Stream IDs are reused across multiple BackEnd mixers, do not reset the
stream mixers if they are not already set for that particular FrontEnd.
Ex:
amixer cset iface=MIXER,name='SLIMBUS_0_RX Audio Mixer MultiMedia1' 1
would set the MultiMedia1 steam for SLIMBUS_0_RX, however doing below
command will reset previously setup MultiMedia1 stream, because both of them
are using MultiMedia1 PCM stream.
amixer cset iface=MIXER,name='SLIMBUS_2_RX Audio Mixer MultiMedia1' 0
reset the FrontEnd Mixers conditionally to fix this issue.
This is more noticeable in desktop setup, where in alsactl tries to restore
the alsa state and overwriting the previous mixer settings.
Fixes: e3a33673e8 ("ASoC: qdsp6: q6routing: Add q6routing driver")
Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org>
Link: https://lore.kernel.org/r/20211116114721.12517-3-srinivas.kandagatla@linaro.org
Signed-off-by: Mark Brown <broonie@kernel.org>
Tony Nguyen says:
====================
Intel Wired LAN Driver Updates 2021-11-15
This series contains updates to iavf driver only.
Mateusz adds a wait for reset completion when changing queue count which
could otherwise cause issues with VF reset.
Nick adds a null check for vf_res in iavf_fix_features(), corrects
ordering of function calls to resolve dependency issues, and prevents
possible freeing of a lock which isn't being held.
Piotr fixes logic that did not allow setting all multicast mode without
promiscuous mode.
Jake prevents possible accidental freeing of filter structure.
Mitch adds null checks for key and indir parameters in iavf_get_rxfh().
Surabhi adds an additional check that would, previously, cause the driver
to print a false error due to values obtained while the VF is in reset.
Grzegorz prevents a queue request of 0 which would cause queue count to
reset to default values.
Akeem restores VLAN filters when bringing the interface back up.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
In vcpu_load_eoi_exitmap(), currently the eoi_exit_bitmap[4] array is
initialized only when Hyper-V context is available, in other path it is
just passed to kvm_x86_ops.load_eoi_exitmap() directly from on the stack,
which would cause unexpected interrupt delivery/handling issues, e.g. an
*old* linux kernel that relies on PIT to do clock calibration on KVM might
randomly fail to boot.
Fix it by passing ioapic_handled_vectors to load_eoi_exitmap() when Hyper-V
context is not available.
Fixes: f2bc14b69c ("KVM: x86: hyper-v: Prepare to meet unallocated Hyper-V context")
Cc: stable@vger.kernel.org
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Huang Le <huangle1@jd.com>
Message-Id: <62115b277dab49ea97da5633f8522daf@jd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Thread creation requires taking the mmap_sem in write mode, which causes
vCPU threads running in guest mode to block while they are populating
memory. Fix this by waiting for all vCPU threads to be created and start
running before entering guest mode on any one vCPU thread.
This substantially improves the "Populate memory time" when using 1GiB
pages since it allows all vCPUs to zero pages in parallel rather than
blocking because a writer is waiting (which is waiting for another vCPU
that is busy zeroing a 1GiB page).
Before:
$ ./dirty_log_perf_test -v256 -s anonymous_hugetlb_1gb
...
Populate memory time: 52.811184013s
After:
$ ./dirty_log_perf_test -v256 -s anonymous_hugetlb_1gb
...
Populate memory time: 10.204573342s
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20211111001257.1446428-4-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move vCPU thread creation and joining to common helper functions. This
is in preparation for the next commit which ensures that all vCPU
threads are fully created before entering guest mode on any one
vCPU.
No functional change intended.
Signed-off-by: David Matlack <dmatlack@google.com>
Reviewed-by: Ben Gardon <bgardon@google.com>
Message-Id: <20211111001257.1446428-3-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Start at iteration 0 instead of -1 to avoid having to initialize
vcpu_last_completed_iteration when setting up vCPU threads. This
simplifies the next commit where we move vCPU thread initialization
out to a common helper.
No functional change intended.
