Zygo Blaxell [Tue, 30 Nov 2021 16:37:05 +0000 (11:37 -0500)]
zygo: btrfs: 'btrfs replace' hangs at end of replacing a device (v5.10.82)
From: Nikolay Borisov <nborisov@suse.com>
Date: Tue, 30 Nov 2021 15:55:12 +0200
I have a working hypothesis what might be going wrong, however without a
crash dump to investigate I can't really confirm it. Basically I think
btrfs_rm_dev_replace_blocked is not seeing the decrement aka the store
to running bios count since it's using cond_wake_up_nomb. If I'm right
then the following should fix it:
Can you apply it and see if it can reproduce, I don't know what's the
incident rate of this bug so you have to decide at what point it should
be fixed. In any case this patch can't have any negative functional
impact, it just makes the ordering slightly stronger to ensure the write
happens before possibly waking up someone on the queue.
Josef Bacik [Thu, 18 Nov 2021 21:33:15 +0000 (16:33 -0500)]
btrfs: index free space entries on size
Currently we index free space on offset only, because usually we have a
hint from the allocator that we want to honor for locality reasons.
However if we fail to use this hint we have to go back to a brute force
search through the free space entries to find a large enough extent.
With sufficiently fragmented free space this becomes quite expensive, as
we have to linearly search all of the free space entries to find if we
have a part that's long enough.
To fix this add a cached rb tree to index based on free space entry
bytes. This will allow us to quickly look up the largest chunk in the
free space tree for this block group, and stop searching once we've
found an entry that is too small to satisfy our allocation. We simply
choose to use this tree if we're searching from the beginning of the
block group, as we know we do not care about locality at that point.
I wrote an allocator test that creates a 10TiB ram backed null block
device and then fallocates random files until the file system is full.
I think go through and delete all of the odd files. Then I spawn 8
threads that fallocate 64MiB files (1/2 our extent size cap) until the
file system is full again. I use bcc's funclatency to measure the
latency of find_free_extent. The baseline results are
There's a little variation in the amount of calls done because of timing
of the threads with metadata requirements, but the avg, total, and
count's are relatively consistent between runs (usually within 2-5% of
each other). As you can see here we have around a 30% decrease in
average latency with a 30% decrease in overall time spent in
find_free_extent.
Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: David Sterba <dsterba@suse.com>
(cherry picked from commit 6b9db64552eb9af65108fe2edfbf6fb66dbf98b2)
Josef Bacik [Thu, 18 Nov 2021 21:33:14 +0000 (16:33 -0500)]
btrfs: only use ->max_extent_size if it is set in the bitmap
While adding self tests for my space index change I was hitting a
problem where the space indexed tree wasn't returning the expected
->max_extent_size. This is because we will skip searching any entry
that doesn't have ->bytes >= the amount of bytes we want. However we'll
still set the max_extent_size based on that entry. The problem is if we
don't search the bitmap we won't have ->max_extent_size set properly, so
we can't really trust it.
This doesn't really result in a problem per-se, it can just result in us
not finding contiguous area that may exist. Fix the max_extent_size
helper to return ->bytes if ->max_extent_size isn't set, and add a big
comment explaining why we're doing this.
Josef Bacik [Thu, 18 Nov 2021 15:26:16 +0000 (10:26 -0500)]
btrfs: index free space entries on size
Currently we index free space on offset only, because usually we have a
hint from the allocator that we want to honor for locality reasons.
However if we fail to use this hint we have to go back to a brute force
search through the free space entries to find a large enough extent.
With sufficiently fragmented free space this becomes quite expensive, as
we have to linearly search all of the free space entries to find if we
have a part that's long enough.
To fix this add a cached rb tree to index based on free space entry
bytes. This will allow us to quickly look up the largest chunk in the
free space tree for this block group, and stop searching once we've
found an entry that is too small to satisfy our allocation. We simply
choose to use this tree if we're searching from the beginning of the
block group, as we know we do not care about locality at that point.
I wrote an allocator test that creates a 10TiB ram backed null block
device and then fallocates random files until the file system is full.
I think go through and delete all of the odd files. Then I spawn 8
threads that fallocate 64mib files (1/2 our extent size cap) until the
file system is full again. I use bcc's funclatency to measure the
latency of find_free_extent. The baseline results are
There's a little variation in the amount of calls done because of timing
of the threads with metadata requirements, but the avg, total, and
count's are relatively consistent between runs (usually within 2-5% of
each other). As you can see here we have around a 30% decrease in
average latency with a 30% decrease in overall time spent in
find_free_extent.
Signed-off-by: Josef Bacik <josef@toxicpanda.com> Reviewed-by: Filipe Manana <fdmanana@suse.com>
(cherry picked from commit b51c6ac16d18e4682871df29d8998addcf17fb52)
Josef Bacik [Thu, 18 Nov 2021 15:26:15 +0000 (10:26 -0500)]
btrfs: only use ->max_extent_size if it is set in the bitmap
While adding self tests for my space index change I was hitting a
problem where the space indexed tree wasn't returning the expected
->max_extent_size. This is because we will skip searching any entry
that doesn't have ->bytes >= the amount of bytes we want. However we'll
still set the max_extent_size based on that entry. The problem is if we
don't search the bitmap we won't have ->max_extent_size set properly, so
we can't really trust it.
