Michael reports that since linux-next-20210211, the AER messages for ECC
errors have started reappearing, and this time they can be reliably
reproduced with the first ping on one of his LS1028A boards.
$ ping 1[ 33.258069] pcieport 0000:00:1f.0: AER: Multiple Corrected error received: 0000:00:00.0
72.16.0.1
PING [ 33.267050] pcieport 0000:00:1f.0: AER: can't find device of ID0000
172.16.0.1 (172.16.0.1): 56 data bytes
64 bytes from 172.16.0.1: seq=0 ttl=64 time=17.124 ms
64 bytes from 172.16.0.1: seq=1 ttl=64 time=0.273 ms
$ devmem 0x1f8010e10 32
0xC0000006
It isn't clear why this is necessary, but it seems that for the errors
to go away, we must clear the entire RFS and RSS memory, not just for
the ports in use.
Sadly the code is structured in such a way that we can't have unified
logic for the used and unused ports. For the minimal initialization of
an unused port, we need just to enable and ioremap the PF memory space,
and a control buffer descriptor ring. Unused ports must then free the
CBDR because the driver will exit, but used ports can not pick up from
where that code path left, since the CBDR API does not reinitialize a
ring when setting it up, so its producer and consumer indices are out of
sync between the software and hardware state. So a separate
enetc_init_unused_port function was created, and it gets called right
after the PF memory space is enabled.
Fixes: 07bf34a50e32 ("net: enetc: initialize the RFS and RSS memories") Reported-by: Michael Walle <michael@walle.cc> Cc: Jesse Brandeburg <jesse.brandeburg@intel.com> Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com> Tested-by: Michael Walle <michael@walle.cc> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
which zero-initializes the entire port RSS table in order to avoid ECC errors.
The trouble really is that enetc_init_port_rss_memory really neads
enetc_alloc_si_resources to be called, because it depends upon
enetc_alloc_cbdr and enetc_setup_cbdr. But that whole enetc_configure_si
thing could have been better thought out, it has nothing to do in a
function called "alloc_si_resources", especially since its counterpart,
"free_si_resources", does nothing to unwind the configuration of the SI.
The point is, we need to pull out enetc_configure_si out of
enetc_alloc_resources, and move it after enetc_init_port_rss_memory.
This allows us to set up the default RSS indirection table after
initializing the memory.
Fixes: 07bf34a50e32 ("net: enetc: initialize the RFS and RSS memories") Cc: Jesse Brandeburg <jesse.brandeburg@intel.com> Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
According to the SH7710, SH7712, SH7713 Group User's Manual: Hardware,
Rev. 3.00, the TRSCER register actually has only bit 7 valid (and named
differently), with all the other bits reserved. Apparently, this was not
the case with some early revisions of the manual as we have the other
bits declared (and set) in the original driver. Follow the suit and add
the explicit sh_eth_cpu_data::trscer_err_mask initializer for SH771x...
Fixes: 86a74ff21a7a ("net: sh_eth: add support for Renesas SuperH Ethernet") Signed-off-by: Sergey Shtylyov <s.shtylyov@omprussia.ru> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
Commit 86dd9868b878 has several issues, but was accepted too soon
before anyone could take a look.
- Double free. dsa_slave_xmit() will free the skb if the xmit function
returns NULL, but the skb is already freed by eth_skb_pad(). Use
__skb_put_padto() to avoid that.
- Unnecessary allocation. It has been done by DSA core since commit a3b0b6479700.
- A u16 pointer points to skb data. It should be __be16 for network
byte order.
- Typo in comments. "numer" -> "number".
Fixes: 86dd9868b878 ("net: dsa: tag_rtl4_a: Support also egress tags") Signed-off-by: DENG Qingfang <dqfext@gmail.com> Reviewed-by: Florian Fainelli <f.fainelli@gmail.com> Reviewed-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
The kernel test robot reports a huge performance regression due to the
commit, and the reason seems fairly straightforward: when there is
contention on the page list (which is what causes acquire_slab() to
fail), we do _not_ want to just loop and try again, because that will
transfer the contention to the 'n->list_lock' spinlock we hold, and
just make things even worse.
This is admittedly likely a problem only on big machines - the kernel
test robot report comes from a 96-thread dual socket Intel Xeon Gold
6252 setup, but the regression there really is quite noticeable:
-47.9% regression of stress-ng.rawpkt.ops_per_sec
and the commit that was marked as being fixed (7ced37197196: "slub:
Acquire_slab() avoid loop") actually did the loop exit early very
intentionally (the hint being that "avoid loop" part of that commit
message), exactly to avoid this issue.
The correct thing to do may be to pick some kind of reasonable middle
ground: instead of breaking out of the loop on the very first sign of
contention, or trying over and over and over again, the right thing may
be to re-try _once_, and then give up on the second failure (or pick
your favorite value for "once"..).
Reported-by: kernel test robot <oliver.sang@intel.com> Link: https://lore.kernel.org/lkml/20210301080404.GF12822@xsang-OptiPlex-9020/ Cc: Jann Horn <jannh@google.com> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Acked-by: Christoph Lameter <cl@linux.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
In cifs_statfs(), if server->ops->queryfs is not NULL, then we should
use its return value rather than always returning 0. Instead, use rc
variable as it is properly set to 0 in case there is no
server->ops->queryfs.
Signed-off-by: Paulo Alcantara (SUSE) <pc@cjr.nz> Reviewed-by: Aurelien Aptel <aaptel@suse.com> Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com> CC: <stable@vger.kernel.org> Signed-off-by: Steve French <stfrench@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
With multichannel, operations like the queries
from "ls -lR" can cause all credits to be used and
errors to be returned since max_credits was not
being set correctly on the secondary channels and
thus the client was requesting 0 credits incorrectly
in some cases (which can lead to not having
enough credits to perform any operation on that
channel).
Signed-off-by: Aurelien Aptel <aaptel@suse.com> CC: <stable@vger.kernel.org> # v5.8+ Reviewed-by: Shyam Prasad N <sprasad@microsoft.com> Signed-off-by: Steve French <stfrench@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
Creating a series of detached mounts, attaching them to the filesystem,
and unmounting them can be used to trigger an integer overflow in
ns->mounts causing the kernel to block any new mounts in count_mounts()
and returning ENOSPC because it falsely assumes that the maximum number
of mounts in the mount namespace has been reached, i.e. it thinks it
can't fit the new mounts into the mount namespace anymore.
Depending on the number of mounts in your system, this can be reproduced
on any kernel that supportes open_tree() and move_mount() by compiling
and running the following program:
new_argv = &argv[optind];
new_argc = argc - optind;
if (new_argc < 1)
exit_usage("Missing base directory\n");
base_dir = new_argv[0];
if (*base_dir != '/')
exit_log("Please specify an absolute path");
/* Ensure that target is a shared mountpoint. */
if (!is_shared_mountpoint(base_dir))
exit_log("Please ensure that \"%s\" is a shared mountpoint", base_dir);
dfd = open(base_dir, O_RDONLY | O_DIRECTORY | O_CLOEXEC);
if (dfd < 0)
exit_log("%m - Failed to open base directory \"%s\"", base_dir);
ret = mkdirat(dfd, "detached-move-mount", 0755);
if (ret < 0)
exit_log("%m - Failed to create required temporary directories");
ret = snprintf(target, sizeof(target), "%s/detached-move-mount", base_dir);
if (ret < 0 || (size_t)ret >= sizeof(target))
exit_log("%m - Failed to assemble target path");
/*
* Having a mount table with 10000 mounts is already quite excessive
* and shoult account even for weird test systems.
*/
for (size_t i = 0; i < 10000; i++) {
fd_tree = sys_open_tree(dfd, "detached-move-mount",
OPEN_TREE_CLONE |
OPEN_TREE_CLOEXEC |
AT_EMPTY_PATH);
if (fd_tree < 0) {
fprintf(stderr, "%m - Failed to open %d(detached-move-mount)", dfd);
exit_code = EXIT_FAILURE;
break;
}
ret = sys_move_mount(fd_tree, "", dfd, "detached-move-mount", MOVE_MOUNT_F_EMPTY_PATH);
if (ret < 0) {
if (errno == ENOSPC)
fprintf(stderr, "%m - Buggy mount counting");
else
fprintf(stderr, "%m - Failed to attach mount to %d(detached-move-mount)", dfd);
exit_code = EXIT_FAILURE;
break;
}
close(fd_tree);
ret = umount2(target, MNT_DETACH);
if (ret < 0) {
fprintf(stderr, "%m - Failed to unmount %s", target);
exit_code = EXIT_FAILURE;
break;
}
}
and wait for the kernel to refuse any new mounts by returning ENOSPC.
