Juergen Gross [Thu, 28 Mar 2024 11:24:48 +0000 (12:24 +0100)]
x86/pat: fix W^X violation false-positives when running as Xen PV guest
When running as Xen PV guest in some cases W^X violation WARN()s have
been observed. Those WARN()s are produced by verify_rwx(), which looks
into the PTE to verify that writable kernel pages have the NX bit set
in order to avoid code modifications of the kernel by rogue code.
As the NX bits of all levels of translation entries are or-ed and the
RW bits of all levels are and-ed, looking just into the PTE isn't enough
for the decision that a writable page is executable, too. When running
as a Xen PV guest, kernel initialization will set the NX bit in PMD
entries of the initial page tables covering the .data segment.
When finding the PTE to have set the RW bit but no NX bit, higher level
entries must be looked at. Only when all levels have the RW bit set and
no NX bit set, the W^X violation should be flagged.
Additionally show_fault_oops() has a similar problem: it will issue the
"kernel tried to execute NX-protected page" message only if it finds
the NX bit set in the leaf translation entry, while any NX bit in
non-leaf entries are being ignored for issuing the message.
Modify lookup_address_in_pgd() to return the effective NX and RW bit
values of the non-leaf translation entries and evaluate those as well
in verify_rwx() and show_fault_oops().
lock_task_sighand() can trigger a hard lockup. If NR_CPUS threads call
do_task_stat() at the same time and the process has NR_THREADS, it will
spin with irqs disabled O(NR_CPUS * NR_THREADS) time.
Change do_task_stat() to use sig->stats_lock to gather the statistics
outside of ->siglock protected section, in the likely case this code will
run lockless.
Link: https://lkml.kernel.org/r/20240123153357.GA21857@redhat.com Signed-off-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Dylan Hatch <dylanbhatch@google.com> Cc: Eric W. Biederman <ebiederm@xmission.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
do/while_each_thread should be avoided when possible.
Link: https://lkml.kernel.org/r/20230909164501.GA11581@redhat.com Signed-off-by: Oleg Nesterov <oleg@redhat.com> Cc: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Stable-dep-of: 7601df8031fd ("fs/proc: do_task_stat: use sig->stats_lock to gather the threads/children stats") Signed-off-by: Sasha Levin <sashal@kernel.org>
lock_task_sighand() can trigger a hard lockup. If NR_CPUS threads call
getrusage() at the same time and the process has NR_THREADS, spin_lock_irq
will spin with irqs disabled O(NR_CPUS * NR_THREADS) time.
Change getrusage() to use sig->stats_lock, it was specifically designed
for this type of use. This way it runs lockless in the likely case.
TODO:
- Change do_task_stat() to use sig->stats_lock too, then we can
remove spin_lock_irq(siglock) in wait_task_zombie().
- Turn sig->stats_lock into seqcount_rwlock_t, this way the
readers in the slow mode won't exclude each other. See
https://lore.kernel.org/all/20230913154907.GA26210@redhat.com/
- stats_lock has to disable irqs because ->siglock can be taken
in irq context, it would be very nice to change __exit_signal()
to avoid the siglock->stats_lock dependency.
Link: https://lkml.kernel.org/r/20240122155053.GA26214@redhat.com Signed-off-by: Oleg Nesterov <oleg@redhat.com> Reported-by: Dylan Hatch <dylanbhatch@google.com> Tested-by: Dylan Hatch <dylanbhatch@google.com> Cc: Eric W. Biederman <ebiederm@xmission.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
do/while_each_thread should be avoided when possible.
Plus this change allows to avoid lock_task_sighand(), we can use rcu
and/or sig->stats_lock instead.
Link: https://lkml.kernel.org/r/20230909172629.GA20454@redhat.com Signed-off-by: Oleg Nesterov <oleg@redhat.com> Cc: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Stable-dep-of: f7ec1cd5cc7e ("getrusage: use sig->stats_lock rather than lock_task_sighand()") Signed-off-by: Sasha Levin <sashal@kernel.org>
Patch series "getrusage: use sig->stats_lock", v2.
This patch (of 2):
thread_group_cputime() does its own locking, we can safely shift
thread_group_cputime_adjusted() which does another for_each_thread loop
outside of ->siglock protected section.
This is also preparation for the next patch which changes getrusage() to
use stats_lock instead of siglock, thread_group_cputime() takes the same
lock. With the current implementation recursive read_seqbegin_or_lock()
is fine, thread_group_cputime() can't enter the slow mode if the caller
holds stats_lock, yet this looks more safe and better performance-wise.
With a second DP monitor connected, drm_atomic_state in dm atomic check
sequence does not include the connector state for the old/existing/first
DP monitor. In such case, dsc determination policy would hit a null ptr
when it tries to iterate the old/existing stream that does not have a
valid connector state attached to it. When that happens, dm atomic check
should call drm_atomic_get_connector_state for a new connector state.
Existing dm has already done that, except for RV due to it does not have
official support of dsc where .num_dsc is not defined in dcn10 resource
cap, that prevent from getting drm_atomic_get_connector_state called.
So, skip dsc determination policy for ASICs that don't have DSC support.
Cc: stable@vger.kernel.org # 6.1+ Link: https://gitlab.freedesktop.org/drm/amd/-/issues/2314 Reviewed-by: Wayne Lin <wayne.lin@amd.com> Acked-by: Hamza Mahfooz <hamza.mahfooz@amd.com> Signed-off-by: Fangzhi Zuo <jerry.zuo@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
[Why]
For FreeSync HDR, native color space flag in AMD VSIF(BT.709) should be
used when intepreting content and color space flag in VSC or AVI
infoFrame should be ignored. However, it turned out some userspace
application still use color flag in VSC or AVI infoFrame which is
incorrect.
[How]
Transfer function is used when building the VSC and AVI infoFrame. Set
colorimetry to BT.709 when all the following match:
1. Pixel format is YCbCr;
2. In FreeSync 2 HDR, color is COLOR_SPACE_2020_YCBCR;
3. Transfer function is TRANSFER_FUNC_GAMMA_22;
Tested-by: Mark Broadworth <mark.broadworth@amd.com> Reviewed-by: Krunoslav Kovac <Krunoslav.Kovac@amd.com> Acked-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com> Signed-off-by: Ma Hanghong <hanghong.ma@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Stable-dep-of: e6a7df96facd ("drm/amd/display: Fix MST Null Ptr for RV") Signed-off-by: Sasha Levin <sashal@kernel.org>
When running the simult_flow selftest in slow environments -- e.g. QEmu
without KVM support --, the results can be unstable. This selftest
checks if the aggregated bandwidth is (almost) fully used as expected.
To help improving the stability while still keeping the same validation
in place, the BW and the delay are reduced to lower the pressure on the
CPU.
Fixes: 1a418cb8e888 ("mptcp: simult flow self-tests") Fixes: 219d04992b68 ("mptcp: push pending frames when subflow has free space") Cc: stable@vger.kernel.org Suggested-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org> Link: https://lore.kernel.org/r/20240131-upstream-net-20240131-mptcp-ci-issues-v1-6-4c1c11e571ff@kernel.org Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Mitigation for RFDS requires RFDS_CLEAR capability which is enumerated
by MSR_IA32_ARCH_CAPABILITIES bit 27. If the host has it set, export it
to guests so that they can deploy the mitigation.
