Drop the vfio_file_iommu_group() stub and instead unconditionally declare
the function to fudge around a KVM wart where KVM tries to do symbol_get()
on vfio_file_iommu_group() (and other VFIO symbols) even if CONFIG_VFIO=n.
Ensuring the symbol is always declared fixes a PPC build error when
modules are also disabled, in which case symbol_get() simply points at the
address of the symbol (with some attributes shenanigans). Because KVM
does symbol_get() instead of directly depending on VFIO, the lack of a
fully defined symbol is not problematic (ugly, but "fine").
arch/powerpc/kvm/../../../virt/kvm/vfio.c:89:7:
error: attribute declaration must precede definition [-Werror,-Wignored-attributes]
fn = symbol_get(vfio_file_iommu_group);
^
include/linux/module.h:805:60: note: expanded from macro 'symbol_get'
#define symbol_get(x) ({ extern typeof(x) x __attribute__((weak,visibility("hidden"))); &(x); })
^
include/linux/vfio.h:294:35: note: previous definition is here
static inline struct iommu_group *vfio_file_iommu_group(struct file *file)
^
arch/powerpc/kvm/../../../virt/kvm/vfio.c:89:7:
error: attribute declaration must precede definition [-Werror,-Wignored-attributes]
fn = symbol_get(vfio_file_iommu_group);
^
include/linux/module.h:805:65: note: expanded from macro 'symbol_get'
#define symbol_get(x) ({ extern typeof(x) x __attribute__((weak,visibility("hidden"))); &(x); })
^
include/linux/vfio.h:294:35: note: previous definition is here
static inline struct iommu_group *vfio_file_iommu_group(struct file *file)
^
2 errors generated.
Although KVM is firmly in the wrong (there is zero reason for KVM to build
virt/kvm/vfio.c when VFIO is disabled), fudge around the error in VFIO as
the stub is unnecessary and doesn't serve its intended purpose (KVM is the
only external user of vfio_file_iommu_group()), and there is an in-flight
series to clean up the entire KVM<->VFIO interaction, i.e. fixing this in
KVM would result in more churn in the long run, and the stub needs to go
away regardless.
Reported-by: kernel test robot <lkp@intel.com> Closes: https://lore.kernel.org/oe-kbuild-all/202308251949.5IiaV0sz-lkp@intel.com Closes: https://lore.kernel.org/oe-kbuild-all/202309030741.82aLACDG-lkp@intel.com Closes: https://lore.kernel.org/oe-kbuild-all/202309110914.QLH0LU6L-lkp@intel.com Link: https://lore.kernel.org/all/0-v1-08396538817d+13c5-vfio_kvm_kconfig_jgg@nvidia.com Link: https://lore.kernel.org/all/20230916003118.2540661-1-seanjc@google.com Cc: Nick Desaulniers <ndesaulniers@google.com> Cc: Jason Gunthorpe <jgg@nvidia.com> Tested-by: Michael Ellerman <mpe@ellerman.id.au> Fixes: c1cce6d079b8 ("vfio: Compile vfio_group infrastructure optionally") Signed-off-by: Sean Christopherson <seanjc@google.com> Reviewed-by: Jason Gunthorpe <jgg@nvidia.com> Link: https://lore.kernel.org/r/20231130001000.543240-1-seanjc@google.com Signed-off-by: Alex Williamson <alex.williamson@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Today the percpu struct vcpu_info is allocated via DEFINE_PER_CPU(),
meaning that it could cross a page boundary. In this case registering
it with the hypervisor will fail, resulting in a panic().
This can easily be fixed by using DEFINE_PER_CPU_ALIGNED() instead,
as struct vcpu_info is guaranteed to have a size of 64 bytes, matching
the cache line size of x86 64-bit processors (Xen doesn't support
32-bit processors).
This can happen if pds_vfio_put_restore_file() and/or
pds_vfio_put_save_file() grab the mutex_lock(&lm_file->lock)
while the spin_lock(&pds_vfio->reset_lock) is held, which can
happen during while calling pds_vfio_state_mutex_unlock().
Fix this by changing the reset_lock to reset_mutex so there are no such
conerns. Also, make sure to destroy the reset_mutex in the driver specific
VFIO device release function.
This also fixes a spinlock bad magic BUG that was caused
by not calling spinlock_init() on the reset_lock. Since, the lock is
being changed to a mutex, make sure to call mutex_init() on it.
Reported-by: Dan Carpenter <dan.carpenter@linaro.org> Closes: https://lore.kernel.org/kvm/1f9bc27b-3de9-4891-9687-ba2820c1b390@moroto.mountain/ Fixes: bb500dbe2ac6 ("vfio/pds: Add VFIO live migration support") Signed-off-by: Brett Creeley <brett.creeley@amd.com> Reviewed-by: Shannon Nelson <shannon.nelson@amd.com> Link: https://lore.kernel.org/r/20231122192532.25791-3-brett.creeley@amd.com Signed-off-by: Alex Williamson <alex.williamson@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
As shown, lock->magic != lock. This is because
mutex_init(&pds_vfio->state_mutex) is called in the VFIO open path. So,
if a reset is initiated before the VFIO device is opened the mutex will
have never been initialized. Fix this by calling
mutex_init(&pds_vfio->state_mutex) in the VFIO init path.
Also, don't destroy the mutex on close because the device may
be re-opened, which would cause mutex to be uninitialized. Fix this by
implementing a driver specific vfio_device_ops.release callback that
destroys the mutex before calling vfio_pci_core_release_dev().
Fixes: bb500dbe2ac6 ("vfio/pds: Add VFIO live migration support") Signed-off-by: Brett Creeley <brett.creeley@amd.com> Reviewed-by: Shannon Nelson <shannon.nelson@amd.com> Link: https://lore.kernel.org/r/20231122192532.25791-2-brett.creeley@amd.com Signed-off-by: Alex Williamson <alex.williamson@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
[Why]
Remove the brightness cache in DC. It uses a single value to represent
the brightness for both SDR and HDR mode. This leads to flash in HDR
on/off. It also unconditionally programs brightness as in HDR mode. This
may introduce garbage on SDR mode in miniLED panel.
[How]
Simplify the initialization flow by removing the DC cache and taking
what panel has as default. Expand the mechanism for PWM to DPCD Aux to
restore cached brightness value generally.
Cc: stable@vger.kernel.org # 6.1+ Reviewed-by: Krunoslav Kovac <krunoslav.kovac@amd.com> Acked-by: Hamza Mahfooz <hamza.mahfooz@amd.com> Signed-off-by: Camille Cho <camille.cho@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
[Why & How]
Currently set_default_brightness_aux function uses 5 nits as lower limit
to check for valid default_backlight setting. However some newer panels
can support even lower default settings
Reviewed-by: Agustin Gutierrez <agustin.gutierrez@amd.com> Acked-by: Roman Li <roman.li@amd.com> Signed-off-by: Swapnil Patel <swapnil.patel@amd.com> Tested-by: Daniel Wheeler <daniel.wheeler@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Stable-dep-of: d9e865826c20 ("drm/amd/display: Simplify brightness initialization") Signed-off-by: Sasha Levin <sashal@kernel.org>
The variable phys is defined as (struct resource *) which aligns with
the printk format specifier %pr. Taking the address of it results in a
value of type (struct resource **) which is incompatible with the format
specifier %pr. Therefore, remove the address of operator (&).
When amd_pstate is running, writing to scaling_min_freq and
scaling_max_freq has no effect. These values are only passed to the
policy level, but not to the platform level. This means that the
platform does not know about the frequency limits set by the user.
