Once the qp has been flushed, it cannot be flushed again. The user qp
flush logic wasn't enforcing it however. The bug can cause
touch-after-free crashes like:
Unable to handle kernel paging request for data at address 0x000001ec
Faulting instruction address: 0xc008000016069100
Oops: Kernel access of bad area, sig: 11 [#1]
...
NIP [c008000016069100] flush_qp+0x80/0x480 [iw_cxgb4]
LR [c00800001606cd6c] c4iw_modify_qp+0x71c/0x11d0 [iw_cxgb4]
Call Trace:
[c00800001606cd6c] c4iw_modify_qp+0x71c/0x11d0 [iw_cxgb4]
[c00800001606e868] c4iw_ib_modify_qp+0x118/0x200 [iw_cxgb4]
[c0080000119eae80] ib_security_modify_qp+0xd0/0x3d0 [ib_core]
[c0080000119c4e24] ib_modify_qp+0xc4/0x2c0 [ib_core]
[c008000011df0284] iwcm_modify_qp_err+0x44/0x70 [iw_cm]
[c008000011df0fec] destroy_cm_id+0xcc/0x370 [iw_cm]
[c008000011ed4358] rdma_destroy_id+0x3c8/0x520 [rdma_cm]
[c0080000134b0540] ucma_close+0x90/0x1b0 [rdma_ucm]
[c000000000444da4] __fput+0xe4/0x2f0
So fix flush_qp() to only flush the wq once.
Cc: stable@vger.kernel.org Signed-off-by: Steve Wise <swise@opengridcomputing.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The PCIE I/O and MEM resource allocation mechanism is that root bus
goes through the following steps:
1. Check PCI bridges' range and computes I/O and Mem base/limits.
2. Sort all subordinate devices I/O and MEM resource requirements and
allocate the resources and writes/updates subordinate devices'
requirements to PCI bridges I/O and Mem MEM/limits registers.
Currently, PCI Aardvark driver only handles the second step and lacks
the first step, so there is an I/O and MEM resource allocation failure
when using a PCI switch. This commit fixes that by sizing bridges
before doing the resource allocation.
When a task which previously ran on a given CPU is remotely queued to
wake up on that same CPU, there is a period where the task's state is
TASK_WAKING and its vruntime is not normalized. This is not accounted
for in vruntime_normalized() which will cause an error in the task's
vruntime if it is switched from the fair class during this time.
For example if it is boosted to RT priority via rt_mutex_setprio(),
rq->min_vruntime will not be subtracted from the task's vruntime but
it will be added again when the task returns to the fair class. The
task's vruntime will have been erroneously doubled and the effective
priority of the task will be reduced.
Note this will also lead to inflation of all vruntimes since the doubled
vruntime value will become the rq's min_vruntime when other tasks leave
the rq. This leads to repeated doubling of the vruntime and priority
penalty.
Fix this by recognizing a WAKING task's vruntime as normalized only if
sched_remote_wakeup is true. This indicates a migration, in which case
the vruntime would have been normalized in migrate_task_rq_fair().
Based on a similar patch from John Dias <joaodias@google.com>.
Suggested-by: Peter Zijlstra <peterz@infradead.org> Tested-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Signed-off-by: Steve Muckle <smuckle@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Chris Redpath <Chris.Redpath@arm.com> Cc: John Dias <joaodias@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Miguel de Dios <migueldedios@google.com> Cc: Morten Rasmussen <Morten.Rasmussen@arm.com> Cc: Patrick Bellasi <Patrick.Bellasi@arm.com> Cc: Paul Turner <pjt@google.com> Cc: Quentin Perret <quentin.perret@arm.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Todd Kjos <tkjos@google.com> Cc: kernel-team@android.com Fixes: b5179ac70de8 ("sched/fair: Prepare to fix fairness problems on migration") Link: http://lkml.kernel.org/r/20180831224217.169476-1-smuckle@google.com Signed-off-by: Ingo Molnar <mingo@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Marking mmp bh dirty before writing it will make writeback
pick up mmp block later and submit a write, we don't want the
duplicate write as kmmpd thread should have full control of
reading and writing the mmp block.
Another reason is we will also have random I/O error on
the writeback request when blk integrity is enabled, because
kmmpd could modify the content of the mmp block(e.g. setting
new seq and time) while the mmp block is under I/O requested
by writeback.
Signed-off-by: Li Dongyang <dongyangli@ddn.com> Signed-off-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Andreas Dilger <adilger@dilger.ca> Cc: stable@vger.kernel.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
An online resize of a file system with the bigalloc feature enabled
and a 1k block size would be refused since ext4_resize_begin() did not
understand s_first_data_block is 0 for all bigalloc file systems, even
when the block size is 1k.
Avoid growing the file system to an extent so that the last block
group is too small to hold all of the metadata that must be stored in
the block group.
This problem can be triggered with the following reproducer:
When mounting the superblock, ext4_fill_super() calculates the free
blocks and free inodes and stores them in the superblock. It's not
strictly necessary, since we don't use them any more, but it's nice to
keep them roughly aligned to reality.
Since it's not critical for file system correctness, the code doesn't
call ext4_commit_super(). The problem is that it's in
ext4_commit_super() that we recalculate the superblock checksum. So
if we're not going to call ext4_commit_super(), we need to call
ext4_superblock_csum_set() to make sure the superblock checksum is
consistent.
Most of the time, this doesn't matter, since we end up calling
ext4_commit_super() very soon thereafter, and definitely by the time
the file system is unmounted. However, it doesn't work in this
sequence:
A specially crafted file system can trick empty_inline_dir() into
reading past the last valid entry in a inline directory, and then run
into the end of xattr marker. This will trigger a divide by zero
fault. Fix this by using the size of the inline directory instead of
dir->i_size.
Also clean up error reporting in __ext4_check_dir_entry so that the
message is clearer and more understandable --- and avoids the division
by zero trap if the size passed in is zero. (I'm not sure why we
coded it that way in the first place; printing offset % size is
actually more confusing and less useful.)
If the destination of the rename(2) system call exists, the inode's
link count (i_nlinks) must be non-zero. If it is, the inode can end
up on the orphan list prematurely, leading to all sorts of hilarity,
including a use-after-free.
Fix this by sanitizing vsa.console before using it to index vc_cons
Notice that given that speculation windows are large, the policy is
to kill the speculation on the first load and not worry if it can be
completed with a dependent load/store [1].
When there's no scaling requested ->is_unity should be true no matter
the format.
Also, when no scaling is requested and we have a multi-planar YUV
format, we should leave ->y_scaling[0] to VC4_SCALING_NONE and only
set ->x_scaling[0] to VC4_SCALING_PPF.
