Filipe Manana [Tue, 8 Jan 2019 11:43:07 +0000 (11:43 +0000)]
Btrfs: fix race between reflink/dedupe and relocation
The recent rework that makes btrfs' remap_file_range operation use the
generic helper generic_remap_file_range_prep() introduced a race between
relocation and reflinking (for both cloning and deduplication) the file
extents between the source and destination inodes.
This happens because we no longer lock the source range anymore, and we do
not lock it anymore because we wait for direct IO writes and writeback to
complete early on the code path right after locking the inodes, which
guarantees no other file operations interfere with the reflinking. However
there is one exception which is relocation, since it replaces the byte
number of file extents items in the fs tree after locking the range the
file extent items represent. This is a problem because after finding each
file extent to clone in the fs tree, the reflink process copies the file
extent item into a local buffer, releases the search path, inserts new
file extent items in the destination range and then increments the
reference count for the extent mentioned in the file extent item that it
previously copied to the buffer. If right after copying the file extent
item into the buffer and releasing the path the relocation process
updates the file extent item to point to the new extent, the reflink
process ends up creating a delayed reference to increment the reference
count of the old extent, for which the relocation process already created
a delayed reference to drop it. This results in failure to run delayed
references because we will attempt to increment the count of a reference
that was already dropped. This is illustrated by the following diagram:
CPU 1 CPU 2
relocation is running
btrfs_clone_files()
btrfs_clone()
--> finds extent item
in source range
point to extent
at bytenr X
--> copies it into a
local buffer
--> releases path
replace_file_extents()
--> successfully locks the
range represented by
the file extent item
--> replaces disk_bytenr
field in the file
extent item with some
other value Y
--> creates delayed reference
to increment reference
count for extent at
bytenr Y
--> creates delayed reference
to drop the extent at
bytenr X
--> starts transaction
--> creates delayed
reference to
increment extent
at bytenr X
<delayed references are run, due to a transaction
commit for example, and the transaction is aborted
with -EIO because we attempt to increment reference
count for the extent at bytenr X after we freed it>
When this race is hit the running transaction ends up getting aborted with
an -EIO error and a trace like the following is produced:
Fix this by locking the source range before searching for the file extent
items in the fs tree, since the relocation process will try to lock the
range a file extent item represents before updating it with the new extent
location.
Fixes: 34a28e3d7753 ("Btrfs: use generic_remap_file_range_prep() for cloning and deduplication") Signed-off-by: Filipe Manana <[email protected]> Signed-off-by: David Sterba <[email protected]>
Filipe Manana [Tue, 8 Jan 2019 11:42:54 +0000 (11:42 +0000)]
Btrfs: fix race between cloning range ending at eof and writeback
The recent rework that makes btrfs' remap_file_range operation use the
generic helper generic_remap_file_range_prep() introduced a race between
writeback and cloning a range that covers the eof extent of the source
file into a destination offset that is greater then the same file's size.
This happens because we now wait for writeback to complete before doing
the truncation of the eof block, while previously we did the truncation
and then waited for writeback to complete. This leads to a race between
writeback of the truncated block and cloning the file extents in the
source range, because we copy each file extent item we find in the fs
root into a buffer, then release the path and then increment the reference
count for the extent referred in that file extent item we copied, which
can no longer exist if writeback of the truncated eof block completes
after we copied the file extent item into the buffer and before we
incremented the reference count. This is illustrated by the following
diagram:
CPU 1 CPU 2
btrfs_clone_files()
btrfs_cont_expand()
btrfs_truncate_block()
--> zeroes part of the
page containg eof,
marking it for
delalloc
btrfs_clone()
--> finds extent item
covering eof,
points to extent
at bytenr X
--> copies it into a
local buffer
--> releases path
writeback starts
btrfs_finish_ordered_io()
insert_reserved_file_extent()
__btrfs_drop_extents()
--> creates delayed
reference to drop
the extent at
bytenr X
--> starts transaction
--> creates delayed
reference to
increment extent
at bytenr X
<delayed references are run, due to a transaction
commit for example, and the transaction is aborted
with -EIO because we attempt to increment reference
count for the extent at bytenr X after we freed it>
When this race is hit the running transaction ends up getting aborted with
an -EIO error and a trace like the following is produced:
Fix this by waiting for writeback to complete after truncating the eof
block.
Fixes: 34a28e3d7753 ("Btrfs: use generic_remap_file_range_prep() for cloning and deduplication") Signed-off-by: Filipe Manana <[email protected]> Signed-off-by: David Sterba <[email protected]>
Quentin Perret [Wed, 9 Jan 2019 10:42:36 +0000 (10:42 +0000)]
cpufreq: scmi: Fix frequency invariance in slow path
The scmi-cpufreq driver calls the arch_set_freq_scale() callback on
frequency changes to provide scale-invariant load-tracking signals to
the scheduler. However, in the slow path, it does so while specifying
the current and max frequencies in different units, hence resulting in a
broken freq_scale factor.
Fix this by passing all frequencies in KHz, as stored in the CPUFreq
frequency table.
Tomi Valkeinen [Thu, 3 Jan 2019 11:59:52 +0000 (13:59 +0200)]
drm/bridge: tc358767: reject modes which require too much BW
The current driver accepts any videomode with pclk < 154MHz. This is not
correct, as with 1 lane and/or 1.62Mbps speed not all videomodes can be
supported.
Add code to reject modes that require more bandwidth that is available.
Tomi Valkeinen [Thu, 3 Jan 2019 11:59:51 +0000 (13:59 +0200)]
drm/bridge: tc358767: fix initial DP0/1_SRCCTRL value
Initially DP0_SRCCTRL is set to a static value which includes
DP0_SRCCTRL_LANES_2 and DP0_SRCCTRL_BW27, even when only 1 lane of
1.62Gbps speed is used. DP1_SRCCTRL is configured to a magic number.
This patch changes the configuration as follows:
Configure DP0_SRCCTRL by using tc_srcctrl() which provides the correct
value.