To fix this, update the min_perf and max_perf values at the platform
level whenever the user changes the scaling_min_freq and scaling_max_freq
values.
Fixes: ffa5096a7c33 ("cpufreq: amd-pstate: implement Pstate EPP support for the AMD processors") Acked-by: Huang Rui <ray.huang@amd.com> Signed-off-by: Wyes Karny <wyes.karny@amd.com> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
For the "starry, 2081101qfh032011-53g" panel, it is stipulated in the
panel spec that MIPI needs to keep the LP11 state before the
lcm_reset pin is pulled high.
The GSC CS is not exposed to the user, so we skipped assigning a uabi
class number for it. However, the trace logs use the uabi class and
instance to identify the engine, so leaving uabi class unset makes the
GSC CS show up as the RCS in those logs.
Given that the engine is not exposed to the user, we can't add a new
case in the uabi enum, so we insted internally define a kernel
internal class as -1.
At the same time remove special handling for the name and complete
the uabi_classes array so internal class is automatically correctly
assigned.
Engine will show as 65535:0 other0 in the logs/traces which should
be unique enough.
v2:
* Fix uabi class u8 vs u16 type confusion.
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Fixes: 194babe26bdc ("drm/i915/mtl: don't expose GSC command streamer to the user") Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com> Cc: Alan Previn <alan.previn.teres.alexis@intel.com> Cc: Matt Roper <matthew.d.roper@intel.com> Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20231116084456.291533-1-tvrtko.ursulin@linux.intel.com
(cherry picked from commit dfed6b58d54f3a5d7e6bc1fb060e2c936330eba2) Signed-off-by: Jani Nikula <jani.nikula@intel.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
In the iommu probe_device path, domain_context_mapping() allows setting
up the context entry for a non-PCI device. However, in the iommu
release_device path, domain_context_clear() only clears context entries
for PCI devices.
Make domain_context_clear() behave consistently with
domain_context_mapping() by clearing context entries for both PCI and
non-PCI devices.
Fixes: 579305f75d34 ("iommu/vt-d: Update to use PCI DMA aliases") Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com> Reviewed-by: Kevin Tian <kevin.tian@intel.com> Link: https://lore.kernel.org/r/20231114011036.70142-4-baolu.lu@linux.intel.com Signed-off-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Sasha Levin <sashal@kernel.org>
When IOMMU hardware operates in legacy mode, the TT field of the context
entry determines the translation type, with three supported types (Section
9.3 Context Entry):
- DMA translation without device TLB support
- DMA translation with device TLB support
- Passthrough mode with translated and translation requests blocked
Device TLB support is absent when hardware is configured in passthrough
mode.
Disable the PCI ATS feature when IOMMU is configured for passthrough
translation type in legacy (non-scalable) mode.
Fixes: 0faa19a1515f ("iommu/vt-d: Decouple PASID & PRI enabling from SVA") Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com> Reviewed-by: Kevin Tian <kevin.tian@intel.com> Link: https://lore.kernel.org/r/20231114011036.70142-3-baolu.lu@linux.intel.com Signed-off-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Sasha Levin <sashal@kernel.org>
The latest VT-d spec indicates that when remapping hardware is disabled
(TES=0 in Global Status Register), upstream ATS Invalidation Completion
requests are treated as UR (Unsupported Request).
Consequently, the spec recommends in section 4.3 Handling of Device-TLB
Invalidations that software refrain from submitting any Device-TLB
invalidation requests when address remapping hardware is disabled.
Verify address remapping hardware is enabled prior to submitting Device-
TLB invalidation requests.
Fixes: 792fb43ce2c9 ("iommu/vt-d: Enable Intel IOMMU scalable mode by default") Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com> Reviewed-by: Kevin Tian <kevin.tian@intel.com> Link: https://lore.kernel.org/r/20231114011036.70142-2-baolu.lu@linux.intel.com Signed-off-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Sasha Levin <sashal@kernel.org>
For a 900MHz i.MX6ULL CPU the 792MHz OPP is disabled. There is no
convincing reason to disable this OPP. If a CPU can run at 900MHz,
it should also be able to cope with 792MHz. Looking at the voltage
level of 792MHz in [1] (page 24, table 10. "Operating Ranges") the
current defined OPP is above the minimum. So the voltage level
shouldn't be a problem. However in [2] (page 24, table 10.
"Operating Ranges"), it is not mentioned that 792MHz OPP isn't
allowed. Change it to only disable 792MHz OPP for i.MX6ULL types
below 792 MHz.
[Why & How]
To organize the edp power control a bit:
1. add flag in dc_link to indicate dc to skip all implicit eDP power control.
2. add edp_set_panel_power link service for DM to call.
Reviewed-by: Aric Cyr <aric.cyr@amd.com> Reviewed-by: Jun Lei <jun.lei@amd.com> Acked-by: Wayne Lin <wayne.lin@amd.com> Signed-off-by: Ian Chen <ian.chen@amd.com> Tested-by: Daniel Wheeler <daniel.wheeler@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Stable-dep-of: b0399e22ada0 ("drm/amd/display: Remove power sequencing check") Signed-off-by: Sasha Levin <sashal@kernel.org>
If the cmma no-dat feature is available the kernel page tables are walked
to identify and mark all pages which are used for address translation (all
region, segment, and page tables). In a subsequent loop all other pages are
marked as "no-dat" pages with the ESSA instruction.
This information is visible to the hypervisor, so that the hypervisor can
optimize purging of guest TLB entries. All pages used for swapper_pg_dir
and invalid_pg_dir are incorrectly marked as no-dat, which in turn can
result in incorrect guest TLB flushes.
Fix this by marking those pages correctly as being used for DAT.
When a device is initialized, the driver invokes dma_supported() twice -
first for streaming mappings followed by coherent mappings. For an
SR-IOV device, default window is deleted and DDW created. With vPMEM
enabled, TCE mappings are dynamically created for both vPMEM and SR-IOV
device. There are no direct mappings.
First time when dma_supported() is called with 64 bit mask, DDW is created
and marked as dynamic window. The second time dma_supported() is called,
enable_ddw() finds existing window for the device and incorrectly returns
it as "direct mapping".
This only happens when size of DDW is big enough to map max LPAR memory.
This results in streaming TCEs to not get dynamically mapped, since code
incorrently assumes these are already pre-mapped. The adapter initially
comes up but goes down due to EEH.
Fixes: 381ceda88c4c ("powerpc/pseries/iommu: Make use of DDW for indirect mapping") Cc: stable@vger.kernel.org # v5.15+ Signed-off-by: Gaurav Batra <gbatra@linux.vnet.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://msgid.link/20231003030802.47914-1-gbatra@linux.vnet.ibm.com Signed-off-by: Sasha Levin <sashal@kernel.org>
On RZ/G3S SMARC Carrier II board having RGMII connections b/w Ethernet
MACs and PHYs it has been discovered that doing unbind/bind for ravb
driver in a loop leads to wrong speed and duplex for Ethernet links and
broken connectivity (the connectivity cannot be restored even with
bringing interface down/up). Before doing unbind/bind the Ethernet
interfaces were configured though systemd. The sh instructions used to
do unbind/bind were:
$ cd /sys/bus/platform/drivers/ravb/
$ while :; do echo 11c30000.ethernet > unbind ; \
echo 11c30000.ethernet > bind; done
It has been discovered that there is a race b/w IOCTLs initialized by
systemd at the response of success binding and the
"ravb_write(ndev, CCC_OPC_RESET, CCC)" call in ravb_remove() as
follows:
1/ as a result of bind success the user space open/configures the
interfaces tough an IOCTL; the following stack trace has been
identified on RZ/G3S:
2/ this call may execute concurrently with ravb_remove() as the
unbind/bind operation was executed in a loop
3/ if the operation mode is changed to RESET (through
ravb_write(ndev, CCC_OPC_RESET, CCC) call in ravb_remove())
while the above ravb_open() is in progress it may lead to MAC
(or PHY, or MAC-PHY connection, the right point hasn't been identified
at the moment) to be broken, thus the Ethernet connectivity fails to
restore.