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20211111001257.1446428-2-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Copy perf_test_args to the guest during VM creation instead of relying on
the caller to do so at their leisure. Ideally, tests wouldn't even be
able to modify perf_test_args, i.e. they would have no motivation to do
the sync, but enforcing that is arguably a net negative for readability.
No functional change intended.
[Set wr_fract=1 by default and add helper to override it since the new
access_tracking_perf_test needs to set it dynamically.]
Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Reviewed-by: Ben Gardon <bgardon@google.com>
Message-Id: <20211111000310.1435032-13-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Fill the per-vCPU args when creating the perf_test VM instead of having
the caller do so. This helps ensure that any adjustments to the number
of pages (and thus vcpu_memory_bytes) are reflected in the per-VM args.
Automatically filling the per-vCPU args will also allow a future patch
to do the sync to the guest during creation.
Signed-off-by: Sean Christopherson <seanjc@google.com>
[Updated access_tracking_perf_test as well.]
Signed-off-by: David Matlack <dmatlack@google.com>
Reviewed-by: Ben Gardon <bgardon@google.com>
Message-Id: <20211111000310.1435032-12-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Remove perf_test_args.host_page_size and instead use getpagesize() so
that it's somewhat obvious that, for tests that care about the host page
size, they care about the system page size, not the hardware page size,
e.g. that the logic is unchanged if hugepages are in play.
No functional change intended.
Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Ben Gardon <bgardon@google.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20211111000310.1435032-10-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move the per-VM GPA into perf_test_args instead of storing it as a
separate global variable. It's not obvious that guest_test_phys_mem
holds a GPA, nor that it's connected/coupled with per_vcpu->gpa.
No functional change intended.
Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Ben Gardon <bgardon@google.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20211111000310.1435032-9-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Grab the per-vCPU GPA and number of pages from perf_util in the demand
paging test instead of duplicating perf_util's calculations.
Note, this may or may not result in a functional change. It's not clear
that the test's calculations are guaranteed to yield the same value as
perf_util, e.g. if guest_percpu_mem_size != vcpu_args->pages.
Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Ben Gardon <bgardon@google.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20211111000310.1435032-8-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Assert that the GPA for a memslot backed by a hugepage is aligned to
the hugepage size and fix perf_test_util accordingly. Lack of GPA
alignment prevents KVM from backing the guest with hugepages, e.g. x86's
write-protection of hugepages when dirty logging is activated is
otherwise not exercised.
Add a comment explaining that guest_page_size is for non-huge pages to
try and avoid confusion about what it actually tracks.
Cc: Ben Gardon <bgardon@google.com>
Cc: Yanan Wang <wangyanan55@huawei.com>
Cc: Andrew Jones <drjones@redhat.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Aaron Lewis <aaronlewis@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
[Used get_backing_src_pagesz() to determine alignment dynamically.]
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20211111000310.1435032-5-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Manually padding and aligning the mmap region is only needed when using
THP. When using HugeTLB, mmap will always return an address aligned to
the HugeTLB page size. Add a comment to clarify this and assert the mmap
behavior for HugeTLB.
[Removed requirement that HugeTLB mmaps must be padded per Yanan's
feedback and added assertion that mmap returns aligned addresses
when using HugeTLB.]
Cc: Ben Gardon <bgardon@google.com>
Cc: Yanan Wang <wangyanan55@huawei.com>
Cc: Andrew Jones <drjones@redhat.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Aaron Lewis <aaronlewis@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20211111000310.1435032-4-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Refactor align() to work with non-pointers and split into separate
helpers for aligning up vs. down. Add align_ptr_up() for use with
pointers. Expose all helpers so that they can be used by tests and/or
other utilities. The align_down() helper in particular will be used to
ensure gpa alignment for hugepages.
No functional change intended.
[Added sepearate up/down helpers and replaced open-coded alignment
bit math throughout the KVM selftests.]
Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Reviewed-by: Ben Gardon <bgardon@google.com>
Message-Id: <20211111000310.1435032-3-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
When I first looked at this, there was no support for guest exception
handling in the KVM selftests. In fact it was merged into 5.10 before
the Xen support got merged in 5.11, and I could have used it from the
start.