RFDS_NO indicates that the system is not vulnerable to RFDS, export it
to guests so that they don't deploy the mitigation unnecessarily. When
the host is not affected by X86_BUG_RFDS, but has RFDS_NO=0, synthesize
RFDS_NO to the guest.
Signed-off-by: Pawan Gupta <pawan.kumar.gupta@linux.intel.com> Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Josh Poimboeuf <jpoimboe@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
RFDS is a CPU vulnerability that may allow userspace to infer kernel
stale data previously used in floating point registers, vector registers
and integer registers. RFDS only affects certain Intel Atom processors.
Intel released a microcode update that uses VERW instruction to clear
the affected CPU buffers. Unlike MDS, none of the affected cores support
SMT.
Add RFDS bug infrastructure and enable the VERW based mitigation by
default, that clears the affected buffers just before exiting to
userspace. Also add sysfs reporting and cmdline parameter
"reg_file_data_sampling" to control the mitigation.
For details see:
Documentation/admin-guide/hw-vuln/reg-file-data-sampling.rst
Signed-off-by: Pawan Gupta <pawan.kumar.gupta@linux.intel.com> Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Josh Poimboeuf <jpoimboe@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Currently MMIO Stale Data mitigation for CPUs not affected by MDS/TAA is
to only deploy VERW at VMentry by enabling mmio_stale_data_clear static
branch. No mitigation is needed for kernel->user transitions. If such
CPUs are also affected by RFDS, its mitigation may set
X86_FEATURE_CLEAR_CPU_BUF to deploy VERW at kernel->user and VMentry.
This could result in duplicate VERW at VMentry.
Fix this by disabling mmio_stale_data_clear static branch when
X86_FEATURE_CLEAR_CPU_BUF is enabled.
Signed-off-by: Pawan Gupta <pawan.kumar.gupta@linux.intel.com> Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com> Reviewed-by: Dave Hansen <dave.hansen@linux.intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Allows us to detect subsequent IH ring buffer overflows as well.
Cc: Joshua Ashton <joshua@froggi.es> Cc: Alex Deucher <alexander.deucher@amd.com> Cc: Christian König <christian.koenig@amd.com> Cc: stable@vger.kernel.org Signed-off-by: Friedrich Vock <friedrich.vock@gmx.de> Reviewed-by: Christian König <christian.koenig@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
xHCI 4.9 explicitly forbids assuming that the xHC has released its
ownership of a multi-TRB TD when it reports an error on one of the
early TRBs. Yet the driver makes such assumption and releases the TD,
allowing the remaining TRBs to be freed or overwritten by new TDs.
The xHC should also report completion of the final TRB due to its IOC
flag being set by us, regardless of prior errors. This event cannot
be recognized if the TD has already been freed earlier, resulting in
"Transfer event TRB DMA ptr not part of current TD" error message.
Fix this by reusing the logic for processing isoc Transaction Errors.
This also handles hosts which fail to report the final completion.
Fix transfer length reporting on Babble errors. They may be caused by
device malfunction, no guarantee that the buffer has been filled.
The last TRB of a isoc TD might not trigger an event if there was
an error event for a TRB mid TD. This is seen on a NEC Corporation
uPD720200 USB 3.0 Host
After an error mid a multi-TRB TD the xHC should according to xhci 4.9.1
generate events for passed TRBs with IOC flag set if it proceeds to the
next TD. This event is either a copy of the original error, or a
"success" transfer event.
If that event is missing then the driver and xHC host get out of sync as
the driver is still expecting a transfer event for that first TD, while
xHC host is already sending events for the next TD in the list.
This leads to
"Transfer event TRB DMA ptr not part of current TD" messages.
As a solution we tag the isoc TDs that get error events mid TD.
If an event doesn't match the first TD, then check if the tag is
set, and event points to the next TD.
In that case give back the fist TD and process the next TD normally
Make sure TD status and transferred length stay valid in both cases
with and without final TD completion event.
On systems with 64k page size and 512M huge page sizes, the allocation and
test succeeds but errors out at the munmap. As the comment states, munmap
will failure if its not HUGEPAGE aligned. This is due to the length of
the mapping being 1/2 the size of the hugepage causing the munmap to not
be hugepage aligned. Fix this by making the mapping length the full
hugepage if the hugepage is larger than the length of the mapping.
Running charge_reserved_hugetlb.sh generates errors if sh is set to
dash:
./charge_reserved_hugetlb.sh: 9: [[: not found
./charge_reserved_hugetlb.sh: 19: [[: not found
./charge_reserved_hugetlb.sh: 27: [[: not found
./charge_reserved_hugetlb.sh: 37: [[: not found
./charge_reserved_hugetlb.sh: 45: Syntax error: "(" unexpected
Switch to using /bin/bash instead of /bin/sh. Make the switch for
write_hugetlb_memory.sh as well which is called from
charge_reserved_hugetlb.sh.
Link: https://lkml.kernel.org/r/20240116090455.3407378-1-usama.anjum@collabora.com Signed-off-by: Muhammad Usama Anjum <usama.anjum@collabora.com> Cc: Muhammad Usama Anjum <usama.anjum@collabora.com> Cc: Shuah Khan <shuah@kernel.org> Cc: David Laight <David.Laight@ACULAB.COM> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
ra_alloc_folio() marks a page that should trigger next round of async
readahead. However it rounds up computed index to the order of page being
allocated. This can however lead to multiple consecutive pages being
marked with readahead flag. Consider situation with index == 1, mark ==
1, order == 0. We insert order 0 page at index 1 and mark it. Then we
bump order to 1, index to 2, mark (still == 1) is rounded up to 2 so page
at index 2 is marked as well. Then we bump order to 2, index is
incremented to 4, mark gets rounded to 4 so page at index 4 is marked as
well. The fact that multiple pages get marked within a single readahead
window confuses the readahead logic and results in readahead window being
trimmed back to 1. This situation is triggered in particular when maximum
readahead window size is not a power of two (in the observed case it was
768 KB) and as a result sequential read throughput suffers.
Fix the problem by rounding 'mark' down instead of up. Because the index
is naturally aligned to 'order', we are guaranteed 'rounded mark' == index
iff 'mark' is within the page we are allocating at 'index' and thus
exactly one page is marked with readahead flag as required by the
readahead code and sequential read performance is restored.
This effectively reverts part of commit b9ff43dd2743 ("mm/readahead: Fix
readahead with large folios"). The commit changed the rounding with the
rationale:
"... we were setting the readahead flag on the folio which contains the
last byte read from the block. This is wrong because we will trigger
readahead at the end of the read without waiting to see if a subsequent
read is going to use the pages we just read."
Although this is true, the fact is this was always the case with read
sizes not aligned to folio boundaries and large folios in the page cache
just make the situation more obvious (and frequent). Also for sequential
read workloads it is better to trigger the readahead earlier rather than
later. It is true that the difference in the rounding and thus earlier
triggering of the readahead can result in reading more for semi-random
workloads. However workloads really suffering from this seem to be rare.
In particular I have verified that the workload described in commit b9ff43dd2743 ("mm/readahead: Fix readahead with large folios") of reading
random 100k blocks from a file like:
is not impacted by the rounding change and achieves ~70MB/s in both cases.
