platform/kernel/linux-rpi.git
3 years agomips: Fix non-POSIX regexp
H. Nikolaus Schaller [Thu, 8 Jul 2021 08:57:10 +0000 (10:57 +0200)]
mips: Fix non-POSIX regexp

[ Upstream commit 28bbbb9875a35975904e46f9b06fa689d051b290 ]

When cross compiling a MIPS kernel on a BSD based HOSTCC leads
to errors like

  SYNC    include/config/auto.conf.cmd - due to: .config
egrep: empty (sub)expression
  UPD     include/config/kernel.release
  HOSTCC  scripts/dtc/dtc.o - due to target missing

It turns out that egrep uses this egrep pattern:

(|MINOR_|PATCHLEVEL_)

This is not valid syntax or gives undefined results according
to POSIX 9.5.3 ERE Grammar

https://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap09.html

It seems to be silently accepted by the Linux egrep implementation
while a BSD host complains.

Such patterns can be replaced by a transformation like

"(|p1|p2)" -> "(p1|p2)?"

Fixes: 48c35b2d245f ("[MIPS] There is no __GNUC_MAJOR__")
Signed-off-by: H. Nikolaus Schaller <hns@goldelico.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoMIPS: check return value of pgtable_pmd_page_ctor
Huang Pei [Wed, 21 Jul 2021 09:30:45 +0000 (17:30 +0800)]
MIPS: check return value of pgtable_pmd_page_ctor

[ Upstream commit 6aa32467299e9e12280a6aec9dbc21bf2db830b0 ]

+. According to Documentation/vm/split_page_table_lock, handle failure
of pgtable_pmd_page_ctor

+. Use GFP_KERNEL_ACCOUNT instead of GFP_KERNEL|__GFP_ACCOUNT

+. Adjust coding style

Fixes: ed914d48b6a1 ("MIPS: add PMD table accounting into MIPS')
Reported-by: Joshua Kinard <kumba@gentoo.org>
Signed-off-by: Huang Pei <huangpei@loongson.cn>
Reviewed-by: Joshua Kinard <kumba@gentoo.org>
Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agonet: sched: fix lockdep_set_class() typo error for sch->seqlock
Yunsheng Lin [Tue, 3 Aug 2021 10:58:21 +0000 (18:58 +0800)]
net: sched: fix lockdep_set_class() typo error for sch->seqlock

[ Upstream commit 06f5553e0f0c2182268179b93856187d9cb86dd5 ]

According to comment in qdisc_alloc(), sch->seqlock's lockdep
class key should be set to qdisc_tx_busylock, due to possible
type error, sch->busylock's lockdep class key is set to
qdisc_tx_busylock, which is duplicated because sch->busylock's
lockdep class key is already set in qdisc_alloc().

So fix it by replacing sch->busylock with sch->seqlock.

Fixes: 96009c7d500e ("sched: replace __QDISC_STATE_RUNNING bit with a spin lock")
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agonet: dsa: qca: ar9331: reorder MDIO write sequence
Oleksij Rempel [Tue, 3 Aug 2021 06:37:46 +0000 (08:37 +0200)]
net: dsa: qca: ar9331: reorder MDIO write sequence

[ Upstream commit d1a58c013a5837451e3213e7a426d350fa524ead ]

In case of this switch we work with 32bit registers on top of 16bit
bus. Some registers (for example access to forwarding database) have
trigger bit on the first 16bit half of request and the result +
configuration of request in the second half. Without this patch, we would
trigger database operation and overwrite result in one run.

To make it work properly, we should do the second part of transfer
before the first one is done.

So far, this rule seems to work for all registers on this switch.

Fixes: ec6698c272de ("net: dsa: add support for Atheros AR9331 built-in switch")
Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Link: https://lore.kernel.org/r/20210803063746.3600-1-o.rempel@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agonet: ipv6: fix returned variable type in ip6_skb_dst_mtu
Antoine Tenart [Tue, 3 Aug 2021 10:00:16 +0000 (12:00 +0200)]
net: ipv6: fix returned variable type in ip6_skb_dst_mtu

[ Upstream commit 4039146777a91e1576da2bf38e0d8a1061a1ae47 ]

The patch fixing the returned value of ip6_skb_dst_mtu (int -> unsigned
int) was rebased between its initial review and the version applied. In
the meantime fade56410c22 was applied, which added a new variable (int)
used as the returned value. This lead to a mismatch between the function
prototype and the variable used as the return value.

Fixes: 40fc3054b458 ("net: ipv6: fix return value of ip6_skb_dst_mtu")
Cc: Vadim Fedorenko <vfedorenko@novek.ru>
Signed-off-by: Antoine Tenart <atenart@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agonfp: update ethtool reporting of pauseframe control
Fei Qin [Tue, 3 Aug 2021 10:39:11 +0000 (12:39 +0200)]
nfp: update ethtool reporting of pauseframe control

[ Upstream commit 9fdc5d85a8fe684cdf24dc31c6bc4a727decfe87 ]

Pauseframe control is set to symmetric mode by default on the NFP.
Pause frames can not be configured through ethtool now, but ethtool can
report the supported mode.

Fixes: 265aeb511bd5 ("nfp: add support for .get_link_ksettings()")
Signed-off-by: Fei Qin <fei.qin@corigine.com>
Signed-off-by: Louis Peens <louis.peens@corigine.com>
Signed-off-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agosctp: move the active_key update after sh_keys is added
Xin Long [Sun, 1 Aug 2021 06:25:31 +0000 (02:25 -0400)]
sctp: move the active_key update after sh_keys is added

[ Upstream commit ae954bbc451d267f7d60d7b49db811d5a68ebd7b ]

In commit 58acd1009226 ("sctp: update active_key for asoc when old key is
being replaced"), sctp_auth_asoc_init_active_key() is called to update
the active_key right after the old key is deleted and before the new key
is added, and it caused that the active_key could be found with the key_id.

In Ying Xu's testing, the BUG_ON in sctp_auth_asoc_init_active_key() was
triggered:

  [ ] kernel BUG at net/sctp/auth.c:416!
  [ ] RIP: 0010:sctp_auth_asoc_init_active_key.part.8+0xe7/0xf0 [sctp]
  [ ] Call Trace:
  [ ]  sctp_auth_set_key+0x16d/0x1b0 [sctp]
  [ ]  sctp_setsockopt.part.33+0x1ba9/0x2bd0 [sctp]
  [ ]  __sys_setsockopt+0xd6/0x1d0
  [ ]  __x64_sys_setsockopt+0x20/0x30
  [ ]  do_syscall_64+0x5b/0x1a0

So fix it by moving the active_key update after sh_keys is added.

Fixes: 58acd1009226 ("sctp: update active_key for asoc when old key is being replaced")
Reported-by: Ying Xu <yinxu@redhat.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoRDMA/mlx5: Delay emptying a cache entry when a new MR is added to it recently
Aharon Landau [Tue, 27 Jul 2021 07:16:06 +0000 (10:16 +0300)]
RDMA/mlx5: Delay emptying a cache entry when a new MR is added to it recently

[ Upstream commit d6793ca97b76642b77629dd0783ec64782a50bdb ]

Fixing a typo that causes a cache entry to shrink immediately after adding
to it new MRs if the entry size exceeds the high limit.  In doing so, the
cache misses its purpose to prevent the creation of new mkeys on the
runtime by using the cached ones.

Fixes: b9358bdbc713 ("RDMA/mlx5: Fix locking in MR cache work queue")
Link: https://lore.kernel.org/r/fcb546986be346684a016f5ca23a0567399145fa.1627370131.git.leonro@nvidia.com
Signed-off-by: Aharon Landau <aharonl@nvidia.com>
Reviewed-by: Maor Gottlieb <maorg@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agogpio: tqmx86: really make IRQ optional
Matthias Schiffer [Fri, 16 Jul 2021 10:00:47 +0000 (12:00 +0200)]
gpio: tqmx86: really make IRQ optional

[ Upstream commit 9b87f43537acfa24b95c236beba0f45901356eb2 ]

The tqmx86 MFD driver was passing IRQ 0 for "no IRQ" in the past. This
causes warnings with newer kernels.

Prepare the gpio-tqmx86 driver for the fixed MFD driver by handling a
missing IRQ properly.

Fixes: b868db94a6a7 ("gpio: tqmx86: Add GPIO from for this IO controller")
Signed-off-by: Matthias Schiffer <matthias.schiffer@ew.tq-group.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Bartosz Golaszewski <bgolaszewski@baylibre.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agonet: natsemi: Fix missing pci_disable_device() in probe and remove
Wang Hai [Sat, 31 Jul 2021 06:38:01 +0000 (14:38 +0800)]
net: natsemi: Fix missing pci_disable_device() in probe and remove

[ Upstream commit 7fe74dfd41c428afb24e2e615470832fa997ff14 ]

Replace pci_enable_device() with pcim_enable_device(),
pci_disable_device() and pci_release_regions() will be
called in release automatically.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Wang Hai <wanghai38@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agonet: phy: micrel: Fix detection of ksz87xx switch
Steve Bennett [Fri, 30 Jul 2021 22:57:50 +0000 (08:57 +1000)]
net: phy: micrel: Fix detection of ksz87xx switch

[ Upstream commit a5e63c7d38d548b8dab6c6205e0b6af76899dbf5 ]

The logic for discerning between KSZ8051 and KSZ87XX PHYs is incorrect
such that the that KSZ87XX switch is not identified correctly.

ksz8051_ksz8795_match_phy_device() uses the parameter ksz_phy_id
to discriminate whether it was called from ksz8051_match_phy_device()
or from ksz8795_match_phy_device() but since PHY_ID_KSZ87XX is the
same value as PHY_ID_KSZ8051, this doesn't work.

Instead use a bool to discriminate the caller.

Without this patch, the KSZ8795 switch port identifies as:

ksz8795-switch spi3.1 ade1 (uninitialized): PHY [dsa-0.1:03] driver [Generic PHY]

With the patch, it identifies correctly:

ksz8795-switch spi3.1 ade1 (uninitialized): PHY [dsa-0.1:03] driver [Micrel KSZ87XX Switch]

Fixes: 8b95599c55ed24b36cf4 ("net: phy: micrel: Discern KSZ8051 and KSZ8795 PHYs")
Signed-off-by: Steve Bennett <steveb@workware.net.au>
Reviewed-by: Marek Vasut <marex@denx.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agonet: dsa: sja1105: match FDB entries regardless of inner/outer VLAN tag
Vladimir Oltean [Fri, 30 Jul 2021 17:18:15 +0000 (20:18 +0300)]
net: dsa: sja1105: match FDB entries regardless of inner/outer VLAN tag

[ Upstream commit 47c2c0c2312118a478f738503781de1d1a6020d2 ]

On SJA1105P/Q/R/S and SJA1110, the L2 Lookup Table entries contain a
maskable "inner/outer tag" bit which means:
- when set to 1: match single-outer and double tagged frames
- when set to 0: match untagged and single-inner tagged frames
- when masked off: match all frames regardless of the type of tag

This driver does not make any meaningful distinction between inner tags
(matches on TPID) and outer tags (matches on TPID2). In fact, all VLAN
table entries are installed as SJA1110_VLAN_D_TAG, which means that they
match on both inner and outer tags.