DP1_SRCCTRL needs two bits to be set to the same value as DP0_SRCCTRL:
SSCG and BW27. All other bits can be zero.
Tomi Valkeinen [Thu, 3 Jan 2019 11:59:48 +0000 (13:59 +0200)]
drm/bridge: tc358767: add bus flags
tc358767 driver does not set DRM bus_flags, even if it does configures
the polarity settings into its registers. This means that the DPI source
can't configure the polarities correctly.
Borislav Petkov [Tue, 8 Jan 2019 16:38:29 +0000 (17:38 +0100)]
x86/cache: Rename config option to CONFIG_X86_RESCTRL
CONFIG_RESCTRL is too generic. The final goal is to have a generic
option called like this which is selected by the arch-specific ones
CONFIG_X86_RESCTRL and CONFIG_ARM64_RESCTRL. The generic one will
cover the resctrl filesystem and other generic and shared bits of
functionality.
Zhenyu Wang [Sat, 29 Dec 2018 03:13:10 +0000 (11:13 +0800)]
drm/i915/gvt: Fix workload request allocation before request add
In commit 6bb2a2af8b1b ("drm/i915/gvt: Fix crash after request->hw_context change"),
forgot to handle workload scan path in ELSP handler case which was to
optimize scanning earlier instead of in gvt submission thread, so request
alloc and add was splitting then which is against right process.
This trys to do a partial revert of that commit which still has workload
request alloc helper and make sure shadow state population is handled after
request alloc for target state buffer.
v3: Fix missed workload status setting in request alloc error path
v2: Fix dispatch workload err path that should add request after alloc anyway.
Jeff Moyer [Tue, 8 Jan 2019 21:57:34 +0000 (16:57 -0500)]
block: clarify documentation for blk_{start|finish}_plug
There was some confusion about what these functions did. Make it clear
that this is a hint for upper layers to pass to the block layer, and
that it does not guarantee that I/O will not be submitted between a
start and finish plug.
Linus Torvalds [Wed, 9 Jan 2019 02:58:29 +0000 (18:58 -0800)]
Merge branch 'akpm' (patches from Andrew)
Merge misc fixes from Andrew Morton:
"14 fixes"
* emailed patches from Andrew Morton <[email protected]>:
mm, page_alloc: do not wake kswapd with zone lock held
hugetlbfs: revert "use i_mmap_rwsem for more pmd sharing synchronization"
hugetlbfs: revert "Use i_mmap_rwsem to fix page fault/truncate race"
mm: page_mapped: don't assume compound page is huge or THP
mm/memory.c: initialise mmu_notifier_range correctly
tools/vm/page_owner: use page_owner_sort in the use example
kasan: fix krealloc handling for tag-based mode
kasan: make tag based mode work with CONFIG_HARDENED_USERCOPY
kasan, arm64: use ARCH_SLAB_MINALIGN instead of manual aligning
mm, memcg: fix reclaim deadlock with writeback
mm/usercopy.c: no check page span for stack objects
slab: alien caches must not be initialized if the allocation of the alien cache failed
fork, memcg: fix cached_stacks case
zram: idle writeback fixes and cleanup
Stafford Horne [Tue, 8 Jan 2019 13:15:15 +0000 (22:15 +0900)]
arch/openrisc: Fix issues with access_ok()
The commit 594cc251fdd0 ("make 'user_access_begin()' do 'access_ok()'")
exposed incorrect implementations of access_ok() macro in several
architectures. This change fixes 2 issues found in OpenRISC.
OpenRISC was not properly using parenthesis for arguments and also using
arguments twice. This patch fixes those 2 issues.
I test booted this patch with v5.0-rc1 on qemu and it's working fine.
Mel Gorman [Tue, 8 Jan 2019 23:23:39 +0000 (15:23 -0800)]
mm, page_alloc: do not wake kswapd with zone lock held
syzbot reported the following regression in the latest merge window and
it was confirmed by Qian Cai that a similar bug was visible from a
different context.
======================================================
WARNING: possible circular locking dependency detected
4.20.0+ #297 Not tainted
------------------------------------------------------
syz-executor0/8529 is trying to acquire lock: 000000005e7fb829 (&pgdat->kswapd_wait){....}, at:
__wake_up_common_lock+0x19e/0x330 kernel/sched/wait.c:120
It appears to be a false positive in that the only way the lock ordering
should be inverted is if kswapd is waking itself and the wakeup
allocates debugging objects which should already be allocated if it's
kswapd doing the waking. Nevertheless, the possibility exists and so
it's best to avoid the problem.
This patch flags a zone as needing a kswapd using the, surprisingly,
unused zone flag field. The flag is read without the lock held to do
the wakeup. It's possible that the flag setting context is not the same
as the flag clearing context or for small races to occur. However, each
race possibility is harmless and there is no visible degredation in
fragmentation treatment.
While zone->flag could have continued to be unused, there is potential
for moving some existing fields into the flags field instead.
Particularly read-mostly ones like zone->initialized and
zone->contiguous.
The reverted commit caused issues with migration and poisoning of anon
huge pages. The LTP move_pages12 test will cause an "unable to handle
kernel NULL pointer" BUG would occur with stack similar to:
The purpose of the reverted patch was to fix some long existing races
with huge pmd sharing. It used i_mmap_rwsem for this purpose with the
idea that this could also be used to address truncate/page fault races
with another patch. Further analysis has determined that i_mmap_rwsem
can not be used to address all these hugetlbfs synchronization issues.
Therefore, revert this patch while working an another approach to the
underlying issues.
The reverted commit caused ABBA deadlocks when file migration raced with
file eviction for specific hugetlbfs files. This was discovered with a
modified version of the LTP move_pages12 test.
The purpose of the reverted patch was to close a long existing race
between hugetlbfs file truncation and page faults. After more analysis
of the patch and impacted code, it was determined that i_mmap_rwsem can
not be used for all required synchronization. Therefore, revert this
patch while working an another approach to the underlying issue.