Doing this fixes an hardly visible artifact (seen when using modetest
and a rather big overlay plane in YUV420).
Fixes: fc04023fafec ("drm/vc4: Add support for YUV planes.") Cc: <stable@vger.kernel.org> Signed-off-by: Boris Brezillon <boris.brezillon@bootlin.com> Reviewed-by: Eric Anholt <eric@anholt.net> Link: https://patchwork.freedesktop.org/patch/msgid/20180725122907.13702-1-boris.brezillon@bootlin.com Signed-off-by: Sean Paul <seanpaul@chromium.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
On most systems with ACPI hotplugging support, it seems that we always
receive a hotplug event once we re-enable EC interrupts even if the GPU
hasn't even been resumed yet.
This can cause problems since even though we schedule hpd_work to handle
connector reprobing for us, hpd_work synchronizes on
pm_runtime_get_sync() to wait until the device is ready to perform
reprobing. Since runtime suspend/resume callbacks are disabled before
the PM core calls ->suspend(), any calls to pm_runtime_get_sync() during
this period will grab a runtime PM ref and return immediately with
-EACCES. Because we schedule hpd_work from our ACPI HPD handler, and
hpd_work synchronizes on pm_runtime_get_sync(), this causes us to launch
a connector reprobe immediately even if the GPU isn't actually resumed
just yet. This causes various warnings in dmesg and occasionally, also
prevents some displays connected to the dedicated GPU from coming back
up after suspend. Example:
So, to fix this we attempt to grab a runtime PM reference in the ACPI
handler itself asynchronously. If the GPU is already awake (it will have
normal hotplugging at this point) or runtime PM callbacks are currently
disabled on the device, we drop our reference without updating the
autosuspend delay. We only schedule connector reprobes when we
successfully managed to queue up a resume request with our asynchronous
PM ref.
This also has the added benefit of preventing redundant connector
reprobes from ACPI while the GPU is runtime resumed!
Signed-off-by: Lyude Paul <lyude@redhat.com> Cc: stable@vger.kernel.org Cc: Karol Herbst <kherbst@redhat.com>
Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1477182#c41 Signed-off-by: Lyude Paul <lyude@redhat.com> Signed-off-by: Ben Skeggs <bskeggs@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
It's true we can't resume the device from poll workers in
nouveau_connector_detect(). We can however, prevent the autosuspend
timer from elapsing immediately if it hasn't already without risking any
sort of deadlock with the runtime suspend/resume operations. So do that
instead of entirely avoiding grabbing a power reference.
Signed-off-by: Lyude Paul <lyude@redhat.com> Reviewed-by: Karol Herbst <kherbst@redhat.com> Acked-by: Daniel Vetter <daniel@ffwll.ch> Cc: stable@vger.kernel.org Cc: Lukas Wunner <lukas@wunner.de> Signed-off-by: Ben Skeggs <bskeggs@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Turns out this part is my fault for not noticing when reviewing 9a2eba337cace ("drm/nouveau: Fix drm poll_helper handling"). Currently
we call drm_kms_helper_poll_enable() from nouveau_display_hpd_work().
This makes basically no sense however, because that means we're calling
drm_kms_helper_poll_enable() every time we schedule the hotplug
detection work. This is also against the advice mentioned in
drm_kms_helper_poll_enable()'s documentation:
Note that calls to enable and disable polling must be strictly ordered,
which is automatically the case when they're only call from
suspend/resume callbacks.
Of course, hotplugs can't really be ordered. They could even happen
immediately after we called drm_kms_helper_poll_disable() in
nouveau_display_fini(), which can lead to all sorts of issues.
Additionally; enabling polling /after/ we call
drm_helper_hpd_irq_event() could also mean that we'd miss a hotplug
event anyway, since drm_helper_hpd_irq_event() wouldn't bother trying to
probe connectors so long as polling is disabled.
So; simply move this back into nouveau_display_init() again. The race
condition that both of these patches attempted to work around has
already been fixed properly in
d61a5c106351 ("drm/nouveau: Fix deadlock on runtime suspend")
Fixes: 9a2eba337cace ("drm/nouveau: Fix drm poll_helper handling") Signed-off-by: Lyude Paul <lyude@redhat.com> Acked-by: Karol Herbst <kherbst@redhat.com> Acked-by: Daniel Vetter <daniel@ffwll.ch> Cc: Lukas Wunner <lukas@wunner.de> Cc: Peter Ujfalusi <peter.ujfalusi@ti.com> Cc: stable@vger.kernel.org Signed-off-by: Ben Skeggs <bskeggs@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
While reading block, it is possible that io error return due to underlying
storage issue, in this case, BH_NeedsValidate was left in the buffer head.
Then when reading the very block next time, if it was already linked into
journal, that will trigger the following panic.
This change has the following effects, in order of descreasing importance:
1) Prevent a stack buffer overflow
2) Do not append an unnecessary NULL to an anyway binary buffer, which
is writing one byte past client_digest when caller is:
chap_string_to_hex(client_digest, chap_r, strlen(chap_r));
The latter was found by KASAN (see below) when input value hes expected size
(32 hex chars), and further analysis revealed a stack buffer overflow can
happen when network-received value is longer, allowing an unauthenticated
remote attacker to smash up to 17 bytes after destination buffer (16 bytes
attacker-controlled and one null). As switching to hex2bin requires
specifying destination buffer length, and does not internally append any null,
it solves both issues.
This addresses CVE-2018-14633.
Beyond this:
- Validate received value length and check hex2bin accepted the input, to log
this rejection reason instead of just failing authentication.
- Only log received CHAP_R and CHAP_C values once they passed sanity checks.
==================================================================
BUG: KASAN: stack-out-of-bounds in chap_string_to_hex+0x32/0x60 [iscsi_target_mod]
Write of size 1 at addr ffff8801090ef7c8 by task kworker/0:0/1021
Signed-off-by: Vincent Pelletier <plr.vincent@gmail.com> Reviewed-by: Mike Christie <mchristi@redhat.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Update 'confirmed' timestamp when ARP packet is received. It shouldn't
affect locktime logic and anyway entry can be confirmed by any higher-layer
protocol. Thus it makes sense to confirm it when ARP packet is received.
Fixes: 77d7123342dc ("neighbour: update neigh timestamps iff update is effective") Signed-off-by: Vasily Khoruzhick <vasilykh@arista.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 2abb7cdc0dc8 ("udp: Add support for doing checksum
unnecessary conversion") left out the early demux path for
connected sockets. As a result IP_CMSG_CHECKSUM gives wrong
values for such socket when GRO is not enabled/available.