This doesn't really result in a problem per-se, it can just result in us
not finding contiguous area that may exist. Fix the max_extent_size
helper to return ->bytes if ->max_extent_size isn't set, and add a big
comment explaining why we're doing this.
Josef Bacik [Tue, 5 Oct 2021 20:35:24 +0000 (16:35 -0400)]
btrfs: change error handling for btrfs_delete_*_in_log (backported for 5.14)
Currently we will abort the transaction if we get a random error (like
-EIO) while trying to remove the directory entries from the root log
during rename.
However since these are simply log tree related errors, we can mark the
trans as needing a full commit. Then if the error was truly
catastrophic we'll hit it during the normal commit and abort as
appropriate.
Reviewed-by: Nikolay Borisov <nborisov@suse.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: David Sterba <dsterba@suse.com>
(cherry picked from commit 9a35fc9542fa6c220d69987612b88c54cba2bc33)
(cherry picked from commit 13b0abbe037fd595fb3a5ef4ba043a6290e9cb97)
of_parse_thermal_zones() parses the thermal-zones node and registers a
thermal_zone device for each subnode. However, if a thermal zone is
consuming a thermal sensor and that thermal sensor device hasn't probed
yet, an attempt to set trip_point_*_temp for that thermal zone device
can cause a NULL pointer dereference. Fix it.
While at it, fix the possible NULL pointer dereference in other
functions as well: of_thermal_get_temp(), of_thermal_set_emul_temp(),
of_thermal_get_trend().
Suggested-by: David Collins <quic_collinsd@quicinc.com> Signed-off-by: Subbaraman Narayanamurthy <quic_subbaram@quicinc.com> Acked-by: Daniel Lezcano <daniel.lezcano@linaro.org> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
PEBS PERF_SAMPLE_PHYS_ADDR events use perf_virt_to_phys() to convert PMU
sampled virtual addresses to physical using get_user_page_fast_only()
and page_to_phys().
Some get_user_page_fast_only() error cases return false, indicating no
page reference, but still initialize the output page pointer with an
unreferenced page. In these error cases perf_virt_to_phys() calls
put_page(). This causes page reference count underflow, which can lead
to unintentional page sharing.
Fix perf_virt_to_phys() to only put_page() if get_user_page_fast_only()
returns a referenced page.
The ION AHCI device pretends that MSI masking isn't a thing, while it
actually implements it and needs MSIs to be unmasked to work. Add a quirk
to that effect.
It appears that some devices are lying about their mask capability,
pretending that they don't have it, while they actually do.
The net result is that now that we don't enable MSIs on such
endpoint.
Add a new per-device flag to deal with this. Further patches will
make use of it, sadly.
Signed-off-by: Marc Zyngier <maz@kernel.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Link: https://lore.kernel.org/r/20211104180130.3825416-2-maz@kernel.org Cc: Bjorn Helgaas <helgaas@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
free_msi_irqs() frees the MSI entries before destroying the sysfs entries
which are exposing them. Nothing prevents a concurrent free while a sysfs
file is read and accesses the possibly freed entry.
Move the sysfs release ahead of freeing the entries.
Fixes: 1c51b50c2995 ("PCI/MSI: Export MSI mode using attributes, not kobjects") Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Bjorn Helgaas <helgaas@kernel.org> Cc: stable@vger.kernel.org Link: https://lore.kernel.org/r/87sfw5305m.ffs@tglx Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 8779e05ba8aa ("parisc: Fix ptrace check on syscall return")
fixed testing of TI_FLAGS. This uncovered a bug in the test mask.
syscall_restore_rfi is only used when the kernel needs to exit to
usespace with single or block stepping and the recovery counter
enabled. The test however used _TIF_SYSCALL_TRACE_MASK, which
includes a lot of bits that shouldn't be tested here.
Fix this by using TIF_SINGLESTEP and TIF_BLOCKSTEP directly.
I encountered this bug by enabling syscall tracepoints. Both in qemu and
on real hardware. As soon as i enabled the tracepoint (sys_exit_read,
but i guess it doesn't really matter which one), i got random page
faults in userspace almost immediately.
There are some duplicated codes to validate the block
size in block drivers. This limitation actually comes
from block layer, so this patch tries to add a new block
layer helper for that.
Clang has never correctly compiled the FORTIFY_SOURCE defenses due to
a couple bugs:
Eliding inlines with matching __builtin_* names
https://bugs.llvm.org/show_bug.cgi?id=50322
Incorrect __builtin_constant_p() of some globals
https://bugs.llvm.org/show_bug.cgi?id=41459
In the process of making improvements to the FORTIFY_SOURCE defenses, the
first (silent) bug (coincidentally) becomes worked around, but exposes
the latter which breaks the build. As such, Clang must not be used with
CONFIG_FORTIFY_SOURCE until at least latter bug is fixed (in Clang 13),
and the fortify routines have been rearranged.
