As you can see there are only two common capabilities:
MAC_ASYM_PAUSE | MAC_SYM_PAUSE.
Meanwhile what is currently implemented defines 10/100/1000 link speeds
for all IP-cores, which is definitely incorrect for DW MAC100, DW XGMAC
and DW XLGMAC devices.
Seeing the flow-control is implemented as a callback for each MAC IP-core
(see dwmac100_flow_ctrl(), dwmac1000_flow_ctrl(), sun8i_dwmac_flow_ctrl(),
etc) and since the MAC-specific setup() method is supposed to be called
for each available DW MAC-based device, the capabilities initialization
can be freely moved to these setup() functions, thus correctly setting up
the MAC-capabilities for each IP-core (including the Allwinner Sun8i). A
new stmmac_link::caps field was specifically introduced for that so to
have all link-specific info preserved in a single structure.
Note the suggested change fixes three earlier commits at a time. The
commit 5b0d7d7da64b ("net: stmmac: Add the missing speeds that XGMAC
supports") permitted the 10-100 link speeds and 1G half-duplex mode for DW
XGMAC IP-core even though it doesn't support them. The commit df7699c70c1b
("net: stmmac: Do not cut down 1G modes") incorrectly added the MAC1000
capability to the DW MAC100 IP-core. Similarly to the DW XGMAC the commit 8a880936e902 ("net: stmmac: Add XLGMII support") incorrectly permitted the
10-100 link speeds and 1G half-duplex mode for DW XLGMAC IP-core.
Fixes: 5b0d7d7da64b ("net: stmmac: Add the missing speeds that XGMAC supports") Fixes: df7699c70c1b ("net: stmmac: Do not cut down 1G modes") Fixes: 8a880936e902 ("net: stmmac: Add XLGMII support") Suggested-by: Russell King (Oracle) <linux@armlinux.org.uk> Signed-off-by: Serge Semin <fancer.lancer@gmail.com> Reviewed-by: Romain Gantois <romain.gantois@bootlin.com> Signed-off-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
It's possible to have the maximum link speed being artificially limited on
the platform-specific basis. It's done either by setting up the
plat_stmmacenet_data::max_speed field or by specifying the "max-speed"
DT-property. In such cases it's required that any specific
MAC-capabilities re-initializations would take the limit into account. In
particular the link speed capabilities may change during the number of
active Tx/Rx queues re-initialization. But the currently implemented
procedure doesn't take the speed limit into account.
Fix that by calling phylink_limit_mac_speed() in the
stmmac_reinit_queues() method if the speed limitation was required in the
same way as it's done in the stmmac_phy_setup() function.
Fixes: 95201f36f395 ("net: stmmac: update MAC capabilities when tx queues are updated") Signed-off-by: Serge Semin <fancer.lancer@gmail.com> Reviewed-by: Romain Gantois <romain.gantois@bootlin.com> Signed-off-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
There are three DW MAC IP-cores which can have the multiple Tx/Rx queues
enabled:
DW GMAC v3.7+ with AV feature,
DW QoS Eth v4.x/v5.x,
DW XGMAC/XLGMAC
Based on the respective HW databooks, only the DW QoS Eth IP-core doesn't
support the half-duplex link mode in case if more than one queues enabled:
"In multiple queue/channel configurations, for half-duplex operation,
enable only the Q0/CH0 on Tx and Rx. For single queue/channel in
full-duplex operation, any queue/channel can be enabled."
The rest of the IP-cores don't have such constraint. Thus in order to have
the constraint applied for the DW QoS Eth MACs only, let's move the it'
implementation to the respective MAC-capabilities getter and make sure the
getter is called in the queues re-init procedure.
Fixes: b6cfffa7ad92 ("stmmac: fix DMA channel hang in half-duplex mode") Signed-off-by: Serge Semin <fancer.lancer@gmail.com> Reviewed-by: Romain Gantois <romain.gantois@bootlin.com> Signed-off-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Upon reviewing the flower control flags handling in
this driver, I notice that the key wasn't being used,
only the mask.
Ie. `tc flower ... ip_flags nofrag` was hardware
offloaded as `... ip_flags frag`.
Only compile tested, no access to HW.
Fixes: c672e3727989 ("octeontx2-pf: Add support to filter packet based on IP fragment") Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net> Reviewed-by: Jacob Keller <jacob.e.keller@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
The commit fc8b2a619469
("net: more strict VIRTIO_NET_HDR_GSO_UDP_L4 validation")
adds check of potential number of UDP segments vs
UDP_MAX_SEGMENTS in linux/virtio_net.h.
After this change certification test of USO guest-to-guest
transmit on Windows driver for virtio-net device fails,
for example with packet size of ~64K and mss of 536 bytes.
In general the USO should not be more restrictive than TSO.
Indeed, in case of unreasonably small mss a lot of segments
can cause queue overflow and packet loss on the destination.
Limit of 128 segments is good for any practical purpose,
with minimal meaningful mss of 536 the maximal UDP packet will
be divided to ~120 segments.
The number of segments for UDP packets is validated vs
UDP_MAX_SEGMENTS also in udp.c (v4,v6), this does not affect
quest-to-guest path but does affect packets sent to host, for
example.
It is important to mention that UDP_MAX_SEGMENTS is kernel-only
define and not available to user mode socket applications.
In order to request MSS smaller than MTU the applications
just uses setsockopt with SOL_UDP and UDP_SEGMENT and there is
no limitations on socket API level.
Fixes: fc8b2a619469 ("net: more strict VIRTIO_NET_HDR_GSO_UDP_L4 validation") Signed-off-by: Yuri Benditovich <yuri.benditovich@daynix.com> Reviewed-by: Willem de Bruijn <willemb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
When disabling aRFS under the `priv->state_lock`, any scheduled
aRFS works are canceled using the `cancel_work_sync` function,
which waits for the work to end if it has already started.
However, while waiting for the work handler, the handler will
try to acquire the `state_lock` which is already acquired.
The worker acquires the lock to delete the rules if the state
is down, which is not the worker's responsibility since
disabling aRFS deletes the rules.
Add an aRFS state variable, which indicates whether the aRFS is
enabled and prevent adding rules when the aRFS is disabled.
Kernel log:
======================================================
WARNING: possible circular locking dependency detected
6.7.0-rc4_net_next_mlx5_5483eb2 #1 Tainted: G I
------------------------------------------------------
ethtool/386089 is trying to acquire lock: ffff88810f21ce68 ((work_completion)(&rule->arfs_work)){+.+.}-{0:0}, at: __flush_work+0x74/0x4e0
but task is already holding lock: ffff8884a1808cc0 (&priv->state_lock){+.+.}-{3:3}, at: mlx5e_ethtool_set_channels+0x53/0x200 [mlx5_core]
which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
The cited patch introduces the concept of buckets in LAG in hash mode.
However, the patch doesn't clear the number of buckets in the LAG
deactivation. This results in using the wrong number of buckets in
case user create a hash mode LAG and afterwards create a non-hash
mode LAG.
Hence, restore buckets number to default after hash mode LAG
deactivation.
Fixes: 352899f384d4 ("net/mlx5: Lag, use buckets in hash mode") Signed-off-by: Shay Drory <shayd@nvidia.com> Reviewed-by: Maor Gottlieb <maorg@nvidia.com> Signed-off-by: Tariq Toukan <tariqt@nvidia.com> Link: https://lore.kernel.org/r/20240411115444.374475-2-tariqt@nvidia.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
I noticed that only 3 out of the 4 input bits were used,
mt.key->flags & FLOW_DIS_IS_FRAGMENT was never checked.
In order to avoid a complicated maze, I converted it to
use a 16 byte mapping table.
As shown in the table below the old heuristics doesn't
always do the right thing, ie. when FLOW_DIS_IS_FRAGMENT=1/1
then it used to only match follow-up fragment packets.