Jan Stancek [Tue, 8 Jan 2019 23:23:28 +0000 (15:23 -0800)]
mm: page_mapped: don't assume compound page is huge or THP
LTP proc01 testcase has been observed to rarely trigger crashes
on arm64:
page_mapped+0x78/0xb4
stable_page_flags+0x27c/0x338
kpageflags_read+0xfc/0x164
proc_reg_read+0x7c/0xb8
__vfs_read+0x58/0x178
vfs_read+0x90/0x14c
SyS_read+0x60/0xc0
The issue is that page_mapped() assumes that if compound page is not
huge, then it must be THP. But if this is 'normal' compound page
(COMPOUND_PAGE_DTOR), then following loop can keep running (for
HPAGE_PMD_NR iterations) until it tries to read from memory that isn't
mapped and triggers a panic:
for (i = 0; i < hpage_nr_pages(page); i++) {
if (atomic_read(&page[i]._mapcount) >= 0)
return true;
}
I could replicate this on x86 (v4.20-rc4-98-g60b548237fed) only
with a custom kernel module [1] which:
- allocates compound page (PAGEC) of order 1
- allocates 2 normal pages (COPY), which are initialized to 0xff (to
satisfy _mapcount >= 0)
- 2 PAGEC page structs are copied to address of first COPY page
- second page of COPY is marked as not present
- call to page_mapped(COPY) now triggers fault on access to 2nd COPY
page at offset 0x30 (_mapcount)
Right now tag-based KASAN can retag the memory that is reallocated via
krealloc and return a differently tagged pointer even if the same slab
object gets used and no reallocated technically happens.
There are a few issues with this approach. One is that krealloc callers
can't rely on comparing the return value with the passed argument to
check whether reallocation happened. Another is that if a caller knows
that no reallocation happened, that it can access object memory through
the old pointer, which leads to false positives. Look at
nf_ct_ext_add() to see an example.
Fix this by keeping the same tag if the memory don't actually gets
reallocated during krealloc.
kasan: make tag based mode work with CONFIG_HARDENED_USERCOPY
With CONFIG_HARDENED_USERCOPY enabled __check_heap_object() compares and
then subtracts a potentially tagged pointer with a non-tagged address of
the page that this pointer belongs to, which leads to unexpected
behavior.
Untag the pointer in __check_heap_object() before doing any of these
operations.
He adds
"task1 is waiting for the PageWriteback bit of the page that task2 has
collected in mpd->io_submit->io_bio, and tasks2 is waiting for the
LOCKED bit the page which tasks1 has locked"
More precisely task1 is handling a page fault and it has a page locked
while it charges a new page table to a memcg. That in turn hits a
memory limit reclaim and the memcg reclaim for legacy controller is
waiting on the writeback but that is never going to finish because the
writeback itself is waiting for the page locked in the #PF path. So
this is essentially ABBA deadlock:
This accumulating of more pages to flush is used by several filesystems
to generate a more optimal IO patterns.
Waiting for the writeback in legacy memcg controller is a workaround for
pre-mature OOM killer invocations because there is no dirty IO
throttling available for the controller. There is no easy way around
that unfortunately. Therefore fix this specific issue by pre-allocating
the page table outside of the page lock. We have that handy
infrastructure for that already so simply reuse the fault-around pattern
which already does this.
There are probably other hidden __GFP_ACCOUNT | GFP_KERNEL allocations
from under a fs page locked but they should be really rare. I am not
aware of a better solution unfortunately.
Shakeel Butt [Tue, 8 Jan 2019 23:22:57 +0000 (15:22 -0800)]
fork, memcg: fix cached_stacks case
Commit 5eed6f1dff87 ("fork,memcg: fix crash in free_thread_stack on
memcg charge fail") fixes a crash caused due to failed memcg charge of
the kernel stack. However the fix misses the cached_stacks case which
this patch fixes. So, the same crash can happen if the memcg charge of
a cached stack is failed.
Minchan Kim [Tue, 8 Jan 2019 23:22:53 +0000 (15:22 -0800)]
zram: idle writeback fixes and cleanup
This patch includes some fixes and cleanup for idle-page writeback.
1. writeback_limit interface
Now writeback_limit interface is rather conusing. For example, once
writeback limit budget is exausted, admin can see 0 from
/sys/block/zramX/writeback_limit which is same semantic with disable
writeback_limit at this moment. IOW, admin cannot tell that zero came
from disable writeback limit or exausted writeback limit.
To make the interface clear, let's sepatate enable of writeback limit to
another knob - /sys/block/zram0/writeback_limit_enable
* before:
while true :
# to re-enable writeback limit once previous one is used up
echo 0 > /sys/block/zram0/writeback_limit
echo $((200<<20)) > /sys/block/zram0/writeback_limit
..
.. # used up the writeback limit budget
* new
# To enable writeback limit, from the beginning, admin should
# enable it.
echo $((200<<20)) > /sys/block/zram0/writeback_limit
echo 1 > /sys/block/zram/0/writeback_limit_enable
while true :
echo $((200<<20)) > /sys/block/zram0/writeback_limit
..
.. # used up the writeback limit budget
The mode in writeback_store is not bit opeartion any more so no need to
use bit operations. Furthermore, current condition check is broken in
that it does writeback every pages regardless of huge/idle.
Lyude Paul [Tue, 8 Jan 2019 21:11:29 +0000 (16:11 -0500)]
drm/dp_mst: Add __must_check to drm_dp_mst_topology_mgr_resume()
Since I've had to fix two cases of drivers not checking the return code
from this function, let's make the compiler complain so this doesn't
come up again in the future.
