Commit 863771a28e27 ("powerpc/32s: Convert switch_mmu_context() to C") moved the switch_mmu_context() to C. While in principle a good idea, it meant that the function now uses the stack. The stack is not accessible from real mode though.
So to keep calling the function, let's turn on MSR_DR while we call it. That way, all pointer references to the stack are handled virtually.
In addition, make sure to save/restore r12 in an SPRG, as it may get clobbered by the C function.
Reported-by: Matt Evans matt@ozlabs.org Fixes: 863771a28e27 ("powerpc/32s: Convert switch_mmu_context() to C") Signed-off-by: Alexander Graf graf@amazon.com Cc: stable@vger.kernel.org # v5.14+
---
v1 -> v2:
- Save and restore R12, so that we don't touch volatile registers while calling into C. --- arch/powerpc/kvm/book3s_32_sr.S | 26 +++++++++++++++++++++----- 1 file changed, 21 insertions(+), 5 deletions(-)
diff --git a/arch/powerpc/kvm/book3s_32_sr.S b/arch/powerpc/kvm/book3s_32_sr.S index e3ab9df6cf19..1ce13e3ab072 100644 --- a/arch/powerpc/kvm/book3s_32_sr.S +++ b/arch/powerpc/kvm/book3s_32_sr.S @@ -122,11 +122,27 @@
/* 0x0 - 0xb */
- /* 'current->mm' needs to be in r4 */ - tophys(r4, r2) - lwz r4, MM(r4) - tophys(r4, r4) - /* This only clobbers r0, r3, r4 and r5 */ + /* switch_mmu_context() clobbers r12, rescue it */ + SET_SCRATCH0(r12) + + /* switch_mmu_context() needs paging, let's enable it */ + mfmsr r9 + ori r11, r9, MSR_DR + mtmsr r11 + sync + + /* Calling switch_mmu_context(<inv>, current->mm, <inv>); */ + lwz r4, MM(r2) bl switch_mmu_context
+ /* Disable paging again */ + mfmsr r9 + li r6, MSR_DR + andc r9, r9, r6 + mtmsr r9 + sync + + /* restore r12 */ + GET_SCRATCH0(r12) + .endm
Le 10/05/2022 à 13:18, Alexander Graf a écrit :
Commit 863771a28e27 ("powerpc/32s: Convert switch_mmu_context() to C") moved the switch_mmu_context() to C. While in principle a good idea, it meant that the function now uses the stack. The stack is not accessible from real mode though.
So to keep calling the function, let's turn on MSR_DR while we call it. That way, all pointer references to the stack are handled virtually.
Is the system ready to handle a DSI in case the stack is not mapped ?
In addition, make sure to save/restore r12 in an SPRG, as it may get clobbered by the C function.
Reported-by: Matt Evans matt@ozlabs.org Fixes: 863771a28e27 ("powerpc/32s: Convert switch_mmu_context() to C")
Oops, sorry for that. I didn't realise that there was other callers to switch_mmu_context() than switch_mm_irqs_off().
Christophe
Signed-off-by: Alexander Graf graf@amazon.com Cc: stable@vger.kernel.org # v5.14+
v1 -> v2:
- Save and restore R12, so that we don't touch volatile registers while calling into C.
arch/powerpc/kvm/book3s_32_sr.S | 26 +++++++++++++++++++++----- 1 file changed, 21 insertions(+), 5 deletions(-)
diff --git a/arch/powerpc/kvm/book3s_32_sr.S b/arch/powerpc/kvm/book3s_32_sr.S index e3ab9df6cf19..1ce13e3ab072 100644 --- a/arch/powerpc/kvm/book3s_32_sr.S +++ b/arch/powerpc/kvm/book3s_32_sr.S @@ -122,11 +122,27 @@ /* 0x0 - 0xb */
- /* 'current->mm' needs to be in r4 */
- tophys(r4, r2)
- lwz r4, MM(r4)
- tophys(r4, r4)
- /* This only clobbers r0, r3, r4 and r5 */
- /* switch_mmu_context() clobbers r12, rescue it */
- SET_SCRATCH0(r12)
- /* switch_mmu_context() needs paging, let's enable it */
- mfmsr r9
- ori r11, r9, MSR_DR
- mtmsr r11
- sync
- /* Calling switch_mmu_context(<inv>, current->mm, <inv>); */
- lwz r4, MM(r2) bl switch_mmu_context
- /* Disable paging again */
- mfmsr r9
- li r6, MSR_DR
- andc r9, r9, r6
- mtmsr r9
- sync
- /* restore r12 */
- GET_SCRATCH0(r12)
- .endm
On 10.05.22 13:31, Christophe Leroy wrote:
Le 10/05/2022 à 13:18, Alexander Graf a écrit :
Commit 863771a28e27 ("powerpc/32s: Convert switch_mmu_context() to C") moved the switch_mmu_context() to C. While in principle a good idea, it meant that the function now uses the stack. The stack is not accessible from real mode though.
So to keep calling the function, let's turn on MSR_DR while we call it. That way, all pointer references to the stack are handled virtually.
Is the system ready to handle a DSI in case the stack is not mapped ?
A DSI itself will be an interrupt again which will in turn destroy the SPRG that we're saving. Guess I was trying to be too smart :). I'll use Matt's original suggestion and just put it on the stack.
In addition, make sure to save/restore r12 in an SPRG, as it may get clobbered by the C function.
Reported-by: Matt Evans matt@ozlabs.org Fixes: 863771a28e27 ("powerpc/32s: Convert switch_mmu_context() to C")
Oops, sorry for that. I didn't realise that there was other callers to switch_mmu_context() than switch_mm_irqs_off().
No worries, the compiled C version looks a lot nicer than the previous asm one - and it was a good way to identify whether there still are users of KVM on Book3S 32bit out there :)
Alex
Amazon Development Center Germany GmbH Krausenstr. 38 10117 Berlin Geschaeftsfuehrung: Christian Schlaeger, Jonathan Weiss Eingetragen am Amtsgericht Charlottenburg unter HRB 149173 B Sitz: Berlin Ust-ID: DE 289 237 879
linux-stable-mirror@lists.linaro.org