commit <249fad734a25> ""powerpc/perf: Disable trace_imc pmu"
disables IMC(In-Memory Collection) trace-mode in kernel, since frequent
mode switching between accumulation mode and trace mode via the spr LDBAR
in the hardware can trigger a checkstop(system crash).
Patch to re-enable imc-trace mode in kernel.
The previous patch(1/2) in this series will address the mode switching issue
by implementing a global lock, and will restrict the usage of
accumulation and trace-mode at a time.
Signed-off-by: Anju T Sudhakar <anju@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200313055238.8656-2-anju@linux.vnet.ibm.com
IMC(In-memory Collection Counters) does performance monitoring in
two different modes, i.e accumulation mode(core-imc and thread-imc events),
and trace mode(trace-imc events). A cpu thread can either be in
accumulation-mode or trace-mode at a time and this is done via the LDBAR
register in POWER architecture. The current design does not address the
races between thread-imc and trace-imc events.
Patch implements a global id and lock to avoid the races between
core, trace and thread imc events. With this global id-lock
implementation, the system can either run core, thread or trace imc
events at a time. i.e. to run any core-imc events, thread/trace imc events
should not be enabled/monitored.
Signed-off-by: Anju T Sudhakar <anju@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200313055238.8656-1-anju@linux.vnet.ibm.com
MCE handling on pSeries platform fails as recent rework to use common
code for pSeries and PowerNV in machine check error handling tries to
access per-cpu variables in realmode. The per-cpu variables may be
outside the RMO region on pSeries platform and needs translation to be
enabled for access. Just moving these per-cpu variable into RMO region
did'nt help because we queue some work to workqueues in real mode, which
again tries to touch per-cpu variables. Also fwnmi_release_errinfo()
cannot be called when translation is not enabled.
This patch fixes this by enabling translation in the exception handler
when all required real mode handling is done. This change only affects
the pSeries platform.
Without this fix below kernel crash is seen on injecting
SLB multihit:
BUG: Unable to handle kernel data access on read at 0xc00000027b205950
Faulting instruction address: 0xc00000000003b7e0
Oops: Kernel access of bad area, sig: 11 [#1]
LE PAGE_SIZE=64K MMU=Hash SMP NR_CPUS=2048 NUMA pSeries
Modules linked in: mcetest_slb(OE+) af_packet(E) xt_tcpudp(E) ip6t_rpfilter(E) ip6t_REJECT(E) ipt_REJECT(E) xt_conntrack(E) ip_set(E) nfnetlink(E) ebtable_nat(E) ebtable_broute(E) ip6table_nat(E) ip6table_mangle(E) ip6table_raw(E) ip6table_security(E) iptable_nat(E) nf_nat(E) nf_conntrack(E) nf_defrag_ipv6(E) nf_defrag_ipv4(E) iptable_mangle(E) iptable_raw(E) iptable_security(E) ebtable_filter(E) ebtables(E) ip6table_filter(E) ip6_tables(E) iptable_filter(E) ip_tables(E) x_tables(E) xfs(E) ibmveth(E) vmx_crypto(E) gf128mul(E) uio_pdrv_genirq(E) uio(E) crct10dif_vpmsum(E) rtc_generic(E) btrfs(E) libcrc32c(E) xor(E) zstd_decompress(E) zstd_compress(E) raid6_pq(E) sr_mod(E) sd_mod(E) cdrom(E) ibmvscsi(E) scsi_transport_srp(E) crc32c_vpmsum(E) dm_mod(E) sg(E) scsi_mod(E)
CPU: 34 PID: 8154 Comm: insmod Kdump: loaded Tainted: G OE 5.5.0-mahesh #1
NIP: c00000000003b7e0 LR: c0000000000f2218 CTR: 0000000000000000
REGS: c000000007dcb960 TRAP: 0300 Tainted: G OE (5.5.0-mahesh)
MSR: 8000000000001003 <SF,ME,RI,LE> CR: 28002428 XER: 20040000
CFAR: c0000000000f2214 DAR: c00000027b205950 DSISR: 40000000 IRQMASK: 0
GPR00: c0000000000f2218 c000000007dcbbf0 c000000001544800 c000000007dcbd70
GPR04: 0000000000000001 c000000007dcbc98 c008000000d00258 c0080000011c0000
GPR08: 0000000000000000 0000000300000003 c000000001035950 0000000003000048
GPR12: 000000027a1d0000 c000000007f9c000 0000000000000558 0000000000000000
GPR16: 0000000000000540 c008000001110000 c008000001110540 0000000000000000
GPR20: c00000000022af10 c00000025480fd70 c008000001280000 c00000004bfbb300
GPR24: c000000001442330 c00800000800000d c008000008000000 4009287a77000510