Hook it up now, to exercise the Xen upcall delivery. I'm about to make
things a bit more interesting by handling the full 2level event channel
stuff in-kernel on top of the basic vector injection that we already
have, and I'll want to build more tests on top.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Message-Id: <20211115165030.7422-3-dwmw2@infradead.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
When CONFIG_VMAP_STACK is disabled, the user can enable CONFIG_STACK_CHECK,
which adds a stack overflow check to each C function in the kernel. This is
also done for functions in the vdso page. These functions are run in user
context and user stack sizes are usually different to what the kernel uses.
This might trigger the stack check although the stack size is valid.
Therefore filter the -mstack-guard and -mstack-size flags when compiling
vdso C files.
Cc: stable@kernel.org # 5.10+
Fixes: 4bff8cb545 ("s390: convert to GENERIC_VDSO")
Reported-by: Janosch Frank <frankja@linux.ibm.com>
Signed-off-by: Sven Schnelle <svens@linux.ibm.com>
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
The -nostdlib option requests the compiler to not use the standard
system startup files or libraries when linking. It is effective only
when $(CC) is used as a linker driver.
Since commit 2b2a25845d ("s390/vdso: Use $(LD) instead of $(CC) to
link vDSO"), $(LD) is directly used, hence -nostdlib is unneeded.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Link: https://lore.kernel.org/r/20211107162111.323701-1-masahiroy@kernel.org
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
show() must not use snprintf() when formatting the value to be
returned to user space.
Fix the coccicheck warnings:
WARNING: use scnprintf or sprintf.
Use sysfs_emit instead of scnprintf or sprintf makes more sense.
Signed-off-by: Qing Wang <wangqing@vivo.com>
Acked-by: Vineeth Vijayan <vneethv@linux.ibm.com>
Acked-by: Stefan Haberland <sth@linux.ibm.com>
Link: https://lore.kernel.org/r/1634280655-4908-1-git-send-email-wangqing@vivo.com
[hca@linux.ibm.com: fix indentation]
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Initial KASAN shadow memory range was picked to preserve original kernel
modules area position. With protected execution support, which might
impose addressing limitation on vmalloc area and hence affect modules
area position, current fixed KASAN shadow memory range is only making
kernel memory layout setup more complex. So move it to the very end of
available virtual space and simplify calculations.
At the same time return to previous kernel address space split. In
particular commit 0c4f2623b9 ("s390: setup kernel memory layout
early") introduced precise identity map size calculation and keeping
vmemmap left most starting from a fresh region table entry. This didn't
take into account additional mapping region requirement for potential
DCSS mapping above available physical memory. So go back to virtual
space split between 1:1 mapping & vmemmap array once vmalloc area size
is subtracted.
Cc: stable@vger.kernel.org
Fixes: 0c4f2623b9 ("s390: setup kernel memory layout early")
Reported-by: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
- Avoid using ULONG_MAX in memblock_remove, it has no functional change
but makes memblock_dbg output a range which makes sense.
- Actually finish memblock memory setup before doing amode31/cr/uv
setup.
- Move memblock_dump_all() debug output after memblock memory setup is
complete. This gives us final "memory" regions if they were trimmed
due to addressing limits and still "physmem" regions as original info
which came from mem_detect.
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
There is a difference in how architectures treat "mem=" option. For some
that is an amount of online memory, for s390 and x86 this is the limiting
max address. Some memblock api like memblock_enforce_memory_limit()
take limit argument and explicitly treat it as the size of online memory,
and use __find_max_addr to convert it to an actual max address. Current
s390 usage:
memblock_enforce_memory_limit(memblock_end_of_DRAM());
yields different results depending on presence of memory holes (offline
memory blocks in between online memory). If there are no memory holes
limit == max_addr in memblock_enforce_memory_limit() and it does trim
online memory and reserved memory regions. With memory holes present it
actually does nothing.