Update the Kconfig to reflect the reality of the current situation.
In 64-bit mode, x86 instruction encoding allows us to use the low 8 bits
of any GPR as an 8-bit operand. In 32-bit mode, however, we can only use
the [abcd] registers. For which, GCC has the "q" constraint instead of
the less restrictive "r".
Also fix st->preempted, which is an input/output operand rather than an
input.
Fixes: 7e2175ebd695 ("KVM: x86: Fix recording of guest steal time / preempted status") Reported-by: kernel test robot <lkp@intel.com> Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Message-Id: <89bf72db1b859990355f9c40713a34e0d2d86c98.camel@infradead.org> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
btrfs: add flags to give an hint to the chunk allocator
Add the following flags to give an hint about which chunk should be
allocated in which a disk.
The following flags are created:
- BTRFS_DEV_ALLOCATION_PREFERRED_DATA
preferred data chunk, but metadata chunk allowed
- BTRFS_DEV_ALLOCATION_PREFERRED_METADATA
preferred metadata chunk, but data chunk allowed
- BTRFS_DEV_ALLOCATION_METADATA_ONLY
only metadata chunk allowed
- BTRFS_DEV_ALLOCATION_DATA_ONLY
only data chunk allowed
When this mode is enabled, the chunk allocation policy is modified as
follow.
Each disk may have a different tag:
- BTRFS_DEV_ALLOCATION_PREFERRED_METADATA
- BTRFS_DEV_ALLOCATION_METADATA_ONLY
- BTRFS_DEV_ALLOCATION_DATA_ONLY
- BTRFS_DEV_ALLOCATION_PREFERRED_DATA (default)
Where:
- ALLOCATION_PREFERRED_X means that it is preferred to use this disk for
the X chunk type (the other type may be allowed when the space is low)
- ALLOCATION_X_ONLY means that it is used *only* for the X chunk type.
This means also that it is a preferred choice.
Each time the allocator allocates a chunk of type X , first it takes the
disks tagged as ALLOCATION_X_ONLY or ALLOCATION_PREFERRED_X; if the space
is not enough, it uses also the disks tagged as ALLOCATION_METADATA_ONLY;
if the space is not enough, it uses also the other disks, with the
exception of the one marked as ALLOCATION_PREFERRED_Y, where Y the other
type of chunk (i.e. not X).
Zygo Blaxell [Fri, 19 Nov 2021 12:19:19 +0000 (07:19 -0500)]
Revert "zygo: preferred_metadata is either type 1 (preferred_metadata) or 2 (metadata_only). 3 (data_only) and 0 (preferred_data) are not preferred metadata, and our patch does not implement the preferred modes"
It has been reported to be causing problems, and is being reworked
upstream and has been dropped from the current 5.15.y stable queue until
it gets resolved.
It has been reported to be causing problems, and is being reworked
upstream and has been dropped from the current 5.15.y stable queue until
it gets resolved.
Reported-by: Chris Rankin <rankincj@gmail.com> Reported-by: Thorsten Leemhuis <linux@leemhuis.info> Link: https://lore.kernel.org/r/ed000478-2a60-0066-c337-a04bffc112b1@leemhuis.info Cc: Stephen Rothwell <sfr@canb.auug.org.au> Cc: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Kees Cook <keescook@chromium.org> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Russell King (Oracle) <rmk+kernel@armlinux.org.uk> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Sasha Levin <sashal@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
It has been reported to be causing problems, and is being reworked
upstream and has been dropped from the current 5.15.y stable queue until
it gets resolved.
Commit a4b83deb3e76 ("media: videobuf2: rework vb2_mem_ops API")
added a new vb member to struct vb2_dma_sg_buf, but it only added
code setting this to the vb2_dma_sg_alloc() function and not to the
vb2_dma_sg_get_userptr() and vb2_dma_sg_attach_dmabuf() which also
create vb2_dma_sg_buf objects.
This is causing a crash due to a NULL pointer deref when using
libcamera on devices with an Intel IPU3 (qcam app).
Fix these crashes by assigning buf->vb in the other 2 functions too,
note libcamera tests the vb2_dma_sg_get_userptr() path, the change
to the vb2_dma_sg_attach_dmabuf() path is untested.
Fixes: a4b83deb3e76 ("media: videobuf2: rework vb2_mem_ops API") Cc: Sergey Senozhatsky <senozhatsky@chromium.org> Signed-off-by: Hans de Goede <hdegoede@redhat.com> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The size of the exception stacks was increased by the commit in Fixes,
resulting in stack sizes greater than a page in size. The #VC exception
handling was only mapping the first (bottom) page, resulting in an
SEV-ES guest failing to boot.
Make the #VC exception stacks part of the default exception stacks
storage and allocate them with a CONFIG_AMD_MEM_ENCRYPT=y .config. Map
them only when a SEV-ES guest has been detected.
Rip out the custom VC stacks mapping and storage code.
In preparation for other confidential computing technologies, introduce
a generic helper function, cc_platform_has(), that can be used to
check for specific active confidential computing attributes, like
memory encryption. This is intended to eliminate having to add multiple
technology-specific checks to the code (e.g. if (sev_active() ||
tdx_active() || ... ).