GPR28: 0000000000000000 0000000000000002 c000000001033d30 0000000000000001
NIP [c00000000003b7e0] save_mce_event+0x30/0x240
LR [c0000000000f2218] pseries_machine_check_realmode+0x2c8/0x4f0
Call Trace:
Instruction dump:
3c4c0151 38429050 7c0802a6 60000000 fbc1fff0 fbe1fff8 f821ffd1 3d42ffaf
3fc2ffaf e98d0030 394a1150 3bdef530 <7d6a62aa> 1d2b0048 2f8b0063 380b0001
---[ end trace 46fd63f36bbdd940 ]---
Fixes: 9ca766f989 ("powerpc/64s/pseries: machine check convert to use common event code")
Reviewed-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Ganesh Goudar <ganeshgr@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200320110119.10207-1-ganeshgr@linux.ibm.com
Commit 3282a3da25 ("powerpc/64: Implement soft interrupt replay in C")
broke the doorbell wakeup optimisation introduced by commit a9af97aa0a
("powerpc/64s: msgclr when handling doorbell exceptions from system
reset").
This patch restores the msgclr, in C code. It's now done in the system
reset wakeup path rather than doorbell interrupt replay where it used
to be, because it is always the right thing to do in the wakeup case,
but it may be rarely of use in other interrupt replay situations in
which case it's wasted work - we would have to run measurements to see
if that was a worthwhile optimisation, and I suspect it would not be.
The results are similar to those in the original commit, test on POWER8
of context_switch selftests benchmark with polling idle disabled (e.g.,
always nap, giving cross-CPU IPIs) gives the following results:
broken patched
Different threads, same core: 317k/s 375k/s +18.7%
Different cores: 280k/s 282k/s +1.0%
Fixes: 3282a3da25 ("powerpc/64: Implement soft interrupt replay in C")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200402121212.1118218-1-npiggin@gmail.com
Pull hmm updates from Jason Gunthorpe:
"This series focuses on corner case bug fixes and general clarity
improvements to hmm_range_fault(). It arose from a review of
hmm_range_fault() by Christoph, Ralph and myself.
hmm_range_fault() is being used by these 'SVM' style drivers to
non-destructively read the page tables. It is very similar to
get_user_pages() except that the output is an array of PFNs and
per-pfn flags, and it has various modes of reading.
This is necessary before RDMA ODP can be converted, as we don't want
to have weird corner case regressions, which is still a looking
forward item. Ralph has a nice tester for this routine, but it is
waiting for feedback from the selftests maintainers.
Summary:
- 9 bug fixes
- Allow pgmap to track the 'owner' of a DEVICE_PRIVATE - in this case
the owner tells the driver if it can understand the DEVICE_PRIVATE
page or not. Use this to resolve a bug in nouveau where it could
touch DEVICE_PRIVATE pages from other drivers.
- Remove a bunch of dead, redundant or unused code and flags
- Clarity improvements to hmm_range_fault()"
* tag 'for-linus-hmm' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (25 commits)
mm/hmm: return error for non-vma snapshots
mm/hmm: do not set pfns when returning an error code
mm/hmm: do not unconditionally set pfns when returning EBUSY
mm/hmm: use device_private_entry_to_pfn()
mm/hmm: remove HMM_FAULT_SNAPSHOT
mm/hmm: remove unused code and tidy comments
mm/hmm: return the fault type from hmm_pte_need_fault()
mm/hmm: remove pgmap checking for devmap pages
mm/hmm: check the device private page owner in hmm_range_fault()
mm: simplify device private page handling in hmm_range_fault
mm: handle multiple owners of device private pages in migrate_vma
memremap: add an owner field to struct dev_pagemap
mm: merge hmm_vma_do_fault into into hmm_vma_walk_hole_
mm/hmm: don't handle the non-fault case in hmm_vma_walk_hole_()
mm/hmm: simplify hmm_vma_walk_hugetlb_entry()
mm/hmm: remove the unused HMM_FAULT_ALLOW_RETRY flag
mm/hmm: don't provide a stub for hmm_range_fault()
mm/hmm: do not check pmd_protnone twice in hmm_vma_handle_pmd()
mm/hmm: add missing call to hmm_pte_need_fault in HMM_PFN_SPECIAL handling
mm/hmm: return -EFAULT when setting HMM_PFN_ERROR on requested valid pages
...