This change addresses the issue by moving the csum conversion to a
common helper and using such helper in both the default and the
early demux rx path.
Fixes: 2abb7cdc0dc8 ("udp: Add support for doing checksum unnecessary conversion") Signed-off-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The operation ~(p100_inb(VG_LAN_CFG_1) & HP100_LINK_UP) returns a value
that is always non-zero and hence the wait for the link to drop always
terminates prematurely. Fix this by using a logical not operator instead
of a bitwise complement. This issue has been in the driver since
pre-2.6.12-rc2.
Detected by CoverityScan, CID#114157 ("Logical vs. bitwise operator")
Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Fields ->dev and ->next of struct ipddp_route may be copied to
userspace on the SIOCFINDIPDDPRT ioctl. This is only accessible
to CAP_NET_ADMIN though. Let's manually copy the relevant fields
instead of using memcpy().
BugLink: http://blog.infosectcbr.com.au/2018/09/linux-kernel-infoleaks.html Cc: Jann Horn <jannh@google.com> Signed-off-by: Willy Tarreau <w@1wt.eu> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In the unlikely case ip6_xmit() has to call skb_realloc_headroom(),
we need to call skb_set_owner_w() before consuming original skb,
otherwise we risk a use-after-free.
Bring IPv6 in line with what we do in IPv4 to fix this.
Fixes: 1da177e4c3f41 ("Linux-2.6.12-rc2") Signed-off-by: Eric Dumazet <edumazet@google.com> Reported-by: syzbot <syzkaller@googlegroups.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When splitting a GSO segment that consists of encapsulated packets, the
skb->mac_len of the segments can end up being set wrong, causing packet
drops in particular when using act_mirred and ifb interfaces in
combination with a qdisc that splits GSO packets.
This happens because at the time skb_segment() is called, network_header
will point to the inner header, throwing off the calculation in
skb_reset_mac_len(). The network_header is subsequently adjust by the
outer IP gso_segment handlers, but they don't set the mac_len.
Fix this by adding skb_reset_mac_len() calls to both the IPv4 and IPv6
gso_segment handlers, after they modify the network_header.
Many thanks to Eric Dumazet for his help in identifying the cause of
the bug.
Acked-by: Dave Taht <dave.taht@gmail.com> Reviewed-by: Eric Dumazet <edumazet@google.com> Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Directories and inodes don't necessarily need to be in the same lockdep
class. For ex, hugetlbfs splits them out too to prevent false positives
in lockdep. Annotate correctly after new inode creation. If its a
directory inode, it will be put into a different class.
This should fix a lockdep splat reported by syzbot:
When reducing ring buffer size, pages are removed by scheduling a work
item on each CPU for the corresponding CPU ring buffer. After the pages
are removed from ring buffer linked list, the pages are free()d in a
tight loop. The loop does not give up CPU until all pages are removed.
In a worst case behavior, when lot of pages are to be freed, it can
cause system stall.
After the pages are removed from the list, the free() can happen while
the work is rescheduled. Call cond_resched() in the loop to prevent the
system hangup.
Link: http://lkml.kernel.org/r/20180907223129.71994-1-vnagarnaik@google.com Cc: stable@vger.kernel.org Fixes: 83f40318dab00 ("ring-buffer: Make removal of ring buffer pages atomic") Reported-by: Jason Behmer <jbehmer@google.com> Signed-off-by: Vaibhav Nagarnaik <vnagarnaik@google.com> Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This reverts f154a718e6cc ("PCI: Add ACS quirk for Intel 300 series").
It turns out that erratum "PCH PCIe* Controller Root Port (ACSCTLR) Appear
As Read Only" has been fixed in 300 series chipsets, even though the
datasheet [1] claims otherwise. To make ACS work properly on 300 series
root ports, revert the faulty commit.
Commit 57f230ab04d291 ("xen/netfront: raise max number of slots in
xennet_get_responses()") raised the max number of allowed slots by one.
This seems to be problematic in some configurations with netback using
a larger MAX_SKB_FRAGS value (e.g. old Linux kernel with MAX_SKB_FRAGS
defined as 18 instead of nowadays 17).
Instead of BUG_ON() in this case just fall back to retransmission.
Fixes: 57f230ab04d291 ("xen/netfront: raise max number of slots in xennet_get_responses()") Cc: stable@vger.kernel.org Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Although private data of sound card instance is usually allocated in the
tail of the instance, drivers in ALSA firewire stack allocate the private
data before allocating the instance. In this case, the private data
should be released explicitly at .private_free callback of the instance.
This commit fixes memory leak following to the above design.
Although private data of sound card instance is usually allocated in the
tail of the instance, drivers in ALSA firewire stack allocate the private
data before allocating the instance. In this case, the private data
should be released explicitly at .private_free callback of the instance.
This commit fixes memory leak following to the above design.
Although private data of sound card instance is usually allocated in the
tail of the instance, drivers in ALSA firewire stack allocate the private
data before allocating the instance. In this case, the private data
should be released explicitly at .private_free callback of the instance.
This commit fixes memory leak following to the above design.
snd_emu10k1_fx8010_ioctl(SNDRV_EMU10K1_IOCTL_INFO) allocates
memory using kmalloc() and partially fills it by calling
snd_emu10k1_fx8010_info() before returning the resulting
structure to userspace, leaving uninitialized holes. Let's
just use kzalloc() here.
When executing 'fw_run_transaction()' with 'TCODE_WRITE_BLOCK_REQUEST',
an address of 'payload' argument is used for streaming DMA mapping by
'firewire_ohci' module if 'size' argument is larger than 8 byte.
Although in this case the address should not be on kernel stack, current
implementation of ALSA bebob driver uses data in kernel stack for a cue
to boot M-Audio devices. This often brings unexpected result, especially
for a case of CONFIG_VMAP_STACK=y.
According to ETSI TS 102 622 specification chapter 4.4 pipe identifier
is 7 bits long which allows for 128 unique pipe IDs. Because
NFC_HCI_MAX_PIPES is used as the number of pipes supported and not
as the max pipe ID, its value should be 128 instead of 127.
nfc_hci_recv_from_llc extracts pipe ID from packet header using
NFC_HCI_FRAGMENT(0x7F) mask which allows for pipe ID value of 127.
Same happens when NCI_HCP_MSG_GET_PIPE() is being used. With
pipes array having only 127 elements and pipe ID of 127 the OOB memory
access will result.
Cc: Samuel Ortiz <sameo@linux.intel.com> Cc: Allen Pais <allen.pais@oracle.com> Cc: "David S. Miller" <davem@davemloft.net> Suggested-by: Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by: Suren Baghdasaryan <surenb@google.com> Reviewed-by: Kees Cook <keescook@chromium.org> Cc: stable <stable@vger.kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When handling SHDLC I-Frame commands "pipe" field used for indexing
into an array should be checked before usage. If left unchecked it
might access memory outside of the array of size NFC_HCI_MAX_PIPES(127).