Changes since v1:
* Remove unneeded __must_check in function declaration - danvet
Lyude Paul [Tue, 8 Jan 2019 21:11:28 +0000 (16:11 -0500)]
drm/amdgpu: Don't fail resume process if resuming atomic state fails
This is an ugly one unfortunately. Currently, all DRM drivers supporting
atomic modesetting will save the state that userspace had set before
suspending, then attempt to restore that state on resume. This probably
worked very well at one point, like many other things, until DP MST came
into the picture. While it's easy to restore state on normal display
connectors that were disconnected during suspend regardless of their
state post-resume, this can't really be done with MST because of the
fact that setting up a downstream sink requires performing sideband
transactions between the source and the MST hub, sending out the ACT
packets, etc.
Because of this, there isn't really a guarantee that we can restore the
atomic state we had before suspend once we've resumed. This sucks pretty
bad, but so far I haven't run into any compositors that this actually
causes serious issues with. Most compositors will notice the hotplug we
send afterwards, and then reprobe state.
Since nouveau and i915 also don't fail the suspend/resume process due to
failing to restore the atomic state, let's make amdgpu match this
behavior. Better to resume the GPU properly, then to stop the process
half way because of a potentially unavoidable atomic commit failure.
Eventually, we'll have a real fix for this problem on the DRM level. But
we've got some more important low-hanging fruit to deal with first.
Lyude Paul [Tue, 8 Jan 2019 21:11:27 +0000 (16:11 -0500)]
drm/amdgpu: Don't ignore rc from drm_dp_mst_topology_mgr_resume()
drm_dp_mst_topology_mgr_resume() returns whether or not it managed to
find the topology in question after a suspend resume cycle, and the
driver is supposed to check this value and disable MST accordingly if
it's gone-in addition to sending a hotplug in order to notify userspace
that something changed during suspend.
Currently, amdgpu just makes the mistake of ignoring the return code
from drm_dp_mst_topology_mgr_resume() which means that if a topology was
removed in suspend, amdgpu never notices and assumes it's still
connected which leads to all sorts of problems.
So, fix this by actually checking the rc from
drm_dp_mst_topology_mgr_resume(). Also, reformat the rest of the
function while we're at it to fix the over-indenting.
There is no code that decreases the reference count of stateful objects
in error path of the nft_add_set_elem(). this causes a leak of reference
count of stateful objects.
Test commands:
$nft add table ip filter
$nft add counter ip filter c1
$nft add map ip filter m1 { type ipv4_addr : counter \;}
$nft add element ip filter m1 { 1 : c1 }
$nft add element ip filter m1 { 1 : c1 }
$nft delete element ip filter m1 { 1 }
$nft delete counter ip filter c1
Result:
Error: Could not process rule: Device or resource busy
delete counter ip filter c1
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
At the second 'nft add element ip filter m1 { 1 : c1 }', the reference
count of the 'c1' is increased then it tries to insert into the 'm1'. but
the 'm1' already has same element so it returns -EEXIST.
But it doesn't decrease the reference count of the 'c1' in the error path.
Due to a leak of the reference count of the 'c1', the 'c1' can't be
removed by 'nft delete counter ip filter c1'.
Fixes: 8aeff920dcc9 ("netfilter: nf_tables: add stateful object reference to set elements") Signed-off-by: Taehee Yoo <[email protected]> Signed-off-by: Pablo Neira Ayuso <[email protected]>
Phil Sutter [Sun, 30 Dec 2018 16:27:43 +0000 (17:27 +0100)]
netfilter: nf_tables: Fix for endless loop when dumping ruleset
__nf_tables_dump_rules() stores the current idx value into cb->args[0]
before returning to caller. With multiple chains present, cb->args[0] is
therefore updated after each chain's rules have been traversed. This
though causes the final nf_tables_dump_rules() run (which should return
an skb->len of zero since no rules are left to dump) to continue dumping
rules for each but the first chain. Fix this by moving the cb->args[0]
update to nf_tables_dump_rules().
With no final action to be performed anymore in
__nf_tables_dump_rules(), drop 'out_unfinished' jump label and 'rc'
variable - instead return the appropriate value directly.
Fixes: 241faeceb849c ("netfilter: nf_tables: Speed up selective rule dumps") Signed-off-by: Phil Sutter <[email protected]> Signed-off-by: Pablo Neira Ayuso <[email protected]>
David S. Miller [Tue, 8 Jan 2019 21:53:54 +0000 (16:53 -0500)]
Merge branch 'mlxsw-fixes'
Daniel Borkmann says:
====================
pull-request: bpf 2019-01-08
The following pull-request contains BPF updates for your *net* tree.
The main changes are:
1) Fix BSD'ism in sendmsg(2) to rewrite unspecified IPv6 dst for
unconnected UDP sockets with [::1] _after_ cgroup BPF invocation,
from Andrey.
2) Follow-up fix to the speculation fix where we need to reject a
corner case for sanitation when ptr and scalars are mixed in the
same alu op. Also, some unrelated minor doc fixes, from Daniel.
3) Fix BPF kselftest's incorrect uses of create_and_get_cgroup()
by not assuming fd of zero value to be the result of an error
case, from Stanislav.
====================
Ido Schimmel [Tue, 8 Jan 2019 16:48:12 +0000 (16:48 +0000)]
selftests: forwarding: Fix test for different devices
When running the test on the Spectrum ASIC the generated packets are
counted on the ingress filter and injected back to the pipeline because
of the 'pass' action. The router block then drops the packets due to
checksum error, as the test generates packets with zero checksum.
When running the test on an emulator that is not as strict about
checksum errors the test fails since packets are counted twice. Once by
the emulated ASIC on its ingress filter and again by the kernel as the
emulator does not perform checksum validation and allows the packets to
be trapped by a matching host route.
Fix this by changing the action to 'drop', which will prevent the packet
from continuing further in the pipeline to the router block.
For veth pairs this change is essentially a NOP given packets are only
processed once (by the kernel).