Currently, CONFIG_VHOST depends on CONFIG_VIRTUALIZATION. But vhost is
not necessarily for VM since it's a generic userspace and kernel
communication protocol. Such dependency may prevent archs without
virtualization support from using vhost.
To solve this, a dedicated vhost menu is created under drivers so
CONIFG_VHOST can be decoupled out of CONFIG_VIRTUALIZATION.
While at it, also squash Kconfig.vringh into vhost Kconfig file. This
avoids the trick of conditional inclusion from VOP or CAIF. Then it
will be easier to introduce new vringh users and common dependency for
both vringh and vhost.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Link: https://lore.kernel.org/r/20200326140125.19794-2-jasowang@redhat.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Declaring setjmp()/longjmp() as taking longs makes the signature
non-standard, and makes clang complain. In the past, this has been
worked around by adding -ffreestanding to the compile flags.
The implementation looks like it only ever propagates the value
(in longjmp) or sets it to 1 (in setjmp), and we only call longjmp
with integer parameters.
This allows removing -ffreestanding from the compilation flags.
Fixes: c9029ef9c9 ("powerpc: Avoid clang warnings around setjmp and longjmp")
Cc: stable@vger.kernel.org # v4.14+
Signed-off-by: Clement Courbet <courbet@google.com>
Reviewed-by: Nathan Chancellor <natechancellor@gmail.com>
Tested-by: Nathan Chancellor <natechancellor@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200330080400.124803-1-courbet@google.com
Relocatable kernel builds produce a warning about .gnu.hash being an
orphan section:
ld: warning: orphan section `.gnu.hash' from `linker stubs' being placed in section `.gnu.hash'
If we try to discard it the build fails:
ld -EL -m elf64lppc -pie --orphan-handling=warn --build-id -o
.tmp_vmlinux1 -T ./arch/powerpc/kernel/vmlinux.lds --whole-archive
arch/powerpc/kernel/head_64.o arch/powerpc/kernel/entry_64.o
...
sound/built-in.a net/built-in.a virt/built-in.a --no-whole-archive
--start-group lib/lib.a --end-group
ld: could not find section .gnu.hash
So add an entry to explicitly retain it, as we do for .hash.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200227045933.22967-1-mpe@ellerman.id.au
Regular interrupt return restores NVGPRS whereas lite returns do not.
This is clumsy: most interrupts can return without restoring NVGPRS in
most of the time, but there are special cases that require it (when
registers have been modified by the kernel). So change interrupt
return to not restore NVGPRS, and have interrupt handlers restore them
explicitly in the cases that requires it.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200225173541.1549955-30-npiggin@gmail.com
Implement the bulk of interrupt return logic in C. The asm return code
must handle a few cases: restoring full GPRs, and emulating stack
store.
The stack store emulation is significantly simplfied, rather than
creating a new return frame and switching to that before performing
the store, it uses the PACA to keep a scratch register around to
perform the store.
The asm return code is moved into 64e for now. The new logic has made
allowance for 64e, but I don't have a full environment that works well
to test it, and even booting in emulated qemu is not great for stress
testing. 64e shouldn't be too far off working with this, given a bit
more testing and auditing of the logic.