How many iterations are needed depends on the number of mounts in your
system. Assuming you have something like 50 mounts on a standard system
it should be almost instantaneous.
The root cause of this is that detached mounts aren't handled correctly
when source and target mount are identical and reside on a shared mount
causing a broken mount tree where the detached source itself is
propagated which propagation prevents for regular bind-mounts and new
mounts. This ultimately leads to a miscalculation of the number of
mounts in the mount namespace.
Detached mounts created via
open_tree(fd, path, OPEN_TREE_CLONE)
are essentially like an unattached new mount, or an unattached
bind-mount. They can then later on be attached to the filesystem via
move_mount() which calls into attach_recursive_mount(). Part of
attaching it to the filesystem is making sure that mounts get correctly
propagated in case the destination mountpoint is MS_SHARED, i.e. is a
shared mountpoint. This is done by calling into propagate_mnt() which
walks the list of peers calling propagate_one() on each mount in this
list making sure it receives the propagation event.
The propagate_one() functions thereby skips both new mounts and bind
mounts to not propagate them "into themselves". Both are identified by
checking whether the mount is already attached to any mount namespace in
mnt->mnt_ns. The is what the IS_MNT_NEW() helper is responsible for.
However, detached mounts have an anonymous mount namespace attached to
them stashed in mnt->mnt_ns which means that IS_MNT_NEW() doesn't
realize they need to be skipped causing the mount to propagate "into
itself" breaking the mount table and causing a disconnect between the
number of mounts recorded as being beneath or reachable from the target
mountpoint and the number of mounts actually recorded/counted in
ns->mounts ultimately causing an overflow which in turn prevents any new
mounts via the ENOSPC issue.
So teach propagation to handle detached mounts by making it aware of
them. I've been tracking this issue down for the last couple of days and
then verifying that the fix is correct by
unmounting everything in my current mount table leaving only /proc and
/sys mounted and running the reproducer above overnight verifying the
number of mounts counted in ns->mounts. With this fix the counts are
correct and the ENOSPC issue can't be reproduced.
This change will only have an effect on mounts created with the new
mount API since detached mounts cannot be created with the old mount API
so regressions are extremely unlikely.
Link: https://lore.kernel.org/r/20210306101010.243666-1-christian.brauner@ubuntu.com Fixes: 2db154b3ea8e ("vfs: syscall: Add move_mount(2) to move mounts around") Cc: David Howells <dhowells@redhat.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: linux-fsdevel@vger.kernel.org Cc: <stable@vger.kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
Make sure to hold the gpio_lock when removing the gpio device from the
gpio_devices list (when dropping the last reference) to avoid corrupting
the list when there are concurrent accesses.
Fixes: ff2b13592299 ("gpio: make the gpiochip a real device") Cc: stable@vger.kernel.org # 4.6 Reviewed-by: Saravana Kannan <saravanak@google.com> Signed-off-by: Johan Hovold <johan@kernel.org> Signed-off-by: Bartosz Golaszewski <bgolaszewski@baylibre.com>
[ johan: adjust context to 5.11 ] Signed-off-by: Johan Hovold <johan@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
Commit b102f0c522cf6 ("mt76: fix array overflow on receiving too many
fragments for a packet") fixes a possible OOB access but it introduces a
memory leak since the pending frame is not released to page_frag_cache
if the frag array of skb_shared_info is full. Commit 93a1d4791c10
("mt76: dma: fix a possible memory leak in mt76_add_fragment()") fixes
the issue but does not free the truncated skb that is forwarded to
mac80211 layer. Fix the leftover issue discarding even truncated skbs.
Fixes: 93a1d4791c10 ("mt76: dma: fix a possible memory leak in mt76_add_fragment()") Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org> Signed-off-by: Kalle Valo <kvalo@codeaurora.org> Link: https://lore.kernel.org/r/a03166fcc8214644333c68674a781836e0f57576.1612697217.git.lorenzo@kernel.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
As reported by coccinelle, the second _irqsave() overwrites the value
saved in 'flags' by the first _irqsave(), therefore when the second
_irqrestore() comes,the value in 'flags' is not valid,the value saved
by the first _irqsave() has been lost.
This likely leads to IRQs remaining disabled. So remove the second
_irqsave():
spin_lock_irqsave(&adapter->state_lock, flags);
spin_lock(&adapter->rwi_lock);
Generated by: ./scripts/coccinelle/locks/flags.cocci
./drivers/net/ethernet/ibm/ibmvnic.c:5413:1-18:
ERROR: nested lock+irqsave that reuses flags from line 5404.
Fixes: 4a41c421f367 ("ibmvnic: serialize access to work queue on remove") Signed-off-by: Junlin Yang <yangjunlin@yulong.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
The last change to ibmvnic_set_mac(), 8fc3672a8ad3, meant to prevent
users from setting an invalid MAC address on an ibmvnic interface
that has not been brought up yet. The change also prevented the
requested MAC address from being stored by the adapter object for an
ibmvnic interface when the state of the ibmvnic interface is
VNIC_PROBED - that is after probing has finished but before the
ibmvnic interface is brought up. The MAC address stored by the
adapter object is used and sent to the hypervisor for checking when
an ibmvnic interface is brought up.
The ibmvnic driver ignoring the requested MAC address when in
VNIC_PROBED state caused LACP bonds (bonds in 802.3ad mode) with more
than one slave to malfunction. The bonding code must be able to
change the MAC address of its slaves before they are brought up
during enslaving. The inability of kernels with 8fc3672a8ad3 to set
the MAC addresses of bonding slaves is observable in the output of
"ip address show". The MAC addresses of the slaves are the same as
the MAC address of the bond on a working system whereas the slaves
retain their original MAC addresses on a system with a malfunctioning
LACP bond.
Fixes: 8fc3672a8ad3 ("ibmvnic: fix ibmvnic_set_mac") Signed-off-by: Jiri Wiesner <jwiesner@suse.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
GCC 7.5 reports:
../drivers/net/ethernet/ibm/ibmvnic.c: In function 'ibmvnic_reset_init':
../drivers/net/ethernet/ibm/ibmvnic.c:5373:51: warning: 'old_num_tx_queues' may be used uninitialized in this function [-Wmaybe-uninitialized]
../drivers/net/ethernet/ibm/ibmvnic.c:5373:6: warning: 'old_num_rx_queues' may be used uninitialized in this function [-Wmaybe-uninitialized]
The variable is initialized only if(reset) and used only if(reset &&
something) so this is a false positive. However, there is no reason to
not initialize the variables unconditionally avoiding the warning.
Fixes: 635e442f4a48 ("ibmvnic: merge ibmvnic_reset_init and ibmvnic_init") Signed-off-by: Michal Suchanek <msuchanek@suse.de> Reviewed-by: Sukadev Bhattiprolu <sukadev@linux.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
xsk_lookup_bpf_maps, based on prog_fd, looks whether current prog has a
reference to XSKMAP. BPF prog can include insns that work on various BPF
maps and this is covered by iterating through map_ids.
The bpf_map_info that is passed to bpf_obj_get_info_by_fd for filling
needs to be cleared at each iteration, so that it doesn't contain any
outdated fields and that is currently missing in the function of
interest.
To fix that, zero-init map_info via memset before each
bpf_obj_get_info_by_fd call.
Also, since the area of this code is touched, in general strcmp is
considered harmful, so let's convert it to strncmp and provide the
size of the array name for current map_info.
While at it, do s/continue/break/ once we have found the xsks_map to
terminate the search.
Fixes: 5750902a6e9b ("libbpf: proper XSKMAP cleanup") Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Björn Töpel <bjorn.topel@intel.com> Link: https://lore.kernel.org/bpf/20210303185636.18070-4-maciej.fijalkowski@intel.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
The verifier test labelled "valid read map access into a read-only array
2" calls the bpf_csum_diff() helper and checks its return value. However,
architecture implementations of csum_partial() (which is what the helper
uses) differ in whether they fold the return value to 16 bit or not. For
example, x86 version has ...
if (unlikely(odd)) {
result = from32to16(result);
result = ((result >> 8) & 0xff) | ((result & 0xff) << 8);
}
... while generic lib/checksum.c does:
result = from32to16(result);
if (odd)
result = ((result >> 8) & 0xff) | ((result & 0xff) << 8);
This makes the helper return different values on different architectures,
breaking the test on non-x86. To fix this, add an additional instruction
to always mask the return value to 16 bits, and update the expected return
value accordingly.