[jack@suse.cz: fix one more place where mark rounding was done as well] Link: https://lkml.kernel.org/r/20240123153254.5206-1-jack@suse.cz Link: https://lkml.kernel.org/r/20240104085839.21029-1-jack@suse.cz Fixes: b9ff43dd2743 ("mm/readahead: Fix readahead with large folios") Signed-off-by: Jan Kara <jack@suse.cz> Cc: Matthew Wilcox <willy@infradead.org> Cc: Guo Xuenan <guoxuenan@huawei.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
The nfp offload flow pay will not allocate a mask id when the out port
is openvswitch internal port. This is because these flows are used to
configure the pre_tun table and are never actually send to the firmware
as an add-flow message. When a tc rule which action contains ct and
the post ct entry's out port is openvswitch internal port, the merge
offload flow pay with the wrong mask id of 0 will be send to the
firmware. Actually, the nfp can not support hardware offload for this
situation, so return EOPNOTSUPP.
Fixes: bd0fe7f96a3c ("nfp: flower-ct: add zone table entry when handling pre/post_ct flows") CC: stable@vger.kernel.org # 5.14+ Signed-off-by: Hui Zhou <hui.zhou@corigine.com> Signed-off-by: Louis Peens <louis.peens@corigine.com> Link: https://lore.kernel.org/r/20240124151909.31603-2-louis.peens@corigine.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
The chain_index has different means in pre ct entry and post ct entry.
In pre ct entry, it means chain index, but in post ct entry, it means
goto chain index, it is confused.
chain_index and goto_chain_index may be present in one flow rule, It
cannot be distinguished by one field chain_index, both chain_index
and goto_chain_index are required in the follow-up patch to support
multiple ct zones
Another field goto_chain_index is added to record the goto chain index.
If no goto action in post ct entry, goto_chain_index is 0.
Signed-off-by: Wentao Jia <wentao.jia@corigine.com> Acked-by: Simon Horman <simon.horman@corigine.com> Signed-off-by: Louis Peens <louis.peens@corigine.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Stable-dep-of: cefa98e806fd ("nfp: flower: add hardware offload check for post ct entry") Signed-off-by: Sasha Levin <sashal@kernel.org>
The 'status' variable in 'core_link_read_dpcd()' &
'core_link_write_dpcd()' was uninitialized.
Thus, initializing 'status' variable to 'DC_ERROR_UNEXPECTED' by default.
Fixes the below:
drivers/gpu/drm/amd/amdgpu/../display/dc/link/protocols/link_dpcd.c:226 core_link_read_dpcd() error: uninitialized symbol 'status'.
drivers/gpu/drm/amd/amdgpu/../display/dc/link/protocols/link_dpcd.c:248 core_link_write_dpcd() error: uninitialized symbol 'status'.
Cc: stable@vger.kernel.org Cc: Jerry Zuo <jerry.zuo@amd.com> Cc: Jun Lei <Jun.Lei@amd.com> Cc: Wayne Lin <Wayne.Lin@amd.com> Cc: Aurabindo Pillai <aurabindo.pillai@amd.com> Cc: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com> Cc: Hamza Mahfooz <hamza.mahfooz@amd.com> Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com> Reviewed-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
The lowest headphones volume setting does not mute so the leave the TLV
mute flag unset.
This is specifically needed to let the sound server use the lowest gain
setting.
Fixes: c03226ba15fe ("ASoC: codecs: wcd938x: fix dB range for HPHL and HPHR") Cc: <stable@vger.kernel.org> # 6.5 Cc: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> Signed-off-by: Johan Hovold <johan+linaro@kernel.org> Link: https://msgid.link/r/20240122091130.27463-1-johan+linaro@kernel.org Signed-off-by: Mark Brown <broonie@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Right now it is possible to see gmap->private being zero in
kvm_s390_vsie_gmap_notifier resulting in a crash. This is due to the
fact that we add gmap->private == kvm after creation:
The shadow gmap tracks memory of nested guests (guest-3). In certain
scenarios, the shadow gmap needs to be rebuilt, which is a costly operation
since it involves a SIE exit into guest-1 for every entry in the respective
shadow level.
Add kvm stat counters when new shadow structures are created at various
levels. Also add a counter gmap_shadow_create when a completely fresh
shadow gmap is created as well as a counter gmap_shadow_reuse when an
existing gmap is being reused.
Note that when several levels are shadowed at once, counters on all
affected levels will be increased.
Also note that not all page table levels need to be present and a ASCE
can directly point to e.g. a segment table. In this case, a new segment
table will always be equivalent to a new shadow gmap and hence will be
counted as gmap_shadow_create and not as gmap_shadow_segment.
Signed-off-by: Nico Boehr <nrb@linux.ibm.com> Reviewed-by: David Hildenbrand <david@redhat.com> Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com> Reviewed-by: Janosch Frank <frankja@linux.ibm.com> Signed-off-by: Janosch Frank <frankja@linux.ibm.com> Link: https://lore.kernel.org/r/20231009093304.2555344-2-nrb@linux.ibm.com
Message-Id: <20231009093304.2555344-2-nrb@linux.ibm.com>
Stable-dep-of: fe752331d4b3 ("KVM: s390: vsie: fix race during shadow creation") Signed-off-by: Sasha Levin <sashal@kernel.org>
There are mainly two reasons that thp_get_unmapped_area() should be
used for EROFS as other filesystems:
- It's needed to enable PMD mappings as a FSDAX filesystem, see
commit 74d2fad1334d ("thp, dax: add thp_get_unmapped_area for pmd
mappings");
- It's useful together with large folios and
CONFIG_READ_ONLY_THP_FOR_FS which enable THPs for mmapped files
(e.g. shared libraries) even without FSDAX. See commit 1854bc6e2420
("mm/readahead: Align file mappings for non-DAX").
Fixes: 06252e9ce05b ("erofs: dax support for non-tailpacking regular file") Fixes: ce529cc25b18 ("erofs: enable large folios for iomap mode") Fixes: e6687b89225e ("erofs: enable large folios for fscache mode") Reviewed-by: Jingbo Xu <jefflexu@linux.alibaba.com> Reviewed-by: Chao Yu <chao@kernel.org> Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com> Link: https://lore.kernel.org/r/20240306053138.2240206-1-hsiangkao@linux.alibaba.com Signed-off-by: Sasha Levin <sashal@kernel.org>
Due to abnormal data in skb->data, the extension bitmap length
exceeds 32 when decoding ras message then uses the length to make
a shift operation. It will change into negative after several loop.
UBSAN load could detect a negative shift as an undefined behaviour
and reports exception.
So we add the protection to avoid the length exceeding 32. Or else
it will return out of range error and stop decoding.
Fixes: 5e35941d9901 ("[NETFILTER]: Add H.323 conntrack/NAT helper") Signed-off-by: Lena Wang <lena.wang@mediatek.com> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
If connection isn't established yet, get_mr() will fail, trigger connection after
get_mr().
Fixes: 584a8279a44a ("RDS: RDMA: return appropriate error on rdma map failures") Reported-and-tested-by: syzbot+d4faee732755bba9838e@syzkaller.appspotmail.com Signed-off-by: Edward Adam Davis <eadavis@qq.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
This bug was noticed while re-implementing parts of the kernel
driver in userspace using spidev. The goal was to enable some
of the errata workarounds that Microchip describes in their
errata sheet [1].
Both the errata sheet and the regular datasheet of e.g. the KSZ8795
imply that you need to do this for indirect register accesses:
- write a 16-bit value to a control register pair (this value
consists of the indirect register table, and the offset inside
the table)
- either read or write an 8-bit value from the data storage
register (indicated by REG_IND_BYTE in the kernel)
The current implementation has the order swapped. It can be
proven, by reading back some indirect register with known content
(the EEE register modified in ksz8_handle_global_errata() is one of
these), that this implementation does not work.
