[PATCH] KVM: PPC: Book3S HV: Handle virtual mode in XIVE VCPU push code

[PATCH] KVM: PPC: Book3S HV: Handle virtual mode in XIVE VCPU push code

[Paul Mackerras]

From: Suraj Jitindar Singh <sjitindarsingh@gmail.com>

The code in book3s_hv_rmhandlers.S that pushes the XIVE virtual CPU
context to the hardware currently assumes it is being called in real
mode, which is usually true.  There is however a path by which it can
be executed in virtual mode, in the case where indep_threads_mode = N.
A virtual CPU executing on an offline secondary thread can take a
hypervisor interrupt in virtual mode and return from the
kvmppc_hv_entry() call after the kvm_secondary_got_guest label.
It is possible for it to be given another vcpu to execute before it
gets to execute the stop instruction.  In that case it will call
kvmppc_hv_entry() for the second VCPU in virtual mode, and the XIVE
vCPU push code will be executed in virtual mode.  The result in that
case will be a host crash due to an unexpected data storage interrupt
caused by executing the stdcix instruction in virtual mode.

This fixes it by adding a code path for virtual mode, which uses the
virtual TIMA pointer and normal load/store instructions.

[paulus@ozlabs.org - wrote patch description]

Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
---
 arch/powerpc/kvm/book3s_hv_rmhandlers.S | 36 +++++++++++++++++++++++----------
 1 file changed, 25 insertions(+), 11 deletions(-)

diff --git a/arch/powerpc/kvm/book3s_hv_rmhandlers.S b/arch/powerpc/kvm/book3s_hv_rmhandlers.S
index 5df137d..3f19f8b 100644
--- a/arch/powerpc/kvm/book3s_hv_rmhandlers.S
+++ b/arch/powerpc/kvm/book3s_hv_rmhandlers.S
@@ -944,17 +944,27 @@ ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
 
 #ifdef CONFIG_KVM_XICS
 	/* We are entering the guest on that thread, push VCPU to XIVE */
-	ld	r10, HSTATE_XIVE_TIMA_PHYS(r13)
-	cmpldi	cr0, r10, 0
-	beq	no_xive
 	ld	r11, VCPU_XIVE_SAVED_STATE(r4)
 	li	r9, TM_QW1_OS
+	lwz	r8, VCPU_XIVE_CAM_WORD(r4)
+	li	r7, TM_QW1_OS + TM_WORD2
+	mfmsr	r0
+	andi.	r0, r0, MSR_DR		/* in real mode? */
+	beq	2f
+	ld	r10, HSTATE_XIVE_TIMA_VIRT(r13)
+	cmpldi	cr1, r10, 0
+	beq     cr1, no_xive
+	eieio
+	stdx	r11,r9,r10
+	stwx	r8,r7,r10
+	b	3f
+2:	ld	r10, HSTATE_XIVE_TIMA_PHYS(r13)
+	cmpldi	cr1, r10, 0
+	beq	cr1, no_xive
 	eieio
 	stdcix	r11,r9,r10
-	lwz	r11, VCPU_XIVE_CAM_WORD(r4)
-	li	r9, TM_QW1_OS + TM_WORD2
-	stwcix	r11,r9,r10
-	li	r9, 1
+	stwcix	r8,r7,r10
+3:	li	r9, 1
 	stb	r9, VCPU_XIVE_PUSHED(r4)
 	eieio
 
@@ -973,12 +983,16 @@ ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
 	 * on, we mask it.
 	 */
 	lbz	r0, VCPU_XIVE_ESC_ON(r4)
-	cmpwi	r0,0
-	beq	1f
-	ld	r10, VCPU_XIVE_ESC_RADDR(r4)
+	cmpwi	cr1, r0,0
+	beq	cr1, 1f
 	li	r9, XIVE_ESB_SET_PQ_01
+	beq	4f			/* in real mode? */
+	ld	r10, VCPU_XIVE_ESC_VADDR(r4)
+	ldx	r0, r10, r9
+	b	5f
+4:	ld	r10, VCPU_XIVE_ESC_RADDR(r4)
 	ldcix	r0, r10, r9
-	sync
+5:	sync
 
 	/* We have a possible subtle race here: The escalation interrupt might
 	 * have fired and be on its way to the host queue while we mask it,
-- 
2.7.4

Re: [PATCH] KVM: PPC: Book3S HV: Handle virtual mode in XIVE VCPU push code

[Paul Mackerras]

On Mon, Apr 29, 2019 at 06:57:45PM +1000, Paul Mackerras wrote:
> From: Suraj Jitindar Singh <sjitindarsingh@gmail.com>
> 
> The code in book3s_hv_rmhandlers.S that pushes the XIVE virtual CPU
> context to the hardware currently assumes it is being called in real
> mode, which is usually true.  There is however a path by which it can
> be executed in virtual mode, in the case where indep_threads_mode = N.
> A virtual CPU executing on an offline secondary thread can take a
> hypervisor interrupt in virtual mode and return from the
> kvmppc_hv_entry() call after the kvm_secondary_got_guest label.
> It is possible for it to be given another vcpu to execute before it
> gets to execute the stop instruction.  In that case it will call
> kvmppc_hv_entry() for the second VCPU in virtual mode, and the XIVE
> vCPU push code will be executed in virtual mode.  The result in that
> case will be a host crash due to an unexpected data storage interrupt
> caused by executing the stdcix instruction in virtual mode.
> 
> This fixes it by adding a code path for virtual mode, which uses the
> virtual TIMA pointer and normal load/store instructions.
> 
> [paulus@ozlabs.org - wrote patch description]
> 
> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>

Patch applied to my kvm-ppc-next tree.

Paul.