Malformed NFC HCI frames could be injected by a malicious NFC device
communicating with the device being attacked (remote attack vector),
or even by an attacker with physical access to the I2C bus such that
they could influence the data transfers on that bus (local attack vector).
skb->data is controlled by the attacker and has only been sanitized in
the most trivial ways (CRC check), therefore we can consider the
create_info struct and all of its members to tainted. 'create_info->pipe'
with max value of 255 (uint8) is used to take an offset of the
hdev->pipes array of 127 elements which can lead to OOB write.
Cc: Samuel Ortiz <sameo@linux.intel.com> Cc: Allen Pais <allen.pais@oracle.com> Cc: "David S. Miller" <davem@davemloft.net> Suggested-by: Kevin Deus <kdeus@google.com> Signed-off-by: Suren Baghdasaryan <surenb@google.com> Acked-by: Kees Cook <keescook@chromium.org> Cc: stable <stable@vger.kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
link_active = true
/* link_active is true, wrongly, and stays so because
* get_link_status is false */
Avoid this problem by making sure that we don't set get_link_status = false
after having checked the link.
It seems this problem has been present since the introduction of e1000e.
Link: https://lkml.org/lkml/2018/1/29/338 Reported-by: Alexander Duyck <alexander.duyck@gmail.com> Signed-off-by: Benjamin Poirier <bpoirier@suse.com> Acked-by: Alexander Duyck <alexander.h.duyck@intel.com> Tested-by: Aaron Brown <aaron.f.brown@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com> Cc: Yanhui He <yanhuih@vmware.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit 19110cfbb34d ("e1000e: Separate signaling for link check/link up")
changed what happens to the link status when there is an error which
happens after "get_link_status = false" in the copper check_for_link
callbacks. Previously, such an error would be ignored and the link
considered up. After that commit, any error implies that the link is down.
Revert commit 19110cfbb34d ("e1000e: Separate signaling for link check/link
up") and its followups. After reverting, the race condition described in
the log of commit 19110cfbb34d is reintroduced. It may still be triggered
by LSC events but this should keep the link down in case the link is
electrically unstable, as discussed. The race may no longer be
triggered by RXO events because commit 4aea7a5c5e94 ("e1000e: Avoid
receiver overrun interrupt bursts") restored reading icr in the Other
handler.
Link: https://lkml.org/lkml/2018/3/1/789 Signed-off-by: Benjamin Poirier <bpoirier@suse.com> Acked-by: Alexander Duyck <alexander.h.duyck@intel.com> Tested-by: Aaron Brown <aaron.f.brown@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com> Cc: Yanhui He <yanhuih@vmware.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The 82574 specification update errata 12 states that interrupts may be
missed if ICR is read while INT_ASSERTED is not set. Avoid that problem by
setting all bits related to events that can trigger the Other interrupt in
IMS.
The Other interrupt is raised for such events regardless of whether or not
they are set in IMS. However, only when they are set is the INT_ASSERTED
bit also set in ICR.
By doing this, we ensure that INT_ASSERTED is always set when we read ICR
in e1000_msix_other() and steer clear of the errata. This also ensures that
ICR will automatically be cleared on read, therefore we no longer need to
clear bits explicitly.
Signed-off-by: Benjamin Poirier <bpoirier@suse.com> Acked-by: Alexander Duyck <alexander.h.duyck@intel.com> Tested-by: Aaron Brown <aaron.f.brown@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
[bwh: Backported to 4.9: adjust context] Cc: Yanhui He <yanhuih@vmware.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Restores the ICS write for Rx/Tx queue interrupts which was present before
commit 16ecba59bc33 ("e1000e: Do not read ICR in Other interrupt", v4.5-rc1)
but was not restored in commit 4aea7a5c5e94
("e1000e: Avoid receiver overrun interrupt bursts", v4.15-rc1).
This re-raises the queue interrupts in case the txq or rxq bits were set in
ICR and the Other interrupt handler read and cleared ICR before the queue
interrupt was raised.
Fixes: 4aea7a5c5e94 ("e1000e: Avoid receiver overrun interrupt bursts") Signed-off-by: Benjamin Poirier <bpoirier@suse.com> Acked-by: Alexander Duyck <alexander.h.duyck@intel.com> Tested-by: Aaron Brown <aaron.f.brown@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com> Cc: Yanhui He <yanhuih@vmware.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We keep the fix for the first part of the problem (1) described in the log
of that commit, that is to read ICR in the other interrupt handler. We
remove the fix for the second part of the problem (2), Other interrupt
throttling.
Bursts of "Other" interrupts may once again occur during rxo (receive
overflow) traffic conditions. This is deemed acceptable in the interest of
avoiding unforeseen fallout from changes that are not strictly necessary.
As discussed, the e1000e driver should be in "maintenance mode".
Link: https://www.spinics.net/lists/netdev/msg480675.html Signed-off-by: Benjamin Poirier <bpoirier@suse.com> Acked-by: Alexander Duyck <alexander.h.duyck@intel.com> Tested-by: Aaron Brown <aaron.f.brown@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com> Cc: Yanhui He <yanhuih@vmware.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
It was reported that emulated e1000e devices in vmware esxi 6.5 Build 7526125 do not link up after commit 4aea7a5c5e94 ("e1000e: Avoid receiver
overrun interrupt bursts", v4.15-rc1). Some tracing shows that after
e1000e_trigger_lsc() is called, ICR reads out as 0x0 in e1000_msix_other()
on emulated e1000e devices. In comparison, on real e1000e 82574 hardware,
icr=0x80000004 (_INT_ASSERTED | _LSC) in the same situation.
Some experimentation showed that this flaw in vmware e1000e emulation can
be worked around by not setting Other in EIAC. This is how it was before 16ecba59bc33 ("e1000e: Do not read ICR in Other interrupt", v4.5-rc1).
Fixes: 4aea7a5c5e94 ("e1000e: Avoid receiver overrun interrupt bursts") Signed-off-by: Benjamin Poirier <bpoirier@suse.com> Tested-by: Aaron Brown <aaron.f.brown@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
[bwh: Backported to 4.9: adjust context] Cc: Yanhui He <yanhuih@vmware.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When a system suffers from dcache aliasing a user program may observe
stale VDSO data from an aliased cache line. Notably this can break the
expectation that clock_gettime(CLOCK_MONOTONIC, ...) is, as its name
suggests, monotonic.