Private discussion with Oleksij Rempel of Pengutronix has revealed
that the workaround was apparantly never tested on actual hardware.
When running an XDP program that is attached to a cpumap entry, we don't
initialise the xdp_rxq_info data structure being used in the xdp_buff
that backs the XDP program invocation. Tobias noticed that this leads to
random values being returned as the xdp_md->rx_queue_index value for XDP
programs running in a cpumap.
This means we're basically returning the contents of the uninitialised
memory, which is bad. Fix this by zero-initialising the rxq data
structure before running the XDP program.
Fixes: 9216477449f3 ("bpf: cpumap: Add the possibility to attach an eBPF program to cpumap") Reported-by: Tobias Böhm <tobias@aibor.de> Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com> Link: https://lore.kernel.org/r/20240305213132.11955-1-toke@redhat.com Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
The buggy address belongs to the object at ffff88809a07fc00
which belongs to the cache kmalloc-512 of size 512
The buggy address is located 100 bytes inside of
freed 512-byte region [ffff88809a07fc00, ffff88809a07fe00)
Memory state around the buggy address: ffff88809a07fb00: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc ffff88809a07fb80: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
>ffff88809a07fc00: fa fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
^ ffff88809a07fc80: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ffff88809a07fd00: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
Fixes: 3b1137fe7482 ("net: ipv6: Change notifications for multipath add to RTA_MULTIPATH") Reported-by: syzbot <syzkaller@googlegroups.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Reviewed-by: David Ahern <dsahern@kernel.org> Link: https://lore.kernel.org/r/20240303144801.702646-1-edumazet@google.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
When a frame can not be transmitted in XDP_REDIRECT
(e.g. due to a full queue), it is necessary to free
it by calling xdp_return_frame_rx_napi.
However, this is the responsibility of the caller of
the ndo_xdp_xmit (see for example bq_xmit_all in
kernel/bpf/devmap.c) and thus calling it inside
igc_xdp_xmit (which is the ndo_xdp_xmit of the igc
driver) as well will lead to memory corruption.
In fact, bq_xmit_all expects that it can return all
frames after the last successfully transmitted one.
Therefore, break for the first not transmitted frame,
but do not call xdp_return_frame_rx_napi in igc_xdp_xmit.
This is equally implemented in other Intel drivers
such as the igb.
There are two alternatives to this that were rejected:
1. Return num_frames as all the frames would have been
transmitted and release them inside igc_xdp_xmit.
While it might work technically, it is not what
the return value is meant to represent (i.e. the
number of SUCCESSFULLY transmitted packets).
2. Rework kernel/bpf/devmap.c and all drivers to
support non-consecutively dropped packets.
Besides being complex, it likely has a negative
performance impact without a significant gain
since it is anyway unlikely that the next frame
can be transmitted if the previous one was dropped.
The memory corruption can be reproduced with
the following script which leads to a kernel panic
after a few seconds. It basically generates more
traffic than a i225 NIC can transmit and pushes it
via XDP_REDIRECT from a virtual interface to the
physical interface where frames get dropped.
The function ice_bridge_setlink() may encounter a NULL pointer dereference
if nlmsg_find_attr() returns NULL and br_spec is dereferenced subsequently
in nla_for_each_nested(). To address this issue, add a check to ensure that
br_spec is not NULL before proceeding with the nested attribute iteration.
Fixes: b1edc14a3fbf ("ice: Implement ice_bridge_getlink and ice_bridge_setlink") Signed-off-by: Rand Deeb <rand.sec96@gmail.com> Reviewed-by: Simon Horman <horms@kernel.org> Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
The E800 series hardware uses the same iAVF driver as older devices,
including the virtchnl negotiation scheme.
This negotiation scheme includes a mechanism to determine what type of RSS
should be supported, including RSS over PF virtchnl messages, RSS over
firmware AdminQ messages, and RSS via direct register access.
The PF driver will always prefer VIRTCHNL_VF_OFFLOAD_RSS_PF if its
supported by the VF driver. However, if an older VF driver is loaded, it
may request only VIRTCHNL_VF_OFFLOAD_RSS_REG or VIRTCHNL_VF_OFFLOAD_RSS_AQ.
The ice driver happily agrees to support these methods. Unfortunately, the
underlying hardware does not support these mechanisms. The E800 series VFs
don't have the appropriate registers for RSS_REG. The mailbox queue used by
VFs for VF to PF communication blocks messages which do not have the
VF-to-PF opcode.
Stop lying to the VF that it could support RSS over AdminQ or registers, as
these interfaces do not work when the hardware is operating on an E800
series device.
In practice this is unlikely to be hit by any normal user. The iAVF driver
has supported RSS over PF virtchnl commands since 2016, and always defaults
to using RSS_PF if possible.
In principle, nothing actually stops the existing VF from attempting to
access the registers or send an AQ command. However a properly coded VF
will check the capability flags and will report a more useful error if it
detects a case where the driver does not support the RSS offloads that it
does.
Fixes: 1071a8358a28 ("ice: Implement virtchnl commands for AVF support") Signed-off-by: Jacob Keller <jacob.e.keller@intel.com> Reviewed-by: Alan Brady <alan.brady@intel.com> Tested-by: Rafal Romanowski <rafal.romanowski@intel.com> Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Based on the static analyzis of the code it looks like when an entry
from the MAC table was removed, the entry was still used after being
freed. More precise the vid of the mac_entry was used after calling
devm_kfree on the mac_entry.
The fix consists in first using the vid of the mac_entry to delete the
entry from the HW and after that to free it.
Issue is similar to the one I fixed in commit 8d975c15c0cd
("ip6_tunnel: make sure to pull inner header in __ip6_tnl_rcv()")
We have to save skb->network_header in a temporary variable
in order to be able to recompute the network_header pointer
after a pskb_inet_may_pull() call.
pskb_inet_may_pull() makes sure the needed headers are in skb->head.
I'm updating __assign_str() and will be removing the second parameter. To
make sure that it does not break anything, I make sure that it matches the
__string() field, as that is where the string is actually going to be
saved in. To make sure there's nothing that breaks, I added a WARN_ON() to
make sure that what was used in __string() is the same that is used in
__assign_str().
In doing this change, an error was triggered as __assign_str() now expects
the string passed in to be a char * value. I instead had the following
warning:
include/trace/events/qdisc.h: In function ‘trace_event_raw_event_qdisc_reset’:
include/trace/events/qdisc.h:91:35: error: passing argument 1 of 'strcmp' from incompatible pointer type [-Werror=incompatible-pointer-types]
91 | __assign_str(dev, qdisc_dev(q));
That's because the qdisc_enqueue() and qdisc_reset() pass in qdisc_dev(q)
to __assign_str() and to __string(). But that function returns a pointer
to struct net_device and not a string.
It appears that these events are just saving the pointer as a string and
then reading it as a string as well.
Use qdisc_dev(q)->name to save the device instead.
Fixes: a34dac0b90552 ("net_sched: add tracepoints for qdisc_reset() and qdisc_destroy()") Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org> Reviewed-by: Jamal Hadi Salim <jhs@mojatatu.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
ice_qp_dis() currently does things in very mixed way. Tx is stopped
before disabling IRQ on related queue vector, then it takes care of
disabling Rx and finally NAPI is disabled.