Fixes: fb2abb73e575 ("bpf, selftest: test {rd, wr}only flags and direct value access") Signed-off-by: Yauheni Kaliuta <yauheni.kaliuta@redhat.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Link: https://lore.kernel.org/bpf/20210228103017.320240-1-yauheni.kaliuta@redhat.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
In bpf geneve tunnel test we set geneve option on tx side. On rx side we
only call bpf_skb_get_tunnel_opt(). Since commit 9c2e14b48119 ("ip_tunnels:
Set tunnel option flag when tunnel metadata is present") geneve_rx() will
not add TUNNEL_GENEVE_OPT flag if there is no geneve option, which cause
bpf_skb_get_tunnel_opt() return ENOENT and _geneve_get_tunnel() in
test_tunnel_kern.c drop the packet.
As it should be valid that bpf_skb_get_tunnel_opt() return error when
there is not tunnel option, there is no need to drop the packet and
break all geneve rx traffic. Just set opt_class to 0 in this test and
keep returning TC_ACT_OK.
Fixes: 933a741e3b82 ("selftests/bpf: bpf tunnel test.") Signed-off-by: Hangbin Liu <liuhangbin@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: William Tu <u9012063@gmail.com> Link: https://lore.kernel.org/bpf/20210224081403.1425474-1-liuhangbin@gmail.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
If phy uses generic driver and autoneg is on, enter command
"ethtool -s eth0 speed 50" will not change phy speed actually, but
command "ethtool eth0" shows speed is 50Mb/s because phydev->speed
has been set to 50 and no update later.
And duplex setting has same problem too.
However, if autoneg is on, phy only changes speed and duplex according to
phydev->advertising, but not phydev->speed and phydev->duplex. So in this
case, phydev->speed and phydev->duplex don't need to be set in function
phy_ethtool_ksettings_set() if autoneg is on.
Fixes: 51e2a3846eab ("PHY: Avoid unnecessary aneg restarts") Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com> Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
There were a few remaining tunnel drivers that didn't receive the prior
conversion to icmp{,v6}_ndo_send. Knowing now that this could lead to
memory corrution (see ee576c47db60 ("net: icmp: pass zeroed opts from
icmp{,v6}_ndo_send before sending") for details), there's even more
imperative to have these all converted. So this commit goes through the
remaining cases that I could find and does a boring translation to the
ndo variety.
The Fixes: line below is the merge that originally added icmp{,v6}_
ndo_send and converted the first batch of icmp{,v6}_send users. The
rationale then for the change applies equally to this patch. It's just
that these drivers were left out of the initial conversion because these
network devices are hiding in net/ rather than in drivers/net/.
Cc: Florian Westphal <fw@strlen.de> Cc: Willem de Bruijn <willemb@google.com> Cc: David S. Miller <davem@davemloft.net> Cc: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org> Cc: David Ahern <dsahern@kernel.org> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Fixes: 803381f9f117 ("Merge branch 'icmp-account-for-NAT-when-sending-icmps-from-ndo-layer'") Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com> Acked-by: Willem de Bruijn <willemb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
nested target/match_revfn() calls work with xt[NFPROTO_UNSPEC] lists
without taking xt[NFPROTO_UNSPEC].mutex. This can race with module unload
and cause host to crash:
Under extremely rare conditions TCP early demux will retrieve the wrong
socket.
1. local machine establishes a connection to a remote server, S, on port
p.
This gives:
laddr:lport -> S:p
... both in tcp and conntrack.
2. local machine establishes a connection to host H, on port p2.
2a. TCP stack choses same laddr:lport, so we have
laddr:lport -> H:p2 from TCP point of view.
2b). There is a destination NAT rewrite in place, translating
H:p2 to S:p. This results in following conntrack entries:
NAT engine has rewritten laddr:lport to laddr:lport2 to map
the reply packet to the correct origin.
When server sends SYN/ACK to laddr:lport2, the PREROUTING hook
will undo-the SNAT transformation, rewriting IP header to
S:p -> laddr:lport
This causes TCP early demux to associate the skb with the TCP socket
of the first connection.
The INPUT hook will then reverse the DNAT transformation, rewriting
the IP header to H:p2 -> laddr:lport.
Because packet ends up with the wrong socket, the new connection
never completes: originator stays in SYN_SENT and conntrack entry
remains in SYN_RECV until timeout, and responder retransmits SYN/ACK
until it gives up.
To resolve this, orphan the skb after the input rewrite:
Because the source IP address changed, the socket must be incorrect.
We can't move the DNAT undo to prerouting due to backwards
compatibility, doing so will make iptables/nftables rules to no longer
match the way they did.
After orphan, the packet will be handed to the next protocol layer
(tcp, udp, ...) and that will repeat the socket lookup just like as if
early demux was disabled.
getsockopt(TCP_ZEROCOPY_RECEIVE) has a bug where we read a
user-provided "len" field of type signed int, and then compare the
value to the result of an "offsetofend" operation, which is unsigned.
Negative values provided by the user will be promoted to large
positive numbers; thus checking that len < offsetofend() will return
false when the intention was that it return true.
Note that while len is originally checked for negative values earlier
on in do_tcp_getsockopt(), subsequent calls to get_user() re-read the
value from userspace which may have changed in the meantime.
Therefore, re-add the check for negative values after the call to
get_user in the handler code for TCP_ZEROCOPY_RECEIVE.
Fixes: c8856c051454 ("tcp-zerocopy: Return inq along with tcp receive zerocopy.") Reported-by: kernel test robot <lkp@intel.com> Reported-by: Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by: Arjun Roy <arjunroy@google.com> Link: https://lore.kernel.org/r/20210225232628.4033281-1-arjunroy.kdev@gmail.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
This patch prevents a potentially destructive race condition. The
device is fully operational on the bus after entering Normal Mode, so
zeroing the MRAM after entering this mode may lead to loss of
information, e.g. new received messages.
This patch fixes the problem by first initializing the MRAM, then
bringing the device into Normale Mode.
Fixes: 5443c226ba91 ("can: tcan4x5x: Add tcan4x5x driver to the kernel") Link: https://lore.kernel.org/r/20210226163440.313628-1-torin@maxiluxsystems.com Suggested-by: Marc Kleine-Budde <mkl@pengutronix.de> Signed-off-by: Torin Cooper-Bennun <torin@maxiluxsystems.com> Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
Assert HALT bit to enter freeze mode, there is a premise that FRZ bit is
asserted. This patch asserts FRZ bit in flexcan_chip_freeze, although
the reset value is 1b'1. This is a prepare patch, later patch will
invoke flexcan_chip_freeze() to enter freeze mode, which polling freeze
mode acknowledge.
Fixes: b1aa1c7a2165b ("can: flexcan: fix transition from and to freeze mode in chip_{,un}freeze") Link: https://lore.kernel.org/r/20210218110037.16591-2-qiangqing.zhang@nxp.com Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com> Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
The commit 0ea683931adb ("gpio: dwapb: Convert driver to using the
GPIO-lib-based IRQ-chip") indeliberately made a regression on how
IRQ line from GPIO I²C expander is handled. I.e. it reveals that
the quirk for Intel Galileo Gen 2 misses the part of setting IRQ type
which previously was predefined by gpio-dwapb driver. Now, we have to
reorganize the approach to call necessary parts, which can be done via
ACPI_GPIO_QUIRK_ABSOLUTE_NUMBER quirk.
Without this fix and with above mentioned change the kernel hangs
on the first IRQ event with:
gpio gpiochip3: Persistence not supported for GPIO 1
irq 32, desc: 62f8fb50, depth: 0, count: 0, unhandled: 0
->handle_irq(): 41c7b0ab, handle_bad_irq+0x0/0x40
->irq_data.chip(): e03f1e72, 0xc2539218
->action(): 0ecc7e6f
->action->handler(): 8a3db21e, irq_default_primary_handler+0x0/0x10
IRQ_NOPROBE set
unexpected IRQ trap at vector 20
Fixes: ba8c90c61847 ("gpio: pca953x: Override IRQ for one of the expanders on Galileo Gen 2")
Depends-on: 0ea683931adb ("gpio: dwapb: Convert driver to using the GPIO-lib-based IRQ-chip") Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Acked-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
There are two ref count variables controlling the free()ing of a socket:
- struct sock::sk_refcnt - which is changed by sock_hold()/sock_put()
- struct sock::sk_wmem_alloc - which accounts the memory allocated by
the skbs in the send path.