The new function validate_hash_algo() assumed that ima_get_hash_algo()
always return a valid 'enum hash_algo', but it returned the
user-supplied value present in the digital signature without
any bounds checks.
Update ima_get_hash_algo() to always return a valid hash algorithm,
defaulting on 'ima_hash_algo' when the user-supplied value inside
the xattr is invalid.
Signed-off-by: THOBY Simon <Simon.THOBY@viveris.fr> Reported-by: syzbot+e8bafe7b82c739eaf153@syzkaller.appspotmail.com Fixes: 50f742dd9147 ("IMA: block writes of the security.ima xattr with unsupported algorithms") Reviewed-by: Lakshmi Ramasubramanian <nramas@linux.microsoft.com> Signed-off-by: Mimi Zohar <zohar@linux.ibm.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Previous fix aded bpf_clamp_umax() helper use to re-validate boundaries.
While that works correctly, it introduces more branches, which blows up
past 1 million instructions in no-alu32 variant of strobemeta selftests.
Switching len variable from u32 to u64 also fixes the issue and reduces
the number of validated instructions, so use that instead. Fix this
patch and bpf_clamp_umax() removed, both alu32 and no-alu32 selftests
pass.
Commit in Fixes changed the iopl emulation to not #GP on CLI and STI
because it would break some insane luserspace tools which would toggle
interrupts.
The corresponding selftest would rely on the fact that executing CLI/STI
would trigger a #GP and thus detect it this way but since that #GP is
not happening anymore, the detection is now wrong too.
Extend the test to actually look at the IF flag and whether executing
those insns had any effect on it. The STI detection needs to have the
fact that interrupts were previously disabled, passed in so do that from
the previous CLI test, i.e., STI test needs to follow a previous CLI one
for it to make sense.
There are several error return paths that dereference the null pointer
host because the pointer has not yet been set to a valid value.
Fix this by adding a new out_mmc label and exiting via this label
to avoid the host clean up and hence the null pointer dereference.
Addresses-Coverity: ("Explicit null dereference") Fixes: 8105c2abbf36 ("mmc: moxart: Fix reference count leaks in moxart_probe") Signed-off-by: Colin Ian King <colin.king@canonical.com> Link: https://lore.kernel.org/r/20211013100052.125461-1-colin.king@canonical.com Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Using a kernel pointer in place of a dma_addr_t token can
lead to undefined behavior if that makes it into cache
management functions. The compiler caught one such attempt
in a cast:
drivers/net/wireless/ath/ath10k/mac.c: In function 'ath10k_add_interface':
drivers/net/wireless/ath/ath10k/mac.c:5586:47: error: cast from pointer to integer of different size [-Werror=pointer-to-int-cast]
5586 | arvif->beacon_paddr = (dma_addr_t)arvif->beacon_buf;
| ^
Looking through how this gets used down the way, I'm fairly
sure that beacon_paddr is never accessed again for ATH10K_DEV_TYPE_HL
devices, and if it was accessed, that would be a bug.
Change the assignment to use a known-invalid address token
instead, which avoids the warning and makes it easier to catch
bugs if it does end up getting used.
Fixes: e263bdab9c0e ("ath10k: high latency fixes for beacon buffer") Signed-off-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Kalle Valo <kvalo@codeaurora.org> Link: https://lore.kernel.org/r/20211014075153.3655910-1-arnd@kernel.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Fixes: 7be3248f3139 ("cifs: To match file servers, make sure the server hostname matches") Signed-off-by: Paulo Alcantara (SUSE) <pc@cjr.nz> Signed-off-by: Steve French <stfrench@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Adding kfree(dvb) to vidtv_bridge_remove() will remove the memory
too soon: if an application still has an open filehandle to the device
when the driver is unloaded, then when that filehandle is closed, a
use-after-free access takes place to the freed memory.
Move the kfree(dvb) to vidtv_bridge_dev_release() instead.
commit 652de07addd2 ("drm/amd/display: Fully switch to dmub for all dcn21
asics") switched over to using dmub on Renoir to fix Gitlab 1735, but this
implied a new dependency on newer firmware which might not be met on older
kernel versions.
Since sw_init runs before hw_init, there is an opportunity to determine
whether or not the firmware version is new to adjust the behavior.
Cc: Roman.Li@amd.com BugLink: https://gitlab.freedesktop.org/drm/amd/-/issues/1772 BugLink: https://gitlab.freedesktop.org/drm/amd/-/issues/1735 Fixes: 652de07addd2 ("drm/amd/display: Fully switch to dmub for all dcn21 asics") Signed-off-by: Mario Limonciello <mario.limonciello@amd.com> Acked-by: Alex Deucher <alexander.deucher@amd.com> Reviewed-by: Roman Li <Roman.Li@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The premise of commit 6f9f17287e78 ("SUNRPC: Mitigate cond_resched() in
xprt_transmit()") was that cond_resched() is expensive and unnecessary
when there has been just a single send.