Fixes: a0b61f3d8ebf ("selftests: forwarding: vxlan_bridge_1d: Add an ECN decap test") Signed-off-by: Ido Schimmel <[email protected]> Reviewed-by: Petr Machata <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Ido Schimmel [Tue, 8 Jan 2019 16:48:11 +0000 (16:48 +0000)]
net: bridge: Fix VLANs memory leak
When adding / deleting VLANs to / from a bridge port, the bridge driver
first tries to propagate the information via switchdev and falls back to
the 8021q driver in case the underlying driver does not support
switchdev. This can result in a memory leak [1] when VXLAN and mlxsw
ports are enslaved to the bridge:
$ ip link set dev vxlan0 master br0
# No mlxsw ports are enslaved to 'br0', so mlxsw ignores the switchdev
# notification and the bridge driver adds the VLAN on 'vxlan0' via the
# 8021q driver
$ bridge vlan add vid 10 dev vxlan0 pvid untagged
# mlxsw port is enslaved to the bridge
$ ip link set dev swp1 master br0
# mlxsw processes the switchdev notification and the 8021q driver is
# skipped
$ bridge vlan del vid 10 dev vxlan0
This results in 'struct vlan_info' and 'struct vlan_vid_info' being
leaked, as they were allocated by the 8021q driver during VLAN addition,
but never freed as the 8021q driver was skipped during deletion.
Fix this by introducing a new VLAN private flag that indicates whether
the VLAN was added on the port by switchdev or the 8021q driver. If the
VLAN was added by the 8021q driver, then we make sure to delete it via
the 8021q driver as well.
Ido Schimmel [Tue, 8 Jan 2019 16:48:10 +0000 (16:48 +0000)]
selftests: mlxsw: Add a test case for VLAN addition error flow
Add a test case for the issue fixed by previous commit. In case the
offloading of an unsupported VxLAN tunnel was triggered by adding the
mapped VLAN to a local port, then error should be returned to the user.
Ido Schimmel [Tue, 8 Jan 2019 16:48:08 +0000 (16:48 +0000)]
mlxsw: spectrum_nve: Replace error code with EINVAL
Adding a VLAN on a port can trigger the offload of a VXLAN tunnel which
is already a member in the VLAN. In case the configuration of the VXLAN
is not supported, the driver would return -EOPNOTSUPP.
This is problematic since bridge code does not interpret this as error,
but rather that it should try to setup the VLAN using the 8021q driver
instead of switchdev.
Fixes: d70e42b22dd4 ("mlxsw: spectrum: Enable VxLAN enslavement to VLAN-aware bridges") Signed-off-by: Ido Schimmel <[email protected]> Reviewed-by: Petr Machata <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Ido Schimmel [Tue, 8 Jan 2019 16:48:07 +0000 (16:48 +0000)]
mlxsw: spectrum_switchdev: Avoid returning errors in commit phase
Drivers are not supposed to return errors in switchdev commit phase if
they returned OK in prepare phase. Otherwise, a WARNING is emitted.
However, when the offloading of a VXLAN tunnel is triggered by the
addition of a VLAN on a local port, it is not possible to guarantee that
the commit phase will succeed without doing a lot of work.
In these cases, the artificial division between prepare and commit phase
does not make sense, so simply do the work in the prepare phase.
Fixes: d70e42b22dd4 ("mlxsw: spectrum: Enable VxLAN enslavement to VLAN-aware bridges") Signed-off-by: Ido Schimmel <[email protected]> Reviewed-by: Petr Machata <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Ido Schimmel [Tue, 8 Jan 2019 16:48:06 +0000 (16:48 +0000)]
mlxsw: spectrum: Add VXLAN dependency for spectrum
When VXLAN is a loadable module, MLXSW_SPECTRUM must not be built-in:
drivers/net/ethernet/mellanox/mlxsw/spectrum_switchdev.c:2547: undefined
reference to `vxlan_fdb_find_uc'
Add Kconfig dependency to enforce usable configurations.
Fixes: 1231e04f5bba ("mlxsw: spectrum_switchdev: Add support for VxLAN encapsulation") Signed-off-by: Ido Schimmel <[email protected]> Reported-by: kbuild test robot <[email protected]> Reviewed-by: Petr Machata <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Jiri Pirko [Tue, 8 Jan 2019 16:48:05 +0000 (16:48 +0000)]
mlxsw: spectrum: Disable lag port TX before removing it
Make sure that lag port TX is disabled before mlxsw_sp_port_lag_leave()
is called and prevent from possible EMAD error.
Fixes: 0d65fc13042f ("mlxsw: spectrum: Implement LAG port join/leave") Signed-off-by: Jiri Pirko <[email protected]> Signed-off-by: Ido Schimmel <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Nir Dotan [Tue, 8 Jan 2019 16:48:04 +0000 (16:48 +0000)]
mlxsw: spectrum_acl: Remove ASSERT_RTNL()s in module removal flow
Removal of the mlxsw driver on Spectrum-2 platforms hits an ASSERT_RTNL()
in Spectrum-2 ACL Bloom filter and in ERP removal paths. This happens
because the multicast router implementation in Spectrum-2 relies on ACLs.
Taking the RTNL lock upon driver removal is useless since the driver first
removes its ports and unregisters from notifiers so concurrent writes
cannot happen at that time. The assertions were originally put as a
reminder for future work involving ERP background optimization, but having
these assertions only during addition serves this purpose as well.
Therefore remove the ASSERT_RTNL() in both places related to ERP and Bloom
filter removal.
Fixes: cf7221a4f5a5 ("mlxsw: spectrum_router: Add Multicast routing support for Spectrum-2") Signed-off-by: Nir Dotan <[email protected]> Reviewed-by: Jiri Pirko <[email protected]> Signed-off-by: Ido Schimmel <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Nir Dotan [Tue, 8 Jan 2019 16:48:03 +0000 (16:48 +0000)]
mlxsw: spectrum_acl: Add cleanup after C-TCAM update error condition
When writing to C-TCAM, mlxsw driver uses cregion->ops->entry_insert().
In case of C-TCAM HW insertion error, the opposite action should take
place.
Add error handling case in which the C-TCAM region entry is removed, by
calling cregion->ops->entry_remove().