Here are all the combinations, and their resulting new/old
VCAP key/mask filter:
/- FLOW_DIS_IS_FRAGMENT (key/mask)
| /- FLOW_DIS_FIRST_FRAG (key/mask)
| | /-- new VCAP fragment (key/mask)
v v v v- old VCAP fragment (key/mask)
0/0 0/0 -/- -/- impossible (due to entry cond. on mask)
0/0 0/1 -/- 0/3 !! invalid (can't match non-fragment + follow-up frag)
0/0 1/0 -/- -/- impossible (key > mask)
0/0 1/1 1/3 1/3 first fragment
0/1 0/0 0/3 3/3 !! not fragmented
0/1 0/1 0/3 3/3 !! not fragmented (+ not first fragment)
0/1 1/0 -/- -/- impossible (key > mask)
0/1 1/1 -/- 1/3 !! invalid (non-fragment and first frag)
1/1 0/0 1/1 3/3 !! some fragment
1/1 0/1 3/3 3/3 follow-up fragment
1/1 1/0 -/- -/- impossible (key > mask)
1/1 1/1 1/3 1/3 first fragment
In the datasheet the VCAP fragment values are documented as:
0 = no fragment
1 = initial fragment
2 = suspicious fragment
3 = valid follow-up fragment
Result: 3 combinations match the old behavior,
3 combinations have been corrected,
2 combinations are now invalid, and fail,
8 combinations are impossible.
It should now be aligned with how FLOW_DIS_IS_FRAGMENT
and FLOW_DIS_FIRST_FRAG is set in __skb_flow_dissect() in
net/core/flow_dissector.c
Since the VCAP fragment values are not a bitfield, we have
to ignore the suspicious fragment value, eg. when matching
on any kind of fragment with FLOW_DIS_IS_FRAGMENT=1/1.
Only compile tested, and logic tested in userspace, as I
unfortunately don't have access to this switch chip (yet).
Fixes: d6c2964db3fe ("net: microchip: sparx5: Adding more tc flower keys for the IS2 VCAP") Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net> Reviewed-by: Steen Hegelund <Steen.Hegelund@microchip.com> Tested-by: Daniel Machon <daniel.machon@microchip.com> Reviewed-by: Jacob Keller <jacob.e.keller@intel.com> Link: https://lore.kernel.org/r/20240411111321.114095-1-ast@fiberby.net Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
If manage_oob() is called when no data has been copied, we only
check if the socket enables SO_OOBINLINE or MSG_PEEK is not used.
Otherwise, the skb is returned as is.
However, here we should return NULL if MSG_PEEK is set and no data
has been copied.
Also, in such a case, we should not jump to the redo label because
we will be caught in the loop and hog the CPU until normal data
comes in.
Then, we need to handle skb == NULL case with the if-clause below
the manage_oob() block.
When we call recv() for AF_UNIX socket, we first peek one skb and
calls manage_oob() to check if the skb is sent with MSG_OOB.
However, when we fetch the next (and the following) skb, manage_oob()
is not called now, leading a wrong behaviour.
Let's say a socket send()s "hello" with MSG_OOB and the peer tries
to recv() 5 bytes with MSG_PEEK. Here, we should get only "hell"
without 'o', but actually not:
pppoe traffic reaching ingress path does not match the flowtable entry
because the pppoe header is expected to be at the network header offset.
This bug causes a mismatch in the flow table lookup, so pppoe packets
enter the classical forwarding path.
Ensure there is sufficient room to access the protocol field of the
PPPoe header. Validate it once before the flowtable lookup, then use a
helper function to access protocol field.
Pablo reports a crash with large batches of elements with a
back-to-back add/remove pattern. Quoting Pablo:
add_elem("00000000") timeout 100 ms
...
add_elem("0000000X") timeout 100 ms
del_elem("0000000X") <---------------- delete one that was just added
...
add_elem("00005000") timeout 100 ms
1) nft_pipapo_remove() removes element 0000000X
Then, KASAN shows a splat.
Looking at the remove function there is a chance that we will drop a
rule that maps to a non-deactivated element.
Removal happens in two steps, first we do a lookup for key k and return the
to-be-removed element and mark it as inactive in the next generation.
Then, in a second step, the element gets removed from the set/map.
The _remove function does not work correctly if we have more than one
element that share the same key.
This can happen if we insert an element into a set when the set already
holds an element with same key, but the element mapping to the existing
key has timed out or is not active in the next generation.
In such case its possible that removal will unmap the wrong element.
If this happens, we will leak the non-deactivated element, it becomes
unreachable.
The element that got deactivated (and will be freed later) will
remain reachable in the set data structure, this can result in
a crash when such an element is retrieved during lookup (stale
pointer).
Add a check that the fully matching key does in fact map to the element
that we have marked as inactive in the deactivation step.
If not, we need to continue searching.
Add a bug/warn trap at the end of the function as well, the remove
function must not ever be called with an invisible/unreachable/non-existent
element.
v2: avoid uneeded temporary variable (Stefano)
Fixes: 3c4287f62044 ("nf_tables: Add set type for arbitrary concatenation of ranges") Reported-by: Pablo Neira Ayuso <pablo@netfilter.org> Reviewed-by: Stefano Brivio <sbrivio@redhat.com> Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
For historical reasons, when bridge device is in promisc mode, packets
that are directed to the taps follow bridge input hook path. This patch
adds a workaround to reset conntrack for these packets.
Jianbo Liu reports warning splats in their test infrastructure where
cloned packets reach the br_netfilter input hook to confirm the
conntrack object.
Scratch one bit from BR_INPUT_SKB_CB to annotate that this packet has
reached the input hook because it is passed up to the bridge device to
reach the taps.
Fixes: 62e7151ae3eb ("netfilter: bridge: confirm multicast packets before passing them up the stack") Reported-by: Jianbo Liu <jianbol@nvidia.com> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
nft_unregister_obj() can concurrent with __nft_obj_type_get(),
and there is not any protection when iterate over nf_tables_objects
list in __nft_obj_type_get(). Therefore, there is potential data-race
of nf_tables_objects list entry.
Use list_for_each_entry_rcu() to iterate over nf_tables_objects
list in __nft_obj_type_get(), and use rcu_read_lock() in the caller
nft_obj_type_get() to protect the entire type query process.
Fixes: e50092404c1b ("netfilter: nf_tables: add stateful objects") Signed-off-by: Ziyang Xuan <william.xuanziyang@huawei.com> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
nft_unregister_expr() can concurrent with __nft_expr_type_get(),
and there is not any protection when iterate over nf_tables_expressions
list in __nft_expr_type_get(). Therefore, there is potential data-race
of nf_tables_expressions list entry.
Use list_for_each_entry_rcu() to iterate over nf_tables_expressions
list in __nft_expr_type_get(), and use rcu_read_lock() in the caller
nft_expr_type_get() to protect the entire type query process.
Fixes: ef1f7df9170d ("netfilter: nf_tables: expression ops overloading") Signed-off-by: Ziyang Xuan <william.xuanziyang@huawei.com> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
KVM/arm64 relies on TLBI RANGE feature to flush TLBs when the dirty
pages are collected by VMM and the page table entries become write
protected during live migration. Unfortunately, the operand passed
to the TLBI RANGE instruction isn't correctly sorted out due to the
commit 117940aa6e5f ("KVM: arm64: Define kvm_tlb_flush_vmid_range()").
It leads to crash on the destination VM after live migration because
TLBs aren't flushed completely and some of the dirty pages are missed.
For example, I have a VM where 8GB memory is assigned, starting from
0x40000000 (1GB). Note that the host has 4KB as the base page size.
In the middile of migration, kvm_tlb_flush_vmid_range() is executed
to flush TLBs. It passes MAX_TLBI_RANGE_PAGES as the argument to
__kvm_tlb_flush_vmid_range() and __flush_s2_tlb_range_op(). SCALE#3
and NUM#31, corresponding to MAX_TLBI_RANGE_PAGES, isn't supported
by __TLBI_RANGE_NUM(). In this specific case, -1 has been returned
from __TLBI_RANGE_NUM() for SCALE#3/2/1/0 and rejected by the loop
in the __flush_tlb_range_op() until the variable @scale underflows
and becomes -9, 0xffff708000040000 is set as the operand. The operand
is wrong since it's sorted out by __TLBI_VADDR_RANGE() according to
invalid @scale and @num.