In case there are still TX skbs on the fly and the socket() is closed,
the struct sock::sk_refcnt reaches 0. In the TX-path the CAN stack
clones an "echo" skb, calls sock_hold() on the original socket and
references it. This produces the following back trace:
Currently only search by index is supported. However, in some cases
we might need to pass the quirks to the acpi_dev_gpio_irq_get().
For this, split out acpi_dev_gpio_irq_get_by() and replace
acpi_dev_gpio_irq_get() by calling above with NULL for name parameter.
Fixes: ba8c90c61847 ("gpio: pca953x: Override IRQ for one of the expanders on Galileo Gen 2")
Depends-on: 0ea683931adb ("gpio: dwapb: Convert driver to using the GPIO-lib-based IRQ-chip") Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Acked-by: Mika Westerberg <mika.westerberg@linux.intel.com> Acked-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
On some systems the ACPI tables has wrong pin number and instead of
having a relative one it provides an absolute one in the global GPIO
number space.
Add ACPI_GPIO_QUIRK_ABSOLUTE_NUMBER quirk to cope with such cases.
Fixes: ba8c90c61847 ("gpio: pca953x: Override IRQ for one of the expanders on Galileo Gen 2")
Depends-on: 0ea683931adb ("gpio: dwapb: Convert driver to using the GPIO-lib-based IRQ-chip") Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Acked-by: Mika Westerberg <mika.westerberg@linux.intel.com> Acked-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
Commit 5ee759cda51b ("l2tp: use standard API for warning log messages")
changed a number of warnings about invalid packets in the receive path
so that they are always shown, instead of only when a special L2TP debug
flag is set. Even with rate limiting these warnings can easily cause
significant log spam - potentially triggered by a malicious party
sending invalid packets on purpose.
In addition these warnings were noticed by projects like Tunneldigger [1],
which uses L2TP for its data path, but implements its own control
protocol (which is sufficiently different from L2TP data packets that it
would always be passed up to userspace even with future extensions of
L2TP).
Some of the warnings were already redundant, as l2tp_stats has a counter
for these packets. This commit adds one additional counter for invalid
packets that are passed up to userspace. Packets with unknown session are
not counted as invalid, as there is nothing wrong with the format of
these packets.
With the additional counter, all of these messages are either redundant
or benign, so we reduce them to pr_debug_ratelimited().
Fixes: 5ee759cda51b ("l2tp: use standard API for warning log messages") Signed-off-by: Matthias Schiffer <mschiffer@universe-factory.net> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
Commit c134d1f8c436 ("ath11k: Handle errors if peer creation fails") completely
broke AP mode on QCA6390:
kernel: [ 151.230734] ath11k_pci 0000:06:00.0: failed to create peer after vdev start delay: -22
wpa_supplicant[2307]: Failed to set beacon parameters
wpa_supplicant[2307]: Interface initialization failed
wpa_supplicant[2307]: wlan0: interface state UNINITIALIZED->DISABLED
wpa_supplicant[2307]: wlan0: AP-DISABLED
wpa_supplicant[2307]: wlan0: Unable to setup interface.
wpa_supplicant[2307]: Failed to initialize AP interface
This was because commit c134d1f8c436 ("ath11k: Handle errors if peer creation
fails") added error handling for ath11k_peer_create(), which had been failing
all along but was unnoticed due to the missing error handling. The actual bug
was introduced already in commit aa44b2f3ecd4 ("ath11k: start vdev if a bss peer is
already created").
ath11k_peer_create() was failing because for AP mode the peer is created
already earlier op_add_interface() and we should skip creation here, but the
check for modes was wrong. Fixing that makes AP mode work again.
This shouldn't affect IPQ8074 nor QCN9074 as they have hw_params.vdev_start_delay disabled.
A packet with skb_inner_network_header(skb) == skb_network_header(skb)
and ETH_P_MPLS_UC will prevent mpls_gso_segment from pulling any headers
from the packet. Subsequently, the call to skb_mac_gso_segment will
again call mpls_gso_segment with the same packet leading to an infinite
loop. In addition, ensure that the header length is a multiple of four,
which should hold irrespective of the number of stacked labels.
Signed-off-by: Balazs Nemeth <bnemeth@redhat.com> Acked-by: Willem de Bruijn <willemb@google.com> Reviewed-by: David Ahern <dsahern@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
For gso packets, virtio_net_hdr_set_proto sets the protocol (if it isn't
set) based on the type in the virtio net hdr, but the skb could contain
anything since it could come from packet_snd through a raw socket. If
there is a mismatch between what virtio_net_hdr_set_proto sets and
the actual protocol, then the skb could be handled incorrectly later
on.
An example where this poses an issue is with the subsequent call to
skb_flow_dissect_flow_keys_basic which relies on skb->protocol being set
correctly. A specially crafted packet could fool
skb_flow_dissect_flow_keys_basic preventing EINVAL to be returned.
Avoid blindly trusting the information provided by the virtio net header
by checking that the protocol in the packet actually matches the
protocol set by virtio_net_hdr_set_proto. Note that since the protocol
is only checked if skb->dev implements header_ops->parse_protocol,
packets from devices without the implementation are not checked at this
stage.
Fixes: 9274124f023b ("net: stricter validation of untrusted gso packets") Signed-off-by: Balazs Nemeth <bnemeth@redhat.com> Acked-by: Willem de Bruijn <willemb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
We noticed a GRO issue for UDP-based encaps such as vxlan/geneve when the
csum for the UDP header itself is 0. In that case, GRO aggregation does
not take place on the phys dev, but instead is deferred to the vxlan/geneve
driver (see trace below).
The reason is essentially that GRO aggregation bails out in udp_gro_receive()
for such case when drivers marked the skb with CHECKSUM_UNNECESSARY (ice, i40e,
others) where for non-zero csums 2abb7cdc0dc8 ("udp: Add support for doing
checksum unnecessary conversion") promotes those skbs to CHECKSUM_COMPLETE
and napi context has csum_valid set. This is however not the case for zero
UDP csum (here: csum_cnt is still 0 and csum_valid continues to be false).
At the same time 57c67ff4bd92 ("udp: additional GRO support") added matches
on !uh->check ^ !uh2->check as part to determine candidates for aggregation,
so it certainly is expected to handle zero csums in udp_gro_receive(). The
purpose of the check added via 662880f44203 ("net: Allow GRO to use and set
levels of checksum unnecessary") seems to catch bad csum and stop aggregation
right away.
One way to fix aggregation in the zero case is to only perform the !csum_valid
check in udp_gro_receive() if uh->check is infact non-zero.
# netperf -H 10.55.10.4 -t TCP_STREAM -l 20
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 10.55.10.4 () port 0 AF_INET : demo
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
87380 16384 16384 20.01 24576.53
Fixes: 57c67ff4bd92 ("udp: additional GRO support") Fixes: 662880f44203 ("net: Allow GRO to use and set levels of checksum unnecessary") Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Cc: Eric Dumazet <edumazet@google.com> Cc: Jesse Brandeburg <jesse.brandeburg@intel.com> Cc: Tom Herbert <tom@herbertland.com> Acked-by: Willem de Bruijn <willemb@google.com> Acked-by: John Fastabend <john.fastabend@gmail.com> Link: https://lore.kernel.org/r/20210226212248.8300-1-daniel@iogearbox.net Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
When transmitting to a receiver in dynamic SMPS mode, all transmissions that
use multiple spatial streams need to be sent using CTS-to-self or RTS/CTS to
give the receiver's extra chains some time to wake up.
This fixes the tx rate getting stuck at <= MCS7 for some clients, especially
Intel ones, which make aggressive use of SMPS.
Cc: stable@vger.kernel.org Reported-by: Martin Kennedy <hurricos@gmail.com> Signed-off-by: Felix Fietkau <nbd@nbd.name> Signed-off-by: Kalle Valo <kvalo@codeaurora.org> Link: https://lore.kernel.org/r/20210214184911.96702-1-nbd@nbd.name Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
in current Linux, MPTCP peers advertising endpoints with port numbers use
a sub-option length that wrongly accounts for the trailing TCP NOP. Also,
receivers will only process incoming ADD_ADDR with port having such wrong
sub-option length. Fix this, making ADD_ADDR compliant to RFC8684 §3.4.1.
this can be verified running tcpdump on the kselftests artifacts:
The MIPS Poly1305 implementation is generic MIPS code written such as to
support down to the original MIPS I and MIPS III ISA for the 32-bit and
64-bit variant respectively. Lift the current limitation then to enable
code for MIPSr1 ISA or newer processors only and have it available for
all MIPS processors.