Heiner Kallweit [Mon, 7 Jan 2019 20:49:09 +0000 (21:49 +0100)]
r8169: load Realtek PHY driver module before r8169
This soft dependency works around an issue where sometimes the genphy
driver is used instead of the dedicated PHY driver. The root cause of
the issue isn't clear yet. People reported the unloading/re-loading
module r8169 helps, and also configuring this soft dependency in
the modprobe config files. Important just seems to be that the
realtek module is loaded before r8169.
Once this has been applied preliminary fix 38af4b903210 ("net: phy:
add workaround for issue where PHY driver doesn't bind to the device")
will be removed.
Fixes: f1e911d5d0df ("r8169: add basic phylib support") Signed-off-by: Heiner Kallweit <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Yu Zhao [Mon, 7 Jan 2019 22:51:15 +0000 (15:51 -0700)]
drm/amdgpu: validate user GEM object size
When creating frame buffer, userspace may request to attach to a
previously allocated GEM object that is smaller than what GPU
requires. Validation must be done to prevent out-of-bound DMA,
otherwise it could be exploited to reveal sensitive data.
This fix is not done in a common code path because individual
driver might have different requirement.
Yu Zhao [Mon, 7 Jan 2019 22:51:14 +0000 (15:51 -0700)]
drm/amdgpu: validate user pitch alignment
Userspace may request pitch alignment that is not supported by GPU.
Some requests 32, but GPU ignores it and uses default 64 when cpp is
4. If GEM object is allocated based on the smaller alignment, GPU
DMA will go out of bound.
David Herrmann [Tue, 8 Jan 2019 12:58:52 +0000 (13:58 +0100)]
fork: record start_time late
This changes the fork(2) syscall to record the process start_time after
initializing the basic task structure but still before making the new
process visible to user-space.
Technically, we could record the start_time anytime during fork(2). But
this might lead to scenarios where a start_time is recorded long before
a process becomes visible to user-space. For instance, with
userfaultfd(2) and TLS, user-space can delay the execution of fork(2)
for an indefinite amount of time (and will, if this causes network
access, or similar).
By recording the start_time late, it much closer reflects the point in
time where the process becomes live and can be observed by other
processes.
Lastly, this makes it much harder for user-space to predict and control
the start_time they get assigned. Previously, user-space could fork a
process and stall it in copy_thread_tls() before its pid is allocated,
but after its start_time is recorded. This can be misused to later-on
cycle through PIDs and resume the stalled fork(2) yielding a process
that has the same pid and start_time as a process that existed before.
This can be used to circumvent security systems that identify processes
by their pid+start_time combination.
Even though user-space was always aware that start_time recording is
flaky (but several projects are known to still rely on start_time-based
identification), changing the start_time to be recorded late will help
mitigate existing attacks and make it much harder for user-space to
control the start_time a process gets assigned.
tools include uapi: Sync linux/vhost.h with the kernel sources
To get the changes in:
4b86713236e4 ("vhost: split structs into a separate header file")
Silencing this perf build warning:
Warning: Kernel ABI header at 'tools/include/uapi/linux/vhost.h' differs from latest version at 'include/uapi/linux/vhost.h'
diff -u tools/include/uapi/linux/vhost.h include/uapi/linux/vhost.h
Those didn't touch things used in tools, i.e. the following continues
working:
tools include uapi: Sync linux/fs.h copy with the kernel sources
To get the changes in:
e262e32d6bde ("vfs: Suppress MS_* flag defs within the kernel unless explicitly enabled")
That made the mount flags string table generator to switch to using
mount.h instead.
This silences the following perf build warning:
Warning: Kernel ABI header at 'tools/include/uapi/linux/fs.h' differs from latest version at 'include/uapi/linux/fs.h'
diff -u tools/include/uapi/linux/fs.h include/uapi/linux/fs.h
We were using a copy of uapi/linux/fs.h to create the mount syscall
'flags' string table to use in 'perf trace', to convert from the number
obtained via the raw_syscalls:sys_enter into a string, using
tools/perf/trace/beauty/mount_flags.sh, but in e262e32d6bde ("vfs:
Suppress MS_* flag defs within the kernel unless explicitly enabled")
those defines got moved to linux/mount.h, so grab a copy of mount.h too.
Keep the uapi/linux/fs.h as we'll use it for the SEEK_ constants.
Alex Williamson [Tue, 8 Jan 2019 05:13:22 +0000 (22:13 -0700)]
vfio/type1: Fix unmap overflow off-by-one
The below referenced commit adds a test for integer overflow, but in
doing so prevents the unmap ioctl from ever including the last page of
the address space. Subtract one to compare to the last address of the
unmap to avoid the overflow and wrap-around.
perf top: Lift restriction on using callchains without "sym" in --sort
This restriction is not present in 'perf report' and since 'perf top'
uses the same hists browser, remove it from it as well.
With this we create per event buckets with callchain trees, so that
# perf top --sort dso -g --no-children
Bucketizes samples by DSO and below it shows the callchains leading to
functions in this DSO.
Try also:
# perf top -e sched:*switch -g --no-children
To see the callchains leading to sched switches, pressing 'E' to expand
all one can quickly see the most common scheduler switches and what
leads to them, for instance, calls to IO, futexes, etc.
tools lib traceevent: Remove tep_data_event_from_type() API
In order to make libtraceevent into a proper library, its API
should be straightforward.
After discussion with Steven Rostedt, we decided to remove the
tep_data_event_from_type() API and to replace it with tep_find_event(),
as it does the same.
tools lib traceevent: Changed return logic of tep_register_event_handler() API
In order to make libtraceevent into a proper library, its API
should be straightforward.
The tep_register_event_handler() functions returns -1 in case it
successfully registers the new event handler. Such return code is used
by the other library APIs in case of an error.
To unify the return logic of tep_register_event_handler() with the other
APIs, this patch introduces enum tep_reg_handler, which is used by this
function as return value, to handle all possible successful return
cases.
tools lib traceevent: Changed return logic of trace_seq_printf() and trace_seq_vprintf() APIs
In order to make libtraceevent into a proper library, its API should be
straightforward.