So it does not make sense that we install FDB entries with the IOTAG bit
set to 1.

In VLAN-unaware mode, we set both TPID and TPID2 to 0xdadb, so the
switch will see frames as outer-tagged or double-tagged (never inner).
So the FDB entries will match if IOTAG is set to 1.

In VLAN-aware mode, we set TPID to 0x8100 and TPID2 to 0x88a8. So the
switch will see untagged and 802.1Q-tagged packets as inner-tagged, and
802.1ad-tagged packets as outer-tagged. So untagged and 802.1Q-tagged
packets will not match FDB entries if IOTAG is set to 1, but 802.1ad
tagged packets will. Strange.

To fix this, simply mask off the IOTAG bit from FDB entries, and make
them match regardless of whether the VLAN tag is inner or outer.

Fixes: 1da73821343c ("net: dsa: sja1105: Add FDB operations for P/Q/R/S series")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agonet: dsa: sja1105: be stateless with FDB entries on SJA1105P/Q/R/S/SJA1110 too
Vladimir Oltean [Fri, 30 Jul 2021 17:18:14 +0000 (20:18 +0300)]
net: dsa: sja1105: be stateless with FDB entries on SJA1105P/Q/R/S/SJA1110 too

[ Upstream commit 589918df93226a1e5f104306c185b6dcf2bd8051 ]

Similar but not quite the same with what was done in commit b11f0a4c0c81
("net: dsa: sja1105: be stateless when installing FDB entries") for
SJA1105E/T, it is desirable to drop the priv->vlan_aware check and
simply go ahead and install FDB entries in the VLAN that was given by
the bridge.

As opposed to SJA1105E/T, in SJA1105P/Q/R/S and SJA1110, the FDB is a
maskable TCAM, and we are installing VLAN-unaware FDB entries with the
VLAN ID masked off. However, such FDB entries might completely obscure
VLAN-aware entries where the VLAN ID is included in the search mask,
because the switch looks up the FDB from left to right and picks the
first entry which results in a masked match. So it depends on whether
the bridge installs first the VLAN-unaware or the VLAN-aware FDB entries.

Anyway, if we had a VLAN-unaware FDB entry towards one set of DESTPORTS
and a VLAN-aware one towards other set of DESTPORTS, the result is that
the packets in VLAN-aware mode will be forwarded towards the DESTPORTS
specified by the VLAN-unaware entry.

To solve this, simply do not use the masked matching ability of the FDB
for VLAN ID, and always match precisely on it. In VLAN-unaware mode, we
configure the switch for shared VLAN learning, so the VLAN ID will be
ignored anyway during lookup, so it is redundant to mask it off in the
TCAM.

This patch conflicts with net-next commit 0fac6aa098ed ("net: dsa: sja1105:
delete the best_effort_vlan_filtering mode") which changed this line:
if (priv->vlan_state != SJA1105_VLAN_UNAWARE) {
into:
if (priv->vlan_aware) {

When merging with net-next, the lines added by this patch should take
precedence in the conflict resolution (i.e. the "if" condition should be
deleted in both cases).

Fixes: 1da73821343c ("net: dsa: sja1105: Add FDB operations for P/Q/R/S series")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agonet: dsa: sja1105: invalidate dynamic FDB entries learned concurrently with staticall...
Vladimir Oltean [Fri, 30 Jul 2021 17:18:12 +0000 (20:18 +0300)]
net: dsa: sja1105: invalidate dynamic FDB entries learned concurrently with statically added ones

[ Upstream commit 6c5fc159e0927531707895709eee1f8bfa04289f ]

The procedure to add a static FDB entry in sja1105 is concurrent with
dynamic learning performed on all bridge ports and the CPU port.

The switch looks up the FDB from left to right, and also learns
dynamically from left to right, so it is possible that between the
moment when we pick up a free slot to install an FDB entry, another slot
to the left of that one becomes free due to an address ageing out, and
that other slot is then immediately used by the switch to learn
dynamically the same address as we're trying to add statically.

The result is that we succeeded to add our static FDB entry, but it is
being shadowed by a dynamic FDB entry to its left, and the switch will
behave as if our static FDB entry did not exist.

We cannot really prevent this from happening unless we make the entire
process to add a static FDB entry a huge critical section where address
learning is temporarily disabled on _all_ ports, and then re-enabled
according to the configuration done by sja1105_port_set_learning.
However, that is kind of disruptive for the operation of the network.

What we can do alternatively is to simply read back the FDB for dynamic
entries located before our newly added static one, and delete them.
This will guarantee that our static FDB entry is now operational. It
will still not guarantee that there aren't dynamic FDB entries to the
_right_ of that static FDB entry, but at least those entries will age
out by themselves since they aren't hit, and won't bother anyone.

Fixes: 291d1e72b756 ("net: dsa: sja1105: Add support for FDB and MDB management")
Fixes: 1da73821343c ("net: dsa: sja1105: Add FDB operations for P/Q/R/S series")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agonet: dsa: sja1105: overwrite dynamic FDB entries with static ones in .port_fdb_add
Vladimir Oltean [Fri, 30 Jul 2021 17:18:11 +0000 (20:18 +0300)]
net: dsa: sja1105: overwrite dynamic FDB entries with static ones in .port_fdb_add

[ Upstream commit e11e865bf84e3c6ea91563ff3e858cfe0e184bd2 ]

The SJA1105 switch family leaves it up to software to decide where
within the FDB to install a static entry, and to concatenate destination
ports for already existing entries (the FDB is also used for multicast
entries), it is not as simple as just saying "please add this entry".

This means we first need to search for an existing FDB entry before
adding a new one. The driver currently manages to fool itself into
thinking that if an FDB entry already exists, there is nothing to be
done. But that FDB entry might be dynamically learned, case in which it
should be replaced with a static entry, but instead it is left alone.

This patch checks the LOCKEDS ("locked/static") bit from found FDB
entries, and lets the code "goto skip_finding_an_index;" if the FDB
entry was not static. So we also need to move the place where we set
LOCKEDS = true, to cover the new case where a dynamic FDB entry existed
but was dynamic.

Fixes: 291d1e72b756 ("net: dsa: sja1105: Add support for FDB and MDB management")
Fixes: 1da73821343c ("net: dsa: sja1105: Add FDB operations for P/Q/R/S series")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agonet, gro: Set inner transport header offset in tcp/udp GRO hook
Jakub Sitnicki [Thu, 29 Jul 2021 13:48:20 +0000 (15:48 +0200)]
net, gro: Set inner transport header offset in tcp/udp GRO hook

[ Upstream commit d51c5907e9809a803b276883d203f45849abd4d6 ]

GSO expects inner transport header offset to be valid when
skb->encapsulation flag is set. GSO uses this value to calculate the length
of an individual segment of a GSO packet in skb_gso_transport_seglen().

However, tcp/udp gro_complete callbacks don't update the
skb->inner_transport_header when processing an encapsulated TCP/UDP
segment. As a result a GRO skb has ->inner_transport_header set to a value
carried over from earlier skb processing.

This can have mild to tragic consequences. From miscalculating the GSO
segment length to triggering a page fault [1], when trying to read TCP/UDP
header at an address past the skb->data page.

The latter scenario leads to an oops report like so:

  BUG: unable to handle page fault for address: ffff9fa7ec00d008
  #PF: supervisor read access in kernel mode
  #PF: error_code(0x0000) - not-present page
  PGD 123f201067 P4D 123f201067 PUD 123f209067 PMD 0
  Oops: 0000 [#1] SMP NOPTI
  CPU: 44 PID: 0 Comm: swapper/44 Not tainted 5.4.53-cloudflare-2020.7.21 #1
  Hardware name: HYVE EDGE-METAL-GEN10/HS-1811DLite1, BIOS V2.15 02/21/2020
  RIP: 0010:skb_gso_transport_seglen+0x44/0xa0
  Code: c0 41 83 e0 11 f6 87 81 00 00 00 20 74 30 0f b7 87 aa 00 00 00 0f [...]
  RSP: 0018:ffffad8640bacbb8 EFLAGS: 00010202
  RAX: 000000000000feda RBX: ffff9fcc8d31bc00 RCX: ffff9fa7ec00cffc
  RDX: ffff9fa7ebffdec0 RSI: 000000000000feda RDI: 0000000000000122
  RBP: 00000000000005c4 R08: 0000000000000001 R09: 0000000000000000
  R10: ffff9fe588ae3800 R11: ffff9fe011fc92f0 R12: ffff9fcc8d31bc00
  R13: ffff9fe0119d4300 R14: 00000000000005c4 R15: ffff9fba57d70900
  FS:  0000000000000000(0000) GS:ffff9fe68df00000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: ffff9fa7ec00d008 CR3: 0000003e99b1c000 CR4: 0000000000340ee0
  Call Trace:
   <IRQ>
   skb_gso_validate_network_len+0x11/0x70
   __ip_finish_output+0x109/0x1c0
   ip_sublist_rcv_finish+0x57/0x70
   ip_sublist_rcv+0x2aa/0x2d0
   ? ip_rcv_finish_core.constprop.0+0x390/0x390
   ip_list_rcv+0x12b/0x14f
   __netif_receive_skb_list_core+0x2a9/0x2d0
   netif_receive_skb_list_internal+0x1b5/0x2e0
   napi_complete_done+0x93/0x140
   veth_poll+0xc0/0x19f [veth]
   ? mlx5e_napi_poll+0x221/0x610 [mlx5_core]
   net_rx_action+0x1f8/0x790
   __do_softirq+0xe1/0x2bf
   irq_exit+0x8e/0xc0
   do_IRQ+0x58/0xe0
   common_interrupt+0xf/0xf
   </IRQ>

The bug can be observed in a simple setup where we send IP/GRE/IP/TCP
packets into a netns over a veth pair. Inside the netns, packets are
forwarded to dummy device:

  trafgen -> [veth A]--[veth B] -forward-> [dummy]

For veth B to GRO aggregate packets on receive, it needs to have an XDP
program attached (for example, a trivial XDP_PASS). Additionally, for UDP,
we need to enable GSO_UDP_L4 feature on the device:

  ip netns exec A ethtool -K AB rx-udp-gro-forwarding on

The last component is an artificial delay to increase the chances of GRO
batching happening:

  ip netns exec A tc qdisc add dev AB root \
     netem delay 200us slot 5ms 10ms packets 2 bytes 64k

With such a setup in place, the bug can be observed by tracing the skb
outer and inner offsets when GSO skb is transmitted from the dummy device:

tcp:

FUNC              DEV   SKB_LEN  NH  TH ENC INH ITH GSO_SIZE GSO_TYPE
ip_finish_output  dumB     2830 270 290   1 294 254     1383 (tcpv4,gre,)
                                                ^^^
udp:

FUNC              DEV   SKB_LEN  NH  TH ENC INH ITH GSO_SIZE GSO_TYPE
ip_finish_output  dumB     2818 270 290   1 294 254     1383 (gre,udp_l4,)
                                                ^^^

Fix it by updating the inner transport header offset in tcp/udp
gro_complete callbacks, similar to how {inet,ipv6}_gro_complete callbacks
update the inner network header offset, when skb->encapsulation flag is
set.