In order to ensure that users observe updates to the VDSO data page as
intended, align the user mappings of the VDSO data page such that their
cache colouring matches that of the virtual address range which the
kernel will use to update the data page - typically its unmapped address
within kseg0.
This ensures that we don't introduce aliasing cache lines for the VDSO
data page, and therefore that userland will observe updates without
requiring cache invalidation.
We accidentally removed the check for negative returns
without considering the issue of type promotion.
The "if_version_length" variable is type size_t so if __mei_cl_recv()
returns a negative then "bytes_recv" is type promoted
to a high positive value and treated as success.
Cc: <stable@vger.kernel.org> Fixes: 582ab27a063a ("mei: bus: fix received data size check in NFC fixup") Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by: Tomas Winkler <tomas.winkler@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If you do this on an sdm845 board:
grep "" /sys/kernel/debug/pinctrl/*spmi:pmic*/pinconf-groups
...it looks like nonsense. For every pin you see listed:
input bias disabled, input bias high impedance, input bias pull down, input bias pull up, ...
That's because pmic_gpio_config_get() isn't complying with the rules
that pinconf_generic_dump_one() expects. Specifically for boolean
parameters (anything with a "struct pin_config_item" where has_arg is
false) the function expects that the function should return its value
not through the "config" parameter but should return "0" if the value
is set and "-EINVAL" if the value isn't set.
Let's fix this.
>From a quick sample of other pinctrl drivers, it appears to be
tradition to also return 1 through the config parameter for these
boolean parameters when they exist. I'm not one to knock tradition,
so I'll follow tradition and return 1 in these cases. While I'm at
it, I'll also continue searching for four leaf clovers, kocking on
wood three times, and trying not to break mirrors.
NOTE: This also fixes an apparent typo for reading
PIN_CONFIG_BIAS_DISABLE where the old driver was accidentally
using "=" instead of "==" and thus was setting some internal
state when you tried to query PIN_CONFIG_BIAS_DISABLE. Oops.
The ARRAY_SIZE() macro is type size_t. If s6e8aa0_dcs_read() returns a
negative error code, then "ret < ARRAY_SIZE(id)" is false because the
negative error code is type promoted to a high positive value.
In the past we've warned when ADJ_OFFSET was in progress, usually
caused by ntpd or some other time adjusting daemon running in non
steady sate, which can cause the skew calculations to be
incorrect.
Thus, this patch checks to see if the clock was being adjusted
when we fail so that we don't cause false negatives.
Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@kernel.org> Cc: Miroslav Lichvar <mlichvar@redhat.com> Cc: Richard Cochran <richardcochran@gmail.com> Cc: Prarit Bhargava <prarit@redhat.com> Cc: Stephen Boyd <sboyd@kernel.org> Cc: Shuah Khan <shuah@kernel.org> Cc: linux-kselftest@vger.kernel.org Suggested-by: Miroslav Lichvar <mlichvar@redhat.com> Signed-off-by: John Stultz <john.stultz@linaro.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Without this commit the following intervals [x y), (x y) were be
replaced to (y-1 y) by snd_interval_refine_last(). This was also done
if y-1 is part of the previous interval.
With this changes it will be replaced with [y-1 y) in case of y-1 is
part of the previous interval. A similar behavior will be used for
snd_interval_refine_first().
This commit adapts the changes for alsa-lib of commit 9bb985c ("pcm: snd_interval_refine_first/last: exclude value only if
also excluded before")
Return ERR_PTR(-EINVAL) if kfd_get_process fails to find the process.
This fixes kernel oopses when a child process calls KFD ioctls with
a file descriptor inherited from the parent process.
Signed-off-by: Wei Lu <wei.lu2@amd.com> Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com> Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com> Acked-by: Christian König <christian.koenig@amd.com> Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com> Signed-off-by: Sasha Levin <alexander.levin@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Locking the root adapter for __i2c_transfer will deadlock if the
device sits behind a mux-locked I2C mux. Switch to the finer-grained
i2c_lock_bus with the I2C_LOCK_SEGMENT flag. If the device does not
sit behind a mux-locked mux, the two locking variants are equivalent.
Signed-off-by: Peter Rosin <peda@axentia.se> Acked-by: Dmitry Torokhov <dmitry.torokhov@gmail.com> Signed-off-by: Wolfram Sang <wsa@the-dreams.de> Signed-off-by: Sasha Levin <alexander.levin@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Locking the root adapter for __i2c_transfer will deadlock if the
device sits behind a mux-locked I2C mux. Switch to the finer-grained
i2c_lock_bus with the I2C_LOCK_SEGMENT flag. If the device does not
sit behind a mux-locked mux, the two locking variants are equivalent.
Signed-off-by: Peter Rosin <peda@axentia.se> Acked-by: Lee Jones <lee.jones@linaro.org> Signed-off-by: Wolfram Sang <wsa@the-dreams.de> Signed-off-by: Sasha Levin <alexander.levin@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Since we put static variable to a header file it's copied to each module
that includes the header. But not all of them are actually used it.
Mark gpio_suffixes array with __maybe_unused to hide a compiler warning:
In file included from
drivers/gpio/gpiolib-legacy.c:6:0:
drivers/gpio/gpiolib.h:95:27: warning: ‘gpio_suffixes’ defined but not used [-Wunused-const-variable=]
static const char * const gpio_suffixes[] = { "gpios", "gpio" };
^~~~~~~~~~~~~
In file included from drivers/gpio/gpiolib-devprop.c:17:0:
drivers/gpio/gpiolib.h:95:27: warning: ‘gpio_suffixes’ defined but not used [-Wunused-const-variable=]
static const char * const gpio_suffixes[] = { "gpios", "gpio" };
^~~~~~~~~~~~~
platform_get_resource() may fail and return NULL, so we should
better check it's return value to avoid a NULL pointer dereference
a bit later in the code.
This is detected by Coccinelle semantic patch.
@@
expression pdev, res, n, t, e, e1, e2;
@@
res = platform_get_resource(pdev, t, n);
+ if (!res)
+ return -EINVAL;
... when != res == NULL
e = devm_ioremap(e1, res->start, e2);
Probing the TPIU driver under UBSan triggers an out-of-bounds shift
warning in coresight_timeout():
...
[ 5.677530] UBSAN: Undefined behaviour in drivers/hwtracing/coresight/coresight.c:929:16
[ 5.685542] shift exponent 64 is too large for 64-bit type 'long unsigned int'
...
On closer inspection things are exponentially out of whack because we're
passing a bitmask where a bit number should be. Amusingly, it seems that
both calls will find their expected values by sheer luck and appear to
succeed: 1 << FFCR_FON_MAN ends up at bit 64 which whilst undefined
evaluates as zero in practice, while 1 << FFSR_FT_STOPPED finds bit 2
(TCPresent) which apparently is usually tied high.