The trace_seq_printf() and trace_seq_vprintf() APIs have inconsistent
returned values with the other trace_seq_* APIs.
This path changes the return logic of trace_seq_printf() and
trace_seq_vprintf() to return the number of printed characters, as the
other trace_seq_* related APIs.
tools lib traceevent: Rename struct cmdline to struct tep_cmdline
In order to make libtraceevent a proper library, variables, data
structures and functions should have a unique prefix to prevent name
space conflicts. That prefix will be "tep_".
This patch renames 'struct cmdline' to 'struct tep_cmdline'.
tools lib traceevent: Initialize host_bigendian at tep_handle allocation
This patch initializes the host_bigendian member of the tep_handle
structure with the byte order of the current host, when this handler is
created - in tep_alloc() API. We need this in order to remove the
tep_set_host_bigendian() API.
tools lib traceevent: Introduce new libtracevent API: tep_override_comm()
This patch adds a new API of tracevent library: tep_override_comm() It
registers a pid / command mapping. If a mapping with the same pid
already exists, the entry is updated with the new command.
Florian Fainelli [Fri, 21 Dec 2018 03:43:37 +0000 (19:43 -0800)]
perf tests: Add a test for the ARM 32-bit [vectors] page
perf on ARM requires CONFIG_KUSER_HELPERS to be turned on to allow some
independance with respect to the ARM CPU being used. Add a test which
tries to locate the [vectors] page, created when CONFIG_KUSER_HELPERS is
turned on to help asses the system's health.
Florian Fainelli [Fri, 21 Dec 2018 03:43:36 +0000 (19:43 -0800)]
perf tools: Make find_vdso_map() more modular
In preparation for checking that the vectors page on the ARM
architecture, refactor the find_vdso_map() function to accept finding an
arbitrary string and create a dedicated helper function for that under
util/find-map.c and update the filename to find-map.c and all references
to it: perf-read-vdso.c and util/vdso.c.
At some point we'll get that poll sys_exit event and will print a "[continued]" line.
While making the sizing of the alignment after the syscall arg list and
its result configurable, so that we can mimic strace, which uses a
smaller alingment by default, a bug was introduced where the closing
parens appeared before the syscall name and its arg list, fix it.
Guo Ren [Tue, 8 Jan 2019 12:49:24 +0000 (20:49 +0800)]
irqchip/csky: fixup handle_irq_perbit break irq
The handle_irq_perbit function loop every bit in hwirq local variable.
handle_irq_perbit(hwirq) {
for_everyt_bit_in(hwirq) {
handle_domain_irq()
->irq_exit()
->invoke_softirq()
->__do_softirq()
->local_irq_enable() // Here will cause new interrupt.
}
}
When new interrupt coming at local_irq_enable, it will finish another
interrupt handler and pull down the interrupt source. But hwirq is the
local variable for handle_irq_perbit(), it can't get new interrupt
controller pending reg status. So we need update hwirq with pending reg
in every loop.
Also change write_relax to writel could prevent stw from fast retire.
When local_irq is enabled, intc regs is really set-in.
Masahiro Yamada [Tue, 8 Jan 2019 03:08:03 +0000 (12:08 +0900)]
vfio/pci: set TRACE_INCLUDE_PATH to fix the build error
drivers/vfio/pci/vfio_pci_nvlink2.c cannot be compiled for in-tree
building.
CC drivers/vfio/pci/vfio_pci_nvlink2.o
In file included from drivers/vfio/pci/trace.h:102,
from drivers/vfio/pci/vfio_pci_nvlink2.c:29:
./include/trace/define_trace.h:89:42: fatal error: ./trace.h: No such file or directory
#include TRACE_INCLUDE(TRACE_INCLUDE_FILE)
^
compilation terminated.
make[1]: *** [scripts/Makefile.build;277: drivers/vfio/pci/vfio_pci_nvlink2.o] Error 1
To fix the build error, let's tell include/trace/define_trace.h the
location of drivers/vfio/pci/trace.h
Hauke Mehrtens [Sun, 6 Jan 2019 18:50:37 +0000 (19:50 +0100)]
serial: lantiq: Do not swap register read/writes
The ltq_r32() and ltq_w32() macros use the __raw_readl() and
__raw_writel() functions which do not swap the value to little endian.
On the big endian vrx200 SoC the UART is operated in big endian IO mode,
the readl() and write() functions convert the value to little endian
first and then the driver does not work any more on this SoC.
Currently the vrx200 SoC selects the CONFIG_SWAP_IO_SPACE option,
without this option the serial driver would work, but PCI devices do not
work any more.
This patch makes the driver use the __raw_readl() and __raw_writel()
functions which do not swap the endianness. On big endian system it is
assumed that the device should be access in big endian IO mode and on a
little endian system it would be access in little endian mode.
Fixes: 89b8bd2082bb ("serial: lantiq: Use readl/writel instead of ltq_r32/ltq_w32") Signed-off-by: Hauke Mehrtens <[email protected]> Acked-by: John Crispin <[email protected]> Signed-off-by: Greg Kroah-Hartman <[email protected]>
stw is fast retire instruction. When PC is run at enable interrupt
stage, the clear interrupt source hasn't finished. It will cause another
wrong irq-enter.
