The system call FLIH (first-level interrupt handler) at 0xc00
unconditionally sets hardware priority to medium. For hypercalls, this
means we lose guest OS priority. The front end (do_kvm_0x**) to the
KVM interrupt handler always assumes that PPR priority is saved in
PACA exception save area, so it copies this to the kvm_hstate
structure. For hypercalls, this would be the saved priority from any
previous exception. Eventually, the guest gets resumed with an
incorrect priority.
The fix is to save the PPR priority in PACA exception save area before
switching HMT priorities in the FLIH so that existing code described above
in the KVM interrupt handler can copy it from there into the VCPU's saved
context.
Signed-off-by: Suresh Warrier <warrier@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
[mpe: Dropped HMT_MEDIUM_PPR_DISCARD and reworded comment]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
. = 0xc00
.globl system_call_pSeries
system_call_pSeries:
- HMT_MEDIUM
+ /*
+ * If CONFIG_KVM_BOOK3S_64_HANDLER is set, save the PPR (on systems
+ * that support it) before changing to HMT_MEDIUM. That allows the KVM
+ * code to save that value into the guest state (it is the guest's PPR
+ * value). Otherwise just change to HMT_MEDIUM as userspace has
+ * already saved the PPR.
+ */
#ifdef CONFIG_KVM_BOOK3S_64_HANDLER
SET_SCRATCH0(r13)
GET_PACA(r13)
std r9,PACA_EXGEN+EX_R9(r13)
+ OPT_GET_SPR(r9, SPRN_PPR, CPU_FTR_HAS_PPR);
+ HMT_MEDIUM;
std r10,PACA_EXGEN+EX_R10(r13)
+ OPT_SAVE_REG_TO_PACA(PACA_EXGEN+EX_PPR, r9, CPU_FTR_HAS_PPR);
mfcr r9
KVMTEST(0xc00)
GET_SCRATCH0(r13)
+#else
+ HMT_MEDIUM;
#endif
SYSCALL_PSERIES_1
SYSCALL_PSERIES_2_RFID