The point of cond_resched() is to ensure that tasks that should pre-empt
this one get a chance to do so when it is safe to do so. The code prior
to commit 6f9f17287e78 failed to take into account that it was keeping a
rpc_task pinned for longer than it needed to, and so rather than doing a
full revert, let's just move the cond_resched.
Change PCIe Max Payload Size setting in PCIe Device Control register to 512
bytes to align with PCIe Link Initialization sequence as defined in Marvell
Armada 3700 Functional Specification. According to the specification,
maximal Max Payload Size supported by this device is 512 bytes.
Without this kernel prints suspicious line:
pci 0000:01:00.0: Upstream bridge's Max Payload Size set to 256 (was 16384, max 512)
With this change it changes to:
pci 0000:01:00.0: Upstream bridge's Max Payload Size set to 256 (was 512, max 512)
Macros SUN8I_CSC_CTRL() and SUN8I_CSC_COEFF() don't follow usual
recommendation of having arguments enclosed in parenthesis. While that
didn't change anything for quite sometime, it actually become important
after CSC code rework with commit ea067aee45a8 ("drm/sun4i: de2/de3:
Remove redundant CSC matrices").
Without this fix, colours are completely off for supported YVU formats
on SoCs with DE2 (A64, H3, R40, etc.).
Fix the issue by enclosing macro arguments in parenthesis.
When CONFIG_SMP=y, timebase synchronization is required when the second
kernel is started.
arch/powerpc/kernel/smp.c:
int __cpu_up(unsigned int cpu, struct task_struct *tidle)
{
...
if (smp_ops->give_timebase)
smp_ops->give_timebase();
...
}
void start_secondary(void *unused)
{
...
if (smp_ops->take_timebase)
smp_ops->take_timebase();
...
}
When CONFIG_HOTPLUG_CPU=n and CONFIG_KEXEC_CORE=n,
smp_85xx_ops.give_timebase is NULL,
smp_85xx_ops.take_timebase is NULL,
As a result, the timebase is not synchronized.
Timebase synchronization does not depend on CONFIG_HOTPLUG_CPU.
Fixes: 56f1ba280719 ("powerpc/mpc85xx: refactor the PM operations") Cc: stable@vger.kernel.org # v4.6+ Signed-off-by: Xiaoming Ni <nixiaoming@huawei.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20210929033646.39630-3-nixiaoming@huawei.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
On VMs with NX encryption, compression, and/or RNG offload, these
capabilities are described by nodes in the ibm,platform-facilities device
tree hierarchy:
The acceleration functions that these nodes describe are not disrupted by
live migration, not even temporarily.
But the post-migration ibm,update-nodes sequence firmware always sends
"delete" messages for this hierarchy, followed by an "add" directive to
reconstruct it via ibm,configure-connector (log with debugging statements
enabled in mobility.c):
Note we receive a single "add" message for the entire hierarchy, and what
we receive from the ibm,configure-connector sequence is the top-level
platform-facilities node along with its three children. The debug message
simply reports the parent node and not the whole subtree.
Also, significantly, the nodes added are almost completely equivalent to
the ones removed; even phandles are unchanged. ibm,shared-interrupt-pool in
the leaf nodes is the only property I've observed to differ, and Linux does
not use that. So in practice, the sum of update messages Linux receives for
this hierarchy is equivalent to minor property updates.
We succeed in removing the original hierarchy from the device tree. But the
vio bus code is ignorant of this, and does not unbind or relinquish its
references. The leaf nodes, still reachable through sysfs, of course still
refer to the now-freed ibm,platform-facilities parent node, which makes
use-after-free possible:
Moreover, the "new" replacement subtree is not correctly added to the
device tree, resulting in ibm,platform-facilities parent node without the
appropriate leaf nodes, and broken symlinks in the sysfs device hierarchy:
$ tree -d /sys/firmware/devicetree/base/ibm,platform-facilities/
/sys/firmware/devicetree/base/ibm,platform-facilities/
0 directories
$ cd /sys/devices/vio ; find . -xtype l -exec file {} +
./ibm,sym-encryption-v1/of_node: broken symbolic link to
../../../firmware/devicetree/base/ibm,platform-facilities/ibm,sym-encryption-v1
./ibm,random-v1/of_node: broken symbolic link to
../../../firmware/devicetree/base/ibm,platform-facilities/ibm,random-v1
./ibm,compression-v1/of_node: broken symbolic link to
../../../firmware/devicetree/base/ibm,platform-facilities/ibm,compression-v1
This is because add_dt_node() -> dlpar_attach_node() attaches only the
parent node returned from configure-connector, ignoring any children. This
should be corrected for the general case, but fixing that won't help with
the stale OF node references, which is the more urgent problem.
One way to address that would be to make the drivers respond to node
removal notifications, so that node references can be dropped
appropriately. But this would likely force the drivers to disrupt active
clients for no useful purpose: equivalent nodes are immediately re-added.
And recall that the acceleration capabilities described by the nodes remain
available throughout the whole process.
The solution I believe to be robust for this situation is to convert
remove+add of a node with an unchanged phandle to an update of the node's
properties in the Linux device tree structure. That would involve changing
and adding a fair amount of code, and may take several iterations to land.