[1] https://lore.kernel.org/netdev/CAKxSbF01cLpZem2GFaUaifh0S-5WYViZemTicAg7FCHOnh6kug@mail.gmail.com/

Fixes: bf296b125b21 ("tcp: Add GRO support")
Fixes: f993bc25e519 ("net: core: handle encapsulation offloads when computing segment lengths")
Fixes: e20cf8d3f1f7 ("udp: implement GRO for plain UDP sockets.")
Reported-by: Alex Forster <aforster@cloudflare.com>
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agodmaengine: imx-dma: configure the generic DMA type to make it work
Juergen Borleis [Thu, 29 Jul 2021 07:18:21 +0000 (09:18 +0200)]
dmaengine: imx-dma: configure the generic DMA type to make it work

[ Upstream commit 7199ddede9f0f2f68d41e6928e1c6c4bca9c39c0 ]

Commit dea7a9fbb009 ("dmaengine: imx-dma: remove dma_slave_config
direction usage") changes the method from a "configuration when called"
to an "configuration when used". Due to this, only the cyclic DMA type
gets configured correctly, while the generic DMA type is left
non-configured.

Without this additional call, the struct imxdma_channel::word_size member
is stuck at DMA_SLAVE_BUSWIDTH_UNDEFINED and imxdma_prep_slave_sg() always
returns NULL.

Signed-off-by: Juergen Borleis <jbe@pengutronix.de>
Fixes: dea7a9fbb009 ("dmaengine: imx-dma: remove dma_slave_config direction usage")
Link: https://lore.kernel.org/r/20210729071821.9857-1-jbe@pengutronix.de
Signed-off-by: Vinod Koul <vkoul@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoARM: dts: stm32: Fix touchscreen IRQ line assignment on DHCOM
Marek Vasut [Wed, 21 Jul 2021 18:10:40 +0000 (20:10 +0200)]
ARM: dts: stm32: Fix touchscreen IRQ line assignment on DHCOM

[ Upstream commit 15f68f027ebd961b99a1c420f96ff3838c5e4450 ]

While 7e5f3155dcbb4 ("ARM: dts: stm32: Fix LED5 on STM32MP1 DHCOM PDK2")
fixed the LED0 assignment on the PDK2 board, the same commit did not
update the touchscreen IRQ line assignment, which is the same GPIO line,
shared between the LED0 output and touchscreen IRQ input. To make this
more convoluted, the same EXTI input (not the same GPIO line) is shared
between Button B which is Active-Low IRQ, and touchscreen IRQ which is
Edge-Falling IRQ, which cannot be used at the same time. In case the LCD
board with touchscreen is in use, which is the case here, LED0 must be
disabled, Button B must be polled, so the touchscreen interrupt works as
it should.

Update the touchscreen IRQ line assignment, disable LED0 and use polled
GPIO button driver for Button B, since the DT here describes baseboard
with LCD board.

Fixes: 7e5f3155dcbb4 ("ARM: dts: stm32: Fix LED5 on STM32MP1 DHCOM PDK2")
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Alexandre Torgue <alexandre.torgue@foss.st.com>
Cc: Patrice Chotard <patrice.chotard@foss.st.com>
Cc: Patrick Delaunay <patrick.delaunay@foss.st.com>
Cc: linux-stm32@st-md-mailman.stormreply.com
To: linux-arm-kernel@lists.infradead.org
Signed-off-by: Alexandre Torgue <alexandre.torgue@foss.st.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoARM: dts: stm32: Disable LAN8710 EDPD on DHCOM
Marek Vasut [Wed, 21 Jul 2021 18:12:53 +0000 (20:12 +0200)]
ARM: dts: stm32: Disable LAN8710 EDPD on DHCOM

[ Upstream commit 36862c1ebc92a7e6fcc55002965c44b8ad17d4ca ]

The LAN8710 Energy Detect Power Down (EDPD) functionality might cause
unreliable cable detection. There are multiple accounts of this in the
SMSC PHY driver patches which attempted to make EDPD reliable, however
it seems there is always some sort of corner case left. Unfortunatelly,
there is no errata documented which would confirm this to be a silicon
bug on the LAN87xx series of PHYs (LAN8700, LAN8710, LAN8720 at least).

Disable EDPD on the DHCOM SoM, just like multiple other boards already
do as well, to make the cable detection reliable.

Fixes: 34e0c7847dcf ("ARM: dts: stm32: Add DH Electronics DHCOM STM32MP1 SoM and PDK2 board")
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Alexandre Torgue <alexandre.torgue@foss.st.com>
Cc: Patrice Chotard <patrice.chotard@foss.st.com>
Cc: Patrick Delaunay <patrick.delaunay@foss.st.com>
Cc: linux-stm32@st-md-mailman.stormreply.com
To: linux-arm-kernel@lists.infradead.org
Signed-off-by: Alexandre Torgue <alexandre.torgue@foss.st.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agomedia: videobuf2-core: dequeue if start_streaming fails
Hans Verkuil [Wed, 30 Jun 2021 07:58:23 +0000 (09:58 +0200)]
media: videobuf2-core: dequeue if start_streaming fails

[ Upstream commit c592b46907adbeb81243f7eb7a468c36692658b8 ]

If a vb2_queue sets q->min_buffers_needed then when the number of
queued buffers reaches q->min_buffers_needed, vb2_core_qbuf() will call
the start_streaming() callback. If start_streaming() returns an error,
then that error was just returned by vb2_core_qbuf(), but the buffer
was still queued. However, userspace expects that if VIDIOC_QBUF fails,
the buffer is returned dequeued.

So if start_streaming() fails, then remove the buffer from the queue,
thus avoiding this unwanted side-effect.

Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Tested-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Fixes: b3379c6201bb ("[media] vb2: only call start_streaming if sufficient buffers are queued")
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoscsi: sr: Return correct event when media event code is 3
Li Manyi [Mon, 26 Jul 2021 11:49:13 +0000 (19:49 +0800)]
scsi: sr: Return correct event when media event code is 3

[ Upstream commit 5c04243a56a7977185b00400e59ca7e108004faf ]

Media event code 3 is defined in the MMC-6 spec as follows:

  "MediaRemoval: The media has been removed from the specified slot, and
   the Drive is unable to access the media without user intervention. This
   applies to media changers only."

This indicated that treating the condition as an EJECT_REQUEST was
appropriate. However, doing so had the unfortunate side-effect of causing
the drive tray to be physically ejected on resume. Instead treat the event
as a MEDIA_CHANGE request.

Fixes: 7dd753ca59d6 ("scsi: sr: Return appropriate error code when disk is ejected")
Link: https://bugzilla.kernel.org/show_bug.cgi?id=213759
Link: https://lore.kernel.org/r/20210726114913.6760-1-limanyi@uniontech.com
Signed-off-by: Li Manyi <limanyi@uniontech.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agospi: imx: mx51-ecspi: Fix low-speed CONFIGREG delay calculation
Marek Vasut [Mon, 26 Jul 2021 10:01:02 +0000 (12:01 +0200)]
spi: imx: mx51-ecspi: Fix low-speed CONFIGREG delay calculation

[ Upstream commit 53ca18acbe645656132fb5a329833db711067e54 ]

The spi_imx->spi_bus_clk may be uninitialized and thus also zero in
mx51_ecspi_prepare_message(), which would lead to division by zero
in kernel. Since bitbang .setup_transfer callback which initializes
the spi_imx->spi_bus_clk is called after bitbang prepare_message
callback, iterate over all the transfers in spi_message, find the
one with lowest bus frequency, and use that bus frequency for the
delay calculation.

Note that it is not possible to move this CONFIGREG delay back into
the .setup_transfer callback, because that is invoked too late, after
the GPIO chipselects were already configured.

Fixes: 135cbd378eab ("spi: imx: mx51-ecspi: Reinstate low-speed CONFIGREG delay")
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Cc: Mark Brown <broonie@kernel.org>
Link: https://lore.kernel.org/r/20210726100102.5188-1-marex@denx.de
Signed-off-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agospi: imx: mx51-ecspi: Reinstate low-speed CONFIGREG delay
Marek Vasut [Sat, 3 Jul 2021 02:23:00 +0000 (04:23 +0200)]
spi: imx: mx51-ecspi: Reinstate low-speed CONFIGREG delay

[ Upstream commit 135cbd378eab336da15de9c84bbb22bf743b38a5 ]

Since 00b80ac935539 ("spi: imx: mx51-ecspi: Move some initialisation to
prepare_message hook."), the MX51_ECSPI_CONFIG write no longer happens
in prepare_transfer hook, but rather in prepare_message hook, however
the MX51_ECSPI_CONFIG delay is still left in prepare_transfer hook and
thus has no effect. This leads to low bus frequency operation problems
described in 6fd8b8503a0dc ("spi: spi-imx: Fix out-of-order CS/SCLK
operation at low speeds") again.

Move the MX51_ECSPI_CONFIG write delay into the prepare_message hook
as well, thus reinstating the low bus frequency fix.

Fixes: 00b80ac935539 ("spi: imx: mx51-ecspi: Move some initialisation to prepare_message hook.")
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Cc: Mark Brown <broonie@kernel.org>
Link: https://lore.kernel.org/r/20210703022300.296114-1-marex@denx.de
Signed-off-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agodmaengine: stm32-dmamux: Fix PM usage counter unbalance in stm32 dmamux ops
Zhang Qilong [Mon, 7 Jun 2021 06:46:39 +0000 (14:46 +0800)]
dmaengine: stm32-dmamux: Fix PM usage counter unbalance in stm32 dmamux ops

[ Upstream commit baa16371c9525f24d508508e4d296c031e1de29c ]

pm_runtime_get_sync will increment pm usage counter
even it failed. Forgetting to putting operation will
result in reference leak here. We fix it by replacing
it with pm_runtime_resume_and_get to keep usage counter
balanced.

Fixes: 4f3ceca254e0f ("dmaengine: stm32-dmamux: Add PM Runtime support")
Signed-off-by: Zhang Qilong <zhangqilong3@huawei.com>
Link: https://lore.kernel.org/r/20210607064640.121394-3-zhangqilong3@huawei.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agodmaengine: stm32-dma: Fix PM usage counter imbalance in stm32 dma ops
Zhang Qilong [Mon, 7 Jun 2021 06:46:38 +0000 (14:46 +0800)]
dmaengine: stm32-dma: Fix PM usage counter imbalance in stm32 dma ops

[ Upstream commit d54db74ad6e0dea8c253fb68c689b836657ab914 ]

pm_runtime_get_sync will increment pm usage counter
even it failed. Forgetting to putting operation will
result in reference leak here. We fix it by replacing
it with pm_runtime_resume_and_get to keep usage counter
balanced.

Fixes: 48bc73ba14bcd ("dmaengine: stm32-dma: Add PM Runtime support")
Fixes: 05f8740a0e6fc ("dmaengine: stm32-dma: add suspend/resume power management support")
Signed-off-by: Zhang Qilong <zhangqilong3@huawei.com>
Link: https://lore.kernel.org/r/20210607064640.121394-2-zhangqilong3@huawei.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoclk: tegra: Implement disable_unused() of tegra_clk_sdmmc_mux_ops
Dmitry Osipenko [Sat, 17 Jul 2021 11:27:42 +0000 (14:27 +0300)]
clk: tegra: Implement disable_unused() of tegra_clk_sdmmc_mux_ops

[ Upstream commit 2bcc025ab9bbd029b1730cde71cb4e4f0ed35d0f ]

Implement disable_unused() callback of tegra_clk_sdmmc_mux_ops to fix
imbalanced disabling of the unused MMC clock on Tegra210 Jetson Nano.