Fix it by extending __TLBI_RANGE_NUM() to support the combination of
SCALE#3 and NUM#31. With the changes, [-1 31] instead of [-1 30] can
be returned from the macro, meaning the TLBs for 0x200000 pages in the
above example can be flushed in one shoot with SCALE#3 and NUM#31. The
macro TLBI_RANGE_MASK is dropped since no one uses it any more. The
comments are also adjusted accordingly.
In preparation for adding support for LPA2 to the tlb invalidation
routines, modify the algorithm used by range-based tlbi to start at the
highest 'scale' and decrement instead of starting at the lowest 'scale'
and incrementing. This new approach makes it possible to maintain 64K
alignment as we work through the range, until the last op (at scale=0).
This is required when LPA2 is enabled. (This part will be added in a
subsequent commit).
This change is separated into its own patch because it will also impact
non-LPA2 systems, and I want to make it easy to bisect in case it leads
to performance regression (see below for benchmarks that suggest this
should not be a problem).
The original commit (d1d3aa98 "arm64: tlb: Use the TLBI RANGE feature in
arm64") stated this as the reason for _incrementing_ scale:
However, in most scenarios, the pages = 1 when flush_tlb_range() is
called. Start from scale = 3 or other proper value (such as scale
=ilog2(pages)), will incur extra overhead. So increase 'scale' from 0
to maximum.
But pages=1 is already special cased by the non-range invalidation path,
which will take care of it the first time through the loop (both in the
original commit and in my change), so I don't think switching to
decrement scale should have any extra performance impact after all.
Indeed benchmarking kernel compilation, a TLBI-heavy workload, suggests
that this new approach actually _improves_ performance slightly (using a
virtual machine on Apple M2):
Table shows time to execute kernel compilation workload with 8 jobs,
relative to baseline without this patch (more negative number is
bigger speedup). Repeated 9 times across 3 system reboots:
Reviewed-by: Oliver Upton <oliver.upton@linux.dev> Signed-off-by: Ryan Roberts <ryan.roberts@arm.com> Signed-off-by: Marc Zyngier <maz@kernel.org> Link: https://lore.kernel.org/r/20231127111737.1897081-2-ryan.roberts@arm.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
After the commit d2689b6a86b9 ("net: usb: ax88179_178a: avoid two
consecutive device resets"), reset operation, in which the default mac
address from the device is read, is not executed from bind operation and
the random address, that is pregenerated just in case, is direclty written
the first time in the device, so the default one from the device is not
even read. This writing is not dangerous because is volatile and the
default mac address is not missed.
In order to avoid this and keep the simplification to have only one
reset and reduce the delays, restore the reset from bind operation and
remove the reset that is commanded from open operation. The behavior is
the same but everything is ready for usbnet_probe.
Tested with ASIX AX88179 USB Gigabit Ethernet devices.
Restore the old behavior for the rest of possible devices because I don't
have the hardware to test.
cc: stable@vger.kernel.org # 6.6+ Fixes: d2689b6a86b9 ("net: usb: ax88179_178a: avoid two consecutive device resets") Reported-by: Jarkko Palviainen <jarkko.palviainen@gmail.com> Signed-off-by: Jose Ignacio Tornos Martinez <jtornosm@redhat.com> Link: https://lore.kernel.org/r/20240417085524.219532-1-jtornosm@redhat.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
There is code in the SCSI core that sets the SCMD_FAIL_IF_RECOVERING
flag but there is no code that clears this flag. Instead of only clearing
SCMD_INITIALIZED in scsi_end_request(), clear all flags. It is never
necessary to preserve any command flags inside scsi_end_request().
Cc: stable@vger.kernel.org Fixes: 310bcaef6d7e ("scsi: core: Support failing requests while recovering") Signed-off-by: Bart Van Assche <bvanassche@acm.org> Link: https://lore.kernel.org/r/20240325224417.1477135-1-bvanassche@acm.org Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The entropy accounting changes a static key when the RNG has
initialized, since it only ever initializes once. Static key changes,
however, cannot be made from atomic context, so depending on where the
last creditable entropy comes from, the static key change might need to
be deferred to a worker.
Previously the code used the execute_in_process_context() helper
function, which accounts for whether or not the caller is
in_interrupt(). However, that doesn't account for the case where the
caller is actually in process context but is holding a spinlock.
This turned out to be the case with input_handle_event() in
drivers/input/input.c contributing entropy:
According to Guoyong, it's not really possible to refactor the various
drivers to never hold a spinlock there. And in_atomic() isn't reliable.
So, rather than trying to be too fancy, just punt the change in the
static key to a workqueue always. There's basically no drawback of doing
this, as the code already needed to account for the static key not
changing immediately, and given that it's just an optimization, there's
not exactly a hurry to change the static key right away, so deferal is
fine.
Reported-by: Guoyong Wang <guoyong.wang@mediatek.com> Cc: stable@vger.kernel.org Fixes: f5bda35fba61 ("random: use static branch for crng_ready()") Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
While sched* events being traced and sched* events continuously happen,
"[xx] event tracing - enable/disable with subsystem level files" would
not stop as on some slower systems it seems to take forever.
Select the first 100 lines of output would be enough to judge whether
there are more than 3 types of sched events.
The rpcgss_context trace event acceptor field is a dynamically sized
string that records the "data" parameter. But this parameter is also
dependent on the "len" field to determine the size of the data.
It needs to use __string_len() helper macro where the length can be passed
in. It also incorrectly uses strncpy() to save it instead of
__assign_str(). As these macros can change, it is not wise to open code
them in trace events.
As of commit c759e609030c ("tracing: Remove __assign_str_len()"),
__assign_str() can be used for both __string() and __string_len() fields.
Before that commit, __assign_str_len() is required to be used. This needs
to be noted for backporting. (In actuality, commit c1fa617caeb0 ("tracing:
Rework __assign_str() and __string() to not duplicate getting the string")
is the commit that makes __string_str_len() obsolete).
Cc: stable@vger.kernel.org Fixes: 0c77668ddb4e ("SUNRPC: Introduce trace points in rpc_auth_gss.ko") Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
That has been tracked down to be happening when another thread is
deactivating the VMA inside __active_retire() helper, after the VMA's
active counter has been already decremented to 0, but before deactivation
of the VMA's object is reported to the object debugging tool.
We could prevent from that race by serializing i915_active_fini() with
__active_retire() via ref->tree_lock, but that wouldn't stop the VMA from
being used, e.g. from __i915_vma_retire() called at the end of
__active_retire(), after that VMA has been already freed by a concurrent
i915_vma_destroy() on return from the i915_active_fini(). Then, we should
rather fix the issue at the VMA level, not in i915_active.
Since __i915_vma_parked() is called from __gt_park() on last put of the
GT's wakeref, the issue could be addressed by holding the GT wakeref long
enough for __active_retire() to complete before that wakeref is released
and the GT parked.
I believe the issue was introduced by commit d93939730347 ("drm/i915:
Remove the vma refcount") which moved a call to i915_active_fini() from
a dropped i915_vma_release(), called on last put of the removed VMA kref,
to i915_vma_parked() processing path called on last put of a GT wakeref.
However, its visibility to the object debugging tool was suppressed by a
bug in i915_active that was fixed two weeks later with commit e92eb246feb9
("drm/i915/active: Fix missing debug object activation").
A VMA associated with a request doesn't acquire a GT wakeref by itself.
Instead, it depends on a wakeref held directly by the request's active
intel_context for a GT associated with its VM, and indirectly on that
intel_context's engine wakeref if the engine belongs to the same GT as the
VMA's VM. Those wakerefs are released asynchronously to VMA deactivation.