Since we already use memblock_remove() explicitly to trim online memory
regions to potential limit (think mem=, kdump, addressing limits, etc.)
drop the usage of memblock_enforce_memory_limit() altogether. Trimming
reserved regions should not be required, since we now use
memblock_set_current_limit() to limit allocations and any explicit memory
reservations above the limit is an actual problem we should not hide.
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Such reserved memory region, if not cleaned up later causes problems when
memblock_free_all() is called to release free pages to the buddy allocator
and those reserved regions are carried over to reserve_bootmem_region()
which marks the pages as PageReserved.
Instead use memblock_set_current_limit() to make sure memblock allocations
do not go over identity mapping (which could happen when "mem=" option
is used or during kdump).
Cc: stable@vger.kernel.org
Fixes: 73045a08cf ("s390: unify identity mapping limits handling")
Reported-by: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
As spotted and explained in commit c12ab8dbc4 ("powerpc/8xx: Fix
Oops with STRICT_KERNEL_RWX without DEBUG_RODATA_TEST"), the selection
of STRICT_KERNEL_RWX without selecting DEBUG_RODATA_TEST has spotted
the lack of the DIRTY bit in the pinned kernel data TLBs.
This problem should have been detected a lot earlier if things had
been working as expected. But due to an incredible level of chance or
mishap, this went undetected because of a set of bugs: In fact the
DTLBs were not pinned, because instead of setting the reserve bit
in MD_CTR, it was set in MI_CTR that is the register for ITLBs.
But then, another huge bug was there: the physical address was
reset to 0 at the boundary between RO and RW areas, leading to the
same physical space being mapped at both 0xc0000000 and 0xc8000000.
This had by miracle no consequence until now because the entry was
not really pinned so it was overwritten soon enough to go undetected.
Of course, now that we really pin the DTLBs, it must be fixed as well.
Fixes: f76c8f6d25 ("powerpc/8xx: Add function to set pinned TLBs")
Cc: stable@vger.kernel.org # v5.8+
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Depends-on: c12ab8dbc4 ("powerpc/8xx: Fix Oops with STRICT_KERNEL_RWX without DEBUG_RODATA_TEST")
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/a21e9a057fe2d247a535aff0d157a54eefee017a.1636963688.git.christophe.leroy@csgroup.eu
The conversion from __copy_from_user() to __get_user() by
commit d3ccc97815 ("powerpc/signal: Use __get_user() to copy
sigset_t") introduced a regression in __get_user_sigset() for
powerpc/32. The bug was subsequently moved into
unsafe_get_user_sigset().
The bug is due to the copied 64 bit value being truncated to
32 bits while being assigned to dst->sig[0]
The regression was reported by users of the Xorg packages distributed in
Debian/powerpc --
"The symptoms are that the fb screen goes blank, with the backlight
remaining on and no errors logged in /var/log; wdm (or startx) run
with no effect (I tried logging in in the blind, with no effect).
And they are hard to kill, requiring 'kill -KILL ...'"
Fix the regression by copying each word of the sigset, not only the
first one.
__get_user_sigset() was tentatively optimised to copy 64 bits at once
in order to minimise KUAP unlock/lock impact, but the unsafe variant
doesn't suffer that, so it can just copy words.
Fixes: 887f3ceb51 ("powerpc/signal32: Convert do_setcontext[_tm]() to user access block")
Cc: stable@vger.kernel.org # v5.13+
Reported-by: Finn Thain <fthain@linux-m68k.org>
Reported-and-tested-by: Stan Johnson <userm57@yahoo.com>
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/99ef38d61c0eb3f79c68942deb0c35995a93a777.1636966353.git.christophe.leroy@csgroup.eu
trans_pgd_create_copy() can hit "VM_BUG_ON(mm != &init_mm)" in the
function pmd_populate_kernel().
This is the combined consequence of commit 5de59884ac ("arm64:
trans_pgd: pass NULL instead of init_mm to *_populate functions"), which
replaced &init_mm with NULL and commit 59511cfd08 ("arm64: mm: use XN
table mapping attributes for user/kernel mappings"), which introduced
the VM_BUG_ON.