The simple fix for this is to move ravb_write(ndev, CCC_OPC_RESET, CCC))
after unregister_netdev() to avoid resetting the controller while the
netdev interface is still registered.
To avoid future issues in ravb_remove(), the patch follows the proper order
of operations in ravb_remove(): reverse order compared with ravb_probe().
This avoids described races as the IOCTLs as well as unregister_netdev()
(called now at the beginning of ravb_remove()) calls rtnl_lock() before
continuing and IOCTLs check (though devinet_ioctl()) if device is still
registered just after taking the lock:
int devinet_ioctl(struct net *net, unsigned int cmd, struct ifreq *ifr)
{
// ...
rtnl_lock();
ret = -ENODEV;
dev = __dev_get_by_name(net, ifr->ifr_name);
if (!dev)
goto done;
In case ravb_phy_start() returns with error the settings applied in
ravb_dmac_init() are not reverted (e.g. config mode). For this call
ravb_stop_dma() on failure path of ravb_open().
ravb_phy_start() may fail. If that happens, the TX queues will remain
started. Thus, move the netif_tx_start_all_queues() after PHY is
successfully initialized.
Hardware manual of RZ/G3S (and RZ/G2L) specifies the following on the
description of CXR35 register (chapter "PHY interface select register
(CXR35)"): "After release reset, make write-access to this register before
making write-access to other registers (except MDIOMOD). Even if not need
to change the value of this register, make write-access to this register
at least one time. Because RGMII/MII MODE is recognized by accessing this
register".
The setup procedure for EMAC module (chapter "Setup procedure" of RZ/G3S,
RZ/G2L manuals) specifies the E-MAC.CXR35 register is the first EMAC
register that is to be configured.
Note [A] from chapter "PHY interface select register (CXR35)" specifies
the following:
[A] The case which CXR35 SEL_XMII is used for the selection of RGMII/MII
in APB Clock 100 MHz.
(1) To use RGMII interface, Set ‘H’03E8_0000’ to this register.
(2) To use MII interface, Set ‘H’03E8_0002’ to this register.
pm_runtime_get_sync() may return an error. In case it returns with an error
dev->power.usage_count needs to be decremented. pm_runtime_resume_and_get()
takes care of this. Thus use it.
reset_control_deassert() could return an error. Some devices cannot work
if reset signal de-assert operation fails. To avoid this check the return
code of reset_control_deassert() in ravb_probe() and take proper action.
Along with it, the free_netdev() call from the error path was moved after
reset_control_assert() on its own label (out_free_netdev) to free
netdev in case reset_control_deassert() fails.
There is an error when an interface has the following conditions:
- PF is in an aggregate (bond)
- PF has VFs created on it
- bond is in a state where it is failed-over to the secondary interface
- A VF reset is issued on one or more of those VFs
The issue is generated by the originating PF trying to rebuild or
reconfigure the VF resources. Since the bond is failed over to the
secondary interface the queue contexts are in a modified state.
To fix this issue, have the originating interface reclaim its resources
prior to the tear-down and rebuild or reconfigure. Then after the process
is complete, move the resources back to the currently active interface.
There are multiple paths that can be used depending on what triggered the
event, so create a helper function to move the queues and use paired calls
to the helper (back to origin, process, then move back to active interface)
under the same lag_mutex lock.
Fixes: 1e0f9881ef79 ("ice: Flesh out implementation of support for SRIOV on bonded interface") Signed-off-by: Dave Ertman <david.m.ertman@intel.com> Tested-by: Sujai Buvaneswaran <sujai.buvaneswaran@intel.com> Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com> Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com> Link: https://lore.kernel.org/r/20231127212340.1137657-1-anthony.l.nguyen@intel.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
AF_UNIX stream sockets are a paired socket. So sending on one of the pairs
will lookup the paired socket as part of the send operation. It is possible
however to put just one of the pairs in a BPF map. This currently increments
the refcnt on the sock in the sockmap to ensure it is not free'd by the
stack before sockmap cleans up its state and stops any skbs being sent/recv'd
to that socket.
But we missed a case. If the peer socket is closed it will be free'd by the
stack. However, the paired socket can still be referenced from BPF sockmap
side because we hold a reference there. Then if we are sending traffic through
BPF sockmap to that socket it will try to dereference the free'd pair in its
send logic creating a use after free. And following splat:
To fix let BPF sockmap hold a refcnt on both the socket in the sockmap and its
paired socket. It wasn't obvious how to contain the fix to bpf_unix logic. The
primarily problem with keeping this logic in bpf_unix was: In the sock close()
we could handle the deref by having a close handler. But, when we are destroying
the psock through a map delete operation we wouldn't have gotten any signal
thorugh the proto struct other than it being replaced. If we do the deref from
the proto replace its too early because we need to deref the sk_pair after the
backlog worker has been stopped.
Given all this it seems best to just cache it at the end of the psock and eat 8B
for the af_unix and vsock users. Notice dgram sockets are OK because they handle
locking already.
Fixes: 94531cfcbe79 ("af_unix: Add unix_stream_proto for sockmap") Signed-off-by: John Fastabend <john.fastabend@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Reviewed-by: Jakub Sitnicki <jakub@cloudflare.com> Link: https://lore.kernel.org/bpf/20231129012557.95371-2-john.fastabend@gmail.com Signed-off-by: Sasha Levin <sashal@kernel.org>
The default dump handler needs to clear ret before returning.
Otherwise if the last interface returns an inconsequential
error this error will propagate to user space.
This may confuse user space (ethtool CLI seems to ignore it,
but YNL doesn't). It will also terminate the dump early
for mutli-skb dump, because netlink core treats EOPNOTSUPP
as a real error.
Fixes: 728480f12442 ("ethtool: default handlers for GET requests") Reviewed-by: Simon Horman <horms@kernel.org> Link: https://lore.kernel.org/r/20231126225806.2143528-1-kuba@kernel.org Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Fix races between ravb_tx_timeout_work() and functions of net_device_ops
and ethtool_ops by using rtnl_trylock() and rtnl_unlock(). Note that
since ravb_close() is under the rtnl lock and calls cancel_work_sync(),
ravb_tx_timeout_work() should calls rtnl_trylock(). Otherwise, a deadlock
may happen in ravb_tx_timeout_work() like below:
CPU0 CPU1
ravb_tx_timeout()
schedule_work()
...
__dev_close_many()
// Under rtnl lock
ravb_close()
cancel_work_sync()
// Waiting
ravb_tx_timeout_work()
rtnl_lock()
// This is possible to cause a deadlock
If rtnl_trylock() fails, rescheduling the work with sleep for 1 msec.
ndo_stop() is RTNL-protected by net core, and the worker function takes
RTNL as well. Therefore we will deadlock when trying to execute a
pending work synchronously. To fix this execute any pending work
asynchronously. This will do no harm because netif_running() is false
in ndo_stop(), and therefore the work function is effectively a no-op.
However we have to ensure that no task is running or pending after
rtl_remove_one(), therefore add a call to cancel_work_sync().