Fix the issue by getting a wakeref for the VMA's GT when activating it,
and putting that wakeref only after the VMA is deactivated. However,
exclude global GTT from that processing path, otherwise the GPU never goes
idle. Since __i915_vma_retire() may be called from atomic contexts, use
async variant of wakeref put. Also, to avoid circular locking dependency,
take care of acquiring the wakeref before VM mutex when both are needed.
v7: Add inline comments with justifications for:
- using untracked variants of intel_gt_pm_get/put() (Nirmoy),
- using async variant of _put(),
- not getting the wakeref in case of a global GTT,
- always getting the first wakeref outside vm->mutex.
v6: Since __i915_vma_active/retire() callbacks are not serialized, storing
a wakeref tracking handle inside struct i915_vma is not safe, and
there is no other good place for that. Use untracked variants of
intel_gt_pm_get/put_async().
v5: Replace "tile" with "GT" across commit description (Rodrigo),
- avoid mentioning multi-GT case in commit description (Rodrigo),
- explain why we need to take a temporary wakeref unconditionally inside
i915_vma_pin_ww() (Rodrigo).
v4: Refresh on top of commit 5e4e06e4087e ("drm/i915: Track gt pm
wakerefs") (Andi),
- for more easy backporting, split out removal of former insufficient
workarounds and move them to separate patches (Nirmoy).
- clean up commit message and description a bit.
v3: Identify root cause more precisely, and a commit to blame,
- identify and drop former workarounds,
- update commit message and description.
v2: Get the wakeref before VM mutex to avoid circular locking dependency,
- drop questionable Fixes: tag.
Fixes: d93939730347 ("drm/i915: Remove the vma refcount") Closes: https://gitlab.freedesktop.org/drm/intel/issues/8875 Signed-off-by: Janusz Krzysztofik <janusz.krzysztofik@linux.intel.com> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com> Cc: Nirmoy Das <nirmoy.das@intel.com> Cc: Andi Shyti <andi.shyti@linux.intel.com> Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> Cc: stable@vger.kernel.org # v5.19+ Reviewed-by: Nirmoy Das <nirmoy.das@intel.com> Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20240305143747.335367-6-janusz.krzysztofik@linux.intel.com
(cherry picked from commit f3c71b2ded5c4367144a810ef25f998fd1d6c381) Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com> Signed-off-by: Janusz Krzysztofik <janusz.krzysztofik@linux.intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This bug was introduced in commit 950e79dd7313 ("io_uring: minor
io_cqring_wait() optimization"), which was made in preparation for adc8682ec690 ("io_uring: Add support for napi_busy_poll"). The latter
got reverted in cb3182167325 ("Revert "io_uring: Add support for
napi_busy_poll""), so simply undo the former as well.
The vb2 read support requests 1 buffer, leaving it to the driver
to increase this number to something that works.
Unfortunately, drivers do not deal with this reliably, and in fact
this caused problems for the bttv driver and reading from /dev/vbiX,
causing every other VBI frame to be all 0.
Instead, request as the number of buffers whatever is the maximum of
2 and q->min_buffers_needed+1.
In order to start streaming you need at least q->min_buffers_needed
queued buffers, so add 1 buffer for processing. And if that field
is 0, then choose 2 (again, one buffer is being filled while the
other one is being processed).
This certainly makes more sense than requesting just 1 buffer, and
the VBI bttv support is now working again.
It turns out that the old videobuf1 behavior of bttv was to allocate
8 (video) and 4 (vbi) buffers when used with read(). After the vb2
conversion that changed to 2 for both. With this patch it is 3, which
is really all you need.
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Fixes: b7ec3212a73a ("media: bttv: convert to vb2") Tested-by: Dr. David Alan Gilbert <dave@treblig.org> Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
The function _dpu_hw_sspp_setup_scaler3() passes and
dpu_hw_setup_scaler3() uses scaler_blk.version to determine in which way
the scaler (QSEED3) block should be programmed. However up to now we
were not setting this field. Set it now, splitting the vig_sblk data
which has different version fields.
Reported-by: Marijn Suijten <marijn.suijten@somainline.org> Fixes: 9b6f4fedaac2 ("drm/msm/dpu: Add SM6125 support") Fixes: 27f0df03f3ff ("drm/msm/dpu: Add SM6375 support") Fixes: 3186acba5cdc ("drm/msm/dpu: Add SM6350 support") Fixes: efcd0107727c ("drm/msm/dpu: add support for SM8550") Fixes: 4a352c2fc15a ("drm/msm/dpu: Introduce SC8280XP") Fixes: 0e91bcbb0016 ("drm/msm/dpu: Add SM8350 to hw catalog") Fixes: 100d7ef6995d ("drm/msm/dpu: add support for SM8450") Fixes: 3581b7062cec ("drm/msm/disp/dpu1: add support for display on SM6115") Fixes: dabfdd89eaa9 ("drm/msm/disp/dpu1: add inline rotation support for sc7280") Fixes: f3af2d6ee9ab ("drm/msm/dpu: Add SC8180x to hw catalog") Fixes: 94391a14fc27 ("drm/msm/dpu1: Add MSM8998 to hw catalog") Fixes: af776a3e1c30 ("drm/msm/dpu: add SM8250 to hw catalog") Fixes: 386fced3f76f ("drm/msm/dpu: add SM8150 to hw catalog") Fixes: b75ab05a3479 ("msm:disp:dpu1: add scaler support on SC7180 display") Fixes: 25fdd5933e4c ("drm/msm: Add SDM845 DPU support") Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
Patchwork: https://patchwork.freedesktop.org/patch/570098/ Link: https://lore.kernel.org/r/20231201234234.2065610-2-dmitry.baryshkov@linaro.org Signed-off-by: Sasha Levin <sashal@kernel.org>
After commit 6d029c25b71f ("selftests/timers/posix_timers: Reimplement
check_timer_distribution()") the following warning occurs when building
with an older gcc:
posix_timers.c:250:2: warning: format not a string literal and no format arguments [-Wformat-security]
250 | ksft_print_msg(errmsg);
| ^~~~~~~~~~~~~~
Fix this up by changing it to ksft_print_msg("%s", errmsg)
We need to covert the inode to ceph_client in the following commit,
and will add one new helper for that, here we rename the old helper
to _fs_client().
Link: https://tracker.ceph.com/issues/61590 Signed-off-by: Xiubo Li <xiubli@redhat.com> Reviewed-by: Patrick Donnelly <pdonnell@redhat.com> Reviewed-by: Milind Changire <mchangir@redhat.com> Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Stable-dep-of: b372e96bd0a3 ("ceph: redirty page before returning AOP_WRITEPAGE_ACTIVATE") Signed-off-by: Sasha Levin <sashal@kernel.org>
check_timer_distribution() runs ten threads in a busy loop and tries to
test that the kernel distributes a process posix CPU timer signal to every
thread over time.
There is not guarantee that this is true even after commit bcb7ee79029d
("posix-timers: Prefer delivery of signals to the current thread") because
that commit only avoids waking up the sleeping process leader thread, but
that has nothing to do with the actual signal delivery.
As the signal is process wide the first thread which observes sigpending
and wins the race to lock sighand will deliver the signal. Testing shows
that this hangs on a regular base because some threads never win the race.
The comment "This primarily tests that the kernel does not favour any one."
is wrong. The kernel does favour a thread which hits the timer interrupt
when CLOCK_PROCESS_CPUTIME_ID expires.
Rewrite the test so it only checks that the group leader sleeping in join()
never receives SIGALRM and the thread which burns CPU cycles receives all
signals.
In older kernels which do not have commit bcb7ee79029d ("posix-timers:
Prefer delivery of signals to the current thread") the test-case fails
immediately, the very 1st tick wakes the leader up. Otherwise it quickly
succeeds after 100 ticks.
CI testing wants to use newer selftest versions on stable kernels. In this
case the test is guaranteed to fail.
So check in the failure case whether the kernel version is less than v6.3
and skip the test result in that case.