Fixes: c592c8a28f58 ("clk: tegra: Fix refcounting of gate clocks")
Reported-by: Jon Hunter <jonathanh@nvidia.com> # T210 Nano
Tested-by: Jon Hunter <jonathanh@nvidia.com> # T210 Nano
Acked-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Link: https://lore.kernel.org/r/20210717112742.7196-1-digetx@gmail.com
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agodmaengine: uniphier-xdmac: Use readl_poll_timeout_atomic() in atomic state
Kunihiko Hayashi [Tue, 27 Jul 2021 05:47:32 +0000 (14:47 +0900)]
dmaengine: uniphier-xdmac: Use readl_poll_timeout_atomic() in atomic state

[ Upstream commit 55f24c27b6c1a840b62fe297616f1f9ea3576cb7 ]

The function uniphier_xdmac_chan_stop() is only called in atomic state.
Should use readl_poll_timeout_atomic() there instead of
readl_poll_timeout().

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Fixes: 667b9251440b ("dmaengine: uniphier-xdmac: Add UniPhier external DMA controller driver")
Signed-off-by: Kunihiko Hayashi <hayashi.kunihiko@socionext.com>
Link: https://lore.kernel.org/r/1627364852-28432-1-git-send-email-hayashi.kunihiko@socionext.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoomap5-board-common: remove not physically existing vdds_1v8_main fixed-regulator
H. Nikolaus Schaller [Thu, 1 Jul 2021 14:00:22 +0000 (16:00 +0200)]
omap5-board-common: remove not physically existing vdds_1v8_main fixed-regulator

[ Upstream commit c68ef4ad180e09805fa46965d15e1dfadf09ffa5 ]

This device tree include file describes a fixed-regulator
connecting smps7_reg output (1.8V) to some 1.8V rail and
consumers (vdds_1v8_main).

This regulator does not physically exist.

I assume it was introduced as a wrapper around smps7_reg
to provide a speaking signal name "vdds_1v8_main" as label.

This fixed-regulator without real function was not an issue
in driver code until

  Commit 98e48cd9283d ("regulator: core: resolve supply for boot-on/always-on regulators")

introduced a new check for regulator initialization which
makes Palmas regulator registration fail:

[    5.407712] ldo1: supplied by vsys_cobra
[    5.412748] ldo2: supplied by vsys_cobra
[    5.417603] palmas-pmic 48070000.i2c:palmas@48:palmas_pmic: failed to register 48070000.i2c:palmas@48:palmas_pmic regulator

The reason is that the supply-chain of regulators is too
long and goes from ldo3 through the virtual vdds_1v8_main
regulator and then back to smps7. This adds a cross-dependency
of probing Palmas regulators and the fixed-regulator which
leads to probe deferral by the new check and is no longer
resolved.

Since we do not control what device tree files including this
one reference (either &vdds_1v8_main or &smps7_reg or both)
we keep both labels for smps7 for compatibility.

Fixes: 98e48cd9283d ("regulator: core: resolve supply for boot-on/always-on regulators")
Signed-off-by: H. Nikolaus Schaller <hns@goldelico.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoARM: dts: am437x-l4: fix typo in can@0 node
Dario Binacchi [Mon, 26 Jul 2021 13:15:25 +0000 (15:15 +0200)]
ARM: dts: am437x-l4: fix typo in can@0 node

[ Upstream commit 0162a9964365fd26e34575e121b17d021204c481 ]

Replace clock-name with clock-names.

Fixes: 2a4117df9b43 ("ARM: dts: Fix dcan driver probe failed on am437x platform")
Signed-off-by: Dario Binacchi <dariobin@libero.it>
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoclk: stm32f4: fix post divisor setup for I2S/SAI PLLs
Dario Binacchi [Sun, 25 Jul 2021 16:07:25 +0000 (18:07 +0200)]
clk: stm32f4: fix post divisor setup for I2S/SAI PLLs

[ Upstream commit 24b5b1978cd5a80db58e2a19db2f9c36fe8d4f7a ]

Enabling the framebuffer leads to a system hang. Running, as a debug
hack, the store_pan() function in drivers/video/fbdev/core/fbsysfs.c
without taking the console_lock, allows to see the crash backtrace on
the serial line.

~ # echo 0 0 > /sys/class/graphics/fb0/pan

[    9.719414] Unhandled exception: IPSR = 00000005 LR = fffffff1
[    9.726937] CPU: 0 PID: 49 Comm: sh Not tainted 5.13.0-rc5 #9
[    9.733008] Hardware name: STM32 (Device Tree Support)
[    9.738296] PC is at clk_gate_is_enabled+0x0/0x28
[    9.743426] LR is at stm32f4_pll_div_set_rate+0xf/0x38
[    9.748857] pc : [<0011e4be>]    lr : [<0011f9e3>]    psr: 0100000b
[    9.755373] sp : 00bc7be0  ip : 00000000  fp : 001f3ac4
[    9.760812] r10: 002610d0  r9 : 01efe920  r8 : 00540560
[    9.766269] r7 : 02e7ddb0  r6 : 0173eed8  r5 : 00000000  r4 : 004027c0
[    9.773081] r3 : 0011e4bf  r2 : 02e7ddb0  r1 : 0173eed8  r0 : 1d3267b8
[    9.779911] xPSR: 0100000b
[    9.782719] CPU: 0 PID: 49 Comm: sh Not tainted 5.13.0-rc5 #9
[    9.788791] Hardware name: STM32 (Device Tree Support)
[    9.794120] [<0000afa1>] (unwind_backtrace) from [<0000a33f>] (show_stack+0xb/0xc)
[    9.802421] [<0000a33f>] (show_stack) from [<0000a8df>] (__invalid_entry+0x4b/0x4c)

The `pll_num' field in the post_div_data configuration contained a wrong
value which also referenced an uninitialized hardware clock when
clk_register_pll_div() was called.

Fixes: 517633ef630e ("clk: stm32f4: Add post divisor for I2S & SAI PLLs")
Signed-off-by: Dario Binacchi <dariobin@libero.it>
Reviewed-by: Gabriel Fernandez <gabriel.fernandez@st.com>
Link: https://lore.kernel.org/r/20210725160725.10788-1-dariobin@libero.it
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoALSA: usb-audio: fix incorrect clock source setting
chihhao.chen [Sat, 24 Jul 2021 04:23:41 +0000 (12:23 +0800)]
ALSA: usb-audio: fix incorrect clock source setting

[ Upstream commit 4511781f95da0a3b2bad34f3f5e3967e80cd2d18 ]

The following scenario describes an echo test for
Samsung USBC Headset (AKG) with VID/PID (0x04e8/0xa051).

We first start a capture stream(USB IN transfer) in 96Khz/24bit/1ch mode.
In clock find source function, we get value 0x2 for clock selector
and 0x1 for clock source.

Kernel-4.14 behavior
Since clock source is valid so clock selector was not set again.
We pass through this function and start a playback stream(USB OUT transfer)
in 48Khz/32bit/2ch mode. This time we get value 0x1 for clock selector
and 0x1 for clock source. Finally clock id with this setting is 0x9.

Kernel-5.10 behavior
Clock selector was always set one more time even it is valid.
When we start a playback stream, we will get 0x2 for clock selector
and 0x1 for clock source. In this case clock id becomes 0xA.
This is an incorrect clock source setting and results in severe noises.
We see wrong data rate in USB IN transfer.
(From 288 bytes/ms becomes 144 bytes/ms) It should keep in 288 bytes/ms.

This earphone works fine on older kernel version load because
this is a newly-added behavior.

Fixes: d2e8f641257d ("ALSA: usb-audio: Explicitly set up the clock selector")
Signed-off-by: chihhao.chen <chihhao.chen@mediatek.com>
Link: https://lore.kernel.org/r/1627100621-19225-1-git-send-email-chihhao.chen@mediatek.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoarm64: dts: armada-3720-turris-mox: remove mrvl,i2c-fast-mode
Pali Rohár [Mon, 28 Jun 2021 15:12:29 +0000 (17:12 +0200)]
arm64: dts: armada-3720-turris-mox: remove mrvl,i2c-fast-mode

[ Upstream commit ee7ab3f263f8131722cff3871b9618b1e7478f07 ]

Some SFP modules are not detected when i2c-fast-mode is enabled even when
clock-frequency is already set to 100000. The I2C bus violates the timing
specifications when run in fast mode. So disable fast mode on Turris Mox.

Same change was already applied for uDPU (also Armada 3720 board with SFP)
in commit fe3ec631a77d ("arm64: dts: uDPU: remove i2c-fast-mode").

Fixes: 7109d817db2e ("arm64: dts: marvell: add DTS for Turris Mox")
Signed-off-by: Pali Rohár <pali@kernel.org>
Reviewed-by: Marek Behún <kabel@kernel.org>
Acked-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: Gregory CLEMENT <gregory.clement@bootlin.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoarm64: dts: armada-3720-turris-mox: fixed indices for the SDHC controllers
Vladimir Oltean [Thu, 22 Jul 2021 10:11:34 +0000 (13:11 +0300)]
arm64: dts: armada-3720-turris-mox: fixed indices for the SDHC controllers

[ Upstream commit 923f98929182dfd04e9149be839160b63a3db145 ]

Since drivers/mmc/host/sdhci-xenon.c declares the PROBE_PREFER_ASYNCHRONOUS
probe type, it is not guaranteed whether /dev/mmcblk0 will belong to
sdhci0 or sdhci1. In turn, this will break booting by:

root=/dev/mmcblk0p1

Fix the issue by adding aliases so that the old MMC controller indices
are preserved.

Fixes: 7320915c8861 ("mmc: Set PROBE_PREFER_ASYNCHRONOUS for drivers that existed in v4.14")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Gregory CLEMENT <gregory.clement@bootlin.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoARM: dts: imx: Swap M53Menlo pinctrl_power_button/pinctrl_power_out pins
Marek Vasut [Sun, 18 Jul 2021 21:43:02 +0000 (23:43 +0200)]
ARM: dts: imx: Swap M53Menlo pinctrl_power_button/pinctrl_power_out pins

[ Upstream commit 3d9e30a52047f2d464efdfd1d561ae1f707a0286 ]

The pinctrl_power_button/pinctrl_power_out each define single GPIO
pinmux, except it is exactly the other one than the matching gpio-keys
and gpio-poweroff DT nodes use for that functionality. Swap the two
GPIOs to correct this error.

Fixes: 50d29fdb765d ("ARM: dts: imx53: Add power GPIOs on M53Menlo")
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Shawn Guo <shawnguo@kernel.org>
Cc: Fabio Estevam <festevam@gmail.com>
Cc: NXP Linux Team <linux-imx@nxp.com>
Reviewed-by: Fabio Estevam <festevam@gmail.com>
Signed-off-by: Shawn Guo <shawnguo@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoARM: imx: fix missing 3rd argument in macro imx_mmdc_perf_init
Colin Ian King [Thu, 15 Jul 2021 13:23:21 +0000 (14:23 +0100)]
ARM: imx: fix missing 3rd argument in macro imx_mmdc_perf_init

[ Upstream commit 20fb73911fec01f06592de1cdbca00b66602ebd7 ]

The function imx_mmdc_perf_init recently had a 3rd argument added to
it but the equivalent macro was not updated and is still the older
2 argument version. Fix this by adding in the missing 3rd argumement
mmdc_ipg_clk.