Until that can be realized we have a confirmed use-after-free and the
possibility of memory corruption. So add a limited workaround that
discriminates on the node type, ignoring adds and removes. This should be
amenable to backporting in the meantime.
Fixes: 410bccf97881 ("powerpc/pseries: Partition migration in the kernel") Cc: stable@vger.kernel.org Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20211020194703.2613093-1-nathanl@linux.ibm.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The check_return_regs_valid() can cause a false positive if the return
regs are marked as norestart and they are an HSRR type interrupt,
because the low bit in the bottom of regs->trap causes interrupt type
matching to fail.
This can occcur for example on bare metal with a HV privileged doorbell
interrupt that causes a signal, but do_signal returns early because
get_signal() fails, and takes the "No signal to deliver" path. In this
case no signal was delivered so the return location is not changed so
return SRRs are not invalidated, yet set_trap_norestart is called, which
messes up the match. Building go-1.16.6 is known to reproduce this.
Fix it by using the TRAP() accessor which masks out the low bit.
Fixes: 6eaaf9de3599 ("powerpc/64s/interrupt: Check and fix srr_valid without crashing") Cc: stable@vger.kernel.org # v5.14+ Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20211026122531.3599918-1-npiggin@gmail.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The mitigation-patching.sh script in the powerpc selftests toggles
all mitigations on and off simultaneously, revealing that rfi_flush
and stf_barrier cannot safely operate at the same time due to races
in updating the static key.
On some systems, the static key code throws a warning and the kernel
remains functional. On others, the kernel will hang or crash.
Fix this by slapping on a mutex.
Fixes: 13799748b957 ("powerpc/64: use interrupt restart table to speed up return from interrupt") Cc: stable@vger.kernel.org # v5.14+ Signed-off-by: Russell Currey <ruscur@russell.cc> Acked-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20211027072410.40950-1-ruscur@russell.cc Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit 587164cd, introduced new opal message type (OPAL_MSG_PRD2) and
added opal notifier. But I missed to unregister the notifier during
module unload path. This results in below call trace if you try to
unload and load opal_prd module.
Also add new notifier_block for OPAL_MSG_PRD2 message.
Sample calltrace (modprobe -r opal_prd; modprobe opal_prd)
BUG: Unable to handle kernel data access on read at 0xc0080000192200e0
Faulting instruction address: 0xc00000000018d1cc
Oops: Kernel access of bad area, sig: 11 [#1]
LE PAGE_SIZE=64K MMU=Radix SMP NR_CPUS=2048 NUMA PowerNV
CPU: 66 PID: 7446 Comm: modprobe Kdump: loaded Tainted: G E 5.14.0prd #759
NIP: c00000000018d1cc LR: c00000000018d2a8 CTR: c0000000000cde10
REGS: c0000003c4c0f0a0 TRAP: 0300 Tainted: G E (5.14.0prd)
MSR: 9000000002009033 <SF,HV,VEC,EE,ME,IR,DR,RI,LE> CR: 24224824 XER: 20040000
CFAR: c00000000018d2a4 DAR: c0080000192200e0 DSISR: 40000000 IRQMASK: 1
...
NIP notifier_chain_register+0x2c/0xc0
LR atomic_notifier_chain_register+0x48/0x80
Call Trace:
0xc000000002090610 (unreliable)
atomic_notifier_chain_register+0x58/0x80
opal_message_notifier_register+0x7c/0x1e0
opal_prd_probe+0x84/0x150 [opal_prd]
platform_probe+0x78/0x130
really_probe+0x110/0x5d0
__driver_probe_device+0x17c/0x230
driver_probe_device+0x60/0x130
__driver_attach+0xfc/0x220
bus_for_each_dev+0xa8/0x130
driver_attach+0x34/0x50
bus_add_driver+0x1b0/0x300
driver_register+0x98/0x1a0
__platform_driver_register+0x38/0x50
opal_prd_driver_init+0x34/0x50 [opal_prd]
do_one_initcall+0x60/0x2d0
do_init_module+0x7c/0x320
load_module+0x3394/0x3650
__do_sys_finit_module+0xd4/0x160
system_call_exception+0x140/0x290
system_call_common+0xf4/0x258
A e5500 machine running a 32-bit kernel sometimes hangs at boot,
seemingly going into an infinite loop of instruction storage interrupts.
The ESR (Exception Syndrome Register) has a value of 0x800000 (store)
when this happens, which is likely set by a previous store. An
instruction TLB miss interrupt would then leave ESR unchanged, and if no
PTE exists it calls directly to the instruction storage interrupt
handler without changing ESR.
access_error() does not cause a segfault due to a store to a read-only
vma because is_exec is true. Most subsequent fault handling does not
check for a write fault on a read-only vma, and might do strange things
like create a writeable PTE or call page_mkwrite on a read only vma or
file. It's not clear what happens here to cause the infinite faulting in
this case, a fault handler failure or low level PTE or TLB handling.
In any case this can be fixed by having the instruction storage
interrupt zero regs->dsisr rather than storing the ESR value to it.