Fixes: abe5fc42f9ce ("r8169: use RTNL to protect critical sections") Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com> Link: https://lore.kernel.org/r/12395867-1d17-4cac-aa7d-c691938fcddf@gmail.com Signed-off-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
When a task needs to accept memory it will scan the accepting_list
to see if any ranges already being processed by other tasks overlap
with its range. Due to an off-by-one in the range comparisons, a task
might falsely determine that an overlapping range is being accepted,
leading to an unnecessary delay before it begins processing the range.
Fix the off-by-one in the range comparison to prevent this and slightly
improve performance.
Previously, one-element and zero-length arrays were treated as true
flexible arrays, even though they are actually "fake" flex arrays.
The __randomize_layout would leave them untouched at the end of the
struct, similarly to proper C99 flex-array members.
However, this approach changed with commit 1ee60356c2dc ("gcc-plugins:
randstruct: Only warn about true flexible arrays"). Now, only C99
flexible-array members will remain untouched at the end of the struct,
while one-element and zero-length arrays will be subject to randomization.
Fix a `__randomize_layout` crash in `struct neighbour` by transforming
zero-length array `primary_key` into a proper C99 flexible-array member.
Fixes: 1ee60356c2dc ("gcc-plugins: randstruct: Only warn about true flexible arrays") Closes: https://lore.kernel.org/linux-hardening/20231124102458.GB1503258@e124191.cambridge.arm.com/ Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org> Reviewed-by: Kees Cook <keescook@chromium.org> Tested-by: Joey Gouly <joey.gouly@arm.com> Link: https://lore.kernel.org/r/ZWJoRsJGnCPdJ3+2@work Signed-off-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
TC ingress policer rules depends on interface receive queue
contexts since the bandwidth profiles are attached to RQ
contexts. When an interface is brought down all the queue
contexts are freed. This in turn frees bandwidth profiles in
hardware causing ingress police rules non-functional after
the interface is brought up. Fix this by applying all the ingress
police rules config to hardware in otx2_open. Also allow
adding ingress rules only when interface is running
since no contexts exist for the interface when it is down.
When more than 64 VFs are enabled for a PF then mbox communication
between VF and PF is not working as mbox work queueing for few VFs
are skipped due to wrong calculation of VF numbers.
Commit aeb18dd07692 ("net: stmmac: xgmac: Disable MMC interrupts
by default") tries to disable MMC interrupts to avoid a storm of
unhandled interrupts, but leaves the FPE(Frame Preemption) MMC
interrupts enabled, FPE MMC interrupts can cause the same problem.
Now we mask FPE TX and RX interrupts to disable all MMC interrupts.
A loop in rvu_mbox_handler_nix_bandprof_free() contains
a break if (idx == MAX_BANDPROF_PER_PFFUNC),
but if idx may reach MAX_BANDPROF_PER_PFFUNC
buffer '(*req->prof_idx)[layer]' overflow happens before that check.
The patch moves the break to the
beginning of the loop.
Found by Linux Verification Center (linuxtesting.org) with SVACE.
Fixes: e8e095b3b370 ("octeontx2-af: cn10k: Bandwidth profiles config support"). Signed-off-by: Elena Salomatkina <elena.salomatkina.cmc@gmail.com> Reviewed-by: Simon Horman <horms@kernel.org> Reviewed-by: Subbaraya Sundeep <sbhatta@marvell.com> Link: https://lore.kernel.org/r/20231124210802.109763-1-elena.salomatkina.cmc@gmail.com Signed-off-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Same init_rng() in both tests. The function reads /dev/urandom to
initialize srand(). In case of failure, it falls back onto the
entropy in the uninitialized variable. Not sure if this is on purpose.
But failure reading urandom should be rare, so just fail hard. While
at it, convert to getrandom(). Which man 4 random suggests is simpler
and more robust.
mptcp_inq.c:525:6:
mptcp_connect.c:1131:6:
error: variable 'foo' is used uninitialized
whenever 'if' condition is false
[-Werror,-Wsometimes-uninitialized]
Fixes: 048d19d444be ("mptcp: add basic kselftest for mptcp") Fixes: b51880568f20 ("selftests: mptcp: add inq test case") Cc: Florian Westphal <fw@strlen.de> Signed-off-by: Willem de Bruijn <willemb@google.com>
----
When input is randomized because this is expected to meaningfully
explore edge cases, should we also add
1. logging the random seed to stdout and
2. adding a command line argument to replay from a specific seed
I can do this in net-next, if authors find it useful in this case. Reviewed-by: Matthieu Baerts <matttbe@kernel.org> Link: https://lore.kernel.org/r/20231124171645.1011043-5-willemdebruijn.kernel@gmail.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Signedness of char is signed on x86_64, but unsigned on arm64.
Fix the warning building cmsg_sender.c on signed platforms or
forced with -fsigned-char:
msg_sender.c:455:12:
error: implicit conversion from 'int' to 'char'
changes value from 128 to -128
[-Werror,-Wconstant-conversion]
buf[0] = ICMPV6_ECHO_REQUEST;
nr_process must be a signed long: it is assigned a signed long by
strtol() and is compared against LONG_MIN and LONG_MAX.
ipsec.c:2280:65:
error: result of comparison of constant -9223372036854775808
with expression of type 'unsigned int' is always false
[-Werror,-Wtautological-constant-out-of-range-compare]
Recently the kernel test robot has reported an ARM-specific BUILD_BUG_ON()
in an old and unmaintained wil6210 wireless driver. The problem comes from
the structure packing rules of old ARM ABI ('-mabi=apcs-gnu'). For example,
the following structure is packed to 18 bytes instead of 16:
struct poorly_packed {
unsigned int a;
unsigned int b;
unsigned short c;
union {
struct {
unsigned short d;
unsigned int e;
} __attribute__((packed));
struct {
unsigned short d;
unsigned int e;
} __attribute__((packed)) inner;
};
} __attribute__((packed));
To fit it into 16 bytes, it's required to add packed attribute to the
container union as well:
struct poorly_packed {
unsigned int a;
unsigned int b;
unsigned short c;
union {
struct {
unsigned short d;
unsigned int e;
} __attribute__((packed));
struct {
unsigned short d;
unsigned int e;
} __attribute__((packed)) inner;
} __attribute__((packed));
} __attribute__((packed));
Thanks to Andrew Pinski of GCC team for sorting the things out at
https://gcc.gnu.org/pipermail/gcc/2023-November/242888.html.
bpf_mem_cache_alloc_flags() may call __alloc() directly when there is no
free object in free list, but it doesn't initialize the allocation hint
for the returned pointer. It may lead to bad memory dereference when
freeing the pointer, so fix it by initializing the allocation hint.
Fixes: 822fb26bdb55 ("bpf: Add a hint to allocated objects.") Signed-off-by: Hou Tao <houtao1@huawei.com> Acked-by: Yonghong Song <yonghong.song@linux.dev> Link: https://lore.kernel.org/r/20231111043821.2258513-1-houtao@huaweicloud.com Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
The blamed commit added support for Rx copybreak. This meant that for
certain frame sizes, a new skb was allocated and the initial data buffer
was recycled. Instead of waiting to recycle the Rx buffer only after all
processing was done on it (like accessing the parse results or timestamp
information), the code path just went ahead and re-used the buffer right
away.
This sometimes lead to corrupted HW and SW annotation areas.
Fix this by delaying the moment when the buffer is recycled.
Fixes: 50f826999a80 ("dpaa2-eth: add rx copybreak support") Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
Increase the needed headroom to account for a 64 byte alignment
restriction which, with this patch, we make mandatory on the Tx path.