This is slightly faster on a POWER9 (page fault speed increases about
1.1%), probably due to reduced mtmsrd.
mpe: Includes fixes from Nick for _TIF_EMULATE_STACK_STORE
handling (including the fast_interrupt_return path), to remove
trace_hardirqs_on(), and fixes the interrupt-return part of the
MSR_VSX restore bug caught by tm-unavailable selftest.
mpe: Incorporate fix from Nick:
The return-to-kernel path has to replay any soft-pending interrupts if
it is returning to a context that had interrupts soft-enabled. It has
to do this carefully and avoid plain enabling interrupts if this is an
irq context, which can cause multiple nesting of interrupts on the
stack, and other unexpected issues.
The code which avoided this case got the soft-mask state wrong, and
marked interrupts as enabled before going around again to retry. This
seems to be mostly harmless except when PREEMPT=y, this calls
preempt_schedule_irq with irqs apparently enabled and runs into a BUG
in kernel/sched/core.c
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michal Suchanek <msuchanek@suse.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200225173541.1549955-29-npiggin@gmail.com
When local_irq_enable() finds a pending soft-masked interrupt, it
"replays" it by setting up registers like the initial interrupt entry,
then calls into the low level handler to set up an interrupt stack
frame and process the interrupt.
This is not necessary, and uses more stack than needed. The high level
interrupt handler can be called directly from C, with just pt_regs set
up on stack. This should be faster and use less stack.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200225173541.1549955-28-npiggin@gmail.com
System call entry and particularly exit code is beyond the limit of
what is reasonable to implement in asm.
This conversion moves all conditional branches out of the asm code,
except for the case that all GPRs should be restored at exit.
Null syscall test is about 5% faster after this patch, because the
exit work is handled under local_irq_disable, and the hard mask and
pending interrupt replay is handled after that, which avoids games
with MSR.
mpe: Includes subsequent fixes from Nick:
This fixes 4 issues caught by TM selftests. First was a tm-syscall bug
that hit due to tabort_syscall being called after interrupts were
reconciled (in a subsequent patch), which led to interrupts being
enabled before tabort_syscall was called. Rather than going through an
un-reconciling interrupts for the return, I just go back to putting
the test early in asm, the C-ification of that wasn't a big win
anyway.
Second is the syscall return _TIF_USER_WORK_MASK check would go into
an infinite loop if _TIF_RESTORE_TM became set. The asm code uses
_TIF_USER_WORK_MASK to brach to slowpath which includes
restore_tm_state.
Third is system call return was not calling restore_tm_state, I missed
this completely (alhtough it's in the return from interrupt C
conversion because when the asm syscall code encountered problems it
would branch to the interrupt return code.
Fourth is MSR_VEC missing from restore_math, which was caught by
tm-unavailable selftest taking an unexpected facility unavailable
interrupt when testing VSX unavailble exception with MSR.FP=1
MSR.VEC=1. Fourth case also has a fixup in a subsequent patch.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michal Suchanek <msuchanek@suse.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200225173541.1549955-26-npiggin@gmail.com
powerpc has an optimisation where interrupts avoid saving the
non-volatile (or callee saved) registers to the interrupt stack frame
if they are not required.
Two problems with this are that an interrupt does not always know
whether it will need non-volatiles; and if it does need them, they can
only be saved from the entry-scoped asm code (because we don't control
what the C compiler does with these registers).
system calls are the most difficult: some system calls always require
all registers (e.g., fork, to copy regs into the child). Sometimes
registers are only required under certain conditions (e.g., tracing,
signal delivery). These cases require ugly logic in the call
chains (e.g., ppc_fork), and require a lot of logic to be implemented
in asm.
So remove the optimisation for system calls, and always save NVGPRs on
entry. Modern high performance CPUs are not so sensitive, because the
stores are dense in cache and can be hidden by other expensive work in
the syscall path -- the null syscall selftests benchmark on POWER9 is
not slowed (124.40ns before and 123.64ns after, i.e., within the
noise).
Other interrupts retain the NVGPR optimisation for now.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200225173541.1549955-24-npiggin@gmail.com
The soft NMI handler does not reconcile interrupt state, so it should
not return via the normal ret_from_except path. Return like other NMIs,
using the EXCEPTION_RESTORE_REGS macro.