Fixes: f07ec8536580 ("ARM: imx: add missing clk_disable_unprepare()")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Shawn Guo <shawnguo@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoARM: dts: colibri-imx6ull: limit SDIO clock to 25MHz
Oleksandr Suvorov [Tue, 13 Jul 2021 20:21:07 +0000 (23:21 +0300)]
ARM: dts: colibri-imx6ull: limit SDIO clock to 25MHz

[ Upstream commit 828db68f4ff1ab6982a36a56522b585160dc8c8e ]

NXP and AzureWave don't recommend using SDIO bus mode 3.3V@50MHz due
to noise affecting the wireless throughput. Colibri iMX6ULL uses only
3.3V signaling for Wi-Fi module AW-CM276NF.

Limit the SDIO Clock on Colibri iMX6ULL to 25MHz.

Fixes: c2e4987e0e02 ("ARM: dts: imx6ull: add Toradex Colibri iMX6ULL support")
Signed-off-by: Oleksandr Suvorov <oleksandr.suvorov@toradex.com>
Reviewed-by: Fabio Estevam <festevam@gmail.com>
Signed-off-by: Shawn Guo <shawnguo@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoarm64: dts: ls1028: sl28: fix networking for variant 2
Michael Walle [Fri, 2 Jul 2021 19:49:14 +0000 (21:49 +0200)]
arm64: dts: ls1028: sl28: fix networking for variant 2

[ Upstream commit 29f6a20c21b5bdc7eb623a712bbf7b99612ee746 ]

The PHY configuration for the variant 2 is still missing the flag for
in-band signalling between PHY and MAC. Both sides - MAC and PHY - have
to match the setting. For now, Linux only supports setting the MAC side
and thus it has to match the setting the bootloader is configuring.
Enable in-band signalling to make ethernet work.

Fixes: ab43f0307449 ("arm64: dts: ls1028a: sl28: add support for variant 2")
Signed-off-by: Michael Walle <michael@walle.cc>
Signed-off-by: Shawn Guo <shawnguo@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoARM: dts: imx6qdl-sr-som: Increase the PHY reset duration to 10ms
Maxime Chevallier [Fri, 25 Jun 2021 12:13:53 +0000 (14:13 +0200)]
ARM: dts: imx6qdl-sr-som: Increase the PHY reset duration to 10ms

[ Upstream commit fd8e83884fdd7b5fc411f201a58d8d01890198a2 ]

The AR803x PHY used on this modules seems to require the reset line to
be asserted for around 10ms in order to avoid rare cases where the PHY
gets stuck in an incoherent state that prevents it to function
correctly.

The previous value of 2ms was found to be problematic on some setups,
causing intermittent issues where the PHY would be unresponsive
every once in a while on some sytems, with a low occurrence (it typically
took around 30 consecutive reboots to encounter the issue).

Bumping the delay to the 10ms makes the issue dissapear, with more than
2500 consecutive reboots performed without the issue showing-up.

Fixes: 208d7baf8085 ("ARM: imx: initial SolidRun HummingBoard support")
Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Tested-by: Hervé Codina <herve.codina@bootlin.com>
Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: Shawn Guo <shawnguo@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoARM: imx: add missing clk_disable_unprepare()
Yang Yingliang [Tue, 15 Jun 2021 12:52:39 +0000 (20:52 +0800)]
ARM: imx: add missing clk_disable_unprepare()

[ Upstream commit f07ec85365807b3939f32d0094a6dd5ce065d1b9 ]

clock source is prepared and enabled by clk_prepare_enable()
in probe function, but no disable or unprepare in remove and
error path.

Fixes: 9454a0caff6a ("ARM: imx: add mmdc ipg clock operation for mmdc")
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Reviewed-by: Dong Aisheng <aisheng.dong@nxp.com>
Signed-off-by: Shawn Guo <shawnguo@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoARM: imx: add missing iounmap()
Yang Yingliang [Tue, 15 Jun 2021 12:52:38 +0000 (20:52 +0800)]
ARM: imx: add missing iounmap()

[ Upstream commit f9613aa07f16d6042e74208d1b40a6104d72964a ]

Commit e76bdfd7403a ("ARM: imx: Added perf functionality to mmdc driver")
introduced imx_mmdc_remove(), the mmdc_base need be unmapped in it if
config PERF_EVENTS is enabled.

If imx_mmdc_perf_init() fails, the mmdc_base also need be unmapped.

Fixes: e76bdfd7403a ("ARM: imx: Added perf functionality to mmdc driver")
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Reviewed-by: Dong Aisheng <aisheng.dong@nxp.com>
Signed-off-by: Shawn Guo <shawnguo@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoarm64: dts: ls1028a: fix node name for the sysclk
Vladimir Oltean [Tue, 8 Jun 2021 11:26:58 +0000 (14:26 +0300)]
arm64: dts: ls1028a: fix node name for the sysclk

[ Upstream commit 7e71b85473f863a29eb1c69265ef025389b4091d ]

U-Boot attempts to fix up the "clock-frequency" property of the "/sysclk" node:
https://elixir.bootlin.com/u-boot/v2021.04/source/arch/arm/cpu/armv8/fsl-layerscape/fdt.c#L512

but fails to do so:

  ## Booting kernel from Legacy Image at a1000000 ...
     Image Name:
     Created:      2021-06-08  10:31:38 UTC
     Image Type:   AArch64 Linux Kernel Image (gzip compressed)
     Data Size:    15431370 Bytes = 14.7 MiB
     Load Address: 80080000
     Entry Point:  80080000
     Verifying Checksum ... OK
  ## Flattened Device Tree blob at a0000000
     Booting using the fdt blob at 0xa0000000
     Uncompressing Kernel Image
     Loading Device Tree to 00000000fbb19000, end 00000000fbb22717 ... OK
  Unable to update property /sysclk:clock-frequency, err=FDT_ERR_NOTFOUND

  Starting kernel ...

All Layerscape SoCs except LS1028A use "sysclk" as the node name, and
not "clock-sysclk". So change the node name of LS1028A accordingly.

Fixes: 8897f3255c9c ("arm64: dts: Add support for NXP LS1028A SoC")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Shawn Guo <shawnguo@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agonet: xfrm: fix memory leak in xfrm_user_rcv_msg
Pavel Skripkin [Fri, 25 Jun 2021 10:23:54 +0000 (13:23 +0300)]
net: xfrm: fix memory leak in xfrm_user_rcv_msg

[ Upstream commit 7c1a80e80cde008f271bae630d28cf684351e807 ]

Syzbot reported memory leak in xfrm_user_rcv_msg(). The
problem was is non-freed skb's frag_list.

In skb_release_all() skb_release_data() will be called only
in case of skb->head != NULL, but netlink_skb_destructor()
sets head to NULL. So, allocated frag_list skb should be
freed manualy, since consume_skb() won't take care of it

Fixes: 5106f4a8acff ("xfrm/compat: Add 32=>64-bit messages translator")
Reported-and-tested-by: syzbot+fb347cf82c73a90efcca@syzkaller.appspotmail.com
Signed-off-by: Pavel Skripkin <paskripkin@gmail.com>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agobus: ti-sysc: Fix gpt12 system timer issue with reserved status
Tony Lindgren [Fri, 11 Jun 2021 05:42:50 +0000 (08:42 +0300)]
bus: ti-sysc: Fix gpt12 system timer issue with reserved status

[ Upstream commit 3ff340e24c9dd5cff9fc07d67914c5adf67f80d6 ]

Jarkko Nikula <jarkko.nikula@bitmer.com> reported that Beagleboard
revision c2 stopped booting. Jarkko bisected the issue down to
commit 6cfcd5563b4f ("clocksource/drivers/timer-ti-dm: Fix suspend
and resume for am3 and am4").

Let's fix the issue by tagging system timers as reserved rather than
ignoring them. And let's not probe any interconnect target module child
devices for reserved modules.

This allows PM runtime to keep track of clocks and clockdomains for
the interconnect target module, and prevent the system timer from idling
as we already have SYSC_QUIRK_NO_IDLE and SYSC_QUIRK_NO_IDLE_ON_INIT
flags set for system timers.

Fixes: 6cfcd5563b4f ("clocksource/drivers/timer-ti-dm: Fix suspend and resume for am3 and am4")
Reported-by: Jarkko Nikula <jarkko.nikula@bitmer.com>
Tested-by: Jarkko Nikula <jarkko.nikula@bitmer.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoALSA: seq: Fix racy deletion of subscriber
Takashi Iwai [Tue, 3 Aug 2021 11:43:12 +0000 (13:43 +0200)]
ALSA: seq: Fix racy deletion of subscriber

commit 97367c97226aab8b298ada954ce12659ee3ad2a4 upstream.

It turned out that the current implementation of the port subscription
is racy.  The subscription contains two linked lists, and we have to
add to or delete from both lists.  Since both connection and
disconnection procedures perform the same order for those two lists
(i.e. src list, then dest list), when a deletion happens during a
connection procedure, the src list may be deleted before the dest list
addition completes, and this may lead to a use-after-free or an Oops,
even though the access to both lists are protected via mutex.

The simple workaround for this race is to change the access order for
the disconnection, namely, dest list, then src list.  This assures
that the connection has been established when disconnecting, and also
the concurrent deletion can be avoided.

Reported-and-tested-by: folkert <folkert@vanheusden.com>
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20210801182754.GP890690@belle.intranet.vanheusden.com
Link: https://lore.kernel.org/r/20210803114312.2536-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agoRevert "ACPICA: Fix memory leak caused by _CID repair function"
Rafael J. Wysocki [Tue, 3 Aug 2021 16:14:44 +0000 (18:14 +0200)]
Revert "ACPICA: Fix memory leak caused by _CID repair function"

commit 6511a8b5b7a65037340cd8ee91a377811effbc83 upstream.

Revert commit c27bac0314131 ("ACPICA: Fix memory leak caused by _CID
repair function") which is reported to cause a boot issue on Acer
Swift 3 (SF314-51).

Reported-by: Adrien Precigout <dev@asdrip.fr>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agoLinux 5.10.57
Greg Kroah-Hartman [Sun, 8 Aug 2021 07:05:24 +0000 (09:05 +0200)]
Linux 5.10.57

Link: https://lore.kernel.org/r/20210806081113.126861800@linuxfoundation.org
Tested-by: Fox Chen <foxhlchen@gmail.com>
Tested-by: Pavel Machek (CIP) <pavel@denx.de>
Tested-by: Jon Hunter <jonathanh@nvidia.com>
Tested-by: Guenter Roeck <linux@roeck-us.net>
Tested-by: Rudi Heitbaum <rudi@heitbaum.com>
Tested-by: Salvatore Bonaccorso <carnil@debian.org>
Tested-by: Sudip Mukherjee <sudip.mukherjee@codethink.co.uk>
Tested-by: Linux Kernel Functional Testing <lkft@linaro.org>
Tested-by: Aakash Hemadri <aakashhemadri123@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agospi: mediatek: Fix fifo transfer
Guenter Roeck [Mon, 2 Aug 2021 03:00:23 +0000 (20:00 -0700)]
spi: mediatek: Fix fifo transfer

commit 0d5c3954b35eddff0da0436c31e8d721eceb7dc2 upstream.