Let us start with disabling IRQs in the first place followed by turning
off NAPI. Then it is safe to handle queues.
One subtle change on top of that is that even though ice_qp_ena() looks
more sane, clear ICE_CFG_BUSY as the last thing there.
Fixes: 2d4238f55697 ("ice: Add support for AF_XDP") Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> Tested-by: Chandan Kumar Rout <chandanx.rout@intel.com> (A Contingent Worker at Intel) Acked-by: Magnus Karlsson <magnus.karlsson@intel.com> Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Disable NAPI before shutting down queues that this particular NAPI
contains so that the order of actions in i40e_queue_pair_disable()
mirrors what we do in i40e_queue_pair_enable().
Fixes: 123cecd427b6 ("i40e: added queue pair disable/enable functions") Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> Tested-by: Chandan Kumar Rout <chandanx.rout@intel.com> (A Contingent Worker at Intel) Acked-by: Magnus Karlsson <magnus.karlsson@intel.com> Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Currently routines that are supposed to toggle state of ring pair do not
take care of associated interrupt with queue vector that these rings
belong to. This causes funky issues such as dead interface due to irq
misconfiguration, as per Pavel's report from Closes: tag.
Add a function responsible for disabling single IRQ in EIMC register and
call this as a very first thing when disabling ring pair during xsk_pool
setup. For enable let's reuse ixgbe_irq_enable_queues(). Besides this,
disable/enable NAPI as first/last thing when dealing with closing or
opening ring pair that xsk_pool is being configured on.
Reported-by: Pavel Vazharov <pavel@x3me.net> Closes: https://lore.kernel.org/netdev/CAJEV1ijxNyPTwASJER1bcZzS9nMoZJqfR86nu_3jFFVXzZQ4NA@mail.gmail.com/ Fixes: 024aa5800f32 ("ixgbe: added Rx/Tx ring disable/enable functions") Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> Acked-by: Magnus Karlsson <magnus.karlsson@intel.com> Tested-by: Chandan Kumar Rout <chandanx.rout@intel.com> (A Contingent Worker at Intel) Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Current driver has some asymmetry in the runtime PM calls. On lan78xx_open()
it will call usb_autopm_get() and unconditionally usb_autopm_put(). And
on lan78xx_stop() it will call only usb_autopm_put(). So far, it was
working only because this driver do not activate autosuspend by default,
so it was visible only by warning "Runtime PM usage count underflow!".
Since, with current driver, we can't use runtime PM with active link,
execute lan78xx_open()->usb_autopm_put() only in error case. Otherwise,
keep ref counting high as long as interface is open.
Fixes: 55d7de9de6c3 ("Microchip's LAN7800 family USB 2/3 to 10/100/1000 Ethernet device driver") Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de> Reviewed-by: Jiri Pirko <jiri@nvidia.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
The addition of bal_rank_mask with encoding version 17 was merged
into ceph.git in Oct 2022 and made it into v18.2.0 release normally.
A few months later, the much delayed addition of max_xattr_size got
merged, also with encoding version 17, placed before bal_rank_mask
in the encoding -- but it didn't make v18.2.0 release.
The way this ended up being resolved on the MDS side is that
bal_rank_mask will continue to be encoded in version 17 while
max_xattr_size is now encoded in version 18. This does mean that
older kernels will misdecode version 17, but this is also true for
v18.2.0 and v18.2.1 clients in userspace.
shutdown_pirq and startup_pirq are not taking the
irq_mapping_update_lock because they can't due to lock inversion. Both
are called with the irq_desc->lock being taking. The lock order,
however, is first irq_mapping_update_lock and then irq_desc->lock.
This opens multiple races:
- shutdown_pirq can be interrupted by a function that allocates an event
channel:
Assume here event channel e refers here to the same event channel
number.
After this race the evtchn_to_irq mapping for e is invalid (-1).
- __startup_pirq races with __unbind_from_irq in a similar way. Because
__startup_pirq doesn't take irq_mapping_update_lock it can grab the
evtchn that __unbind_from_irq is currently freeing and cleaning up. In
this case even though the event channel is allocated, its mapping can
be unset in evtchn_to_irq.
The fix is to first cleanup the mappings and then close the event
channel. In this way, when an event channel gets allocated it's
potential previous evtchn_to_irq mappings are guaranteed to be unset already.
This is also the reverse order of the allocation where first the event
channel is allocated and then the mappings are setup.
On a 5.10 kernel prior to commit 3fcdaf3d7634 ("xen/events: modify internal
[un]bind interfaces"), we hit a BUG like the following during probing of NVMe
devices. The issue is that during nvme_setup_io_queues, pci_free_irq
is called for every device which results in a call to shutdown_pirq.
With many nvme devices it's therefore likely to hit this race during
boot because there will be multiple calls to shutdown_pirq and
startup_pirq are running potentially in parallel.
The recently introduced EFI memory attributes protocol should be used
if it exists to ensure that the memory allocation created for the kernel
permits execution. This is needed for compatibility with tightened
requirements related to Windows logo certification for x86 PCs.
Currently, we simply strip the execute protect (XP) attribute from the
entire range, but this might be rejected under some firmware security
policies, and so in a subsequent patch, this will be changed to only
strip XP from the executable region that runs early, and make it
read-only (RO) as well.
In order to catch any issues early, ensure that the memory attribute
protocol works as intended, and give up if it produces spurious errors.
Note that the DXE services based fallback was always based on best
effort, so don't propagate any errors returned by that API.
Extend the bpf_fib_lookup() helper by making it to return the source
IPv4/IPv6 address if the BPF_FIB_LOOKUP_SRC flag is set.
For example, the following snippet can be used to derive the desired
source IP address:
struct bpf_fib_lookup p = { .ipv4_dst = ip4->daddr };
ret = bpf_skb_fib_lookup(skb, p, sizeof(p),
BPF_FIB_LOOKUP_SRC | BPF_FIB_LOOKUP_SKIP_NEIGH);
if (ret != BPF_FIB_LKUP_RET_SUCCESS)
return TC_ACT_SHOT;
/* the p.ipv4_src now contains the source address */
The inability to derive the proper source address may cause malfunctions
in BPF-based dataplanes for hosts containing netdevs with more than one
routable IP address or for multi-homed hosts.
For example, Cilium implements packet masquerading in BPF. If an
egressing netdev to which the Cilium's BPF prog is attached has
multiple IP addresses, then only one [hardcoded] IP address can be used for
masquerading. This breaks connectivity if any other IP address should have
been selected instead, for example, when a public and private addresses
are attached to the same egress interface.
The change was tested with Cilium [1].
Nikolay Aleksandrov helped to figure out the IPv6 addr selection.
[1]: https://github.com/cilium/cilium/pull/28283
Signed-off-by: Martynas Pumputis <m@lambda.lt> Link: https://lore.kernel.org/r/20231007081415.33502-2-m@lambda.lt Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Add ability to specify routing table ID to the `bpf_fib_lookup` BPF
helper.
A new field `tbid` is added to `struct bpf_fib_lookup` used as
parameters to the `bpf_fib_lookup` BPF helper.
When the helper is called with the `BPF_FIB_LOOKUP_DIRECT` and
`BPF_FIB_LOOKUP_TBID` flags the `tbid` field in `struct bpf_fib_lookup`
will be used as the table ID for the fib lookup.