Following the examples of other drivers, define separate FOO and FOO_BIT
macros for masks vs. indices, and put things right.
CC: Robert Walker <robert.walker@arm.com> CC: Mike Leach <mike.leach@linaro.org> CC: Mathieu Poirier <mathieu.poirier@linaro.org> Fixes: 11595db8e17f ("coresight: Fix disabling of CoreSight TPIU") Signed-off-by: Robin Murphy <robin.murphy@arm.com> Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Sasha Levin <alexander.levin@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Depending on the kernel configuration, early ARM architecture setup code
may have attached the GPU to a DMA/IOMMU mapping that transparently uses
the IOMMU to back the DMA API. Tegra requires special handling for IOMMU
backed buffers (a special bit in the GPU's MMU page tables indicates the
memory path to take: via the SMMU or directly to the memory controller).
Transparently backing DMA memory with an IOMMU prevents Nouveau from
properly handling such memory accesses and causes memory access faults.
As a side-note: buffers other than those allocated in instance memory
don't need to be physically contiguous from the GPU's perspective since
the GPU can map them into contiguous buffers using its own MMU. Mapping
these buffers through the IOMMU is unnecessary and will even lead to
performance degradation because of the additional translation. One
exception to this are compressible buffers which need large pages. In
order to enable these large pages, multiple small pages will have to be
combined into one large (I/O virtually contiguous) mapping via the
IOMMU. However, that is a topic outside the scope of this fix and isn't
currently supported. An implementation will want to explicitly create
these large pages in the Nouveau driver, so detaching from a DMA/IOMMU
mapping would still be required.
Signed-off-by: Thierry Reding <treding@nvidia.com> Acked-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Robin Murphy <robin.murphy@arm.com> Tested-by: Nicolas Chauvet <kwizart@gmail.com> Signed-off-by: Ben Skeggs <bskeggs@redhat.com> Signed-off-by: Sasha Levin <alexander.levin@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
For eMMC devices it is valid to only support 1.8V signaling. When
vqmmc is set to a fixed 1.8V regulator the stack tries to set 3.3V
initially and prints the following warning:
mmc1: Switching to 3.3V signalling voltage failed
Clear the MMC_SIGNAL_VOLTAGE_330 flag in case 3.3V is signaling is
not available. This prevents the stack from even trying to use
3.3V signaling and avoids the above warning.
The stack assumes that SDHC controller which support SD3.0 (SDR104) do
support HS200. This is not the case for Tegra 3, which does support SD
3.0
but only supports eMMC spec 4.41.
Use SDHCI_QUIRK2_BROKEN_HS200 to indicate that the controller does not
support HS200.
Note that commit 156e14b126ff ("mmc: sdhci: fix caps2 for HS200") added
the tie between SD3.0 (SDR104) and HS200. I don't think that this is
necessarly true. It is fully legitimate to support SD3.0 and not support
HS200. The quirk naming suggests something is broken in the controller,
but this is not the case: The controller simply does not support HS200.
of_find_compatible_node() returns a device node with refcount incremented
and thus needs an explicit of_node_put(). Further relying on an unchecked
of_iomap() which can return NULL is problematic here, after all ctrl_base
is critical enough for hix5hd2_set_cpu() to call BUG() if not available
so a check seems mandated here.
of_iomap() can return NULL which seems critical here and thus should be
explicitly flagged so that the cause of system halting can be understood.
As of_find_compatible_node() is returning a device node with refcount
incremented it must be explicitly decremented here.
Signed-off-by: Nicholas Mc Guire <hofrat@osadl.org> Fixes: commit 7fda91e73155 ("ARM: hisi: enable smp for HiP01") Signed-off-by: Wei Xu <xuwei5@hisilicon.com> Signed-off-by: Sasha Levin <alexander.levin@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Relying on an unchecked of_iomap() which can return NULL is problematic
here, an explicit check seems mandatory. Also the call to
of_find_compatible_node() returns a device node with refcount incremented
therefor an explicit of_node_put() is needed here.
7e1550b8f208 ("efi: Drop type and attribute checks in efi_mem_desc_lookup()")
refactored the implementation of efi_mem_desc_lookup() so that the type
check is moved to the callers, one of which is the x86 version of
efi_arch_mem_reserve(), where we added a modified check that only takes
EFI_BOOT_SERVICES_DATA regions into account.
This is reasonable, since it is the only memory type that requires this,
but doing so uncovered some unexpected behavior in the ESRT code, which
permits the ESRT table to reside in other types of memory than what the
UEFI spec mandates (i.e., EFI_BOOT_SERVICES_DATA), and unconditionally
calls efi_mem_reserve() on the region in question. This may result in
errors such as
esrt: Reserving ESRT space from 0x000000009c810318 to 0x000000009c810350.
efi: Failed to lookup EFI memory descriptor for 0x000000009c810318
when the ESRT table is not in EFI_BOOT_SERVICES_DATA memory, but we try
to reserve it nonetheless.
So make the call to efi_mem_reserve() conditional on the memory type.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Jones <pjones@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-efi@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Ingo Molnar <mingo@kernel.org> Signed-off-by: Sasha Levin <alexander.levin@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This patch fixes a bug where configfs_register_group had added
a group in a tree, and userspace has done a rmdir on a dir somewhere
above that group and we hit a kernel crash. The problem is configfs_rmdir
will detach everything under it and unlink groups on the default_groups
list. It will not unlink groups added with configfs_register_group so when
configfs_unregister_group is called to drop its references to the group/items
we crash when we try to access the freed dentrys.
The patch just adds a check for if a rmdir has been done above
us and if so just does the unlink part of unregistration.
Sorry if you are getting this multiple times. I thouhgt I sent
this to some of you and lkml, but I do not see it.
Signed-off-by: Mike Christie <mchristi@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Joel Becker <jlbec@evilplan.org> Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Sasha Levin <alexander.levin@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When EVM attempts to appraise a file signed with a crypto algorithm the
kernel doesn't have support for, it will cause the kernel to trigger a
module load. If the EVM policy includes appraisal of kernel modules this
will in turn call back into EVM - since EVM is holding a lock until the
crypto initialisation is complete, this triggers a deadlock. Add a
CRYPTO_NOLOAD flag and skip module loading if it's set, and add that flag
in the EVM case in order to fail gracefully with an error message
instead of deadlocking.