The case in which the amount of headroom needed is not available is
already handled by the driver which instead sends a S/G frame with the
first buffer only holding the SW and HW annotation areas.
Without this patch, we can empirically see data corruption happening
between Tx and Tx confirmation which sometimes leads to the SW
annotation area being overwritten.
Since this is an old IP where the hardware team cannot help to
understand the underlying behavior, we make the Tx alignment mandatory
for all frames to avoid the crash on Tx conf. Also, remove the comment
that suggested that this is just an optimization.
This patch also sets the needed_headroom net device field to the usual
value that the driver would need on the Tx path:
- 64 bytes for the software annotation area
- 64 bytes to account for a 64 byte aligned buffer address
Fixes: 6e2387e8f19e ("staging: fsl-dpaa2/eth: Add Freescale DPAA2 Ethernet driver") Closes: https://lore.kernel.org/netdev/aa784d0c-85eb-4e5d-968b-c8f74fa86be6@gin.de/ Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
As of commit b92143d4420f ("net: dsa: mv88e6xxx: add infrastructure for
phylink_pcs") probing of a Marvell 88e6350 switch causes a NULL pointer
de-reference like this example:
...
mv88e6085 d0072004.mdio-mii:11: switch 0x3710 detected: Marvell 88E6350, revision 2
8<--- cut here ---
Unable to handle kernel NULL pointer dereference at virtual address 00000000 when read
[00000000] *pgd=00000000
Internal error: Oops: 5 [#1] ARM
Modules linked in:
CPU: 0 PID: 8 Comm: kworker/u2:0 Not tainted 6.7.0-rc2-dirty #26
Hardware name: Marvell Armada 370/XP (Device Tree)
Workqueue: events_unbound deferred_probe_work_func
PC is at mv88e6xxx_port_setup+0x1c/0x44
LR is at dsa_port_devlink_setup+0x74/0x154
pc : [<c057ea24>] lr : [<c0819598>] psr: a0000013
sp : c184fce0 ip : c542b8f4 fp : 00000000
r10: 00000001 r9 : c542a540 r8 : c542bc00
r7 : c542b838 r6 : c5244580 r5 : 00000005 r4 : c5244580
r3 : 00000000 r2 : c542b840 r1 : 00000005 r0 : c1a02040
...
The Marvell 6350 switch has no SERDES interface and so has no
corresponding pcs_ops defined for it. But during probing a call is made
to mv88e6xxx_port_setup() which unconditionally expects pcs_ops to exist -
though the presence of the pcs_ops->pcs_init function is optional.
Modify code to check for pcs_ops first, before checking for and calling
pcs_ops->pcs_init. Modify checking and use of pcs_ops->pcs_teardown
which may potentially suffer the same problem.
Fixes: b92143d4420f ("net: dsa: mv88e6xxx: add infrastructure for phylink_pcs") Signed-off-by: Greg Ungerer <gerg@kernel.org> Reviewed-by: Andrew Lunn <andrew@lunn.ch> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
The problem stems from the use of mv88e6185_phylink_get_caps() to get
the device capabilities. Create a new dedicated phylink_get_caps for the
6351 family (which the 6350 is one of) to properly support their set of
capabilities.
According to chip.h the 6351 switch family includes the 6171, 6175, 6350
and 6351 switches, so update each of these to use the correct
phylink_get_caps.
Fixes: de5c9bf40c45 ("net: phylink: require supported_interfaces to be filled") Signed-off-by: Greg Ungerer <gerg@kernel.org> Reviewed-by: Andrew Lunn <andrew@lunn.ch> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
This fixes WARN_ONs when using AP_VLANs after station removal. The flush
call passed AP_VLAN vif to driver, but because these vifs are virtual and
not registered with drivers, we need to translate to the correct AP vif
first.
Closes: https://github.com/openwrt/openwrt/issues/12420 Fixes: 0b75a1b1e42e ("wifi: mac80211: flush queues on STA removal") Fixes: d00800a289c9 ("wifi: mac80211: add flush_sta method") Tested-by: Konstantin Demin <rockdrilla@gmail.com> Tested-by: Koen Vandeputte <koen.vandeputte@citymesh.com> Signed-off-by: Oldřich Jedlička <oldium.pro@gmail.com> Link: https://lore.kernel.org/r/20231104141333.3710-1-oldium.pro@gmail.com Signed-off-by: Johannes Berg <johannes.berg@intel.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
The root causes are as follows:
Thread A Thread B
... netif_receive_skb
br_dev_stop ...
br_multicast_leave_snoopers ...
__ip_mc_dec_group ...
__igmp_group_dropped igmp_rcv
igmp_stop_timer igmp_heard_query //ref = 1
ip_ma_put igmp_mod_timer
refcount_dec_and_test igmp_start_timer //ref = 0
... refcount_inc //ref increases from 0
When the device receives an IGMPv2 Query message, it starts the timer
immediately, regardless of whether the device is running. If the device is
down and has left the multicast group, it will cause the mc list refcount
uaf issue.
Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2") Signed-off-by: Zhengchao Shao <shaozhengchao@huawei.com> Reviewed-by: Eric Dumazet <edumazet@google.com> Reviewed-by: Hangbin Liu <liuhangbin@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
The type of ret in rswitch_start_xmit() should be netdev_tx_t. So,
fix it.
Fixes: 3590918b5d07 ("net: ethernet: renesas: Add support for "Ethernet Switch"") Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com> Reviewed-by: Wojciech Drewek <wojciech.drewek@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
Commit 2b3486bc2d23 ("bpf: Introduce device-bound XDP programs") introduced
device-bound programs by largely reusing existing offloading infrastructure.
This changed the semantics of 'prog->aux->offload' a bit. Now, it's non-NULL
for both offloaded and device-bound programs.
Instead of looking at 'prog->aux->offload' let's call bpf_prog_is_offloaded
which should be true iff the program is offloaded and not merely device-bound.
The "client_cap->capabilities" variable is a u64. The AND operation
is supposed to clear out the V4L2_SUBDEV_CLIENT_CAP_STREAMS flag. But
because it's a 32 bit variable it accidentally clears out the high 32
bits as well.
Currently we only use the first bit and none of the upper bits so this
doesn't affect runtime behavior.
Fixes: f57fa2959244 ("media: v4l2-subdev: Add new ioctl for client capabilities") Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org> Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by: Sasha Levin <sashal@kernel.org>
My prior race fix here broke CQM when ranges aren't used, as
the reporting worker now requires the cqm_config to be set in
the wdev, but isn't set when there's no range configured.
Rather than continuing to special-case the range version, set
the cqm_config always and configure accordingly, also tracking
if range was used or not to be able to clear the configuration
appropriately with the same API, which was actually not right
if both were implemented by a driver for some reason, as is
the case with mac80211 (though there the implementations are
equivalent so it doesn't matter.)
Also, the original multiple-RSSI commit lost checking for the
callback, so might have potentially crashed if a driver had
neither implementation, and userspace tried to use it despite
not being advertised as supported.
Right now we stash any potentially mmap'ed provided ring buffer range
for freeing at release time, regardless of when they get unregistered.
Since we're keeping track of these ranges anyway, keep track of their
registration state as well, and use that to recycle ranges when
appropriate rather than always allocate new ones.
The lookup is a basic scan of entries, checking for the best matching
free entry.
If a provided buffer ring is setup with IOU_PBUF_RING_MMAP, then the
kernel allocates the memory for it and the application is expected to
mmap(2) this memory. However, io_uring uses remap_pfn_range() for this
operation, so we cannot rely on normal munmap/release on freeing them
for us.