Fixes: a01a3f2ddbcd ("powerpc: remove arguments from fault handler functions") Cc: stable@vger.kernel.org # v5.12+ Reported-by: Jacques de Laval <jacques.delaval@protonmail.com> Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Tested-by: Jacques de Laval <jacques.delaval@protonmail.com> Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20211028133043.4159501-1-npiggin@gmail.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Following the introduction of the generic ECC engine infrastructure, it
was necessary to reorganize the code and move the ECC configuration in
the ->attach_chip() hook. Failing to do that properly lead to a first
series of fixes supposed to stabilize the situation. Unfortunately, this
only fixed the use of software ECC engines, preventing any other kind of
engine to be used, including on-die ones.
It is now time to (finally) fix the situation by ensuring that we still
provide a default (eg. software ECC) but will still support different
ECC engines such as on-die ECC engines if properly described in the
device tree.
There are no changes needed on the core side in order to do this, but we
just need to leverage the logic there which allows:
1- a subsystem default (set to Host engines in the raw NAND world)
2- a driver specific default (here set to software ECC engines)
3- any type of engine requested by the user (ie. described in the DT)
As the raw NAND subsystem has not yet been fully converted to the ECC
engine infrastructure, in order to provide a default ECC engine for this
driver we need to set chip->ecc.engine_type *before* calling
nand_scan(). During the initialization step, the core will consider this
entry as the default engine for this driver. This value may of course
be overloaded by the user if the usual DT properties are provided.
Following the introduction of the generic ECC engine infrastructure, it
was necessary to reorganize the code and move the ECC configuration in
the ->attach_chip() hook. Failing to do that properly lead to a first
series of fixes supposed to stabilize the situation. Unfortunately, this
only fixed the use of software ECC engines, preventing any other kind of
engine to be used, including on-die ones.
It is now time to (finally) fix the situation by ensuring that we still
provide a default (eg. software ECC) but will still support different
ECC engines such as on-die ECC engines if properly described in the
device tree.
There are no changes needed on the core side in order to do this, but we
just need to leverage the logic there which allows:
1- a subsystem default (set to Host engines in the raw NAND world)
2- a driver specific default (here set to software ECC engines)
3- any type of engine requested by the user (ie. described in the DT)
As the raw NAND subsystem has not yet been fully converted to the ECC
engine infrastructure, in order to provide a default ECC engine for this
driver we need to set chip->ecc.engine_type *before* calling
nand_scan(). During the initialization step, the core will consider this
entry as the default engine for this driver. This value may of course
be overloaded by the user if the usual DT properties are provided.
Following the introduction of the generic ECC engine infrastructure, it
was necessary to reorganize the code and move the ECC configuration in
the ->attach_chip() hook. Failing to do that properly lead to a first
series of fixes supposed to stabilize the situation. Unfortunately, this
only fixed the use of software ECC engines, preventing any other kind of
engine to be used, including on-die ones.
It is now time to (finally) fix the situation by ensuring that we still
provide a default (eg. software ECC) but will still support different
ECC engines such as on-die ECC engines if properly described in the
device tree.
There are no changes needed on the core side in order to do this, but we
just need to leverage the logic there which allows:
1- a subsystem default (set to Host engines in the raw NAND world)
2- a driver specific default (here set to software ECC engines)
3- any type of engine requested by the user (ie. described in the DT)
As the raw NAND subsystem has not yet been fully converted to the ECC
engine infrastructure, in order to provide a default ECC engine for this
driver we need to set chip->ecc.engine_type *before* calling
nand_scan(). During the initialization step, the core will consider this
entry as the default engine for this driver. This value may of course
be overloaded by the user if the usual DT properties are provided.
Following the introduction of the generic ECC engine infrastructure, it
was necessary to reorganize the code and move the ECC configuration in
the ->attach_chip() hook. Failing to do that properly lead to a first
series of fixes supposed to stabilize the situation. Unfortunately, this
only fixed the use of software ECC engines, preventing any other kind of
engine to be used, including on-die ones.
It is now time to (finally) fix the situation by ensuring that we still
provide a default (eg. software ECC) but will still support different
ECC engines such as on-die ECC engines if properly described in the
device tree.
There are no changes needed on the core side in order to do this, but we
just need to leverage the logic there which allows:
1- a subsystem default (set to Host engines in the raw NAND world)
2- a driver specific default (here set to software ECC engines)
3- any type of engine requested by the user (ie. described in the DT)
As the raw NAND subsystem has not yet been fully converted to the ECC
engine infrastructure, in order to provide a default ECC engine for this
driver we need to set chip->ecc.engine_type *before* calling
nand_scan(). During the initialization step, the core will consider this
entry as the default engine for this driver. This value may of course
be overloaded by the user if the usual DT properties are provided.
Following the introduction of the generic ECC engine infrastructure, it
was necessary to reorganize the code and move the ECC configuration in
the ->attach_chip() hook. Failing to do that properly lead to a first
series of fixes supposed to stabilize the situation. Unfortunately, this
only fixed the use of software ECC engines, preventing any other kind of
engine to be used, including on-die ones.