Since the former sounds reasonable, it is better to work on the later.
From the perspective of trans_pgd, two groups of functions are
considered in the later one:
pmd_populate_kernel()
mm == NULL should be fixed, else it hits VM_BUG_ON()
p?d_populate()
mm == NULL means PXN, that is OK, since trans_pgd only copies a
linear map, no execution will happen on the map.
So it is good enough to just relax VM_BUG_ON() to disregard mm == NULL
Fixes: 59511cfd08 ("arm64: mm: use XN table mapping attributes for user/kernel mappings")
Signed-off-by: Pingfan Liu <kernelfans@gmail.com>
Cc: <stable@vger.kernel.org> # 5.13.x
Cc: Ard Biesheuvel <ardb@kernel.org>
Cc: James Morse <james.morse@arm.com>
Cc: Matthias Brugger <mbrugger@suse.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Pasha Tatashin <pasha.tatashin@soleen.com>
Link: https://lore.kernel.org/r/20211112052214.9086-1-kernelfans@gmail.com
Signed-off-by: Will Deacon <will@kernel.org>
Do not use "/**" to begin a non-kernel-doc comment.
Fixes these build warnings:
drivers/phy/ti/phy-am654-serdes.c:3: warning: This comment starts with '/**', but isn't a kernel-doc comment. Refer Documentation/doc-guide/kernel-doc.rst
* PCIe SERDES driver for AM654x SoC
drivers/phy/ti/phy-j721e-wiz.c:3: warning: This comment starts with '/**', but isn't a kernel-doc comment. Refer Documentation/doc-guide/kernel-doc.rst
* Wrapper driver for SERDES used in J721E
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Reported-by: kernel test robot <lkp@intel.com>
Cc: Kishon Vijay Abraham I <kishon@ti.com>
Cc: Vinod Koul <vkoul@kernel.org>
Cc: linux-phy@lists.infradead.org
Link: https://lore.kernel.org/r/20211115030559.13994-1-rdunlap@infradead.org
Signed-off-by: Vinod Koul <vkoul@kernel.org>
When WWAN device wake from S3 deep, under thinkpad platform,
WWAN would be disabled. This disable status could be checked
by command 'nmcli r wwan' or 'rfkill list'.
Issue analysis as below:
When host resume from S3 deep, thinkpad_acpi driver would
call hotkey_resume() function. Finnaly, it will use
wan_get_status to check the current status of WWAN device.
During this resume progress, wan_get_status would always
return off even WWAN boot up completely.
In patch V2, Hans said 'sw_state should be unchanged
after a suspend/resume. It's better to drop the
tpacpi_rfk_update_swstate call all together from the
resume path'.
And it's confimed by Lenovo that GWAN is no longer
available from WHL generation because the design does not
match with current pin control.
Signed-off-by: Slark Xiao <slark_xiao@163.com>
Link: https://lore.kernel.org/r/20211108060648.8212-1-slark_xiao@163.com
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
A Lenovo ThinkStation S20 (4157CTO BIOS 60KT41AUS) fails to boot on
recent kernels including the think-lmi driver, due to the fact that
errors returned by the tlmi_analyze() function are ignored by
tlmi_probe(), where tlmi_sysfs_init() is called unconditionally.
This results in making use of an array of already freed, non-null
pointers and other uninitialized globals, causing all sorts of nasty
kobject and memory faults.
Make use of the analyze function return value, free a couple leaked
allocations, and remove the settings_count field, which is incremented
but never consumed.
Fixes: a40cd7ef22 ("platform/x86: think-lmi: Add WMI interface support on Lenovo platforms")
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Reviewed-by: Mark Gross <markgross@kernel.org>
Reviewed-by: Mark Pearson <markpearson@lenovo.com>
Link: https://lore.kernel.org/r/163639463588.1330483.15850167112490200219.stgit@omen
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
When CONFIG_FUNCTION_GRAPH_TRACER is selected and the function graph
tracer is in use, unwind_frame() may erroneously associate a traced
function with an incorrect return address. This can happen when starting
an unwind from a pt_regs, or when unwinding across an exception
boundary.