Signed-off-by: Maciej W. Rozycki <macro@orcam.me.uk> Fixes: a11d055e7a64 ("crypto: mips/poly1305 - incorporate OpenSSL/CRYPTOGAMS optimized implementation") Cc: stable@vger.kernel.org # v5.5+ Acked-by: Jason A. Donenfeld <Jason@zx2c4.com> Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
Depending on the number of online CPUs in the original kernel, it is
likely for CPU #0 to be offline in a kdump kernel. The associated IRQs
in the affinity mappings provided by irq_create_affinity_masks() are
thus not started by irq_startup(), as per-design with managed IRQs.
This can be a problem with multi-queue block devices driven by blk-mq :
such a non-started IRQ is very likely paired with the single queue
enforced by blk-mq during kdump (see blk_mq_alloc_tag_set()). This
causes the device to remain silent and likely hangs the guest at
some point.
This is a regression caused by commit 9ea69a55b3b9 ("powerpc/pseries:
Pass MSI affinity to irq_create_mapping()"). Note that this only happens
with the XIVE interrupt controller because XICS has a workaround to bypass
affinity, which is activated during kdump with the "noirqdistrib" kernel
parameter.
The issue comes from a combination of factors:
- discrepancy between the number of queues detected by the multi-queue
block driver, that was used to create the MSI vectors, and the single
queue mode enforced later on by blk-mq because of kdump (i.e. keeping
all queues fixes the issue)
- CPU#0 offline (i.e. kdump always succeed with CPU#0)
Given that I couldn't reproduce on x86, which seems to always have CPU#0
online even during kdump, I'm not sure where this should be fixed. Hence
going for another approach : fine-grained affinity is for performance
and we don't really care about that during kdump. Simply revert to the
previous working behavior of ignoring affinity masks in this case only.
Fixes: 9ea69a55b3b9 ("powerpc/pseries: Pass MSI affinity to irq_create_mapping()") Cc: stable@vger.kernel.org # v5.10+ Signed-off-by: Greg Kurz <groug@kaod.org> Reviewed-by: Laurent Vivier <lvivier@redhat.com> Reviewed-by: Cédric Le Goater <clg@kaod.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20210215094506.1196119-1-groug@kaod.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
The above trace shows that while the CPU was handling a performance
monitor exception, there was a call to security_perf_event_open()
function. In powerpc core-book3s, this function is called from
perf_allow_kernel() check during recording of data address in the
sample via perf_get_data_addr().
Commit da97e18458fb ("perf_event: Add support for LSM and SELinux
checks") introduced security enhancements to perf. As part of this
commit, the new security hook for perf_event_open() was added in all
places where perf paranoid check was previously used. In powerpc
core-book3s code, originally had paranoid checks in
perf_get_data_addr() and power_pmu_bhrb_read(). So
perf_paranoid_kernel() checks were replaced with perf_allow_kernel()
in these PMU helper functions as well.
The intention of paranoid checks in core-book3s was to verify
privilege access before capturing some of the sample data. Along with
paranoid checks, perf_allow_kernel() also does a
security_perf_event_open(). Since these functions are accessed while
recording a sample, we end up calling selinux_perf_event_open() in PMI
context. Some of the security functions use spinlock like
sidtab_sid2str_put(). If a perf interrupt hits under a spin lock and
if we end up in calling selinux hook functions in PMI handler, this
could cause a dead lock.
Since the purpose of this security hook is to control access to
perf_event_open(), it is not right to call this in interrupt context.
The paranoid checks in powerpc core-book3s were done at interrupt time
which is also not correct.
Reference commits:
Commit cd1231d7035f ("powerpc/perf: Prevent kernel address leak via perf_get_data_addr()")
Commit bb19af816025 ("powerpc/perf: Prevent kernel address leak to userspace via BHRB buffer")
We only allow creation of events that have already passed the
privilege checks in perf_event_open(). So these paranoid checks are
not needed at event time. As a fix, patch uses
'event->attr.exclude_kernel' check to prevent exposing kernel address
for userspace only sampling.
Fixes: cd1231d7035f ("powerpc/perf: Prevent kernel address leak via perf_get_data_addr()") Cc: stable@vger.kernel.org # v4.17+ Suggested-by: Michael Ellerman <mpe@ellerman.id.au> Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/1614247839-1428-1-git-send-email-atrajeev@linux.vnet.ibm.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
On book3s/32, page protection is defined by the PP bits in the PTE
which provide the following protection depending on the access
keys defined in the matching segment register:
- PP 00 means RW with key 0 and N/A with key 1.
- PP 01 means RW with key 0 and RO with key 1.
- PP 10 means RW with both key 0 and key 1.
- PP 11 means RO with both key 0 and key 1.
Since the implementation of kernel userspace access protection,
PP bits have been set as follows:
- PP00 for pages without _PAGE_USER
- PP01 for pages with _PAGE_USER and _PAGE_RW
- PP11 for pages with _PAGE_USER and without _PAGE_RW
For kernelspace segments, kernel accesses are performed with key 0
and user accesses are performed with key 1. As PP00 is used for
non _PAGE_USER pages, user can't access kernel pages not flagged
_PAGE_USER while kernel can.
For userspace segments, both kernel and user accesses are performed
with key 0, therefore pages not flagged _PAGE_USER are still
accessible to the user.
This shouldn't be an issue, because userspace is expected to be
accessible to the user. But unlike most other architectures, powerpc
implements PROT_NONE protection by removing _PAGE_USER flag instead of
flagging the page as not valid. This means that pages in userspace
that are not flagged _PAGE_USER shall remain inaccessible.
To get the expected behaviour, just mimic other architectures in the
TLB miss handler by checking _PAGE_USER permission on userspace
accesses as if it was the _PAGE_PRESENT bit.
Note that this problem only is only for 603 cores. The 604+ have
an hash table, and hash_page() function already implement the
verification of _PAGE_USER permission on userspace pages.
Apparently, <linux/netfilter/nfnetlink_cthelper.h> and
<linux/netfilter/nfnetlink_acct.h> could not be included into the same
compilation unit because of a cut-and-paste typo in the former header.
Greentime Hu [Tue, 16 Mar 2021 21:31:15 +0000 (21:31 +0000)]
riscv: dts: Add PCIe support for the SiFive FU740-C000 SoC
Signed-off-by: Greentime Hu <greentime.hu@sifive.com> Signed-off-by: Dimitri John Ledkov <xnox@ubuntu.com> Signed-off-by: Paolo Pisati <paolo.pisati@canonical.com>
Add driver for the SiFive FU740 PCIe host controller.
This controller is based on the DesignWare PCIe core.
Co-developed-by: Henry Styles <hes@sifive.com> Signed-off-by: Henry Styles <hes@sifive.com> Co-developed-by: Erik Danie <erik.danie@sifive.com> Signed-off-by: Erik Danie <erik.danie@sifive.com> Co-developed-by: Greentime Hu <greentime.hu@sifive.com> Signed-off-by: Greentime Hu <greentime.hu@sifive.com> Signed-off-by: Paul Walmsley <paul.walmsley@sifive.com> Signed-off-by: Dimitri John Ledkov <xnox@ubuntu.com> Signed-off-by: Paolo Pisati <paolo.pisati@canonical.com>
Add PCIe host controller DT bindings of SiFive FU740.
Signed-off-by: Greentime Hu <greentime.hu@sifive.com> Signed-off-by: Dimitri John Ledkov <xnox@ubuntu.com> Signed-off-by: Paolo Pisati <paolo.pisati@canonical.com>
Greentime Hu [Tue, 16 Mar 2021 21:31:12 +0000 (21:31 +0000)]
MAINTAINERS: Add maintainers for SiFive FU740 PCIe driver
Here add maintainer information for SiFive FU740 PCIe driver.