Commit 3a70dd2d0503 ("spi: mediatek: fix fifo rx mode") claims that
fifo RX mode was never handled, and adds the presumably missing code
to the FIFO transfer function. However, the claim that receive data
was not handled is incorrect. It was handled as part of interrupt
handling after the transfer was complete. The code added with the above
mentioned commit reads data from the receive FIFO before the transfer
is started, which is wrong. This results in an actual transfer error
on a Hayato Chromebook.

Remove the code trying to handle receive data before the transfer is
started to fix the problem.

Fixes: 3a70dd2d0503 ("spi: mediatek: fix fifo rx mode")
Cc: Peter Hess <peter.hess@ph-home.de>
Cc: Frank Wunderlich <frank-w@public-files.de>
Cc: Tzung-Bi Shih <tzungbi@google.com>
Cc: Hsin-Yi Wang <hsinyi@google.com>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Tested-by: Hsin-Yi Wang <hsinyi@google.com>
Tested-by: Tzung-Bi Shih <tzungbi@google.com>
Link: https://lore.kernel.org/r/20210802030023.1748777-1-linux@roeck-us.net
Signed-off-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agoselftest/bpf: Verifier tests for var-off access
Andrei Matei [Wed, 4 Aug 2021 17:09:17 +0000 (20:09 +0300)]
selftest/bpf: Verifier tests for var-off access

commit 7a22930c4179b51352f2ec9feb35167cbe79afd9 upstream

Add tests for the new functionality - reading and writing to the stack
through a variable-offset pointer.

Signed-off-by: Andrei Matei <andreimatei1@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20210207011027.676572-4-andreimatei1@gmail.com
Signed-off-by: Ovidiu Panait <ovidiu.panait@windriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agobpf, selftests: Adjust few selftest outcomes wrt unreachable code
Daniel Borkmann [Wed, 4 Aug 2021 17:09:16 +0000 (20:09 +0300)]
bpf, selftests: Adjust few selftest outcomes wrt unreachable code

commit 973377ffe8148180b2651825b92ae91988141b05 upstream

In almost all cases from test_verifier that have been changed in here, we've
had an unreachable path with a load from a register which has an invalid
address on purpose. This was basically to make sure that we never walk this
path and to have the verifier complain if it would otherwise. Change it to
match on the right error for unprivileged given we now test these paths
under speculative execution.

There's one case where we match on exact # of insns_processed. Due to the
extra path, this will of course mismatch on unprivileged. Thus, restrict the
test->insn_processed check to privileged-only.

In one other case, we result in a 'pointer comparison prohibited' error. This
is similarly due to verifying an 'invalid' branch where we end up with a value
pointer on one side of the comparison.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Ovidiu Panait <ovidiu.panait@windriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agobpf: Update selftests to reflect new error states
Daniel Borkmann [Wed, 4 Aug 2021 17:09:15 +0000 (20:09 +0300)]
bpf: Update selftests to reflect new error states

commit d7a5091351756d0ae8e63134313c455624e36a13 upstream

Update various selftest error messages:

 * The 'Rx tried to sub from different maps, paths, or prohibited types'
   is reworked into more specific/differentiated error messages for better
   guidance.

 * The change into 'value -4294967168 makes map_value pointer be out of
   bounds' is due to moving the mixed bounds check into the speculation
   handling and thus occuring slightly later than above mentioned sanity
   check.

 * The change into 'math between map_value pointer and register with
   unbounded min value' is similarly due to register sanity check coming
   before the mixed bounds check.

 * The case of 'map access: known scalar += value_ptr from different maps'
   now loads fine given masks are the same from the different paths (despite
   max map value size being different).

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Ovidiu Panait <ovidiu.panait@windriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agobpf, selftests: Adjust few selftest result_unpriv outcomes
Daniel Borkmann [Wed, 4 Aug 2021 17:09:14 +0000 (20:09 +0300)]
bpf, selftests: Adjust few selftest result_unpriv outcomes

commit 1bad6fd52be4ce12d207e2820ceb0f29ab31fc53 upstream

Given we don't need to simulate the speculative domain for registers with
immediates anymore since the verifier uses direct imm-based rewrites instead
of having to mask, we can also lift a few cases that were previously rejected.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Ovidiu Panait <ovidiu.panait@windriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agoselftest/bpf: Adjust expected verifier errors
Andrei Matei [Wed, 4 Aug 2021 17:09:13 +0000 (20:09 +0300)]
selftest/bpf: Adjust expected verifier errors

commit a680cb3d8e3f4f84205720b90c926579d04eedb6 upstream

The verifier errors around stack accesses have changed slightly in the
previous commit (generally for the better).

Signed-off-by: Andrei Matei <andreimatei1@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20210207011027.676572-3-andreimatei1@gmail.com
Signed-off-by: Ovidiu Panait <ovidiu.panait@windriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agoselftests/bpf: Add a test for ptr_to_map_value on stack for helper access
Yonghong Song [Wed, 4 Aug 2021 17:09:12 +0000 (20:09 +0300)]
selftests/bpf: Add a test for ptr_to_map_value on stack for helper access

commit b4b638c36b7e7acd847b9c4b9c80f268e45ea30c upstream

Change bpf_iter_task.c such that pointer to map_value may appear
on the stack for bpf_seq_printf() to access. Without previous
verifier patch, the bpf_iter test will fail.

Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/bpf/20201210013350.943985-1-yhs@fb.com
Signed-off-by: Ovidiu Panait <ovidiu.panait@windriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agoRevert "watchdog: iTCO_wdt: Account for rebooting on second timeout"
Greg Kroah-Hartman [Fri, 6 Aug 2021 06:28:48 +0000 (08:28 +0200)]
Revert "watchdog: iTCO_wdt: Account for rebooting on second timeout"

This reverts commit 39ed17de8c6ff54c7ed4605a4a8e04a2e2f0b82e which is
commit cb011044e34c293e139570ce5c01aed66a34345c upstream.

It is reported to cause problems with systems and probably should not
have been backported in the first place :(

Link: https://lore.kernel.org/r/20210803165108.4154cd52@endymion
Reported-by: Jean Delvare <jdelvare@suse.de>
Cc: Jan Kiszka <jan.kiszka@siemens.com>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: Wim Van Sebroeck <wim@linux-watchdog.org>
Cc: Sasha Levin <sashal@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agofirmware: arm_scmi: Add delayed response status check
Cristian Marussi [Tue, 8 Jun 2021 10:30:56 +0000 (11:30 +0100)]
firmware: arm_scmi: Add delayed response status check

commit f1748b1ee1fa0fd1a074504045b530b62f949188 upstream.

A successfully received delayed response could anyway report a failure at
the protocol layer in the message status field.

Add a check also for this error condition.

Link: https://lore.kernel.org/r/20210608103056.3388-1-cristian.marussi@arm.com
Fixes: 58ecdf03dbb9 ("firmware: arm_scmi: Add support for asynchronous commands and delayed response")
Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agofirmware: arm_scmi: Ensure drivers provide a probe function
Sudeep Holla [Thu, 24 Jun 2021 09:50:59 +0000 (10:50 +0100)]
firmware: arm_scmi: Ensure drivers provide a probe function

commit 5e469dac326555d2038d199a6329458cc82a34e5 upstream.

The bus probe callback calls the driver callback without further
checking. Better be safe than sorry and refuse registration of a driver
without a probe function to prevent a NULL pointer exception.

Link: https://lore.kernel.org/r/20210624095059.4010157-2-sudeep.holla@arm.com
Fixes: 933c504424a2 ("firmware: arm_scmi: add scmi protocol bus to enumerate protocol devices")
Reported-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Tested-by: Cristian Marussi <cristian.marussi@arm.com>
Reviewed-by: Cristian Marussi <cristian.marussi@arm.com>
Acked-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agoRevert "Bluetooth: Shutdown controller after workqueues are flushed or cancelled"
Greg Kroah-Hartman [Thu, 5 Aug 2021 18:58:57 +0000 (20:58 +0200)]
Revert "Bluetooth: Shutdown controller after workqueues are flushed or cancelled"

This reverts commit 60789afc02f592b8d91217b60930e7a76271ae07 which is
commit 0ea9fd001a14ebc294f112b0361a4e601551d508 upstream.

It has been reported to have problems:
https://lore.kernel.org/linux-bluetooth/8735ryk0o7.fsf@baylibre.com/

Reported-by: Guenter Roeck <linux@roeck-us.net>
Cc: Kai-Heng Feng <kai.heng.feng@canonical.com>
Cc: Marcel Holtmann <marcel@holtmann.org>
Cc: Sasha Levin <sashal@kernel.org>
Link: https://lore.kernel.org/r/efee3a58-a4d2-af22-0931-e81b877ab539@roeck-us.net
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agoACPI: fix NULL pointer dereference
Linus Torvalds [Sat, 24 Jul 2021 22:25:54 +0000 (15:25 -0700)]
ACPI: fix NULL pointer dereference

[ Upstream commit fc68f42aa737dc15e7665a4101d4168aadb8e4c4 ]

Commit 71f642833284 ("ACPI: utils: Fix reference counting in
for_each_acpi_dev_match()") started doing "acpi_dev_put()" on a pointer
that was possibly NULL.  That fails miserably, because that helper
inline function is not set up to handle that case.

Just make acpi_dev_put() silently accept a NULL pointer, rather than
calling down to put_device() with an invalid offset off that NULL
pointer.

Link: https://lore.kernel.org/lkml/a607c149-6bf6-0fd0-0e31-100378504da2@kernel.dk/
Reported-and-tested-by: Jens Axboe <axboe@kernel.dk>
Tested-by: Daniel Scally <djrscally@gmail.com>
Cc: Andy Shevchenko <andy.shevchenko@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agodrm/amd/display: Fix max vstartup calculation for modes with borders
Nicholas Kazlauskas [Wed, 7 Jul 2021 17:19:14 +0000 (13:19 -0400)]
drm/amd/display: Fix max vstartup calculation for modes with borders

[ Upstream commit d7940911fc0754d99b208f0e3098762d39f403a0 ]

[Why]
Vertical and horizontal borders in timings are treated as increasing the
active area - vblank and hblank actually shrink.

Our input into DML does not include these borders so it incorrectly
assumes it has more time than available for vstartup and tmdl
calculations for some modes with borders.

An example of such a timing would be 640x480@72Hz:

h_total: 832
h_border_left: 8
h_addressable: 640
h_border_right: 8
h_front_porch: 16
h_sync_width: 40
v_total: 520
v_border_top: 8
v_addressable: 480
v_border_bottom: 8
v_front_porch: 1
v_sync_width: 3
pix_clk_100hz: 315000

[How]
Include borders as part of destination vactive/hactive.

This change DCN20+ so it has wide impact, but the destination vactive
and hactive are only really used for vstartup calculation anyway.

Most modes do not have vertical or horizontal borders.