[ tglx: Massaged change log, renamed the version check helper ]
Fixes: e797203fb3ba ("selftests/timers/posix_timers: Test delivery of signals across threads") Signed-off-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: stable@vger.kernel.org Link: https://lore.kernel.org/r/20240409133802.GD29396@redhat.com Signed-off-by: Sasha Levin <sashal@kernel.org>
Currently the posix_timers test does not produce KTAP output but rather a
custom format. This means that we only get a pass/fail for the suite, not
for each individual test that the suite does. Convert to using the standard
kselftest output functions which result in KTAP output being generated.
As part of this fix the printing of diagnostics in the unlikely event that
the pthread APIs fail, these were using perror() but the API functions
directly return an error code instead of setting errno.
Signed-off-by: Mark Brown <broonie@kernel.org> Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
Stable-dep-of: 6d029c25b71f ("selftests/timers/posix_timers: Reimplement check_timer_distribution()") Signed-off-by: Sasha Levin <sashal@kernel.org>
All joined pipes share the same transcoder/timing generator.
Currently we just do the commits per-pipe, which doesn't really
work if we need to change the timings at the same time. For
now just disable live M/N updates when bigjoiner is needed.
Make the seamless_m_n flag more like the update_pipe fastset
flag, ie. the flag will only be set if we need to do the seamless
M/N update, and in all other cases the flag is cleared. Also
rename the flag to update_m_n to make it more clear it's similar
to update_pipe.
I believe special casing seamless_m_n like this makes sense
as it also affects eg. vblank evasion. We can potentially avoid
some vblank evasion tricks, simplify some checks, and hopefully
will help with the VRR vs. M/N mess.
In order to reconcile seamless M/N updates with VRR we'll
need to defer the fastset VRR enable to happen after the
seamless M/N update (which happens during the vblank evade
critical section). So just push the VRR enable to be the last
thing during the update.
This will also affect the vblank evasion as the transcoder
will now still be running with the old VRR state during
the vblank evasion. So just grab the timings always from the
old crtc state during any non-modeset commit, and also grab
the current state of VRR from the active timings (as we disable
VRR before vblank evasion during fastsets).
This also fixes vblank evasion for seamless M/N updates as
we now properly account for the fact that the M/N update
happens after vblank evasion.
Cc: Manasi Navare <navaremanasi@chromium.org> Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20230901130440.2085-5-ville.syrjala@linux.intel.com Reviewed-by: Manasi Navare <navaremanasi@chromium.org> Reviewed-by: Mitul Golani <mitulkumar.ajitkumar.golani@intel.com>
Stable-dep-of: 4a36e46df7aa ("drm/i915: Disable live M/N updates when using bigjoiner") Signed-off-by: Sasha Levin <sashal@kernel.org>
We'll need to also look at the old crtc state in
intel_pipe_update_start() so change the calling convention to
just plumb in the full atomic state instead.
Cc: Manasi Navare <navaremanasi@chromium.org> Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20230901130440.2085-3-ville.syrjala@linux.intel.com Reviewed-by: Manasi Navare <navaremanasi@chromium.org> Reviewed-by: Mitul Golani <mitulkumar.ajitkumar.golani@intel.com>
Stable-dep-of: 4a36e46df7aa ("drm/i915: Disable live M/N updates when using bigjoiner") Signed-off-by: Sasha Levin <sashal@kernel.org>
Currently we only consider the relationship of the
old and new CDCLK frequencies when determining whether
to do the repgramming from intel_set_cdclk_pre_plane_update()
or intel_set_cdclk_post_plane_update().
It is technically possible to have a situation where the
CDCLK frequency is decreasing, but the voltage_level is
increasing due a DDI port. In this case we should bump
the voltage level already in intel_set_cdclk_pre_plane_update()
(so that the voltage_level will have been increased by the
time the port gets enabled), while leaving the CDCLK frequency
unchanged (as active planes/etc. may still depend on it).
We can then reduce the CDCLK frequency to its final value
from intel_set_cdclk_post_plane_update().
In order to handle that correctly we shall construct a
suitable amalgam of the old and new cdclk states in
intel_set_cdclk_pre_plane_update().
And we can simply call intel_set_cdclk() unconditionally
in both places as it will not do anything if nothing actually
changes vs. the current hw state.
v2: Handle cdclk_state->disable_pipes
v3: Only synchronize the cd2x update against the pipe's vblank
when the cdclk frequency is changing during the current
commit phase (Gustavo)
On pre-TGL FEC is a port level feature, not a transcoder
level feature, and it's DDI A which doesn't have it, not
trancoder A. Check for the correct thing when determining
whether FEC is supported or not.
Avoid refreshing DFS referral with refpath_lock acquired as the I/O
could block for a while due to a potentially disconnected or slow DFS
root server and then making other threads - that use same @server and
don't require a DFS root server - unable to make any progress.
Cc: stable@vger.kernel.org # 6.4+ Signed-off-by: Paulo Alcantara (Red Hat) <pc@manguebit.com> Signed-off-by: Steve French <stfrench@microsoft.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Avoid potential use-after-free bugs when walking DFS referrals,
mounting and performing DFS failover by ensuring that all children
from parent @tcon->ses are also refcounted. They're all needed across
the entire DFS mount. Get rid of @tcon->dfs_ses_list while we're at
it, too.
Cc: stable@vger.kernel.org # 6.4+ Reported-by: kernel test robot <lkp@intel.com> Closes: https://lore.kernel.org/oe-kbuild-all/202404021527.ZlRkIxgv-lkp@intel.com/ Signed-off-by: Paulo Alcantara (Red Hat) <pc@manguebit.com> Signed-off-by: Steve French <stfrench@microsoft.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
In analyzing problems, one missing piece of debug data is when the
mount occurred. A related problem is when collecting stats we don't
know the period of time the stats covered, ie when this set of stats
for the tcon started to be collected. To make debugging easier track
the stats begin time. Set it when the mount occurred at mount time,
and reset it to current time whenever stats are reset. For example,
...
1) \\localhost\test
SMBs: 14 since 2024-01-17 22:17:30 UTC
Bytes read: 0 Bytes written: 0
Open files: 0 total (local), 0 open on server
TreeConnects: 1 total 0 failed
TreeDisconnects: 0 total 0 failed
...
2) \\localhost\scratch
SMBs: 24 since 2024-01-17 22:16:04 UTC
Bytes read: 0 Bytes written: 0
Open files: 0 total (local), 0 open on server
TreeConnects: 1 total 0 failed
TreeDisconnects: 0 total 0 failed
...
Note the time "since ... UTC" is now displayed in /proc/fs/cifs/Stats
for each share that is mounted.
Suggested-by: Shyam Prasad N <sprasad@microsoft.com> Reviewed-by: Bharath SM <bharathsm@microsoft.com> Signed-off-by: Steve French <stfrench@microsoft.com>
Stable-dep-of: 062a7f0ff46e ("smb: client: guarantee refcounted children from parent session") Signed-off-by: Sasha Levin <sashal@kernel.org>
The UAF bug is due to smb2_reconnect_server() accessing a session that
is already being teared down by another thread that is executing
__cifs_put_smb_ses(). This can happen when (a) the client has
connection to the server but no session or (b) another thread ends up
setting @ses->ses_status again to something different than
SES_EXITING.
To fix this, we need to make sure to unconditionally set
@ses->ses_status to SES_EXITING and prevent any other threads from
setting a new status while we're still tearing it down.
The following can be reproduced by adding some delay to right after
the ipc is freed in __cifs_put_smb_ses() - which will give
smb2_reconnect_server() worker a chance to run and then accessing
@ses->ipc:
The previous check for the is_vsc_sdp_colorimetry_supported flag
for MST sink signals did nothing. Simplify the code and use the
same check for MST and SST.
Cc: stable@vger.kernel.org Reviewed-by: Agustin Gutierrez <agustin.gutierrez@amd.com> Acked-by: Hamza Mahfooz <hamza.mahfooz@amd.com> Signed-off-by: Harry Wentland <harry.wentland@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In order for display colorimetry to work correctly on DP displays
we need to send the VSC SDP packet. We should only do so for
panels with DPCD revision greater or equal to 1.4 as older
receivers might have problems with it.