Signed-off-by: Matthew Garrett <mjg59@google.com> Acked-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: Mimi Zohar <zohar@linux.vnet.ibm.com> Signed-off-by: Sasha Levin <alexander.levin@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The first checks in mtdchar_read() and mtdchar_write() attempt to limit
`count` such that `*ppos + count <= mtd->size`. However, they ignore the
possibility of `*ppos > mtd->size`, allowing the calculation of `count` to
wrap around. `mtdchar_lseek()` prevents seeking beyond mtd->size, but the
pread/pwrite syscalls bypass this.
I haven't found any codepath on which this actually causes dangerous
behavior, but it seems like a sensible change anyway.
audit_add_watch stores locally krule->watch without taking a reference
on watch. Then, it calls audit_add_to_parent, and uses the watch stored
locally.
Unfortunately, it is possible that audit_add_to_parent updates
krule->watch.
When it happens, it also drops a reference of watch which
could free the watch.
The second call to auditctl triggers the use-after-free, because
audit_to_parent updates krule->watch to use a previous existing watch
and drops the reference to the newly created watch.
To fix the issue, we grab a reference of watch and we release it at the
end of the function.
Signed-off-by: Ronny Chevalier <ronny.chevalier@hp.com> Reviewed-by: Richard Guy Briggs <rgb@redhat.com> Signed-off-by: Paul Moore <paul@paul-moore.com> Signed-off-by: Sasha Levin <alexander.levin@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The regset API documented in <linux/regset.h> defines -ENODEV as the
result of the `->active' handler to be used where the feature requested
is not available on the hardware found. However code handling core file
note generation in `fill_thread_core_info' interpretes any non-zero
result from the `->active' handler as the regset requested being active.
Consequently processing continues (and hopefully gracefully fails later
on) rather than being abandoned right away for the regset requested.
Fix the problem then by making the code proceed only if a positive
result is returned from the `->active' handler.
Signed-off-by: Maciej W. Rozycki <macro@mips.com> Signed-off-by: Paul Burton <paul.burton@mips.com> Fixes: 4206d3aa1978 ("elf core dump: notes user_regset")
Patchwork: https://patchwork.linux-mips.org/patch/19332/ Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: James Hogan <jhogan@kernel.org> Cc: Ralf Baechle <ralf@linux-mips.org> Cc: linux-fsdevel@vger.kernel.org Cc: linux-mips@linux-mips.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Sasha Levin <alexander.levin@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The previous fix broke recovery of delegated stateids because it assumes
that if we did not mark the delegation as suspect, then the delegation has
effectively been revoked, and so it removes that delegation irrespectively
of whether or not it is valid and still in use. While this is "mostly
harmless" for ordinary I/O, we've seen pNFS fail with LAYOUTGET spinning
in an infinite loop while complaining that we're using an invalid stateid
(in this case the all-zero stateid).
What we rather want to do here is ensure that the delegation is always
correctly marked as needing testing when that is the case. So we want
to close the loophole offered by nfs4_schedule_stateid_recovery(),
which marks the state as needing to be reclaimed, but not the
delegation that may be backing it.
Perf can record user stack data in response to a synchronous request, such
as a tracepoint firing. If this happens under set_fs(KERNEL_DS), then we
end up reading user stack data using __copy_from_user_inatomic() under
set_fs(KERNEL_DS). I think this conflicts with the intention of using
set_fs(KERNEL_DS). And it is explicitly forbidden by hardware on ARM64
when both CONFIG_ARM64_UAO and CONFIG_ARM64_PAN are used.
So fix this by forcing USER_DS when recording user stack data.
Signed-off-by: Yabin Cui <yabinc@google.com> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: <stable@vger.kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Fixes: 88b0193d9418 ("perf/callchain: Force USER_DS when invoking perf_callchain_user()") Link: http://lkml.kernel.org/r/20180823225935.27035-1-yabinc@google.com Signed-off-by: Ingo Molnar <mingo@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The problem is that "entryptr + next_offset" and "entryptr + len + size"
can wrap. I ended up changing the type of "entryptr" because it makes
the math easier when we don't have to do so much casting.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by: Steve French <stfrench@microsoft.com> Reviewed-by: Aurelien Aptel <aaptel@suse.com> Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com> CC: Stable <stable@vger.kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The "old_entry + le32_to_cpu(pDirInfo->NextEntryOffset)" can wrap
around so I have added a check for integer overflow.
Reported-by: Dr Silvio Cesare of InfoSect <silvio.cesare@gmail.com> Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com> Reviewed-by: Aurelien Aptel <aaptel@suse.com> Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by: Steve French <stfrench@microsoft.com> CC: Stable <stable@vger.kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The patch causes a regression, which I cannot find the reason for.
So let's revert for now, as a revert hurts only performance.
Original report:
I was trying to resolve the problem with Oliver but we don't get any conclusion
for 5 months, so I am now sending this to mail list and cdc_acm authors.
I am using simple request-response protocol to obtain the boiller parameters
in constant intervals.
A simple one transaction is:
1. opening the /dev/ttyACM0
2. sending the following 10-bytes request to the device:
unsigned char req[] = {0x02, 0xfe, 0x01, 0x05, 0x08, 0x02, 0x01, 0x69, 0xab, 0x03};
3. reading response (frame of 74 bytes length).
4. closing the descriptor
I am doing this transaction with 5 seconds intervals.
Before the bad commit everything was working correctly: I've got a requests and
a responses in a timely manner.
After the bad commit more time I am using the kernel module, more problems I have.
The graph [2] is showing the problem.
As you can see after module load all seems fine but after about 30 minutes I've got
a plenty of EAGAINs when doing read()'s and trying to read back the data.
When I rmmod and insmod the cdc_acm module again, then the situation is starting
over again: running ok shortly after load, and more time it is running, more EAGAINs
I have when calling read().
As a bonus I can see the problem on the device itself:
The device is configured as you can see here on this screen [3].
It has two transmision LEDs: TX and RX. Blink duration is set for 100ms.
This is a recording before the bad commit when all is working fine: [4]
And this is with the bad commit: [5]
As you can see the TX led is blinking wrongly long (indicating transmission?)
and I have problems doing read() calls (EAGAIN).
wdm_in_callback() is a completion handler function for the USB driver.
So it should not sleep. But it calls service_outstanding_interrupt(),
which calls usb_submit_urb() with GFP_KERNEL.
To fix this bug, GFP_KERNEL is replaced with GFP_ATOMIC.
This bug is found by my static analysis tool DSAC.
If the written data starts with a digit, yurex_write() tries to parse
it as an integer using simple_strtoull(). This requires a null-
terminator, and currently there's no guarantee that there is one.
(The sample program at
https://github.com/NeoCat/YUREX-driver-for-Linux/blob/master/sample/yurex_clock.pl
writes an integer without a null terminator. It seems like it must
have worked by chance!)