This can be seen when recording with perf while the function graph
tracer is in use. For example:
| # echo function_graph > /sys/kernel/debug/tracing/current_tracer
| # perf record -g -e raw_syscalls:sys_enter:k /bin/true
| # perf report
... reports the callchain erroneously as:
| el0t_64_sync
| el0t_64_sync_handler
| el0_svc_common.constprop.0
| perf_callchain
| get_perf_callchain
| syscall_trace_enter
| syscall_trace_enter
... whereas when the function graph tracer is not in use, it reports:
| el0t_64_sync
| el0t_64_sync_handler
| el0_svc
| do_el0_svc
| el0_svc_common.constprop.0
| syscall_trace_enter
| syscall_trace_enter
The underlying problem is that ftrace_graph_get_ret_stack() takes an
index offset from the most recent entry added to the fgraph return
stack. We start an unwind at offset 0, and increment the offset each
time we encounter a rewritten return address (i.e. when we see
`return_to_handler`). This is broken in two cases:
1) Between creating a pt_regs and starting the unwind, function calls
may place entries on the stack, leaving an arbitrary offset which we
can only determine by performing a full unwind from the caller of the
unwind code (and relying on none of the unwind code being
instrumented).
This can result in erroneous entries being reported in a backtrace
recorded by perf or kfence when the function graph tracer is in use.
Currently show_regs() is unaffected as dump_backtrace() performs an
initial unwind.
2) When unwinding across an exception boundary (whether continuing an
unwind or starting a new unwind from regs), we currently always skip
the LR of the interrupted context. Where this was live and contained
a rewritten address, we won't consume the corresponding fgraph ret
stack entry, leaving subsequent entries off-by-one.
This can result in erroneous entries being reported in a backtrace
performed by any in-kernel unwinder when that backtrace crosses an
exception boundary, with entries after the boundary being reported
incorrectly. This includes perf, kfence, show_regs(), panic(), etc.
To fix this, we need to be able to uniquely identify each rewritten
return address such that we can map this back to the original return
address. We can use HAVE_FUNCTION_GRAPH_RET_ADDR_PTR to associate
each rewritten return address with a unique location on the stack. As
the return address is passed in the LR (and so is not guaranteed a
unique location in memory), we use the FP upon entry to the function
(i.e. the address of the caller's frame record) as the return address
pointer. Any nested call will have a different FP value as the caller
must create its own frame record and update FP to point to this.
Since ftrace_graph_ret_addr() requires the return address with the PAC
stripped, the stripping of the PAC is moved before the fixup of the
rewritten address. As we would unconditionally strip the PAC, moving
this earlier is not harmful, and we can avoid a redundant strip in the
return address fixup code.
I've tested this with the perf case above, the ftrace selftests, and
a number of ad-hoc unwinder tests. The tests all pass, and I have seen
no unexpected behaviour as a result of this change. I've tested with
pointer authentication under QEMU TCG where magic-sysrq+l correctly
recovers the original return addresses.
Note that this doesn't fix the issue of skipping a live LR at an
exception boundary, which is a more general problem and requires more
substantial rework. Were we to consume the LR in all cases this would
result in warnings where the interrupted context's LR contains
`return_to_handler`, but the FP has been altered, e.g.
| func:
| <--- ftrace entry ---> // logs FP & LR, rewrites LR
| STP FP, LR, [SP, #-16]!
| MOV FP, SP
| <--- INTERRUPT --->
... as ftrace_graph_get_ret_stack() fill not find a matching entry,
triggering the WARN_ON_ONCE() in unwind_frame().
Link: https://lore.kernel.org/r/20211025164925.GB2001@C02TD0UTHF1T.local
Link: https://lore.kernel.org/r/20211027132529.30027-1-mark.rutland@arm.com
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Madhavan T. Venkataraman <madvenka@linux.microsoft.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Will Deacon <will@kernel.org>
Reviewed-by: Mark Brown <broonie@kernel.org>
Link: https://lore.kernel.org/r/20211029162245.39761-1-mark.rutland@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