Stash an io_buf_free entry away for each of these, if any, and provide
a helper to free them post ->release().
Cc: stable@vger.kernel.org Fixes: c56e022c0a27 ("io_uring: add support for user mapped provided buffer ring") Reported-by: Jann Horn <jannh@google.com> Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When the send protocol versioning was added in 5.16 e77fbf990316
("btrfs: send: prepare for v2 protocol"), the 32/64bit compat code was
not updated (added by 2351f431f727 ("btrfs: fix send ioctl on 32bit with
64bit kernel")), missing the version struct member. The compat code is
probably rarely used, nobody reported any bugs.
Found by tool https://github.com/jirislaby/clang-struct .
Fixes: e77fbf990316 ("btrfs: send: prepare for v2 protocol") CC: stable@vger.kernel.org # 6.1+ Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[BUG]
If btrfs_alloc_page_array() fail to allocate all pages but part of the
slots, then the partially allocated pages would be leaked in function
btrfs_submit_compressed_read().
[CAUSE]
As explicitly stated, if btrfs_alloc_page_array() returned -ENOMEM,
caller is responsible to free the partially allocated pages.
For the existing call sites, most of them are fine:
- btrfs_raid_bio::stripe_pages
Handled by free_raid_bio().
- scrub_stripe::pages[]
Handled by release_scrub_stripe().
But there is one exception in btrfs_submit_compressed_read(), if
btrfs_alloc_page_array() failed, we didn't cleanup the array and freed
the array pointer directly.
Initially there is still the error handling in commit dd137dd1f2d7
("btrfs: factor out allocating an array of pages"), but later in commit 544fe4a903ce ("btrfs: embed a btrfs_bio into struct compressed_bio"),
the error handling is removed, leading to the possible memory leak.
[FIX]
This patch would add back the error handling first, then to prevent such
situation from happening again, also
Make btrfs_alloc_page_array() to free the allocated pages as a extra
safety net, then we don't need to add the error handling to
btrfs_submit_compressed_read().
Fixes: 544fe4a903ce ("btrfs: embed a btrfs_bio into struct compressed_bio") CC: stable@vger.kernel.org # 6.4+ Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When getting a chunk map, at btrfs_get_chunk_map(), we do some sanity
checks to verify we found a chunk map and that map found covers the
logical address the caller passed in. However the messages aren't very
clear in the sense that don't mention the issue is with a chunk map and
one of them prints the 'length' argument as if it were the end offset of
the requested range (while the in the string format we use %llu-%llu
which suggests a range, and the second %llu-%llu is actually a range for
the chunk map). So improve these two details in the error messages.
CC: stable@vger.kernel.org # 5.4+ Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
kernel_write() requires the caller to ensure that the file is writable.
Let's do that directly after looking up the ->send_fd.
We don't need a separate bailout path because the "out" path already
does fput() if ->send_filp is non-NULL.
This has no security impact for two reasons:
- the ioctl requires CAP_SYS_ADMIN
- __kernel_write() bails out on read-only files - but only since 5.8,
see commit a01ac27be472 ("fs: check FMODE_WRITE in __kernel_write")
At btrfs_get_chunk_map() we get the extent map for the chunk that contains
the given logical address stored in the 'logical' argument. Then we do
sanity checks to verify the extent map contains the logical address. One
of these checks verifies if the extent map covers a range with an end
offset behind the target logical address - however this check has an
off-by-one error since it will consider an extent map whose start offset
plus its length matches the target logical address as inclusive, while
the fact is that the last byte it covers is behind the target logical
address (by 1).
So fix this condition by using '<=' rather than '<' when comparing the
extent map's "start + length" against the target logical address.
CC: stable@vger.kernel.org # 4.14+ Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In btrfs_ref_tree_mod(), when !parent 're' was allocated through
kmalloc(). In the following code, if an error occurs, the execution will
be redirected to 'out' or 'out_unlock' and the function will be exited.
However, on some of the paths, 're' are not deallocated and may lead to
memory leaks.
For example: lookup_block_entry() for 'be' returns NULL, the out label
will be invoked. During that flow ref and 'ra' are freed but not 're',
which can potentially lead to a memory leak.
There is a feature request to add dmesg output when unmounting a btrfs.
There are several alternative methods to do the same thing, but with
their own problems:
- Use eBPF to watch btrfs_put_super()/open_ctree()
Not end user friendly, they have to dip their head into the source
code.
- Watch for directory /sys/fs/<uuid>/
This is way more simple, but still requires some simple device -> uuid
lookups. And a script needs to use inotify to watch /sys/fs/.
Compared to all these, directly outputting the information into dmesg
would be the most simple one, with both device and UUID included.
And since we're here, also add the output when mounting a filesystem for
the first time for parity. A more fine grained monitoring of subvolume
mounts should be done by another layer, like audit.
Now mounting a btrfs with all default mkfs options would look like this:
[81.906566] BTRFS info (device dm-8): first mount of filesystem 633b5c16-afe3-4b79-b195-138fe145e4f2
[81.907494] BTRFS info (device dm-8): using crc32c (crc32c-intel) checksum algorithm
[81.908258] BTRFS info (device dm-8): using free space tree
[81.912644] BTRFS info (device dm-8): auto enabling async discard
[81.913277] BTRFS info (device dm-8): checking UUID tree
[91.668256] BTRFS info (device dm-8): last unmount of filesystem 633b5c16-afe3-4b79-b195-138fe145e4f2
The jump_table stores two 32-bit words and one 32- (on 32-bit kernel)
or one 64-bit word (on 64-bit kernel).
Ensure that the last word is always 64-bit aligned on a 64-bit kernel
by aligning the whole structure on sizeof(long).
Those return codes are only defined for the parisc architecture and
are leftovers from when we wanted to be HP-UX compatible.
They are not returned by any Linux kernel syscall but do trigger
problems with the glibc strerrorname_np() and strerror() functions as
reported in glibc issue #31080.
There is no need to keep them, so simply remove them.
On parisc we need 16-byte alignment for variables which are used for
locking. Mark the __lock_aligned attribute acordingly so that the
.data..lock_aligned section will get that alignment in the generated
object files.
Make sure that the __bug_table section gets 32- or 64-bit aligned,
depending if a 32- or 64-bit kernel is being built.
Mark it non-writeable and use .blockz instead of the .org assembler
directive to pad the struct.
Add an align statement to tell the linker that all ex_table entries and as
such the whole ex_table section should be 32-bit aligned in vmlinux and modules.
Add an align statement to tell the linker that all ex_table entries and as
such the whole ex_table section should be 32-bit aligned in vmlinux and modules.
During floating point and vector save to thread data f0/vs0 are
clobbered by the FPSCR/VSCR store routine. This has been obvserved to
lead to userspace register corruption and application data corruption
with io-uring.
Fix it by restoring f0/vs0 after FPSCR/VSCR store has completed for
all the FP, altivec, VMX register save paths.
Tested under QEMU in kvm mode, running on a Talos II workstation with
dual POWER9 DD2.2 CPUs.
Additional detail (mpe):
Typically save_fpu() is called from __giveup_fpu() which saves the FP
regs and also *turns off FP* in the tasks MSR, meaning the kernel will
reload the FP regs from the thread struct before letting the task use FP
again. So in that case save_fpu() is free to clobber f0 because the FP
regs no longer hold live values for the task.
There is another case though, which is the path via:
sys_clone()
...
copy_process()
dup_task_struct()
arch_dup_task_struct()
flush_all_to_thread()
save_all()
That path saves the FP regs but leaves them live. That's meant as an
optimisation for a process that's using FP/VSX and then calls fork(),
leaving the regs live means the parent process doesn't have to take a
fault after the fork to get its FP regs back. The optimisation was added
in commit 8792468da5e1 ("powerpc: Add the ability to save FPU without
giving it up").