For SOC21 ASICs, there is an issue in re-enabling PM features if a
suspend got aborted. In such cases, reset the device during resume
phase. This is a workaround till a proper solution is finalized.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com> Reviewed-by: Alex Deucher <alexander.deucher@amd.com> Reviewed-by: Yang Wang <kevinyang.wang@amd.com> Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Cc: stable@vger.kernel.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The current modeset sequence can't handle port sync and bigjoiner
at the same time. Refuse port sync when bigjoiner is needed,
at least until we fix the modeset sequence.
Currently we always reprogram CDCLK from the
intel_set_cdclk_pre_plane_update() when using squash/crawl.
The code only works correctly for the cd2x update or full
modeset cases, and it was simply never updated to deal with
squash/crawl.
If the CDCLK frequency is increasing we must reprogram it
before we do anything else that might depend on the new
higher frequency, and conversely we must not decrease
the frequency until everything that might still depend
on the old higher frequency has been dealt with.
Since cdclk_state->pipe is only relevant when doing a cd2x
update we can't use it to determine the correct sequence
during squash/crawl. To that end introduce cdclk_state->disable_pipes
which simply indicates that we must perform the update
while the pipes are disable (ie. during
intel_set_cdclk_pre_plane_update()). Otherwise we use the
same old vs. new CDCLK frequency comparsiong as for cd2x
updates.
The only remaining problem case is when the voltage_level
needs to increase due to a DDI port, but the CDCLK frequency
is decreasing (and not all pipes are being disabled). The
current approach will not bump the voltage level up until
after the port has already been enabled, which is too late.
But we'll take care of that case separately.
v2: Don't break the "must disable pipes case"
v3: Keep the on stack 'pipe' for future use
For consistency with the other CONFIG_MITIGATION_* options, replace the
CONFIG_SPECTRE_BHI_{ON,OFF} options with a single
CONFIG_MITIGATION_SPECTRE_BHI option.
While syscall hardening helps prevent some BHI attacks, there's still
other low-hanging fruit remaining. Don't classify it as a mitigation
and make it clear that the system may still be vulnerable if it doesn't
have a HW or SW mitigation enabled.
The ARCH_CAP_RRSBA check isn't correct: RRSBA may have already been
disabled by the Spectre v2 mitigation (or can otherwise be disabled by
the BHI mitigation itself if needed). In that case retpolines are fine.
So we are using the 'ia32_cap' value in a number of places,
which got its name from MSR_IA32_ARCH_CAPABILITIES MSR register.
But there's very little 'IA32' about it - this isn't 32-bit only
code, nor does it originate from there, it's just a historic
quirk that many Intel MSR names are prefixed with IA32_.
This is already clear from the helper method around the MSR:
x86_read_arch_cap_msr(), which doesn't have the IA32 prefix.
So rename 'ia32_cap' to 'x86_arch_cap_msr' to be consistent with
its role and with the naming of the helper function.
There's no need to keep reading MSR_IA32_ARCH_CAPABILITIES over and
over. It's even read in the BHI sysfs function which is a big no-no.
Just read it once and cache it.
When done from a virtual machine, instructions that touch APIC memory
must be emulated. By convention, MMIO accesses are typically performed
via io.h helpers such as readl() or writeq() to simplify instruction
emulation/decoding (ex: in KVM hosts and SEV guests) [0].
Currently, native_apic_mem_read() does not follow this convention,
allowing the compiler to emit instructions other than the MOV
instruction generated by readl(). In particular, when the kernel is
compiled with clang and run as a SEV-ES or SEV-SNP guest, the compiler
would emit a TESTL instruction which is not supported by the SEV-ES
emulator, causing a boot failure in that environment. It is likely the
same problem would happen in a TDX guest as that uses the same
instruction emulator as SEV-ES.
To make sure all emulators can emulate APIC memory reads via MOV, use
the readl() function in native_apic_mem_read(). It is expected that any
emulator would support MOV in any addressing mode as it is the most
generic and is what is usually emitted currently.
The TESTL instruction is emitted when native_apic_mem_read() is inlined
into apic_mem_wait_icr_idle(). The emulator comes from
insn_decode_mmio() in arch/x86/lib/insn-eval.c. It's not worth it to
extend insn_decode_mmio() to support more instructions since, in theory,
the compiler could choose to output nearly any instruction for such
reads which would bloat the emulator beyond reason.
Building with clang results in the following warning:
posix_timers.c:69:6: warning: absolute value function 'abs' given an
argument of type 'long long' but has parameter of type 'int' which may
cause truncation of value [-Wabsolute-value]
if (abs(diff - DELAY * USECS_PER_SEC) > USECS_PER_SEC / 2) {
^
So switch to using llabs() instead.
Initialize cpu_mitigations to CPU_MITIGATIONS_OFF if the kernel is built
with CONFIG_SPECULATION_MITIGATIONS=n, as the help text quite clearly
states that disabling SPECULATION_MITIGATIONS is supposed to turn off all
mitigations by default.
│ If you say N, all mitigations will be disabled. You really
│ should know what you are doing to say so.
As is, the kernel still defaults to CPU_MITIGATIONS_AUTO, which results in
some mitigations being enabled in spite of SPECULATION_MITIGATIONS=n.
On x86 each struct cpu_hw_events maintains a table for counter assignment but
it missed to update one for the deleted event in x86_pmu_del(). This
can make perf_clear_dirty_counters() reset used counter if it's called
before event scheduling or enabling. Then it would return out of range
data which doesn't make sense.
It should move the contents in the cpuc->assign as well.
Fixes: 5471eea5d3bf ("perf/x86: Reset the dirty counter to prevent the leak for an RDPMC task") Signed-off-by: Namhyung Kim <namhyung@kernel.org> Signed-off-by: Ingo Molnar <mingo@kernel.org> Reviewed-by: Kan Liang <kan.liang@linux.intel.com> Cc: stable@vger.kernel.org Link: https://lore.kernel.org/r/20240306061003.1894224-1-namhyung@kernel.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
A smp_rmb() has been missed in vhost_enable_notify(), inspired by
Will. Otherwise, it's not ensured the available ring entries pushed
by guest can be observed by vhost in time, leading to stale available
ring entries fetched by vhost in vhost_get_vq_desc(), as reported by
Yihuang Yu on NVidia's grace-hopper (ARM64) platform.
Add the missed smp_rmb() in vhost_enable_notify(). When it returns true,
it means there's still pending tx buffers. Since it might read indices,
so it still can bypass the smp_rmb() in vhost_get_vq_desc(). Note that
it should be safe until vq->avail_idx is changed by commit d3bb267bbdcb
("vhost: cache avail index in vhost_enable_notify()").
Fixes: d3bb267bbdcb ("vhost: cache avail index in vhost_enable_notify()") Cc: <stable@kernel.org> # v5.18+ Reported-by: Yihuang Yu <yihyu@redhat.com> Suggested-by: Will Deacon <will@kernel.org> Signed-off-by: Gavin Shan <gshan@redhat.com> Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20240328002149.1141302-3-gshan@redhat.com> Signed-off-by: Michael S. Tsirkin <mst@redhat.com> Reviewed-by: Stefano Garzarella <sgarzare@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
A smp_rmb() has been missed in vhost_vq_avail_empty(), spotted by
Will. Otherwise, it's not ensured the available ring entries pushed
by guest can be observed by vhost in time, leading to stale available
ring entries fetched by vhost in vhost_get_vq_desc(), as reported by
Yihuang Yu on NVidia's grace-hopper (ARM64) platform.
Add the missed smp_rmb() in vhost_vq_avail_empty(). When tx_can_batch()
returns true, it means there's still pending tx buffers. Since it might
read indices, so it still can bypass the smp_rmb() in vhost_get_vq_desc().
Note that it should be safe until vq->avail_idx is changed by commit 275bf960ac697 ("vhost: better detection of available buffers").