Corentin Labbe [Mon, 7 Jan 2019 17:22:23 +0000 (17:22 +0000)]
PCI: amlogic: Fix build failure due to missing gpio header
Building the driver when GPIOLIB=n is not selected is causing the following
compilation failure:
drivers/pci/controller/dwc/pci-meson.c: In function 'meson_pcie_assert_reset':
drivers/pci/controller/dwc/pci-meson.c:290:2: error: implicit declaration of function 'gpiod_set_value_cansleep'; did you mean 'gpio_set_value_cansleep'? [-Werror=implicit-function-declaration]
gpiod_set_value_cansleep(mp->reset_gpio, 0);
^~~~~~~~~~~~~~~~~~~~~~~~
gpio_set_value_cansleep
drivers/pci/controller/dwc/pci-meson.c: In function 'meson_pcie_probe':
drivers/pci/controller/dwc/pci-meson.c:540:19: error: implicit declaration of function 'devm_gpiod_get'; did you mean 'devm_gpio_free'? [-Werror=implicit-function-declaration]
mp->reset_gpio = devm_gpiod_get(dev, "reset", GPIOD_OUT_LOW);
^~~~~~~~~~~~~~
devm_gpio_free
drivers/pci/controller/dwc/pci-meson.c:540:48: error: 'GPIOD_OUT_LOW' undeclared (first use in this function); did you mean 'GPIOF_INIT_LOW'?
mp->reset_gpio = devm_gpiod_get(dev, "reset", GPIOD_OUT_LOW);
^~~~~~~~~~~~~
GPIOF_INIT_LOW
Add the missing linux/gpio/consumer.h header to fix it.
Luis Chamberlain [Wed, 19 Dec 2018 20:30:34 +0000 (12:30 -0800)]
dma-mapping: remove dma_zalloc_coherent()
dma_zalloc_coherent() is no longer needed as it has no users because
dma_alloc_coherent() already zeroes out memory for us.
The Coccinelle grammar rule that used to check for dma_alloc_coherent()
+ memset() is modified so that it just tells the user that the memset is
not needed anymore.
Luis Chamberlain [Wed, 19 Dec 2018 20:24:08 +0000 (12:24 -0800)]
cross-tree: phase out dma_zalloc_coherent() on headers
The last few stragglers coccinelle doesn't pick up are on driver
specific header files. Phase those out as well as dma_alloc_coherent()
zeroes out the memory as well now too.
mtd: rawnand: qcom: fix memory corruption that causes panic
This patch fixes a memory corruption that occurred in the
qcom-nandc driver since it was converted to nand_scan().
On boot, an affected device will panic from a NPE at a weird place:
| Unable to handle kernel NULL pointer dereference at virtual address 0
| pgd = (ptrval)
| [00000000] *pgd=00000000
| Internal error: Oops: 80000005 [#1] SMP ARM
| CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.19.9 #0
| Hardware name: Generic DT based system
| PC is at (null)
| LR is at nand_block_isbad+0x90/0xa4
| pc : [<00000000>] lr : [<c0592240>] psr: 80000013
| sp : cf839d40 ip : 00000000 fp : cfae9e20
| r10: cf815810 r9 : 00000000 r8 : 00000000
| r7 : 00000000 r6 : 00000000 r5 : 00000001 r4 : cf815810
| r3 : 00000000 r2 : cfae9810 r1 : ffffffff r0 : cf815810
| Flags: Nzcv IRQs on FIQs on Mode SVC_32 ISA ARM Segment none
| Control: 10c5387d Table: 8020406a DAC: 00000051
| Process swapper/0 (pid: 1, stack limit = 0x(ptrval))
| [<c0592240>] (nand_block_isbad) from [<c0580a94>]
| [<c0580a94>] (allocate_partition) from [<c05811e4>]
| [<c05811e4>] (add_mtd_partitions) from [<c0581164>]
| [<c0581164>] (parse_mtd_partitions) from [<c057def4>]
| [<c057def4>] (mtd_device_parse_register) from [<c059d274>]
| [<c059d274>] (qcom_nandc_probe) from [<c0567f00>]
The problem is that the nand_scan()'s qcom_nand_attach_chip callback
is updating the nandc->max_cwperpage from 1 to 4. This causes the
sg_init_table of clear_bam_transaction() in the driver's
qcom_nandc_block_bad() to memset much more than what was initially
allocated by alloc_bam_transaction().
This patch restores the old behavior by reallocating the shared bam
transaction alloc_bam_transaction() after the chip was identified,
but before mtd_device_parse_register() (which is an alias for
mtd_device_register() - see panic) gets called. This fixes the
corruption and the driver is working again.
Sudeep Holla [Mon, 7 Jan 2019 18:51:53 +0000 (18:51 +0000)]
cpufreq: check if policy is inactive early in __cpufreq_get()
cpuinfo_cur_freq gets current CPU frequency as detected by hardware
while scaling_cur_freq last known CPU frequency. Some platforms may not
allow checking the CPU frequency of an offline CPU or the associated
resources may have been released via cpufreq_exit when the CPU gets
offlined, in which case the policy would have been invalidated already.
If we attempt to get current frequency from the hardware, it may result
in hang or crash.
For example on Juno, I see:
Unable to handle kernel NULL pointer dereference at virtual address 0000000000000188
[0000000000000188] pgd=0000000000000000
Internal error: Oops: 96000004 [#1] PREEMPT SMP
Modules linked in:
CPU: 5 PID: 4202 Comm: cat Not tainted 4.20.0-08251-ga0f2c0318a15-dirty #87
Hardware name: ARM LTD ARM Juno Development Platform/ARM Juno Development Platform
pstate: 40000005 (nZcv daif -PAN -UAO)
pc : scmi_cpufreq_get_rate+0x34/0xb0
lr : scmi_cpufreq_get_rate+0x34/0xb0
Call trace:
scmi_cpufreq_get_rate+0x34/0xb0
__cpufreq_get+0x34/0xc0
show_cpuinfo_cur_freq+0x24/0x78
show+0x40/0x60
sysfs_kf_seq_show+0xc0/0x148
kernfs_seq_show+0x44/0x50
seq_read+0xd4/0x480
kernfs_fop_read+0x15c/0x208
__vfs_read+0x60/0x188
vfs_read+0x94/0x150
ksys_read+0x6c/0xd8
__arm64_sys_read+0x24/0x30
el0_svc_common+0x78/0x100
el0_svc_handler+0x38/0x78
el0_svc+0x8/0xc
---[ end trace 3d1024e58f77f6b2 ]---
So fix the issue by checking if the policy is invalid early in
__cpufreq_get before attempting to get the current frequency.
projects / linux.git / log