That path does clobber f0, but f0 is volatile across function calls,
and typically programs reach copy_process() from userspace via a syscall
wrapper function. So in normal usage f0 being clobbered across a
syscall doesn't cause visible data corruption.
But there is now a new path, because io-uring can call copy_process()
via create_io_thread() from the signal handling path. That's OK if the
signal is handled as part of syscall return, but it's not OK if the
signal is handled due to some other interrupt.
That path is:
interrupt_return_srr_user()
interrupt_exit_user_prepare()
interrupt_exit_user_prepare_main()
do_notify_resume()
get_signal()
task_work_run()
create_worker_cb()
create_io_worker()
copy_process()
dup_task_struct()
arch_dup_task_struct()
flush_all_to_thread()
save_all()
if (tsk->thread.regs->msr & MSR_FP)
save_fpu()
# f0 is clobbered and potentially live in userspace
Note the above discussion applies equally to save_altivec().
Fixes: 8792468da5e1 ("powerpc: Add the ability to save FPU without giving it up") Cc: stable@vger.kernel.org # v4.6+ Closes: https://lore.kernel.org/all/480932026.45576726.1699374859845.JavaMail.zimbra@raptorengineeringinc.com/ Closes: https://lore.kernel.org/linuxppc-dev/480221078.47953493.1700206777956.JavaMail.zimbra@raptorengineeringinc.com/ Tested-by: Timothy Pearson <tpearson@raptorengineering.com> Tested-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Timothy Pearson <tpearson@raptorengineering.com>
[mpe: Reword change log to describe exact path of corruption & other minor tweaks] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://msgid.link/1921539696.48534988.1700407082933.JavaMail.zimbra@raptorengineeringinc.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Before running a guest, the host process (e.g., QEMU) FP/VEC registers
are saved if they were being used, similarly to when the kernel uses FP
registers. The guest values are then loaded into regs, and the host
process registers will be restored lazily when it uses FP/VEC.
KVM HV has a bug here: the host process registers do get saved, but the
user MSR bits remain enabled, which indicates the registers are valid
for the process. After they are clobbered by running the guest, this
valid indication causes the host process to take on the FP/VEC register
values of the guest.
Fixes: 34e119c96b2b ("KVM: PPC: Book3S HV P9: Reduce mtmsrd instructions required to save host SPRs") Cc: stable@vger.kernel.org # v5.17+ Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://msgid.link/20231122025811.2973-1-npiggin@gmail.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The VT-d spec requires (10.4.4 Global Command Register, TE field) that:
Hardware implementations supporting DMA draining must drain any in-flight
DMA read/write requests queued within the Root-Complex before switching
address translation on or off and reflecting the status of the command
through the TES field in the Global Status register.
Unfortunately, some integrated graphic devices fail to do so after some
kind of power state transition. As the result, the system might stuck in
iommu_disable_translation(), waiting for the completion of TE transition.
Add MTL to the quirk list for those devices and skips TE disabling if the
qurik hits.
Fixes: b1012ca8dc4f ("iommu/vt-d: Skip TE disabling on quirky gfx dedicated iommu") Cc: stable@vger.kernel.org Signed-off-by: Abdul Halim, Mohd Syazwan <mohd.syazwan.abdul.halim@intel.com> Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com> Link: https://lore.kernel.org/r/20231116022324.30120-1-baolu.lu@linux.intel.com Signed-off-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit "filemap: update ki_pos in generic_perform_write", made updating
of ki_pos into common code in generic_perform_write() function.
This also causes generic/091 to fail.
This happened due to an in-flight collision with: fb5de4358e1a ("ext2: Move direct-io to use iomap"). I have chosen fixes tag
based on which commit got landed later to upstream kernel.
Fixes: 182c25e9c157 ("filemap: update ki_pos in generic_perform_write") Cc: stable@vger.kernel.org Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com> Signed-off-by: Jan Kara <jack@suse.cz>
Message-Id: <d595bee9f2475ed0e8a2e7fb94f7afc2c6ffc36a.1700643443.git.ritesh.list@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit 028ddcac477b ("bcache: Remove unnecessary NULL point check in
node allocations") replaced IS_ERR_OR_NULL by IS_ERR. This leads to a
NULL pointer dereference.
for (i = 0; i < nodes; i++) {
if (__bch_keylist_realloc(&keylist, bkey_u64s(&r[i].b->key)))
goto out_nocoalesce;
// ...
out_nocoalesce:
// ...
for (i = 0; i < nodes; i++)
if (!IS_ERR(new_nodes[i])) { // IS_ERR_OR_NULL before 028ddcac477b
btree_node_free(new_nodes[i]); // new_nodes[0] is NULL
rw_unlock(true, new_nodes[i]);
}
This patch replaces IS_ERR() by IS_ERR_OR_NULL() to fix this.
Fixes: 028ddcac477b ("bcache: Remove unnecessary NULL point check in node allocations") Link: https://lore.kernel.org/all/3DF4A87A-2AC1-4893-AE5F-E921478419A9@suse.de/ Cc: stable@vger.kernel.org Cc: Zheng Wang <zyytlz.wz@163.com> Cc: Coly Li <colyli@suse.de> Signed-off-by: Markus Weippert <markus@gekmihesg.de> Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
It turns out there are more subtle races beyond just the main part of
__iommu_probe_device() itself running in parallel - the dev_iommu_free()
on the way out of an unsuccessful probe can still manage to trip up
concurrent accesses to a device's fwspec. Thus, extend the scope of
iommu_probe_device_lock() to also serialise fwspec creation and initial
retrieval.
This flag only applies to the SQ and CQ rings, it's perfectly valid
to use a mmap approach for the provided ring buffers. Move the
check into where it belongs.
Cc: stable@vger.kernel.org Fixes: 03d89a2de25b ("io_uring: support for user allocated memory for rings/sqes") Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
It's valid to add the same fence multiple times to a dma-resv object and
we shouldn't need one extra slot for each.
Signed-off-by: Christian König <christian.koenig@amd.com> Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com> Fixes: a3f7c10a269d5 ("dma-buf/dma-resv: check if the new fence is really later") Cc: stable@vger.kernel.org # v5.19+ Link: https://patchwork.freedesktop.org/patch/msgid/20231115093035.1889-1-christian.koenig@amd.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
cpufreq_driver->fast_switch() callback expects a frequency as a return
value. amd_pstate_fast_switch() was returning the return value of
amd_pstate_update_freq(), which only indicates a success or failure.
Fix this by making amd_pstate_fast_switch() return the target_freq
when the call to amd_pstate_update_freq() is successful, and return
the current frequency from policy->cur when the call to
amd_pstate_update_freq() is unsuccessful.
Fixes: 4badf2eb1e98 ("cpufreq: amd-pstate: Add ->fast_switch() callback") Acked-by: Huang Rui <ray.huang@amd.com> Reviewed-by: Wyes Karny <wyes.karny@amd.com> Reviewed-by: Perry Yuan <perry.yuan@amd.com> Cc: 6.4+ <stable@vger.kernel.org> # v6.4+ Signed-off-by: Gautham R. Shenoy <gautham.shenoy@amd.com> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
With the new uapi we don't have the comp flags on the allocation,
so we shouldn't be using the first size that works, we should be
iterating until we get the correct one.
This reduces allocations from 2MB to 64k in lots of places.