Signed-off-by: Greentime Hu <greentime.hu@sifive.com> Signed-off-by: Dimitri John Ledkov <xnox@ubuntu.com> Signed-off-by: Paolo Pisati <paolo.pisati@canonical.com>
Greentime Hu [Tue, 16 Mar 2021 21:31:11 +0000 (21:31 +0000)]
clk: sifive: Use reset-simple in prci driver for PCIe driver
We use reset-simple in this patch so that pcie driver can use
devm_reset_control_get() to get this reset data structure and use
reset_control_deassert() to deassert pcie_power_up_rst_n.
Signed-off-by: Greentime Hu <greentime.hu@sifive.com> Signed-off-by: Dimitri John Ledkov <xnox@ubuntu.com> Signed-off-by: Paolo Pisati <paolo.pisati@canonical.com>
Greentime Hu [Tue, 16 Mar 2021 21:31:10 +0000 (21:31 +0000)]
clk: sifive: Add pcie_aux clock in prci driver for PCIe driver
We add pcie_aux clock in this patch so that pcie driver can use
clk_prepare_enable() and clk_disable_unprepare() to enable and disable
pcie_aux clock.
Signed-off-by: Greentime Hu <greentime.hu@sifive.com> Signed-off-by: Dimitri John Ledkov <xnox@ubuntu.com> Signed-off-by: Paolo Pisati <paolo.pisati@canonical.com>
Vincent Chen [Tue, 16 Mar 2021 21:31:09 +0000 (21:31 +0000)]
riscv: sifive: apply errata "cip-453" patch
Add sign extension to the $badaddr before addressing the instruction page
fault and instruction access fault to workaround the issue "cip-453".
To avoid affecting the existing code sequence, this patch will creates two
trampolines to add sign extension to the $badaddr. By the "alternative"
mechanism, these two trampolines will replace the original exception
handler of instruction page fault and instruction access fault in the
excp_vect_table. In this case, only the specific SiFive CPU core jumps to
the do_page_fault and do_trap_insn_fault through these two trampolines.
Other CPUs are not affected.
Signed-off-by: Vincent Chen <vincent.chen@sifive.com> Signed-off-by: Dimitri John Ledkov <xnox@ubuntu.com> Signed-off-by: Paolo Pisati <paolo.pisati@canonical.com>
Vincent Chen [Tue, 16 Mar 2021 21:31:08 +0000 (21:31 +0000)]
riscv: Introduce alternative mechanism to apply errata solution
Introduce the "alternative" mechanism from ARM64 and x86 to apply the CPU
vendors' errata solution at runtime. The main purpose of this patch is
to provide a framework. Therefore, the implementation is quite basic for
now so that some scenarios could not use this schemei, such as patching
code to a module, relocating the patching code and heterogeneous CPU
topology.
Users could use the macro ALTERNATIVE to apply an errata to the existing
code flow. In the macro ALTERNATIVE, users need to specify the manufacturer
information(vendorid, archid, and impid) for this errata. Therefore, kernel
will know this errata is suitable for which CPU core. During the booting
procedure, kernel will select the errata required by the CPU core and then
patch it. It means that the kernel only applies the errata to the specified
CPU core. In this case the vendor's errata does not affect each other at
runtime. The above patching procedure only occurs during the booting phase,
so we only take the overhead of the "alternative" mechanism once.
This "alternative" mechanism is enabled by default to ensure that all
required errata will be applied. However, users can disable this feature by
the Kconfig "CONFIG_RISCV_ERRATA_ALTERNATIVE".
Signed-off-by: Vincent Chen <vincent.chen@sifive.com> Signed-off-by: Dimitri John Ledkov <xnox@ubuntu.com> Signed-off-by: Paolo Pisati <paolo.pisati@canonical.com>
Vincent Chen [Tue, 16 Mar 2021 21:31:07 +0000 (21:31 +0000)]
riscv: Get CPU manufacturer information
Issue 3 SBI calls to get the vendor ID, architecture ID and implementation
ID early in boot so we only need to take the SBI call overhead once.
Signed-off-by: Vincent Chen <vincent.chen@sifive.com> Signed-off-by: Dimitri John Ledkov <xnox@ubuntu.com> Signed-off-by: Paolo Pisati <paolo.pisati@canonical.com>
Vincent Chen [Tue, 16 Mar 2021 21:31:06 +0000 (21:31 +0000)]
riscv: Add 3 SBI wrapper functions to get cpu manufacturer information
Add 3 wrapper functions to get vendor id, architecture id and implement id
from M-mode
Signed-off-by: Vincent Chen <vincent.chen@sifive.com> Signed-off-by: Dimitri John Ledkov <xnox@ubuntu.com> Signed-off-by: Paolo Pisati <paolo.pisati@canonical.com>
Signed-off-by: David Abdurachmanov <david.abdurachmanov@sifive.com> Signed-off-by: Dimitri John Ledkov <xnox@ubuntu.com> Signed-off-by: Paolo Pisati <paolo.pisati@canonical.com>
riscv: sifive: fu740: cpu{1, 2, 3, 4} set compatible to sifive, u74-mc
Signed-off-by: David Abdurachmanov <david.abdurachmanov@sifive.com> Signed-off-by: Dimitri John Ledkov <xnox@ubuntu.com> Signed-off-by: Paolo Pisati <paolo.pisati@canonical.com>
Signed-off-by: David Abdurachmanov <david.abdurachmanov@sifive.com> Signed-off-by: Dimitri John Ledkov <xnox@ubuntu.com> Signed-off-by: Paolo Pisati <paolo.pisati@canonical.com>
Kai-Heng Feng [Wed, 3 Feb 2021 16:41:00 +0000 (00:41 +0800)]
UBUNTU: SAUCE: efifb: Ensure graphics device for efifb stays at PCI D0
BugLink: https://bugs.launchpad.net/bugs/1914411
We are seeing root ports on some desktop boards support D3cold for
discrete graphics card. So when efifb is in use while graphics device
isn't bound to a driver, PCI and ACPI will put the graphics to D3cold
when runtime suspend kicks in, makes efifb stop working.
So ensure the graphics device won't be runtime suspended, to keep efifb
work all the time.
Kai-Heng Feng [Tue, 16 Mar 2021 13:13:28 +0000 (21:13 +0800)]
UBUNTU: SAUCE: PCI: Serialize TGL e1000e PM ops
BugLink: https://bugs.launchpad.net/bugs/1919321
On TGL systems, PCI_COMMAND may randomly flip to 0 on system resume.
This is devastating to drivers that use pci_set_master(), like NVMe and
xHCI, to enable DMA in their resume routine, as pci_set_master() can
inadvertently disable PCI_COMMAND_IO and PCI_COMMAND_MEMORY, making
resources inaccessible.
The issue is reproducible on all kernel releases, but obviously the
situation is exacerbated by commit 6cecf02e77ab ('Revert "e1000e:
disable s0ix entry and exit flows for ME systems"').
Seems like ME is out to lunch until it's finally out of ULP polling. So
ensure e1000e PM ops are serialized by enforcing device links to
workaround the issue. This is another hacky hackish hack that we can't
upstream :)
Of course this will make suspend and resume a bit slower, but at least
we protect other PCI devices by keeping ME from going full basket case.
UBUNTU: [Packaging] Force the target updateconfigs to consider do_enforce_all
Ignore:yes
The binary build calls the target config-prepare-check-generic to
ensure the config is valid before compiling the kernel. This target
however is using the flag do_enforce_all to control if all the
annotations should be enforced on now.
However, during development the target updateconfigs is used instead
to verify the config options are matching the annotations file, but
this target currently always assumes only annotations explicitly
marked should be enforced.
Update kernelconfig and the maintainer make file to pass the
do_enforce_all when running the target updateconfigs.
Signed-off-by: Marcelo Henrique Cerri <marcelo.cerri@canonical.com> Acked-by: Thadeu Lima de Souza Cascardo <cascardo@canonical.com> Acked-by: Tim Gardner <tim.gardner@canonical.com> Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
Colin Ian King [Tue, 2 Mar 2021 11:22:57 +0000 (11:22 +0000)]
Revert "SiFive Unleashed CPUFreq"
The SiFive Unleashed CPU frequency is known to be unreliabe especially
when using the latest device tree and the default Ubuntu CPU frequency
schedule settings. Revert the CPU Frequency commit for now until we
get a suitable upstream fix that works more reliably with our default
Ubuntu configs.
BugLink: https://bugs.launchpad.net/bugs/1919123
On some platforms, the EC doesn't support the register reading sequence
for sentelic[1], and then make the EC can't respond commands for a while
when probing. It leads to the keyboard non-responsive for around 10
seconds while waking up from s2idle.
