run_core turned out to be necessary for
good performance, at least with gcc 4.9.2 as packaged with Fedora 21
and a little-endian POWER8 host.
Signed-off-by: NPaul Mackerras <paulus@samba.org>
Signed-off-by: NAlexander Graf <agraf@suse.de>

This replaces the assembler code for kvmhv_commence_exit() with C code
in book3s_hv_builtin.c.  It also moves the IPI sending code that was
in book3s_hv_rm_xics.c into a new kvmhv_rm_send_ipi() function so it
can be used by kvmhv_commence_exit() as well as icp_rm_set_vcpu_irq().
Signed-off-by: NPaul Mackerras <paulus@samba.org>
Signed-off-by: NAlexander Graf <agraf@suse.de>

On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.

On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after.  That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.

This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.

Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement.  Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:

 rm_entry:     avg 4537.3ns (222 - 48444, 1068878 samples)
  rm_exit:     avg 4787.6ns (152 - 165490, 1010717 samples)
  rm_intr:     avg 1673.6ns (12 - 341304, 3818691 samples)

and this after the change:

 rm_entry:     avg 3427.7ns (232 - 68150, 1118921 samples)
  rm_exit:     avg 4716.0ns (12 - 150720, 1119477 samples)
  rm_intr:     avg 1614.8ns (12 - 522436, 3850432 samples)

showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times.  (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: NPaul Mackerras <paulus@samba.org>
Signed-off-by: NAlexander Graf <agraf@suse.de>

Currently, the entry_exit_count field in the kvmppc_vcore struct
contains two 8-bit counts, one of the threads that have started entering
the guest, and one of the threads that have started exiting the guest.
This changes it to an entry_exit_map field which contains two bitmaps
of 8 bits each.  The advantage of doing this is that it gives us a
bitmap of which threads need to be signalled when exiting the guest.
That means that we no longer need to use the trick of setting the
HDEC to 0 to pull the other threads out of the guest, which led in
some cases to a spurious HDEC interrupt on the next guest entry.
Signed-off-by: NPaul Mackerras <paulus@samba.org>
Signed-off-by: NAlexander Graf <agraf@suse.de>