If the `tbid` does not exist the fib lookup will fail with
`BPF_FIB_LKUP_RET_NOT_FWDED`.
The `tbid` field becomes a union over the vlan related output fields
in `struct bpf_fib_lookup` and will be zeroed immediately after usage.
This functionality is useful in containerized environments.
For instance, if a CNI wants to dictate the next-hop for traffic leaving
a container it can create a container-specific routing table and perform
a fib lookup against this table in a "host-net-namespace-side" TC program.
This functionality also allows `ip rule` like functionality at the TC
layer, allowing an eBPF program to pick a routing table based on some
aspect of the sk_buff.
As a concrete use case, this feature will be used in Cilium's SRv6 L3VPN
datapath.
When egress traffic leaves a Pod an eBPF program attached by Cilium will
determine which VRF the egress traffic should target, and then perform a
FIB lookup in a specific table representing this VRF's FIB.
During VMentry VERW is executed to mitigate MDS. After VERW, any memory
access like register push onto stack may put host data in MDS affected
CPU buffers. A guest can then use MDS to sample host data.
Although likelihood of secrets surviving in registers at current VERW
callsite is less, but it can't be ruled out. Harden the MDS mitigation
by moving the VERW mitigation late in VMentry path.
Note that VERW for MMIO Stale Data mitigation is unchanged because of
the complexity of per-guest conditional VERW which is not easy to handle
that late in asm with no GPRs available. If the CPU is also affected by
MDS, VERW is unconditionally executed late in asm regardless of guest
having MMIO access.
Use EFLAGS.CF instead of EFLAGS.ZF to track whether to use VMRESUME versus
VMLAUNCH. Freeing up EFLAGS.ZF will allow doing VERW, which clobbers ZF,
for MDS mitigations as late as possible without needing to duplicate VERW
for both paths.
[ pawan: resolved merge conflict in __vmx_vcpu_run in backport. ]
Signed-off-by: Sean Christopherson <seanjc@google.com> Signed-off-by: Pawan Gupta <pawan.kumar.gupta@linux.intel.com> Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com> Reviewed-by: Nikolay Borisov <nik.borisov@suse.com> Link: https://lore.kernel.org/all/20240213-delay-verw-v8-5-a6216d83edb7%40linux.intel.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The VERW mitigation at exit-to-user is enabled via a static branch
mds_user_clear. This static branch is never toggled after boot, and can
be safely replaced with an ALTERNATIVE() which is convenient to use in
asm.
Switch to ALTERNATIVE() to use the VERW mitigation late in exit-to-user
path. Also remove the now redundant VERW in exc_nmi() and
arch_exit_to_user_mode().
Mitigation for MDS is to use VERW instruction to clear any secrets in
CPU Buffers. Any memory accesses after VERW execution can still remain
in CPU buffers. It is safer to execute VERW late in return to user path
to minimize the window in which kernel data can end up in CPU buffers.
There are not many kernel secrets to be had after SWITCH_TO_USER_CR3.
Add support for deploying VERW mitigation after user register state is
restored. This helps minimize the chances of kernel data ending up into
CPU buffers after executing VERW.
Note that the mitigation at the new location is not yet enabled.
Corner case not handled
=======================
Interrupts returning to kernel don't clear CPUs buffers since the
exit-to-user path is expected to do that anyways. But, there could be
a case when an NMI is generated in kernel after the exit-to-user path
has cleared the buffers. This case is not handled and NMI returning to
kernel don't clear CPU buffers because:
1. It is rare to get an NMI after VERW, but before returning to user.
2. For an unprivileged user, there is no known way to make that NMI
less rare or target it.
3. It would take a large number of these precisely-timed NMIs to mount
an actual attack. There's presumably not enough bandwidth.
4. The NMI in question occurs after a VERW, i.e. when user state is
restored and most interesting data is already scrubbed. Whats left
is only the data that NMI touches, and that may or may not be of
any interest.
[ pawan: resolved conflict for hunk swapgs_restore_regs_and_return_to_usermode in backport ]
MDS mitigation requires clearing the CPU buffers before returning to
user. This needs to be done late in the exit-to-user path. Current
location of VERW leaves a possibility of kernel data ending up in CPU
buffers for memory accesses done after VERW such as:
1. Kernel data accessed by an NMI between VERW and return-to-user can
remain in CPU buffers since NMI returning to kernel does not
execute VERW to clear CPU buffers.
2. Alyssa reported that after VERW is executed,
CONFIG_GCC_PLUGIN_STACKLEAK=y scrubs the stack used by a system
call. Memory accesses during stack scrubbing can move kernel stack
contents into CPU buffers.
3. When caller saved registers are restored after a return from
function executing VERW, the kernel stack accesses can remain in
CPU buffers(since they occur after VERW).
To fix this VERW needs to be moved very late in exit-to-user path.
In preparation for moving VERW to entry/exit asm code, create macros
that can be used in asm. Also make VERW patching depend on a new feature
flag X86_FEATURE_CLEAR_CPU_BUF.
[pawan: - Runtime patch jmp instead of verw in macro CLEAR_CPU_BUFFERS
due to lack of relative addressing support for relocations
in kernels < v6.5.
- Add UNWIND_HINT_EMPTY to avoid warning:
arch/x86/entry/entry.o: warning: objtool: mds_verw_sel+0x0: unreachable instruction]
Reported-by: Alyssa Milburn <alyssa.milburn@intel.com> Suggested-by: Andrew Cooper <andrew.cooper3@citrix.com> Suggested-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Pawan Gupta <pawan.kumar.gupta@linux.intel.com> Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com> Link: https://lore.kernel.org/all/20240213-delay-verw-v8-1-a6216d83edb7%40linux.intel.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
It is reported to cause boot crashes in Android systems, so revert it
from the stable trees for now.
Cc: Rob Clark <robdclark@chromium.org> Cc: Georgi Djakov <djakov@kernel.org> Cc: Guenter Roeck <linux@roeck-us.net> Cc: Jon Hunter <jonathanh@nvidia.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In commit 19416123ab3e ("block: define 'struct bvec_iter' as packed"),
what we need is to save the 4byte padding, and avoid `bio` to spread on
one extra cache line.
It is enough to define it as '__packed __aligned(4)', as '__packed'
alone means byte aligned, and can cause compiler to generate horrible
code on architectures that don't support unaligned access in case that
bvec_iter is embedded in other structures.
Cc: Mikulas Patocka <mpatocka@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Fixes: 19416123ab3e ("block: define 'struct bvec_iter' as packed") Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Chip outputs are enabled[1] before actual reset is performed[2] which might
cause pin output value to flip flop if previous pin value was set to 1.
Fix that behavior by making sure chip is fully reset before all outputs are
enabled.
Flip-flop can be noticed when module is removed and inserted again and one of
the pins was changed to 1 before removal. 100 microsecond flipping is
noticeable on oscilloscope (100khz SPI bus).
For a properly reset chip - output is enabled around 100 microseconds (on 100khz
SPI bus) later during probing process hence should be irrelevant behavioral
change.
At the time of LPAR dump, before kexec hands over control to kdump
kernel, DDWs (Dynamic DMA Windows) are scanned and added to the FDT.
For the SR-IOV case, default DMA window "ibm,dma-window" is removed from
the FDT and DDW added, for the device.
Now, kexec hands over control to the kdump kernel.