A DMI quirk to mark this platform doesn't have aux device could avoid those
commands to be sent. And the system could still using i2c interface to
communicate with the touchpad.
Signed-off-by: Chia-Lin Kao (AceLan) <acelan.kao@canonical.com>
(cherry picked from https://lkml.org/lkml/2021/3/15/126) Signed-off-by: Chia-Lin Kao (AceLan) <acelan.kao@canonical.com> Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
Pavel Skripkin [Fri, 12 Mar 2021 06:56:05 +0000 (14:56 +0800)]
ALSA: usb-audio: fix use after free in usb_audio_disconnect
BugLink: https://bugs.launchpad.net/bugs/1918670
The problem was in wrong "if" placement. chip->quirk_type is freed
in snd_card_free_when_closed(), but inside if statement it's accesed.
Pavel Skripkin [Fri, 12 Mar 2021 06:56:04 +0000 (14:56 +0800)]
ALSA: usb-audio: fix NULL ptr dereference in usb_audio_probe
BugLink: https://bugs.launchpad.net/bugs/1918670
syzbot reported null pointer dereference in usb_audio_probe.
The problem was in case, when quirk == NULL. It's not an
error condition, so quirk must be checked before dereferencing.
Andy Whitcroft [Thu, 11 Mar 2021 12:06:32 +0000 (12:06 +0000)]
UBUNTU: [Packaging] sync dkms-build et al from LRMv4
dkms-build and particularly dkms-build--nvidia-N are consumed in all
derivative linux-restricted-modules packages. Each gets its local
independant copy from its main kernel package, and that is common to all
kernels based on a particular top-level kernel. Sync the latest updated
versions of these back into the kernel.
BugLink: https://bugs.launchpad.net/bugs/1918134 Signed-off-by: Andy Whitcroft <apw@canonical.com> Acked-by: Stefan Bader <stefan.bader@canonical.com> Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
dann frazier [Fri, 12 Mar 2021 18:46:32 +0000 (11:46 -0700)]
UBUNTU: [Packaging] Allow grub-efi-arm* to satisfy recommends on ARM
BugLink: https://bugs.launchpad.net/bugs/1918427
The only package that currently satisfies the bootloader Recommends
relationship on ARM systems is flash-kernel. This ignores EFI-based systems,
which will instead require GRUB. Allow the architecture-appropriate
grub-efi-* package to also satisfy this relationship.
Note: I've ordered the Recommends options such that flash-kernel will
remain the default choice if GRUB is not already installed, except for the
64K page flavor, which is more oriented towards servers that will
almost certainly be using EFI.
Signed-off-by: dann frazier <dann.frazier@canonical.com> Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
This problem was reported on a SVM guest while executing kexec.
Kexec fails to load the new kernel when the PCID feature is enabled.
When kexec starts loading the new kernel, it starts the process by
resetting the vCPU's and then bringing each vCPU online one by one.
The vCPU reset is supposed to reset all the register states before the
vCPUs are brought online. However, the CR4 register is not reset during
this process. If this register is already setup during the last boot,
all the flags can remain intact. The X86_CR4_PCIDE bit can only be
enabled in long mode. So, it must be enabled much later in SMP
initialization. Having the X86_CR4_PCIDE bit set during SMP boot can
cause a boot failures.
Fix the issue by resetting the CR4 register in init_vmcb().
Signed-off-by: Babu Moger <babu.moger@amd.com>
Message-Id: <161471109108.30811.6392805173629704166.stgit@bmoger-ubuntu> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org> Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
The Acer Aspire Switch 10E (SW3-016)'s keyboard-dock uses the same USB-ids
as the Acer One S1003 keyboard-dock. Yet they are not entirely the same:
1. The S1003 keyboard-dock has the same report descriptors as the
S1002 keyboard-dock (which has different USB-ids)
2. The Acer Aspire Switch 10E's keyboard-dock has different
report descriptors from the S1002/S1003 keyboard docks and it
sends 0x00880078 / 0x00880079 usage events when the touchpad is
toggled on/off (which is handled internally).
This means that all Acer kbd-docks handled by the hid-ite.c drivers
report their touchpad being toggled on/off through these custom
usage-codes with the exception of the S1003 dock, which likely is
a bug of that dock.
Add a QUIRK_TOUCHPAD_ON_OFF_REPORT quirk for the Aspire Switch 10E / S1003
usb-id so that the touchpad toggling will get reported to userspace on
the Aspire Switch 10E.
Since the Aspire Switch 10E's kbd-dock has different report-descriptors,
this also requires adding support for fixing those to ite_report_fixup().
Setting the quirk will also cause ite_report_fixup() to hit the
S1002/S1003 descriptors path on the S1003. Since the S1003 kbd-dock
never generates any input-reports for the fixed up part of the
descriptors this does not matter; and if there are versions out there
which do actually send input-reports for the touchpad-toggle then the
fixup should actually help to make things work.
This was tested on both an Acer Aspire Switch 10E and on an Acer One S1003.
Signed-off-by: Hans de Goede <hdegoede@redhat.com> Signed-off-by: Jiri Kosina <jkosina@suse.cz> Signed-off-by: Sasha Levin <sashal@kernel.org> Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
This commit adds mixer quirks for the Pioneer DJM-900NXS2 mixer. This
device has 6 capture channels, 5 of them allow setting the signal
source. This adds controls for these, similar to the DJM-250Mk2.
However, playpack channels are not controllable via software like on the
250Mk2, as they can only be set manually on the mixing console.
Read-only controls showing the currently selected playback channels are
omitted.
This allows for N different devices to use the pioneer mixer quirk for
setting capture/record type and recording level. The impementation has
not changed much with the exception of an additional mask on
private_value to allow storing of a device index:
DEVICE MASK 0xff000000
GROUP_MASK 0x00ff0000
VALUE_MASK 0x0000ffff
This could be improved by changing the arrays of wValues for each
channel to contain named definitions (e.g. SND_DJM_CAP_LINE). It would
improve readability and perhaps would allow using the same array for
multiple channels. The channel number can be specified on the control
next to the wIndex.
Feedback is very much appreciated as I'm not the most proficient C
programmer but am learning as I go.
The PC_DBG_ECO_CNTL register on the Adreno A5xx family gets
programmed to some different values on a per-model basis.
At least, this is what we intend to do here;
Unfortunately, though, this register is being overwritten with a
static magic number, right after applying the GPU-specific
configuration (including the GPU-specific quirks) and that is
effectively nullifying the efforts.
Let's remove the redundant and wrong write to the PC_DBG_ECO_CNTL
register in order to retain the wanted configuration for the
target GPU.
Signed-off-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@somainline.org> Reviewed-by: Jordan Crouse <jcrouse@codeaurora.org> Signed-off-by: Rob Clark <robdclark@chromium.org> Signed-off-by: Sasha Levin <sashal@kernel.org> Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
Exynos needs scatterlist entries aligned to page size because it isn't
capable of transferring data contained in one DATA IN operation to seversal
areas in memory.
Some SoCs require a single scatterlist entry for smaller than page size,
i.e. 4KB. When dispatching commands with more than one scatterlist entry
under 4KB in size the following behavior is observed:
A command to read a block range is dispatched with two scatterlist entries
that are named AAA and BBB. After dispatching, the host builds two PRDT
entries and during transmission, device sends just one DATA IN because
device doesn't care about host DMA. The host then transfers the combined
amount of data from start address of the area named AAA. As a consequence,
the area that follows AAA in memory would be corrupted.
A dummy zero bit is sent preceding the data during a read transfer by the
Microchip 93LC46B eeprom (section 2.7 of[1]). This results in right shift
of data during a read. In order to ignore this bit a quirk can be added to
send an extra zero bit after the read address.
Add a quirk to ignore the zero bit sent before data by adding a zero bit
after the read address.
The UniPro specification states that attribute IDs of the following
parameters are vendor-specific so some SoCs could have no regions at the
defined addresses:
Flush during hibern8 is sufficient on MediaTek platforms, thus enable
UFSHCI_QUIRK_SKIP_MANUAL_WB_FLUSH_CTRL to skip enabling
fWriteBoosterBufferFlush during WriteBooster initialization.