This becomes important when the scv interrupt is implemented, which
must handle soft-masked interrupts that have r13 set to something
other than the PACA -- returning to kernel in this case must restore
r13.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200225173541.1549955-23-npiggin@gmail.com
This adds IRQ_HARD_DIS to irq_happened. Although it doesn't seem to
matter much because we're not allowed to enable irqs in an NMI
handler, the soft-irq debugging code is becoming more strict about
ensuring IRQ_HARD_DIS is in sync with MSR[EE], this may help avoid
asserts or other issues.
Add a comment explaining why MCE does not have this. Early machine
check is generally much smaller and more contained code which will
explode if you look at it wrong anyway as it runs in real mode, though
there's an argument that we should do similar reconciling for the MCE
as well.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200225173541.1549955-22-npiggin@gmail.com
Apart from SRESET, MCE, and syscall (hcall variant), the SRR type
interrupts are not escalated to hypervisor mode, so are delivered to
the OS.
When running PR KVM, the OS is the hypervisor, and the guest runs with
MSR[PR]=1 (ie. usermode), so these interrupts must test if a guest was
running when interrupted. These tests are required at the real-mode
entry points because the PR KVM host runs with LPCR[AIL]=0.
In HV KVM and nested HV KVM, the guest always receives these
interrupts, so there is no need for the host to make this test. So
remove the tests if PR KVM is not configured.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200225173541.1549955-21-npiggin@gmail.com
The hdec interrupt handler is reported to sometimes fire in Linux if
KVM leaves it pending after a guest exists. This is harmless, so there
is a no-op handler for it.
The interrupt handler currently uses the regular kernel stack. Change
this to avoid touching the stack entirely.
This should be the last place where the regular Linux stack can be
accessed with asynchronous interrupts (including PMI) soft-masked.
It might be possible to take advantage of this invariant, e.g., to
context switch the kernel stack SLB entry without clearing MSR[EE].
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200225173541.1549955-17-npiggin@gmail.com
Replace IEARLY=1 and IEARLY=2 with IBRANCH_COMMON, which controls if
the entry code branches to a common handler; and IREALMODE_COMMON,
which controls whether the common handler should remain in real mode.
These special cases no longer avoid loading the SRR registers, there
is no point as most of them load the registers immediately anyway.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200225173541.1549955-14-npiggin@gmail.com
This allows more code to be moved out of unrelocated regions. The
system call KVMTEST is changed to be open-coded and remain in the
tramp area to avoid having to move it to entry_64.S. The custom nature
of the system call entry code means the hcall case can be made more
streamlined than regular interrupt handlers.
mpe: Incorporate fix from Nick:
Moving KVM test to the common entry code missed the case of HMI and
MCE, which do not do __GEN_COMMON_ENTRY (because they don't want to
switch to virt mode).
This means a MCE or HMI exception that is taken while KVM is running a
guest context will not be switched out of that context, and KVM won't
be notified. Found by running sigfuz in guest with patched host on
POWER9 DD2.3, which causes some TM related HMI interrupts (which are
expected and supposed to be handled by KVM).
This fix adds a __GEN_REALMODE_COMMON_ENTRY for those handlers to add
the KVM test. This makes them look a little more like other handlers
that all use __GEN_COMMON_ENTRY.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200225173541.1549955-13-npiggin@gmail.com
The real mode interrupt entry points currently use rfid to branch to
the common handler in virtual mode. This is a significant amount of
code, and forces other code (notably the KVM test) to live in the
real mode handler.
In the interest of minimising the amount of code that runs unrelocated
move the switch to virt mode into the common code, and do it with
mtmsrd, which avoids clobbering SRRs (although the post-KVMTEST
performance of real-mode interrupt handlers is not a big concern these
days).
This requires CTR to always be saved (real-mode needs to reach 0xc...)
but that's not a huge impact these days. It could be optimized away in
future.
mpe: Incorporate fix from Nick:
It's possible for interrupts to be replayed when TM is enabled and
suspended, for example rt_sigreturn, where the mtmsrd MSR_KERNEL in
the real-mode entry point to the common handler causes a TM Bad Thing
exception (due to attempting to clear suspended).
The fix for this is to have replay interrupts go to the _virt entry
point and skip the mtmsrd, which matches what happens before this
patch.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200225173541.1549955-11-npiggin@gmail.com