It is now time to (finally) fix the situation by ensuring that we still
provide a default (eg. software ECC) but will still support different
ECC engines such as on-die ECC engines if properly described in the
device tree.
There are no changes needed on the core side in order to do this, but we
just need to leverage the logic there which allows:
1- a subsystem default (set to Host engines in the raw NAND world)
2- a driver specific default (here set to software ECC engines)
3- any type of engine requested by the user (ie. described in the DT)
As the raw NAND subsystem has not yet been fully converted to the ECC
engine infrastructure, in order to provide a default ECC engine for this
driver we need to set chip->ecc.engine_type *before* calling
nand_scan(). During the initialization step, the core will consider this
entry as the default engine for this driver. This value may of course
be overloaded by the user if the usual DT properties are provided.
Following the introduction of the generic ECC engine infrastructure, it
was necessary to reorganize the code and move the ECC configuration in
the ->attach_chip() hook. Failing to do that properly lead to a first
series of fixes supposed to stabilize the situation. Unfortunately, this
only fixed the use of software ECC engines, preventing any other kind of
engine to be used, including on-die ones.
It is now time to (finally) fix the situation by ensuring that we still
provide a default (eg. software ECC) but will still support different
ECC engines such as on-die ECC engines if properly described in the
device tree.
There are no changes needed on the core side in order to do this, but we
just need to leverage the logic there which allows:
1- a subsystem default (set to Host engines in the raw NAND world)
2- a driver specific default (here set to software ECC engines)
3- any type of engine requested by the user (ie. described in the DT)
As the raw NAND subsystem has not yet been fully converted to the ECC
engine infrastructure, in order to provide a default ECC engine for this
driver we need to set chip->ecc.engine_type *before* calling
nand_scan(). During the initialization step, the core will consider this
entry as the default engine for this driver. This value may of course
be overloaded by the user if the usual DT properties are provided.
Following the introduction of the generic ECC engine infrastructure, it
was necessary to reorganize the code and move the ECC configuration in
the ->attach_chip() hook. Failing to do that properly lead to a first
series of fixes supposed to stabilize the situation. Unfortunately, this
only fixed the use of software ECC engines, preventing any other kind of
engine to be used, including on-die ones.
It is now time to (finally) fix the situation by ensuring that we still
provide a default (eg. software ECC) but will still support different
ECC engines such as on-die ECC engines if properly described in the
device tree.
There are no changes needed on the core side in order to do this, but we
just need to leverage the logic there which allows:
1- a subsystem default (set to Host engines in the raw NAND world)
2- a driver specific default (here set to software ECC engines)
3- any type of engine requested by the user (ie. described in the DT)
As the raw NAND subsystem has not yet been fully converted to the ECC
engine infrastructure, in order to provide a default ECC engine for this
driver we need to set chip->ecc.engine_type *before* calling
nand_scan(). During the initialization step, the core will consider this
entry as the default engine for this driver. This value may of course
be overloaded by the user if the usual DT properties are provided.
Fixes: d525914b5bd8 ("mtd: rawnand: xway: Move the ECC initialization to ->attach_chip()") Cc: stable@vger.kernel.org Cc: Jan Hoffmann <jan@3e8.eu> Cc: Kestrel seventyfour <kestrelseventyfour@gmail.com> Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com> Tested-by: Jan Hoffmann <jan@3e8.eu> Link: https://lore.kernel.org/linux-mtd/20210928222258.199726-10-miquel.raynal@bootlin.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Following the introduction of the generic ECC engine infrastructure, it
was necessary to reorganize the code and move the ECC configuration in
the ->attach_chip() hook. Failing to do that properly lead to a first
series of fixes supposed to stabilize the situation. Unfortunately, this
only fixed the use of software ECC engines, preventing any other kind of
engine to be used, including on-die ones.
It is now time to (finally) fix the situation by ensuring that we still
provide a default (eg. software ECC) but will still support different
ECC engines such as on-die ECC engines if properly described in the
device tree.
There are no changes needed on the core side in order to do this, but we
just need to leverage the logic there which allows:
1- a subsystem default (set to Host engines in the raw NAND world)
2- a driver specific default (here set to software ECC engines)
3- any type of engine requested by the user (ie. described in the DT)
As the raw NAND subsystem has not yet been fully converted to the ECC
engine infrastructure, in order to provide a default ECC engine for this
driver we need to set chip->ecc.engine_type *before* calling
nand_scan(). During the initialization step, the core will consider this
entry as the default engine for this driver. This value may of course
be overloaded by the user if the usual DT properties are provided.
The introduction of the generic ECC engine API lead to a number of
changes in various drivers which broke some of them. Here is a typical
example: I expected the SM_ORDER option to be handled by the Hamming ECC
engine internals. Problem: the fsmc driver does not instantiate (yet) a
real ECC engine object so we had to use a 'bare' ECC helper instead of
the shiny rawnand functions. However, when not intializing this engine
properly and using the bare helpers, we do not get the SM ORDER feature
handled automatically. It looks like this was lost in the process so
let's ensure we use the right SM ORDER now.
projects / linux / log