Fixes dEQP-VK.memory.allocation.basic.size_8KiB.forward.count_4000
on my ampere/gsp system.
mmap_lock nests under uring_lock out of necessity, as we may be doing
user copies with uring_lock held. However, for mmap of provided buffer
rings, we attempt to grab uring_lock with mmap_lock already held from
do_mmap(). This makes lockdep, rightfully, complain:
WARNING: possible circular locking dependency detected 6.7.0-rc1-00009-gff3337ebaf94-dirty #4438 Not tainted
------------------------------------------------------
buf-ring.t/442 is trying to acquire lock: ffff00020e1480a8 (&ctx->uring_lock){+.+.}-{3:3}, at: io_uring_validate_mmap_request.isra.0+0x4c/0x140
but task is already holding lock: ffff0000dc226190 (&mm->mmap_lock){++++}-{3:3}, at: vm_mmap_pgoff+0x124/0x264
which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
From that mmap(2) path, we really just need to ensure that the buffer
list doesn't go away from underneath us. For the lower indexed entries,
they never go away until the ring is freed and we can always sanely
reference those as long as the caller has a file reference. For the
higher indexed ones in our xarray, we just need to ensure that the
buffer list remains valid while we return the address of it.
Free the higher indexed io_buffer_list entries via RCU. With that we can
avoid needing ->uring_lock inside mmap(2), and simply hold the RCU read
lock around the buffer list lookup and address check.
To ensure that the arrayed lookup either returns a valid fully formulated
entry via RCU lookup, add an 'is_ready' flag that we access with store
and release memory ordering. This isn't needed for the xarray lookups,
but doesn't hurt either. Since this isn't a fast path, retain it across
both types. Similarly, for the allocated array inside the ctx, ensure
we use the proper load/acquire as setup could in theory be running in
parallel with mmap.
While in there, add a few lockdep checks for documentation purposes.
Cc: stable@vger.kernel.org Fixes: c56e022c0a27 ("io_uring: add support for user mapped provided buffer ring") Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit 6bbd42e2df8f ("mmu_notifiers: call invalidate_range() when
invalidating TLBs") moved the secondary TLB invalidations into the TLB
invalidation functions to ensure that all secondary TLB invalidations
happen at the same time as the CPU invalidation and added a flush-all
type of secondary TLB invalidation for the batched mode, where a range
of [0, -1UL) is used to indicates that the range extends to the end of
the address space.
However, using an end address of -1UL caused an overflow in the Intel
IOMMU driver, where the end address was rounded up to the next page.
As a result, both the IOTLB and device ATC were not invalidated correctly.
Add a flush all helper function and call it when the invalidation range
is from 0 to -1UL, ensuring that the entire caches are invalidated
correctly.
io_sqes_map() is used rather than io_mem_alloc(), if the application
passes in memory for mapping rather than have the kernel allocate it and
then mmap(2) the ranges. This then calls __io_uaddr_map() to perform the
page mapping and pinning, which checks if we end up with the same pages,
if more than one page is mapped. But this check is incorrect and only
checks if the first and last pages are the same, where it really should
be checking if the mapped pages are contigous. This allows mapping a
single normal page, or a huge page range.
Down the line we can add support for remapping pages to be virtually
contigous, which is really all that io_uring cares about.
Cc: stable@vger.kernel.org Fixes: 03d89a2de25b ("io_uring: support for user allocated memory for rings/sqes") Reported-by: Jann Horn <jannh@google.com> Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The acpi_video code was storing the acpi_video_device as driver_data
in the acpi_device children of the acpi_video_bus acpi_device.
But the acpi_video driver only binds to the bus acpi_device.
It uses, but does not bind to, the children. Since it is not
the driver it should not be using the driver_data of the children's
acpi_device-s.
Since commit 0d16710146a1 ("ACPI: bus: Set driver_data to NULL every
time .add() fails") the childen's driver_data ends up getting set
to NULL after a driver fails to bind to the children leading to a NULL
pointer deref in video_get_max_state when registering the cooling-dev:
Fix this by directly using the acpi_video_device as devdata for
the cooling-device, which avoids the need to set driver-data on
the children at all.
Fixes: 0d16710146a1 ("ACPI: bus: Set driver_data to NULL every time .add() fails") Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/9718 Cc: 6.6+ <stable@vger.kernel.org> # 6.6+ Signed-off-by: Hans de Goede <hdegoede@redhat.com> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The original change results in a deadlock if jumbo mtu mode is used.
Reason is that the phydev lock is held when rtl_reset_work() is called
here, and rtl_jumbo_config() calls phy_start_aneg() which also tries
to acquire the phydev lock. Fix this by calling rtl_reset_work()
asynchronously.
Fixes: 621735f59064 ("r8169: fix rare issue with broken rx after link-down on RTL8125") Reported-by: Ian Chen <free122448@hotmail.com> Tested-by: Ian Chen <free122448@hotmail.com> Cc: stable@vger.kernel.org Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com> Link: https://lore.kernel.org/r/caf6a487-ef8c-4570-88f9-f47a659faf33@gmail.com Signed-off-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When scanning namespaces, it is possible to get valid data from the first
call to nvme_identify_ns() in nvme_alloc_ns(), but not from the second
call in nvme_update_ns_info_block(). In particular, if the NSID becomes
inactive between the two commands, a storage device may return a buffer
filled with zero as per 4.1.5.1. In this case, we can get a kernel crash
due to a divide-by-zero in blk_stack_limits() because ns->lba_shift will
be set to zero.
We found an issue under Android OTA scenario that many BIOs have to do
FEC where the data under dm-verity is 100% complete and no corruption.
Android OTA has many dm-block layers, from upper to lower:
dm-verity
dm-snapshot
dm-origin & dm-cow
dm-linear
ufs
DM tables have to change 2 times during Android OTA merging process.
When doing table change, the dm-snapshot will be suspended for a while.
During this interval, many readahead IOs are submitted to dm_verity
from filesystem. Then the kverity works are busy doing FEC process
which cost too much time to finish dm-verity IO. This causes needless
delay which feels like system is hung.
After adding debugging it was found that each readahead IO needed
around 10s to finish when this situation occurred. This is due to IO
amplification:
dm-snapshot suspend
erofs_readahead // 300+ io is submitted
dm_submit_bio (dm_verity)
dm_submit_bio (dm_snapshot)
bio return EIO
bio got nothing, it's empty
verity_end_io
verity_verify_io
forloop range(0, io->n_blocks) // each io->nblocks ~= 20
verity_fec_decode
fec_decode_rsb
fec_read_bufs
forloop range(0, v->fec->rsn) // v->fec->rsn = 253
new_read
submit_bio (dm_snapshot)
end loop
end loop
dm-snapshot resume
Readahead BIOs get nothing while dm-snapshot is suspended, so all of
them will cause verity's FEC.
Each readahead BIO needs to verify ~20 (io->nblocks) blocks.
Each block needs to do FEC, and every block needs to do 253
(v->fec->rsn) reads.
So during the suspend interval(~200ms), 300 readahead BIOs trigger
~1518000 (300*20*253) IOs to dm-snapshot.
As readahead IO is not required by userspace, and to fix this issue,
it is best to pass readahead errors to upper layer to handle it.
Cc: stable@vger.kernel.org Fixes: a739ff3f543a ("dm verity: add support for forward error correction") Signed-off-by: Wu Bo <bo.wu@vivo.com> Reviewed-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Mike Snitzer <snitzer@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If BIO error, verity_end_io() can call verity_finish_io() before
verity_fec_init_io(). Therefore, fec_io->rs is not initialized and
may crash when doing memory freeing in verity_fec_finish_io().