Reviewed-by: Dmytro Laktyushkin <Dmytro.Laktyushkin@amd.com>
Acked-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agodrm/amd/display: Fix comparison error in dcn21 DML
Victor Lu [Thu, 24 Jun 2021 15:05:42 +0000 (11:05 -0400)]
drm/amd/display: Fix comparison error in dcn21 DML

[ Upstream commit ec3102dc6b36c692104c4a0546d4119de59a3bc1 ]

[why]
A comparison error made it possible to not iterate through all the
specified prefetch modes.

[how]
Correct "<" to "<="

Reviewed-by: Dmytro Laktyushkin <Dmytro.Laktyushkin@amd.com>
Reviewed-by: Yongqiang Sun <Yongqiang.Sun@amd.com>
Acked-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Victor Lu <victorchengchi.lu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agonvme: fix nvme_setup_command metadata trace event
Keith Busch [Mon, 19 Jul 2021 16:44:39 +0000 (09:44 -0700)]
nvme: fix nvme_setup_command metadata trace event

[ Upstream commit 234211b8dd161fa25f192c78d5a8d2dd6bf920a0 ]

The metadata address is set after the trace event, so the trace is not
capturing anything useful. Rather than logging the memory address, it's
useful to know if the command carries a metadata payload, so change the
trace event to log that true/false state instead.

Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoefi/mokvar: Reserve the table only if it is in boot services data
Borislav Petkov [Tue, 20 Jul 2021 07:28:09 +0000 (09:28 +0200)]
efi/mokvar: Reserve the table only if it is in boot services data

[ Upstream commit 47e1e233e9d822dfda068383fb9a616451bda703 ]

One of the SUSE QA tests triggered:

  localhost kernel: efi: Failed to lookup EFI memory descriptor for 0x000000003dcf8000

which comes from x86's version of efi_arch_mem_reserve() trying to
reserve a memory region. Usually, that function expects
EFI_BOOT_SERVICES_DATA memory descriptors but the above case is for the
MOKvar table which is allocated in the EFI shim as runtime services.

That lead to a fix changing the allocation of that table to boot services.

However, that fix broke booting SEV guests with that shim leading to
this kernel fix

  8d651ee9c71b ("x86/ioremap: Map EFI-reserved memory as encrypted for SEV")

which extended the ioremap hint to map reserved EFI boot services as
decrypted too.

However, all that wasn't needed, IMO, because that error message in
efi_arch_mem_reserve() was innocuous in this case - if the MOKvar table
is not in boot services, then it doesn't need to be reserved in the
first place because it is, well, in runtime services which *should* be
reserved anyway.

So do that reservation for the MOKvar table only if it is allocated
in boot services data. I couldn't find any requirement about where
that table should be allocated in, unlike the ESRT which allocation is
mandated to be done in boot services data by the UEFI spec.

Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoASoC: ti: j721e-evm: Check for not initialized parent_clk_id
Peter Ujfalusi [Sat, 17 Jul 2021 12:28:19 +0000 (15:28 +0300)]
ASoC: ti: j721e-evm: Check for not initialized parent_clk_id

[ Upstream commit 82d28b67f780910f816fe1cfb0f676fc38c4cbb3 ]

During probe the parent_clk_id is set to -1 which should not be used to
array index within hsdiv_rates[].

Signed-off-by: Peter Ujfalusi <peter.ujfalusi@gmail.com>
Link: https://lore.kernel.org/r/20210717122820.1467-3-peter.ujfalusi@gmail.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoASoC: ti: j721e-evm: Fix unbalanced domain activity tracking during startup
Peter Ujfalusi [Sat, 17 Jul 2021 12:28:18 +0000 (15:28 +0300)]
ASoC: ti: j721e-evm: Fix unbalanced domain activity tracking during startup

[ Upstream commit 78d2a05ef22e7b5863b01e073dd6a06b3979bb00 ]

In case of an error within j721e_audio_startup() the domain->active must
be decremented to avoid unbalanced counter.

Signed-off-by: Peter Ujfalusi <peter.ujfalusi@gmail.com>
Link: https://lore.kernel.org/r/20210717122820.1467-2-peter.ujfalusi@gmail.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agonet: Fix zero-copy head len calculation.
Pravin B Shelar [Thu, 15 Jul 2021 23:59:00 +0000 (16:59 -0700)]
net: Fix zero-copy head len calculation.

[ Upstream commit a17ad0961706244dce48ec941f7e476a38c0e727 ]

In some cases skb head could be locked and entire header
data is pulled from skb. When skb_zerocopy() called in such cases,
following BUG is triggered. This patch fixes it by copying entire
skb in such cases.
This could be optimized incase this is performance bottleneck.

---8<---
kernel BUG at net/core/skbuff.c:2961!
invalid opcode: 0000 [#1] SMP PTI
CPU: 2 PID: 0 Comm: swapper/2 Tainted: G           OE     5.4.0-77-generic #86-Ubuntu
Hardware name: OpenStack Foundation OpenStack Nova, BIOS 1.13.0-1ubuntu1.1 04/01/2014
RIP: 0010:skb_zerocopy+0x37a/0x3a0
RSP: 0018:ffffbcc70013ca38 EFLAGS: 00010246
Call Trace:
 <IRQ>
 queue_userspace_packet+0x2af/0x5e0 [openvswitch]
 ovs_dp_upcall+0x3d/0x60 [openvswitch]
 ovs_dp_process_packet+0x125/0x150 [openvswitch]
 ovs_vport_receive+0x77/0xd0 [openvswitch]
 netdev_port_receive+0x87/0x130 [openvswitch]
 netdev_frame_hook+0x4b/0x60 [openvswitch]
 __netif_receive_skb_core+0x2b4/0xc90
 __netif_receive_skb_one_core+0x3f/0xa0
 __netif_receive_skb+0x18/0x60
 process_backlog+0xa9/0x160
 net_rx_action+0x142/0x390
 __do_softirq+0xe1/0x2d6
 irq_exit+0xae/0xb0
 do_IRQ+0x5a/0xf0
 common_interrupt+0xf/0xf

Code that triggered BUG:
int
skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
{
        int i, j = 0;
        int plen = 0; /* length of skb->head fragment */
        int ret;
        struct page *page;
        unsigned int offset;

        BUG_ON(!from->head_frag && !hlen);

Signed-off-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoASoC: rt5682: Fix the issue of garbled recording after powerd_dbus_suspend
Oder Chiou [Fri, 16 Jul 2021 08:58:53 +0000 (16:58 +0800)]
ASoC: rt5682: Fix the issue of garbled recording after powerd_dbus_suspend

[ Upstream commit 6a503e1c455316fd0bfd8188c0a62cce7c5525ca ]

While using the DMIC recording, the garbled data will be captured by the
DMIC. It is caused by the critical power of PLL closed in the jack detect
function.

Signed-off-by: Oder Chiou <oder_chiou@realtek.com>
Link: https://lore.kernel.org/r/20210716085853.20170-1-oder_chiou@realtek.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoqed: fix possible unpaired spin_{un}lock_bh in _qed_mcp_cmd_and_union()
Jia He [Thu, 15 Jul 2021 08:08:21 +0000 (16:08 +0800)]
qed: fix possible unpaired spin_{un}lock_bh in _qed_mcp_cmd_and_union()

[ Upstream commit 6206b7981a36476f4695d661ae139f7db36a802d ]

Liajian reported a bug_on hit on a ThunderX2 arm64 server with FastLinQ
QL41000 ethernet controller:
 BUG: scheduling while atomic: kworker/0:4/531/0x00000200
  [qed_probe:488()]hw prepare failed
  kernel BUG at mm/vmalloc.c:2355!
  Internal error: Oops - BUG: 0 [#1] SMP
  CPU: 0 PID: 531 Comm: kworker/0:4 Tainted: G W 5.4.0-77-generic #86-Ubuntu
  pstate: 00400009 (nzcv daif +PAN -UAO)
 Call trace:
  vunmap+0x4c/0x50
  iounmap+0x48/0x58
  qed_free_pci+0x60/0x80 [qed]
  qed_probe+0x35c/0x688 [qed]
  __qede_probe+0x88/0x5c8 [qede]
  qede_probe+0x60/0xe0 [qede]
  local_pci_probe+0x48/0xa0
  work_for_cpu_fn+0x24/0x38
  process_one_work+0x1d0/0x468
  worker_thread+0x238/0x4e0
  kthread+0xf0/0x118
  ret_from_fork+0x10/0x18

In this case, qed_hw_prepare() returns error due to hw/fw error, but in
theory work queue should be in process context instead of interrupt.

The root cause might be the unpaired spin_{un}lock_bh() in
_qed_mcp_cmd_and_union(), which causes botton half is disabled incorrectly.

Reported-by: Lijian Zhang <Lijian.Zhang@arm.com>
Signed-off-by: Jia He <justin.he@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agor8152: Fix potential PM refcount imbalance
Takashi Iwai [Wed, 14 Jul 2021 17:00:21 +0000 (19:00 +0200)]
r8152: Fix potential PM refcount imbalance

[ Upstream commit 9c23aa51477a37f8b56c3c40192248db0663c196 ]

rtl8152_close() takes the refcount via usb_autopm_get_interface() but
it doesn't release when RTL8152_UNPLUG test hits.  This may lead to
the imbalance of PM refcount.  This patch addresses it.

Link: https://bugzilla.suse.com/show_bug.cgi?id=1186194
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoASoC: tlv320aic31xx: fix reversed bclk/wclk master bits
Kyle Russell [Tue, 22 Jun 2021 01:09:41 +0000 (21:09 -0400)]
ASoC: tlv320aic31xx: fix reversed bclk/wclk master bits

[ Upstream commit 9cf76a72af6ab81030dea6481b1d7bdd814fbdaf ]

These are backwards from Table 7-71 of the TLV320AIC3100 spec [1].

This was broken in 12eb4d66ba2e when BCLK_MASTER and WCLK_MASTER
were converted from 0x08 and 0x04 to BIT(2) and BIT(3), respectively.

-#define AIC31XX_BCLK_MASTER 0x08
-#define AIC31XX_WCLK_MASTER 0x04
+#define AIC31XX_BCLK_MASTER BIT(2)
+#define AIC31XX_WCLK_MASTER BIT(3)

Probably just a typo since the defines were not listed in bit order.

[1] https://www.ti.com/lit/gpn/tlv320aic3100

Signed-off-by: Kyle Russell <bkylerussell@gmail.com>
Link: https://lore.kernel.org/r/20210622010941.241386-1-bkylerussell@gmail.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agospi: stm32h7: fix full duplex irq handler handling
Alain Volmat [Wed, 30 Jun 2021 08:45:19 +0000 (10:45 +0200)]
spi: stm32h7: fix full duplex irq handler handling

[ Upstream commit e4a5c19888a5f8a9390860ca493e643be58c8791 ]

In case of Full-Duplex mode, DXP flag is set when RXP and TXP flags are
set. But to avoid 2 different handlings, just add TXP and RXP flag in
the mask instead of DXP, and then keep the initial handling of TXP and
RXP events.
Also rephrase comment about EOTIE which is one of the interrupt enable
bits. It is not triggered by any event.

