KVM: PPC: Book3S HV: Fix bug causing host SLB to be restored incorrectly
authorPaul Mackerras <paulus@ozlabs.org>
Tue, 12 Sep 2017 03:47:23 +0000 (13:47 +1000)
committerPaul Mackerras <paulus@ozlabs.org>
Tue, 12 Sep 2017 06:02:46 +0000 (16:02 +1000)
Aneesh Kumar reported seeing host crashes when running recent kernels
on POWER8.  The symptom was an oops like this:

Unable to handle kernel paging request for data at address 0xf00000000786c620
Faulting instruction address: 0xc00000000030e1e4
Oops: Kernel access of bad area, sig: 11 [#1]
LE SMP NR_CPUS=2048 NUMA PowerNV
Modules linked in: powernv_op_panel
CPU: 24 PID: 6663 Comm: qemu-system-ppc Tainted: G        W 4.13.0-rc7-43932-gfc36c59 #2
task: c000000fdeadfe80 task.stack: c000000fdeb68000
NIP:  c00000000030e1e4 LR: c00000000030de6c CTR: c000000000103620
REGS: c000000fdeb6b450 TRAP: 0300   Tainted: G        W        (4.13.0-rc7-43932-gfc36c59)
MSR:  9000000000009033 <SF,HV,EE,ME,IR,DR,RI,LE>  CR: 24044428  XER: 20000000
CFAR: c00000000030e134 DAR: f00000000786c620 DSISR: 40000000 SOFTE: 0
GPR00: 0000000000000000 c000000fdeb6b6d0 c0000000010bd000 000000000000e1b0
GPR04: c00000000115e168 c000001fffa6e4b0 c00000000115d000 c000001e1b180386
GPR08: f000000000000000 c000000f9a8913e0 f00000000786c600 00007fff587d0000
GPR12: c000000fdeb68000 c00000000fb0f000 0000000000000001 00007fff587cffff
GPR16: 0000000000000000 c000000000000000 00000000003fffff c000000fdebfe1f8
GPR20: 0000000000000004 c000000fdeb6b8a8 0000000000000001 0008000000000040
GPR24: 07000000000000c0 00007fff587cffff c000000fdec20bf8 00007fff587d0000
GPR28: c000000fdeca9ac0 00007fff587d0000 00007fff587c0000 00007fff587d0000
NIP [c00000000030e1e4] __get_user_pages_fast+0x434/0x1070
LR [c00000000030de6c] __get_user_pages_fast+0xbc/0x1070
Call Trace:
[c000000fdeb6b6d0] [c00000000139dab8] lock_classes+0x0/0x35fe50 (unreliable)
[c000000fdeb6b7e0] [c00000000030ef38] get_user_pages_fast+0xf8/0x120
[c000000fdeb6b830] [c000000000112318] kvmppc_book3s_hv_page_fault+0x308/0xf30
[c000000fdeb6b960] [c00000000010e10c] kvmppc_vcpu_run_hv+0xfdc/0x1f00
[c000000fdeb6bb20] [c0000000000e915c] kvmppc_vcpu_run+0x2c/0x40
[c000000fdeb6bb40] [c0000000000e5650] kvm_arch_vcpu_ioctl_run+0x110/0x300
[c000000fdeb6bbe0] [c0000000000d6468] kvm_vcpu_ioctl+0x528/0x900
[c000000fdeb6bd40] [c0000000003bc04c] do_vfs_ioctl+0xcc/0x950
[c000000fdeb6bde0] [c0000000003bc930] SyS_ioctl+0x60/0x100
[c000000fdeb6be30] [c00000000000b96c] system_call+0x58/0x6c
Instruction dump:
7ca81a14 2fa50000 41de0010 7cc8182a 68c60002 78c6ffe2 0b060000 3cc2000a
794a3664 390610d8 e9080000 7d485214 <e90a00207d435378 790507e1 408202f0
---[ end trace fad4a342d0414aa2 ]---

It turns out that what has happened is that the SLB entry for the
vmmemap region hasn't been reloaded on exit from a guest, and it has
the wrong page size.  Then, when the host next accesses the vmemmap
region, it gets a page fault.

Commit a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with
KVM", 2017-07-24) modified the guest exit code so that it now only clears
out the SLB for hash guest.  The code tests the radix flag and puts the
result in a non-volatile CR field, CR2, and later branches based on CR2.

Unfortunately, the kvmppc_save_tm function, which gets called between
those two points, modifies all the user-visible registers in the case
where the guest was in transactional or suspended state, except for a
few which it restores (namely r1, r2, r9 and r13).  Thus the hash/radix indication in CR2 gets corrupted.

This fixes the problem by re-doing the comparison just before the
result is needed.  For good measure, this also adds comments next to
the call sites of kvmppc_save_tm and kvmppc_restore_tm pointing out
that non-volatile register state will be lost.

Cc: stable@vger.kernel.org # v4.13
Fixes: a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with KVM")
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
arch/powerpc/kvm/book3s_hv_rmhandlers.S

index 663a4a861e7f3600903c61df65240fe32e58e5d3..17936f82d3c787c38327f41186f634d043d843a9 100644 (file)
@@ -771,6 +771,9 @@ END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
 
 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
 BEGIN_FTR_SECTION
+       /*
+        * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS INCLUDING CR
+        */
        bl      kvmppc_restore_tm
 END_FTR_SECTION_IFSET(CPU_FTR_TM)
 #endif
@@ -1630,6 +1633,9 @@ END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
 
 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
 BEGIN_FTR_SECTION
+       /*
+        * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS INCLUDING CR
+        */
        bl      kvmppc_save_tm
 END_FTR_SECTION_IFSET(CPU_FTR_TM)
 #endif
@@ -1749,7 +1755,10 @@ END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
        /*
         * Are we running hash or radix ?
         */
-       beq     cr2,3f
+       ld      r5, VCPU_KVM(r9)
+       lbz     r0, KVM_RADIX(r5)
+       cmpwi   cr2, r0, 0
+       beq     cr2, 3f
 
        /* Radix: Handle the case where the guest used an illegal PID */
        LOAD_REG_ADDR(r4, mmu_base_pid)
@@ -2466,6 +2475,9 @@ _GLOBAL(kvmppc_h_cede)            /* r3 = vcpu pointer, r11 = msr, r13 = paca */
 
 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
 BEGIN_FTR_SECTION
+       /*
+        * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS INCLUDING CR
+        */
        ld      r9, HSTATE_KVM_VCPU(r13)
        bl      kvmppc_save_tm
 END_FTR_SECTION_IFSET(CPU_FTR_TM)
@@ -2578,6 +2590,9 @@ kvm_end_cede:
 
 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
 BEGIN_FTR_SECTION
+       /*
+        * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS INCLUDING CR
+        */
        bl      kvmppc_restore_tm
 END_FTR_SECTION_IFSET(CPU_FTR_TM)
 #endif