Link: https://lore.kernel.org/r/20201222072928.32328-1-stanley.chu@mediatek.com Reviewed-by: Avri Altman <avri.altman@wdc.com> Signed-off-by: Stanley Chu <stanley.chu@mediatek.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org> Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
increase_address_space() calls get_zeroed_page(gfp) under spin_lock with
disabled interrupts. gfp flags passed to increase_address_space() may allow
sleeping, so it comes to this:
BUG: sleeping function called from invalid context at mm/page_alloc.c:4342
in_atomic(): 1, irqs_disabled(): 1, pid: 21555, name: epdcbbf1qnhbsd8
Calling btrfs_qgroup_reserve_meta_prealloc from
btrfs_delayed_inode_reserve_metadata can result in flushing delalloc
while holding a transaction and delayed node locks. This is deadlock
prone. In the past multiple commits:
* ae5e070eaca9 ("btrfs: qgroup: don't try to wait flushing if we're
already holding a transaction")
* 6f23277a49e6 ("btrfs: qgroup: don't commit transaction when we already
hold the handle")
Tried to solve various aspects of this but this was always a
whack-a-mole game. Unfortunately those 2 fixes don't solve a deadlock
scenario involving btrfs_delayed_node::mutex. Namely, one thread
can call btrfs_dirty_inode as a result of reading a file and modifying
its atime:
To fully address those cases the complete fix is to never issue any
flushing while holding the transaction or the delayed node lock. This
patch achieves it by calling qgroup_reserve_meta directly which will
either succeed without flushing or will fail and return -EDQUOT. In the
latter case that return value is going to be propagated to
btrfs_dirty_inode which will fallback to start a new transaction. That's
fine as the majority of time we expect the inode will have
BTRFS_DELAYED_NODE_INODE_DIRTY flag set which will result in directly
copying the in-memory state.
Fixes: c53e9653605d ("btrfs: qgroup: try to flush qgroup space when we get -EDQUOT") CC: stable@vger.kernel.org # 5.10+ Reviewed-by: Qu Wenruo <wqu@suse.com> Signed-off-by: Nikolay Borisov <nborisov@suse.com> Signed-off-by: David Sterba <dsterba@suse.com> Signed-off-by: Sudip Mukherjee <sudipm.mukherjee@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
Similar to commit 28187dc8ebd9 ("ARM: 9025/1: Kconfig: CPU_BIG_ENDIAN
depends on !LD_IS_LLD"), ld.lld prior to 13.0.0 does not properly
support aarch64 big endian, leading to the following build error when
CONFIG_CPU_BIG_ENDIAN is selected:
ld.lld: error: unknown emulation: aarch64linuxb
This has been resolved in LLVM 13. To avoid errors like this, only allow
CONFIG_CPU_BIG_ENDIAN to be selected if using ld.bfd or ld.lld 13.0.0
and newer.
While we are here, the indentation of this symbol used spaces since its
introduction in commit a872013d6d03 ("arm64: kconfig: allow
CPU_BIG_ENDIAN to be selected"). Change it to tabs to be consistent with
kernel coding style.
The kernel test robot reported multiple linkage problems like this:
hppa64-linux-ld: init/main.o(.init.text+0x56c): cannot reach printk
init/main.o: in function `unknown_bootoption':
(.init.text+0x56c): relocation truncated to fit: R_PARISC_PCREL22F against
symbol `printk' defined in .text.unlikely section in kernel/printk/printk.o
There are two ways to solve it:
a) Enable the -mlong-call compiler option (CONFIG_MLONGCALLS),
b) Add long branch stub support in 64-bit linker.
While b) is the long-term solution, this patch works around the issue by
automatically enabling the CONFIG_MLONGCALLS option when
CONFIG_COMPILE_TEST is set, which indicates that a non-production kernel
(e.g. 0-day kernel) is built.
Signed-off-by: Helge Deller <deller@gmx.de> Reported-by: kernel test robot <lkp@intel.com> Fixes: 00e35f2b0e8a ("parisc: Enable -mlong-calls gcc option by default when !CONFIG_MODULES") Cc: stable@vger.kernel.org # v5.6+ Signed-off-by: Sudip Mukherjee <sudipm.mukherjee@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
My 2TB SKC2000 showed the exact same symptoms that were provided
in 538e4a8c57 ("nvme-pci: avoid the deepest sleep state on
Kingston A2000 SSDs"), i.e. a complete NVME lockup that needed
cold boot to get it back.
According to some sources, the A2000 is simply a rebadged
SKC2000 with a slightly optimized firmware.
Adding the SKC2000 PCI ID to the quirk list with the same workaround
as the A2000 made my laptop survive a 5 hours long Yocto bootstrap
buildfest which reliably triggered the SSD lockup previously.
Signed-off-by: Zoltán Böszörményi <zboszor@gmail.com> Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Sudip Mukherjee <sudipm.mukherjee@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
According to v4l2 request api specifications, it's allowed to skip
control if its content isn't changed for performance reasons. Cedrus
driver predates that, so it has implemented mechanism to check if all
required controls are included in one request.
Conform to specifications with removing that mechanism.
Note that this mechanism with static required flag isn't very good
anyway because need for control is usually signaled in other controls.
[ 97.866748] a.out/2890 is trying to acquire lock:
[ 97.867829] ffff8881046763e8 (&ctx->uring_lock){+.+.}-{3:3}, at:
io_wq_submit_work+0x155/0x240
[ 97.869735]
[ 97.869735] but task is already holding lock:
[ 97.871033] ffff88810dfe0be8 (&ctx->uring_lock){+.+.}-{3:3}, at:
__x64_sys_io_uring_enter+0x3f0/0x5b0
[ 97.873074]
[ 97.873074] other info that might help us debug this:
[ 97.874520] Possible unsafe locking scenario:
[ 97.874520]
[ 97.875845] CPU0
[ 97.876440] ----
[ 97.877048] lock(&ctx->uring_lock);
[ 97.877961] lock(&ctx->uring_lock);
[ 97.878881]
[ 97.878881] *** DEADLOCK ***
[ 97.878881]
[ 97.880341] May be due to missing lock nesting notation
[ 97.880341]
[ 97.881952] 1 lock held by a.out/2890:
[ 97.882873] #0: ffff88810dfe0be8 (&ctx->uring_lock){+.+.}-{3:3}, at:
__x64_sys_io_uring_enter+0x3f0/0x5b0
[ 97.885108]
[ 97.885108] stack backtrace:
[ 97.890457] Call Trace:
[ 97.891121] dump_stack+0xac/0xe3
[ 97.891972] __lock_acquire+0xab6/0x13a0
[ 97.892940] lock_acquire+0x2c3/0x390
[ 97.894894] __mutex_lock+0xae/0x9f0
[ 97.901101] io_wq_submit_work+0x155/0x240
[ 97.902112] io_wq_cancel_cb+0x162/0x490
[ 97.904126] io_async_find_and_cancel+0x3b/0x140
[ 97.905247] io_issue_sqe+0x86d/0x13e0
[ 97.909122] __io_queue_sqe+0x10b/0x550
[ 97.913971] io_queue_sqe+0x235/0x470
[ 97.914894] io_submit_sqes+0xcce/0xf10
[ 97.917872] __x64_sys_io_uring_enter+0x3fb/0x5b0
[ 97.921424] do_syscall_64+0x2d/0x40
[ 97.922329] entry_SYSCALL_64_after_hwframe+0x44/0xa9
While holding uring_lock, e.g. from inline execution, async cancel
request may attempt cancellations through io_wq_submit_work, which may
try to grab a lock. Delay it to task_work, so we do it from a clean
context and don't have to worry about locking.
Cc: <stable@vger.kernel.org> # 5.5+ Fixes: c07e6719511e ("io_uring: hold uring_lock while completing failed polled io in io_wq_submit_work()") Reported-by: Abaci <abaci@linux.alibaba.com> Reported-by: Hao Xu <haoxu@linux.alibaba.com> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
Saving one lock/unlock for io-wq is not super important, but adds some
ugliness in the code. More important, atomic decs not turning it to zero
for some archs won't give the right ordering/barriers so the
io_steal_work() may pretty easily get subtly and completely broken.
Return back 2-step io-wq work exchange and clean it up.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
We currently split the close into two, in case we have a ->flush op
that we can't safely handle from non-blocking context. This requires
us to flag the op as uncancelable if we do need to punt it async, and
that means special handling for just this op type.
Use __close_fd_get_file() and grab the files lock so we can get the file
and check if we need to go async in one atomic operation. That gets rid
of the need for splitting this into two steps, and hence the need for
IO_WQ_WORK_NO_CANCEL.
Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Andrea Righi <andrea.righi@canonical.com>