Fixes: 275bf960ac69 ("vhost: better detection of available buffers") Cc: <stable@kernel.org> # v4.11+ Reported-by: Yihuang Yu <yihyu@redhat.com> Suggested-by: Will Deacon <will@kernel.org> Signed-off-by: Gavin Shan <gshan@redhat.com> Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20240328002149.1141302-2-gshan@redhat.com> Signed-off-by: Michael S. Tsirkin <mst@redhat.com> Reviewed-by: Stefano Garzarella <sgarzare@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Arg0 is divided by 4 in lpcg driver. <&spi0_lpcg 0> and <&spi0_lpcg 1> are
IMX_SC_PM_CLK_PER. Although code can work, code logic is wrong. It should
use IMX_LPCG_CLK_0 and IMX_LPCG_CLK_4 for lpcg arg0.
Arg0 is divided by 4 in lpcg driver, so index 0 and 1 will be get by pwm
driver, which are same as IMX_LPCG_CLK_6 and IMX_LPCG_CLK_1. Even it can
work, but code logic is wrong. Fixed it by use correct indices.
Cc: stable@vger.kernel.org Fixes: 23fa99b205ea ("arm64: dts: freescale: imx8-ss-lsio: add support for lsio_pwm0-3") Signed-off-by: Frank Li <Frank.Li@nxp.com> Signed-off-by: Shawn Guo <shawnguo@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Arg0 is divided by 4 in lpcg driver. So adc get IMX_SC_PM_CLK_PER by
<&adc0_lpcg 0>, <&adc0_lpcg 1>. Although function can work, code logic is
wrong. Fix it by using correct indices.
Arg0 is divided by 4 in lpcg driver. flexcan driver get IMX_SC_PM_CLK_PER
by <&can0_lpcg 1> and <&can0_lpcg 0>. Although function can work, code
logic is wrong. Fix it by using correct clock indices.
Cc: stable@vger.kernel.org Fixes: 5e7d5b023e03 ("arm64: dts: imx8qxp: add flexcan in adma") Signed-off-by: Frank Li <Frank.Li@nxp.com> Signed-off-by: Shawn Guo <shawnguo@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Arg0 is divided by 4 in lpcg driver. So flexcan get IMX_SC_PM_CLK_PER by
<&can1_lpcg 1> and <&can1_lpcg 0>. Although function work, code logic is
wrong. Fix it by using correct clock indices.
Cc: stable@vger.kernel.org Fixes: be85831de020 ("arm64: dts: imx8qm: add can node in devicetree") Signed-off-by: Frank Li <Frank.Li@nxp.com> Signed-off-by: Shawn Guo <shawnguo@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The modes[] array contains pointers to modes on the connectors'
mode lists, which are protected by dev->mode_config.mutex.
Thus we need to extend modes[] the same protection or by the
time we use it the elements may already be pointing to
freed/reused memory.
Cc: stable@vger.kernel.org Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/10583 Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20240404203336.10454-2-ville.syrjala@linux.intel.com Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org> Reviewed-by: Jani Nikula <jani.nikula@intel.com> Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Subject: drm/panfrost: Fix the error path in panfrost_mmu_map_fault_addr()
If some the pages or sgt allocation failed, we shouldn't release the
pages ref we got earlier, otherwise we will end up with unbalanced
get/put_pages() calls. We should instead leave everything in place
and let the BO release function deal with extra cleanup when the object
is destroyed, or let the fault handler try again next time it's called.
Fixes: 187d2929206e ("drm/panfrost: Add support for GPU heap allocations") Cc: <stable@vger.kernel.org> Reviewed-by: Steven Price <steven.price@arm.com> Reviewed-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com> Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com> Co-developed-by: Dmitry Osipenko <dmitry.osipenko@collabora.com> Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com> Link: https://patchwork.freedesktop.org/patch/msgid/20240105184624.508603-18-dmitry.osipenko@collabora.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
There is a while-loop in ast_dp_set_on_off() that could lead to
infinite-loop. This is because the register, VGACRI-Dx, checked in
this API is a scratch register actually controlled by a MCU, named
DPMCU, in BMC.
These scratch registers are protected by scu-lock. If suc-lock is not
off, DPMCU can not update these registers and then host will have soft
lockup due to never updated status.
DPMCU is used to control DP and relative registers to handshake with
host's VGA driver. Even the most time-consuming task, DP's link
training, is less than 100ms. 200ms should be enough.
Signed-off-by: Jammy Huang <jammy_huang@aspeedtech.com> Fixes: 594e9c04b586 ("drm/ast: Create the driver for ASPEED proprietory Display-Port") Reviewed-by: Jocelyn Falempe <jfalempe@redhat.com> Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de> Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de> Cc: KuoHsiang Chou <kuohsiang_chou@aspeedtech.com> Cc: Thomas Zimmermann <tzimmermann@suse.de> Cc: Dave Airlie <airlied@redhat.com> Cc: Jocelyn Falempe <jfalempe@redhat.com> Cc: dri-devel@lists.freedesktop.org Cc: <stable@vger.kernel.org> # v5.19+ Link: https://patchwork.freedesktop.org/patch/msgid/20240403090246.1495487-1-jammy_huang@aspeedtech.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Currently, with F32 HWS GPU reset is only when unmap queue fails.
However, if compute queue doesn't repond to preemption request in time
unmap will return without any error. In this case, only preemption error
is logged and Reset is not triggered. Call GPU reset in this case also.
All joined pipes share the same transcoder/timing generator.
Currently we just do the commits per-pipe, which doesn't really
work if we need to change switch between non-VRR and VRR timings
generators on the fly, or even when sending the push to the
transcoder. For now just disable VRR when bigjoiner is needed.
ivpu_device->context_xa is locked both in kernel thread and IRQ context.
It requires XA_FLAGS_LOCK_IRQ flag to be passed during initialization
otherwise the lock could be acquired from a thread and interrupted by
an IRQ that locks it for the second time causing the deadlock.
This deadlock was reported by lockdep and observed in internal tests.
Commit 27f58c04a8f4 ("scsi: sg: Avoid sg device teardown race") introduced
an incorrect WARN_ON_ONCE() and missed a sequence where sg_device_destroy()
was used after scsi_device_put().
sg_device_destroy() is accessing the parent scsi_device request_queue which
will already be set to NULL when the preceding call to scsi_device_put()
removed the last reference to the parent scsi_device.
Drop the incorrect WARN_ON_ONCE() - allowing more than one concurrent
access to the sg device - and make sure sg_device_destroy() is not used
after scsi_device_put() in the error handling.
sg_remove_sfp_usercontext() must not use sg_device_destroy() after calling
scsi_device_put().
sg_device_destroy() is accessing the parent scsi_device request_queue which
will already be set to NULL when the preceding call to scsi_device_put()
removed the last reference to the parent scsi_device.
The resulting NULL pointer exception will then crash the kernel.
When unloading a module, its state is changing MODULE_STATE_LIVE ->
MODULE_STATE_GOING -> MODULE_STATE_UNFORMED. Each change will take
a time. `is_module_text_address()` and `__module_text_address()`
works with MODULE_STATE_LIVE and MODULE_STATE_GOING.
If we use `is_module_text_address()` and `__module_text_address()`
separately, there is a chance that the first one is succeeded but the
next one is failed because module->state becomes MODULE_STATE_UNFORMED
between those operations.
In `check_kprobe_address_safe()`, if the second `__module_text_address()`
is failed, that is ignored because it expected a kernel_text address.
But it may have failed simply because module->state has been changed
to MODULE_STATE_UNFORMED. In this case, arm_kprobe() will try to modify
non-exist module text address (use-after-free).
To fix this problem, we should not use separated `is_module_text_address()`
and `__module_text_address()`, but use only `__module_text_address()`
once and do `try_module_get(module)` which is only available with
MODULE_STATE_LIVE.
cac9e4418f4cb ("io_uring/net: save msghdr->msg_control for retries")
reinstatiates msg_control before every __sys_sendmsg_sock(), since the
function can overwrite the value in msghdr. We need to do same for
zerocopy sendmsg.