Always add a null byte after the written data. Enlarge the buffer
to allow for this.
Cc: stable@vger.kernel.org Signed-off-by: Ben Hutchings <ben.hutchings@codethink.co.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Similarly to a recently reported bug in io_ti, a malicious USB device
could set port_number to a negative value and we would underflow the
port array in the interrupt completion handler.
As these devices only have one or two ports, fix this by making sure we
only consider the seventh bit when determining the port number (and
ignore bits 0xb0 which are typically set to 0x30).
async_complete() in uss720.c is a completion handler function for the
USB driver. So it should not sleep, but it is can sleep according to the
function call paths (from bottom to top) in Linux-4.16.
[FUNC] set_1284_register(GFP_KERNEL)
drivers/usb/misc/uss720.c, 372:
set_1284_register in parport_uss720_frob_control
drivers/parport/ieee1284.c, 560:
[FUNC_PTR]parport_uss720_frob_control in parport_ieee1284_ack_data_avail
drivers/parport/ieee1284.c, 577:
parport_ieee1284_ack_data_avail in parport_ieee1284_interrupt
./include/linux/parport.h, 474:
parport_ieee1284_interrupt in parport_generic_irq
drivers/usb/misc/uss720.c, 116:
parport_generic_irq in async_complete
[FUNC] get_1284_register(GFP_KERNEL)
drivers/usb/misc/uss720.c, 382:
get_1284_register in parport_uss720_read_status
drivers/parport/ieee1284.c, 555:
[FUNC_PTR]parport_uss720_read_status in parport_ieee1284_ack_data_avail
drivers/parport/ieee1284.c, 577:
parport_ieee1284_ack_data_avail in parport_ieee1284_interrupt
./include/linux/parport.h, 474:
parport_ieee1284_interrupt in parport_generic_irq
drivers/usb/misc/uss720.c, 116:
parport_generic_irq in async_complete
Note that [FUNC_PTR] means a function pointer call is used.
To fix these bugs, GFP_KERNEL is replaced with GFP_ATOMIC.
These bugs are found by my static analysis tool DSAC.
As reported by Dan Carpenter, a malicious USB device could set
port_number to a negative value and we would underflow the port array in
the interrupt completion handler.
As these devices only have one or two ports, fix this by making sure we
only consider the seventh bit when determining the port number (and
ignore bits 0xb0 which are typically set to 0x30).
Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2") Cc: stable <stable@vger.kernel.org> Reported-by: Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by: Johan Hovold <johan@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit f16443a034c7 ("USB: gadgetfs, dummy-hcd, net2280: fix locking
for callbacks") was based on a serious misunderstanding. It
introduced regressions into both the dummy-hcd and net2280 drivers.
The problem in dummy-hcd was fixed by commit 7dbd8f4cabd9 ("USB:
dummy-hcd: Fix erroneous synchronization change"), but the problem in
net2280 remains. Namely: the ->disconnect(), ->suspend(), ->resume(),
and ->reset() callbacks must be invoked without the private lock held;
otherwise a deadlock will occur when the callback routine tries to
interact with the UDC driver.
This patch largely is a reversion of the relevant parts of f16443a034c7. It also drops the private lock around the calls to
->suspend() and ->resume() (something the earlier patch forgot to do).
This is safe from races with device interrupts because it occurs
within the interrupt handler.
Finally, the patch changes where the ->disconnect() callback is
invoked when net2280_pullup() turns the pullup off. Rather than
making the callback from within stop_activity() at a time when dropping
the private lock could be unsafe, the callback is moved to a point
after the lock has already been dropped.
Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Fixes: f16443a034c7 ("USB: gadgetfs, dummy-hcd, net2280: fix locking for callbacks") Reported-by: D. Ziesche <dziesche@zes.com> Tested-by: D. Ziesche <dziesche@zes.com> CC: <stable@vger.kernel.org> Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This patch fixes an issue that maxpacket size of ep0 is incorrect
for SuperSpeed. Otherwise, CDC NCM class with SuperSpeed doesn't
work correctly on this driver because its control read data size
is more than 64 bytes.
i_usX2Y_subs_startup in usbusx2yaudio.c is a completion handler function
for the USB driver. So it should not sleep, but it is can sleep
according to the function call paths (from bottom to top) in Linux-4.16.
[FUNC] msleep
drivers/usb/host/u132-hcd.c, 2558:
msleep in u132_get_frame
drivers/usb/core/hcd.c, 2231:
[FUNC_PTR]u132_get_frame in usb_hcd_get_frame_number
drivers/usb/core/usb.c, 822:
usb_hcd_get_frame_number in usb_get_current_frame_number
sound/usb/usx2y/usbusx2yaudio.c, 303:
usb_get_current_frame_number in i_usX2Y_urb_complete
sound/usb/usx2y/usbusx2yaudio.c, 366:
i_usX2Y_urb_complete in i_usX2Y_subs_startup
Note that [FUNC_PTR] means a function pointer call is used.
To fix this bug, msleep() is replaced with mdelay().
This bug is found by my static analysis tool DSAC.
The steps taken by usb core to set a new interface is very different from
what is done on the xHC host side.
xHC hardware will do everything in one go. One command is used to set up
new endpoints, free old endpoints, check bandwidth, and run the new
endpoints.
All this is done by xHC when usb core asks the hcd to check for
available bandwidth. At this point usb core has not yet flushed the old
endpoints, which will cause use-after-free issues in xhci driver as
queued URBs are cancelled on a re-allocated endpoint.
To resolve this add a call to usb_disable_interface() which will flush
the endpoints before calling usb_hcd_alloc_bandwidth()
Additional checks in xhci driver will also be implemented to gracefully
handle stale URB cancel on freed and re-allocated endpoints
The hope that UAS devices would be less broken than old style storage
devices has turned out to be unfounded. Make UAS support more of the
quirk flags of the old driver.
This device does not correctly handle the LPM operations.
Also, the device cannot handle ATA pass-through commands
and locks up when attempted while running in super speed.
This patch adds the equivalent quirk logic as found in uas.
Signed-off-by: Tim Anderson <tsa@biglakesoftware.com> Acked-by: Alan Stern <stern@rowland.harvard.edu> Cc: stable <stable@vger.kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Some of the ME clients are available only for BIOS operation and are
removed during hand off to an OS. However the removal is not instant.
A client may be visible on the client list when the mei driver requests
for enumeration, while the subsequent request for properties will be
answered with client not found error value. The default behavior
for an error is to perform client reset while this error is harmless and
the link reset should be prevented. This issue started to be visible due to
suspend/resume timing changes. Currently reported only on the Haswell
based system.