When the kdump kernel initializes, PCI busses are scanned and IOMMU
group/tables created, in pci_dma_bus_setup_pSeriesLP(). For the SR-IOV
case, there is no "ibm,dma-window". The original commit: b1fc44eaa9ba,
fixes the path where memory is pre-mapped (direct mapped) to the DDW.
When TCEs are direct mapped, there is no need to initialize IOMMU
tables.
iommu_table_setparms_lpar() only considers "ibm,dma-window" property
when initiallizing IOMMU table. In the scenario where TCEs are
dynamically allocated for SR-IOV, newly created IOMMU table is not
initialized. Later, when the device driver tries to enter TCEs for the
SR-IOV device, NULL pointer execption is thrown from iommu_area_alloc().
The fix is to initialize the IOMMU table with DDW property stored in the
FDT. There are 2 points to remember:
1. For the dedicated adapter, kdump kernel would encounter both
default and DDW in FDT. In this case, DDW property is used to
initialize the IOMMU table.
2. A DDW could be direct or dynamic mapped. kdump kernel would
initialize IOMMU table and mark the existing DDW as
"dynamic". This works fine since, at the time of table
initialization, iommu_table_clear() makes some space in the
DDW, for some predefined number of TCEs which are needed for
kdump to succeed.
Fixes: b1fc44eaa9ba ("pseries/iommu/ddw: Fix kdump to work in absence of ibm,dma-window") Signed-off-by: Gaurav Batra <gbatra@linux.vnet.ibm.com> Reviewed-by: Brian King <brking@linux.vnet.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://msgid.link/20240125203017.61014-1-gbatra@linux.ibm.com Signed-off-by: Sasha Levin <sashal@kernel.org>
Devicetree spec lists only dashes as valid characters for alias names.
Table 3.2: Valid characters for alias names, Devicee Specification,
Release v0.4
syzbot reported another task hung in __unix_gc(). [0]
The current while loop assumes that all of the left candidates
have oob_skb and calling kfree_skb(oob_skb) releases the remaining
candidates.
However, I missed a case that oob_skb has self-referencing fd and
another fd and the latter sk is placed before the former in the
candidate list. Then, the while loop never proceeds, resulting
the task hung.
__unix_gc() has the same loop just before purging the collected skb,
so we can call kfree_skb(oob_skb) there and let __skb_queue_purge()
release all inflight sockets.
When KASLR is enabled, the KASLR_FLAG bit in boot_params->hdr.loadflags
should be set to 1 to propagate KASLR status from compressed kernel to
kernel, just as the choose_random_location() function does.
Currently, when the kernel is booted via the EFI stub, the KASLR_FLAG
bit in boot_params->hdr.loadflags is not set, even though it should be.
This causes some functions, such as kernel_randomize_memory(), not to
execute as expected. Fix it.
Fixes: a1b87d54f4e4 ("x86/efistub: Avoid legacy decompressor when doing EFI boot") Signed-off-by: Yuntao Wang <ytcoode@gmail.com>
[ardb: drop 'else' branch clearing KASLR_FLAG] Signed-off-by: Ard Biesheuvel <ardb@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Now that the x86 EFI stub calls into some APIs exposed by the
decompressor (e.g., kaslr_get_random_long()), it is necessary to ensure
that the global boot_params variable is set correctly before doing so.
The x86 decompressor is built and linked as a separate executable, but
it shares components with the kernel proper, which are either #include'd
as C files, or linked into the decompresor as a static library (e.g, the
EFI stub)
Both the kernel itself and the decompressor define a global symbol
'boot_params' to refer to the boot_params struct, but in the former
case, it refers to the struct directly, whereas in the decompressor, it
refers to a global pointer variable referring to the struct boot_params
passed by the bootloader or constructed from scratch.
This ambiguity is unfortunate, and makes it impossible to assign this
decompressor variable from the x86 EFI stub, given that declaring it as
extern results in a clash. So rename the decompressor version (whose
scope is limited) to boot_params_ptr.
[ mingo: Renamed 'boot_params_p' to 'boot_params_ptr' for clarity ]
The EFI stub's kernel placement logic randomizes the physical placement
of the kernel by taking all available memory into account, and picking a
region at random, based on a random seed.
When KASLR is disabled, this seed is set to 0x0, and this results in the
lowest available region of memory to be selected for loading the kernel,
even if this is below LOAD_PHYSICAL_ADDR. Some of this memory is
typically reserved for the GFP_DMA region, to accommodate masters that
can only access the first 16 MiB of system memory.
Even if such devices are rare these days, we may still end up with a
warning in the kernel log, as reported by Tom:
River reports boot hangs with v6.6 and v6.7, and the bisect points to
commit
a1b87d54f4e4 ("x86/efistub: Avoid legacy decompressor when doing EFI boot")
which moves the memory allocation and kernel decompression from the
legacy decompressor (which executes *after* ExitBootServices()) to the
EFI stub, using boot services for allocating the memory. The memory
allocation succeeds but the subsequent call to decompress_kernel() never
returns, resulting in a failed boot and a hanging system.
As it turns out, this issue only occurs when physical address
randomization (KASLR) is enabled, and given that this is a feature we
can live without (virtual KASLR is much more important), let's disable
the physical part of KASLR when booting on AMI UEFI firmware claiming to
implement revision v2.0 of the specification (which was released in
2006), as this is the version these systems advertise.
The bare metal decompressor code was never really intended to run in a
hosted environment such as the EFI boot services, and does a few things
that are becoming problematic in the context of EFI boot now that the
logo requirements are getting tighter: EFI executables will no longer be
allowed to consist of a single executable section that is mapped with
read, write and execute permissions if they are intended for use in a
context where Secure Boot is enabled (and where Microsoft's set of
certificates is used, i.e., every x86 PC built to run Windows).
To avoid stepping on reserved memory before having inspected the E820
tables, and to ensure the correct placement when running a kernel build
that is non-relocatable, the bare metal decompressor moves its own
executable image to the end of the allocation that was reserved for it,
in order to perform the decompression in place. This means the region in
question requires both write and execute permissions, which either need
to be given upfront (which EFI will no longer permit), or need to be
applied on demand using the existing page fault handling framework.
However, the physical placement of the kernel is usually randomized
anyway, and even if it isn't, a dedicated decompression output buffer
can be allocated anywhere in memory using EFI APIs when still running in
the boot services, given that EFI support already implies a relocatable
kernel. This means that decompression in place is never necessary, nor
is moving the compressed image from one end to the other.
Since EFI already maps all of memory 1:1, it is also unnecessary to
create new page tables or handle page faults when decompressing the
kernel. That means there is also no need to replace the special
exception handlers for SEV. Generally, there is little need to do
any of the things that the decompressor does beyond
- initialize SEV encryption, if needed,
- perform the 4/5 level paging switch, if needed,
- decompress the kernel
- relocate the kernel
So do all of this from the EFI stub code, and avoid the bare metal
decompressor altogether.
Before refactoring the EFI stub boot flow to avoid the legacy bare metal
decompressor, duplicate the SNP feature check in the EFI stub before
handing over to the kernel proper.
The SNP feature check can be performed while running under the EFI boot
services, which means it can force the boot to fail gracefully and
return an error to the bootloader if the loaded kernel does not
implement support for all the features that the hypervisor enabled.
Currently, the EFI stub relies on DXE services in some cases to clear
non-execute restrictions from page allocations that need to be
executable. This is dodgy, because DXE services are not specified by
UEFI but by PI, and they are not intended for consumption by OS loaders.