The transaction is only able to free PERTRANS reservations for a root
once that root has been recorded with the TRANS tag on the roots radix
tree. Therefore, until we are sure that this root will get tagged, it
isn't safe to convert. Generally, this is not an issue as *some*
transaction will likely tag the root before long and this reservation
will get freed in that transaction, but technically it could stick
around until unmount and result in a warning about leaked metadata
reservation space.
This path is most exercised by running the generic/269 fstest with
CONFIG_BTRFS_DEBUG.
When running delayed inode updates, we do not record the inode's root in
the transaction, but we do allocate PREALLOC and thus converted PERTRANS
space for it. To be sure we free that PERTRANS meta rsv, we must ensure
that we record the root in the transaction.
Fixes: 4f5427ccce5d ("btrfs: delayed-inode: Use new qgroup meta rsv for delayed inode and item") CC: stable@vger.kernel.org # 6.1+ Reviewed-by: Qu Wenruo <wqu@suse.com> Signed-off-by: Boris Burkov <boris@bur.io> Signed-off-by: David Sterba <dsterba@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Create subvolume, create snapshot and delete subvolume all use
btrfs_subvolume_reserve_metadata() to reserve metadata for the changes
done to the parent subvolume's fs tree, which cannot be mediated in the
normal way via start_transaction. When quota groups (squota or qgroups)
are enabled, this reserves qgroup metadata of type PREALLOC. Once the
operation is associated to a transaction, we convert PREALLOC to
PERTRANS, which gets cleared in bulk at the end of the transaction.
However, the error paths of these three operations were not implementing
this lifecycle correctly. They unconditionally converted the PREALLOC to
PERTRANS in a generic cleanup step regardless of errors or whether the
operation was fully associated to a transaction or not. This resulted in
error paths occasionally converting this rsv to PERTRANS without calling
record_root_in_trans successfully, which meant that unless that root got
recorded in the transaction by some other thread, the end of the
transaction would not free that root's PERTRANS, leaking it. Ultimately,
this resulted in hitting a WARN in CONFIG_BTRFS_DEBUG builds at unmount
for the leaked reservation.
The fix is to ensure that every qgroup PREALLOC reservation observes the
following properties:
1. any failure before record_root_in_trans is called successfully
results in freeing the PREALLOC reservation.
2. after record_root_in_trans, we convert to PERTRANS, and now the
transaction owns freeing the reservation.
This patch enforces those properties on the three operations. Without
it, generic/269 with squotas enabled at mkfs time would fail in ~5-10
runs on my system. With this patch, it ran successfully 1000 times in a
row.
Fixes: e85fde5162bf ("btrfs: qgroup: fix qgroup meta rsv leak for subvolume operations") CC: stable@vger.kernel.org # 6.1+ Reviewed-by: Qu Wenruo <wqu@suse.com> Signed-off-by: Boris Burkov <boris@bur.io> Signed-off-by: David Sterba <dsterba@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We use add_root_meta_rsv and sub_root_meta_rsv to track prealloc and
pertrans reservations for subvolumes when quotas are enabled. The
convert function does not properly increment pertrans after decrementing
prealloc, so the count is not accurate.
Note: we check that the fs is not read-only to mirror the logic in
qgroup_convert_meta, which checks that before adding to the pertrans rsv.
Fixes: 8287475a2055 ("btrfs: qgroup: Use root::qgroup_meta_rsv_* to record qgroup meta reserved space") CC: stable@vger.kernel.org # 6.1+ Reviewed-by: Qu Wenruo <wqu@suse.com> Signed-off-by: Boris Burkov <boris@bur.io> Signed-off-by: David Sterba <dsterba@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This patch uses addition assignment operator (+=) to append strings
instead of duplicating the variable name in mptcp_connect.sh and
mptcp_join.sh.
This can make the statements shorter.
Note: in mptcp_connect.sh, add a local variable extra in do_transfer to
save the various extra warning logs, using += to append it. And add a
new variable tc_info to save various tc info, also using += to append it.
This can make the code more readable and prepare for the next commit.
Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn> Reviewed-by: Matthieu Baerts (NGI0) <matttbe@kernel.org> Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org> Link: https://lore.kernel.org/r/20240308-upstream-net-next-20240308-selftests-mptcp-unification-v1-8-4f42c347b653@kernel.org Signed-off-by: Jakub Kicinski <kuba@kernel.org>
[ Conflicts in mptcp_connect.sh: this commit was supposed to be
backported before commit 7a1b3490f47e ("mptcp: don't account accept()
of non-MPC client as fallback to TCP"). The new condition added by
this commit was then not expected, and was in fact at the wrong place
in v6.6: in case of issue, the problem would not have been reported
correctly. ] Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When CONFIG_PERF_EVENTS, a 'make W=1' build produces a warning about the
unused ftrace_event_id_fops variable:
kernel/trace/trace_events.c:2155:37: error: 'ftrace_event_id_fops' defined but not used [-Werror=unused-const-variable=]
2155 | static const struct file_operations ftrace_event_id_fops = {
Hide this in the same #ifdef as the reference to it.
The patch mentioned in the `Fixes` tag removed the explicit assignment
of tx_info->xdpf to NULL with the justification that there's no need
to set tx_info->xdpf to NULL and tx_info->num_of_bufs to 0 in case
of a mapping error. Both values won't be used once the mapping function
returns an error, and their values would be overridden by the next
transmitted packet.
While both values do indeed get overridden in the next transmission
call, the value of tx_info->xdpf is also used to check whether a TX
descriptor's transmission has been completed (i.e. a completion for it
was polled).
An example scenario:
1. Mapping failed, tx_info->xdpf wasn't set to NULL
2. A VF reset occurred leading to IO resource destruction and
a call to ena_free_tx_bufs() function
3. Although the descriptor whose mapping failed was freed by the
transmission function, it still passes the check
if (!tx_info->skb)
(skb and xdp_frame are in a union)
4. The xdp_frame associated with the descriptor is freed twice
This patch returns the assignment of NULL to tx_info->xdpf to make the
cleaning function knows that the descriptor is already freed.
Fixes: 504fd6a5390c ("net: ena: fix DMA mapping function issues in XDP") Signed-off-by: Shay Agroskin <shayagr@amazon.com> Signed-off-by: David Arinzon <darinzon@amazon.com> Reviewed-by: Shannon Nelson <shannon.nelson@amd.com> Signed-off-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
When an XDP program is loaded the existing channels in the driver split
into two halves:
- The first half of the channels contain RX and TX rings, these queues
are used for receiving traffic and sending packets originating from
kernel.
- The second half of the channels contain only a TX ring. These queues
are used for sending packets that were redirected using XDP_TX
or XDP_REDIRECT.
Referring to the queues in the second half of the channels as "xdp_ring"
can be confusing and may give the impression that ENA has the capability
to generate an additional special queue.
This patch ensures that the xdp_ring field is exclusively used to
describe the XDP TX queue that a specific RX queue needs to utilize when
forwarding packets with XDP TX and XDP REDIRECT, preserving the
integrity of the xdp_ring field in ena_ring.
Signed-off-by: Shay Agroskin <shayagr@amazon.com> Signed-off-by: David Arinzon <darinzon@amazon.com> Link: https://lore.kernel.org/r/20240101190855.18739-6-darinzon@amazon.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Stable-dep-of: 36a1ca01f045 ("net: ena: Set tx_info->xdpf value to NULL") Signed-off-by: Sasha Levin <sashal@kernel.org>
This change will enable the ability to use ena_xmit_common()
in functions that don't have a net_device pointer.
While it can be retrieved by dereferencing
ena_adapter (adapter->netdev), there's no reason to do it in
fast path code where this pointer is only needed for
debug prints.
Signed-off-by: Shay Agroskin <shayagr@amazon.com> Signed-off-by: David Arinzon <darinzon@amazon.com> Link: https://lore.kernel.org/r/20240101190855.18739-3-darinzon@amazon.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Stable-dep-of: 36a1ca01f045 ("net: ena: Set tx_info->xdpf value to NULL") Signed-off-by: Sasha Levin <sashal@kernel.org>
projects / thirdparty / kernel / stable.git / log