However, no alternative existed at the time.
Now, there is a new UEFI protocol that should be used instead, so if it
exists, prefer it over the DXE services calls.
Factor out the decompressor sequence that invokes the decompressor,
parses the ELF and applies the relocations so that it can be called
directly from the EFI stub.
In preparation for updating the EFI stub boot flow to avoid the bare
metal decompressor code altogether, implement the support code for
switching between 4 and 5 levels of paging before jumping to the kernel
proper.
This reuses the newly refactored trampoline that the bare metal
decompressor uses, but relies on EFI APIs to allocate 32-bit addressable
memory and remap it with the appropriate permissions. Given that the
bare metal decompressor will no longer call into the trampoline if the
number of paging levels is already set correctly, it is no longer needed
to remove NX restrictions from the memory range where this trampoline
may end up.
x86 will need to limit the kernel memory allocation to the lowest 512
MiB of memory, to match the behavior of the existing bare metal KASLR
physical randomization logic. So in preparation for that, add a limit
parameter to efi_random_alloc() and wire it up.
EFI_MEMORY_ATTRIBUTE_PROTOCOL servers as a better alternative to
DXE services for setting memory attributes in EFI Boot Services
environment. This protocol is better since it is a part of UEFI
specification itself and not UEFI PI specification like DXE
services.
Add EFI_MEMORY_ATTRIBUTE_PROTOCOL definitions.
Support mixed mode properly for its calls.
The so-called EFI handover protocol is value-add from the distros that
permits a loader to simply copy a PE kernel image into memory and call
an alternative entrypoint that is described by an embedded boot_params
structure.
Most implementations of this protocol do not bother to check the PE
header for minimum alignment, section placement, etc, and therefore also
don't clear the image's BSS, or even allocate enough memory for it.
Allocating more memory on the fly is rather difficult, but at least
clear the BSS region explicitly when entering in this manner, so that
the EFI stub code does not get confused by global variables that were
not zero-initialized correctly.
When booting in mixed mode, this BSS clearing must occur before any
global state is created, so clear it in the 32-bit asm entry point.
The native 32-bit or 64-bit EFI handover protocol entrypoint offset
relative to the respective startup_32/64 address is described in
boot_params as handover_offset, so that the special Linux/x86 aware EFI
loader can find it there.
When mixed mode is enabled, this single field has to describe this
offset for both the 32-bit and 64-bit entrypoints, so their respective
relative offsets have to be identical. Given that startup_32 and
startup_64 are 0x200 bytes apart, and the EFI handover entrypoint
resides at a fixed offset, the 32-bit and 64-bit versions of those
entrypoints must be exactly 0x200 bytes apart as well.
Currently, hard-coded fixed offsets are used to ensure this, but it is
sufficient to emit the 64-bit entrypoint 0x200 bytes after the 32-bit
one, wherever it happens to reside. This allows this code (which is now
EFI mixed mode specific) to be moved into efi_mixed.S and out of the
startup code in head_64.S.
Now that the EFI entry code in assembler is only used by the optional
and deprecated EFI handover protocol, and given that the EFI stub C code
no longer returns to it, most of it can simply be dropped.
While at it, clarify the symbol naming, by merging efi_main() and
efi_stub_entry(), making the latter the shared entry point for all
different boot modes that enter via the EFI stub.
The efi32_stub_entry() and efi64_stub_entry() names are referenced
explicitly by the tooling that populates the setup header, so these must
be retained, but can be emitted as aliases of efi_stub_entry() where
appropriate.
Add the missing sanity check to efivars_register() so that it is no
longer possible to override an already registered set of efivar ops
(without first deregistering them).
This can help debug initialisation ordering issues where drivers have so
far unknowingly been relying on overriding the generic ops.
The UEFI spec does not mention or reason about the configured size of
the virtual address space at all, but it does mention that all memory
should be identity mapped using a page size of 4 KiB.
This means that a LPA2 capable system that has any system memory outside
of the 48-bit addressable physical range and follows the spec to the
letter may serve page allocation requests from regions of memory that
the kernel cannot access unless it was built with LPA2 support and
enables it at runtime.
So let's ensure that all page allocations are limited to the 48-bit
range.
The nfs4_file table is global, so shutting it down when a containerized
nfsd is shut down is wrong and can lead to double-frees. Tear down the
nfs4_file_rhltable in nfs4_state_shutdown instead of
nfs4_state_shutdown_net.
Fixes: d47b295e8d76 ("NFSD: Use rhashtable for managing nfs4_file objects") Link: https://bugzilla.redhat.com/show_bug.cgi?id=2169017 Reported-by: JianHong Yin <jiyin@redhat.com> Signed-off-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Since nfsd4_state_shrinker_count always calls mod_delayed_work with
0 delay, we can replace delayed_work with work_struct to save some
space and overhead.
Also add the call to cancel_work after unregister the shrinker
in nfs4_state_shutdown_net.
Signed-off-by: Dai Ngo <dai.ngo@oracle.com> Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Currently the nfsd-client shrinker is registered and unregistered at
the time the nfsd module is loaded and unloaded. The problem with this
is the shrinker is being registered before all of the relevant fields
in nfsd_net are initialized when nfsd is started. This can lead to an
oops when memory is low and the shrinker is called while nfsd is not
running.
This patch moves the register/unregister of nfsd-client shrinker from
module load/unload time to nfsd startup/shutdown time.
Fixes: 44df6f439a17 ("NFSD: add delegation reaper to react to low memory condition") Reported-by: Mike Galbraith <efault@gmx.de> Signed-off-by: Dai Ngo <dai.ngo@oracle.com> Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The premise that "Once an svc thread is scheduled and executing an
RPC, no other processes will touch svc_rqst::rq_flags" is false.
svc_xprt_enqueue() examines the RQ_BUSY flag in scheduled nfsd
threads when determining which thread to wake up next.
Fixes: 9315564747cb ("NFSD: Use only RQ_DROPME to signal the need to drop a reply") Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When built with Control Flow Integrity, function prototypes between
caller and function declaration must match. These mismatches are visible
at compile time with the new -Wcast-function-type-strict in Clang[1].
There were 97 warnings produced by NFS. For example:
fs/nfsd/nfs4xdr.c:2228:17: warning: cast from '__be32 (*)(struct nfsd4_compoundargs *, struct nfsd4_access *)' (aka 'unsigned int (*)(struct nfsd4_compoundargs *, struct nfsd4_access *)') to 'nfsd4_dec' (aka 'unsigned int (*)(struct nfsd4_compoundargs *, void *)') converts to incompatible function type [-Wcast-function-type-strict]
[OP_ACCESS] = (nfsd4_dec)nfsd4_decode_access,
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The enc/dec callbacks were defined as passing "void *" as the second
argument, but were being implicitly cast to a new type. Replace the
argument with union nfsd4_op_u, and perform explicit member selection
in the function body. There are no resulting binary differences.
Changes were made mechanically using the following Coccinelle script,
with minor by-hand fixes for members that didn't already match their
existing argument name:
@find@
identifier func;
type T, opsT;
identifier ops, N;
@@
Clean up: NFSv2 has the only two usages of rpc_drop_reply in the
NFSD code base. Since NFSv2 is going away at some point, replace
these in order to simplify the "drop this reply?" check in
nfsd_dispatch().
Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
projects / people / aperard / linux.git / log