Signed-off-by: Amelie Delaunay <amelie.delaunay@foss.st.com>
Signed-off-by: Alain Volmat <alain.volmat@foss.st.com>
Reviewed-by: Amelie Delaunay <amelie.delaunay@foss.st.com>
Link: https://lore.kernel.org/r/1625042723-661-3-git-send-email-alain.volmat@foss.st.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoregulator: rt5033: Fix n_voltages settings for BUCK and LDO
Axel Lin [Sun, 27 Jun 2021 08:04:18 +0000 (16:04 +0800)]
regulator: rt5033: Fix n_voltages settings for BUCK and LDO

[ Upstream commit 6549c46af8551b346bcc0b9043f93848319acd5c ]

For linear regulators, the n_voltages should be (max - min) / step + 1.

Buck voltage from 1v to 3V, per step 100mV, and vout mask is 0x1f.
If value is from 20 to 31, the voltage will all be fixed to 3V.
And LDO also, just vout range is different from 1.2v to 3v, step is the
same. If value is from 18 to 31, the voltage will also be fixed to 3v.

Signed-off-by: Axel Lin <axel.lin@ingics.com>
Reviewed-by: ChiYuan Huang <cy_huang@richtek.com>
Link: https://lore.kernel.org/r/20210627080418.1718127-1-axel.lin@ingics.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoregulator: rtmv20: Fix wrong mask for strobe-polarity-high
ChiYuan Huang [Sat, 26 Jun 2021 15:58:32 +0000 (23:58 +0800)]
regulator: rtmv20: Fix wrong mask for strobe-polarity-high

[ Upstream commit 2b6a761be079f9fa8abf3157b5679a6f38885db4 ]

Fix wrong mask for strobe-polarity-high.

Signed-off-by: ChiYuan Huang <cy_huang@richtek.com>
In-reply-to: <CAFRkauB=0KwrJW19nJTTagdHhBR=V2R8YFWG3R3oVXt=rBRsqw@mail.gmail.com>
Reviewed-by: Axel Lin <axel.lin@ingics.com>
Link: https://lore.kernel.org/r/1624723112-26653-1-git-send-email-u0084500@gmail.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agobtrfs: fix lost inode on log replay after mix of fsync, rename and inode eviction
Filipe Manana [Tue, 27 Jul 2021 10:24:43 +0000 (11:24 +0100)]
btrfs: fix lost inode on log replay after mix of fsync, rename and inode eviction

[ Upstream commit ecc64fab7d49c678e70bd4c35fe64d2ab3e3d212 ]

When checking if we need to log the new name of a renamed inode, we are
checking if the inode and its parent inode have been logged before, and if
not we don't log the new name. The check however is buggy, as it directly
compares the logged_trans field of the inodes versus the ID of the current
transaction. The problem is that logged_trans is a transient field, only
stored in memory and never persisted in the inode item, so if an inode
was logged before, evicted and reloaded, its logged_trans field is set to
a value of 0, meaning the check will return false and the new name of the
renamed inode is not logged. If the old parent directory was previously
fsynced and we deleted the logged directory entries corresponding to the
old name, we end up with a log that when replayed will delete the renamed
inode.

The following example triggers the problem:

  $ mkfs.btrfs -f /dev/sdc
  $ mount /dev/sdc /mnt

  $ mkdir /mnt/A
  $ mkdir /mnt/B
  $ echo -n "hello world" > /mnt/A/foo

  $ sync

  # Add some new file to A and fsync directory A.
  $ touch /mnt/A/bar
  $ xfs_io -c "fsync" /mnt/A

  # Now trigger inode eviction. We are only interested in triggering
  # eviction for the inode of directory A.
  $ echo 2 > /proc/sys/vm/drop_caches

  # Move foo from directory A to directory B.
  # This deletes the directory entries for foo in A from the log, and
  # does not add the new name for foo in directory B to the log, because
  # logged_trans of A is 0, which is less than the current transaction ID.
  $ mv /mnt/A/foo /mnt/B/foo

  # Now make an fsync to anything except A, B or any file inside them,
  # like for example create a file at the root directory and fsync this
  # new file. This syncs the log that contains all the changes done by
  # previous rename operation.
  $ touch /mnt/baz
  $ xfs_io -c "fsync" /mnt/baz

  <power fail>

  # Mount the filesystem and replay the log.
  $ mount /dev/sdc /mnt

  # Check the filesystem content.
  $ ls -1R /mnt
  /mnt/:
  A
  B
  baz

  /mnt/A:
  bar

  /mnt/B:
  $

  # File foo is gone, it's neither in A/ nor in B/.

Fix this by using the inode_logged() helper at btrfs_log_new_name(), which
safely checks if an inode was logged before in the current transaction.

A test case for fstests will follow soon.

CC: stable@vger.kernel.org # 4.14+
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agobtrfs: fix race causing unnecessary inode logging during link and rename
Filipe Manana [Wed, 25 Nov 2020 12:19:23 +0000 (12:19 +0000)]
btrfs: fix race causing unnecessary inode logging during link and rename

[ Upstream commit de53d892e5c51dfa0a158e812575a75a6c991f39 ]

When we are doing a rename or a link operation for an inode that was logged
in the previous transaction and that transaction is still committing, we
have a time window where we incorrectly consider that the inode was logged
previously in the current transaction and therefore decide to log it to
update it in the log. The following steps give an example on how this
happens during a link operation:

1) Inode X is logged in transaction 1000, so its logged_trans field is set
   to 1000;

2) Task A starts to commit transaction 1000;

3) The state of transaction 1000 is changed to TRANS_STATE_UNBLOCKED;

4) Task B starts a link operation for inode X, and as a consequence it
   starts transaction 1001;

5) Task A is still committing transaction 1000, therefore the value stored
   at fs_info->last_trans_committed is still 999;

6) Task B calls btrfs_log_new_name(), it reads a value of 999 from
   fs_info->last_trans_committed and because the logged_trans field of
   inode X has a value of 1000, the function does not return immediately,
   instead it proceeds to logging the inode, which should not happen
   because the inode was logged in the previous transaction (1000) and
   not in the current one (1001).

This is not a functional problem, just wasted time and space logging an
inode that does not need to be logged, contributing to higher latency
for link and rename operations.

So fix this by comparing the inodes' logged_trans field with the
generation of the current transaction instead of comparing with the value
stored in fs_info->last_trans_committed.

This case is often hit when running dbench for a long enough duration, as
it does lots of rename operations.

This patch belongs to a patch set that is comprised of the following
patches:

  btrfs: fix race causing unnecessary inode logging during link and rename
  btrfs: fix race that results in logging old extents during a fast fsync
  btrfs: fix race that causes unnecessary logging of ancestor inodes
  btrfs: fix race that makes inode logging fallback to transaction commit
  btrfs: fix race leading to unnecessary transaction commit when logging inode
  btrfs: do not block inode logging for so long during transaction commit

Performance results are mentioned in the change log of the last patch.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoRevert "drm/i915: Propagate errors on awaiting already signaled fences"
Jason Ekstrand [Mon, 2 Aug 2021 18:48:02 +0000 (13:48 -0500)]
Revert "drm/i915: Propagate errors on awaiting already signaled fences"

commit 3761baae908a7b5012be08d70fa553cc2eb82305 upstream.

This reverts commit 9e31c1fe45d555a948ff66f1f0e3fe1f83ca63f7.  Ever
since that commit, we've been having issues where a hang in one client
can propagate to another.  In particular, a hang in an app can propagate
to the X server which causes the whole desktop to lock up.

Error propagation along fences sound like a good idea, but as your bug
shows, surprising consequences, since propagating errors across security
boundaries is not a good thing.

What we do have is track the hangs on the ctx, and report information to
userspace using RESET_STATS. That's how arb_robustness works. Also, if my
understanding is still correct, the EIO from execbuf is when your context
is banned (because not recoverable or too many hangs). And in all these
cases it's up to userspace to figure out what is all impacted and should
be reported to the application, that's not on the kernel to guess and
automatically propagate.

What's more, we're also building more features on top of ctx error
reporting with RESET_STATS ioctl: Encrypted buffers use the same, and the
userspace fence wait also relies on that mechanism. So it is the path
going forward for reporting gpu hangs and resets to userspace.

So all together that's why I think we should just bury this idea again as
not quite the direction we want to go to, hence why I think the revert is
the right option here.

For backporters: Please note that you _must_ have a backport of
https://lore.kernel.org/dri-devel/20210602164149.391653-2-jason@jlekstrand.net/
for otherwise backporting just this patch opens up a security bug.

v2: Augment commit message. Also restore Jason's sob that I
accidentally lost.

v3: Add a note for backporters

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reported-by: Marcin Slusarz <marcin.slusarz@intel.com>
Cc: <stable@vger.kernel.org> # v5.6+
Cc: Jason Ekstrand <jason.ekstrand@intel.com>
Cc: Marcin Slusarz <marcin.slusarz@intel.com>
Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/3080
Fixes: 9e31c1fe45d5 ("drm/i915: Propagate errors on awaiting already signaled fences")
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Jon Bloomfield <jon.bloomfield@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20210714193419.1459723-3-jason@jlekstrand.net
(cherry picked from commit 93a2711cddd5760e2f0f901817d71c93183c3b87)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agodrm/i915: Revert "drm/i915/gem: Asynchronous cmdparser"
Jason Ekstrand [Mon, 2 Aug 2021 18:48:01 +0000 (13:48 -0500)]
drm/i915: Revert "drm/i915/gem: Asynchronous cmdparser"

commit c9d9fdbc108af8915d3f497bbdf3898bf8f321b8 upstream.

This reverts 686c7c35abc2 ("drm/i915/gem: Asynchronous cmdparser").  The
justification for this commit in the git history was a vague comment
about getting it out from under the struct_mutex.  While this may
improve perf for some workloads on Gen7 platforms where we rely on the
command parser for features such as indirect rendering, no numbers were
provided to prove such an improvement.  It claims to closed two
gitlab/bugzilla issues but with no explanation whatsoever as to why or
what bug it's fixing.

Meanwhile, by moving command parsing off to an async callback, it leaves
us with a problem of what to do on error.  When things were synchronous,
EXECBUFFER2 would fail with an error code if parsing failed.  When
moving it to async, we needed another way to handle that error and the
solution employed was to set an error on the dma_fence and then trust
that said error gets propagated to the client eventually.  Moving back
to synchronous will help us untangle the fence error propagation mess.

This also reverts most of 0edbb9ba1bfe ("drm/i915: Move cmd parser
pinning to execbuffer") which is a refactor of some of our allocation
paths for asynchronous parsing.  Now that everything is synchronous, we
don't need it.

v2 (Daniel Vetter):
 - Add stabel Cc and Fixes tag

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Cc: <stable@vger.kernel.org> # v5.6+
Fixes: 9e31c1fe45d5 ("drm/i915: Propagate errors on awaiting already signaled fences")
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Jon Bloomfield <jon.bloomfield@intel.com>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20210714193419.1459723-2-jason@jlekstrand.net
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoLinux 5.10.56
Greg Kroah-Hartman [Wed, 4 Aug 2021 10:46:45 +0000 (12:46 +0200)]
Linux 5.10.56

Link: https://lore.kernel.org/r/20210802134339.023067817@linuxfoundation.org
Tested-by: Fox Chen <foxhlchen@gmail.com>
Tested-by: Pavel Machek (CIP) <pavel@denx.de>
Tested-by: Linux Kernel Functional Testing <lkft@linaro.org>
Tested-by: Sudip Mukherjee <sudip.mukherjee@codethink.co.uk>
Tested-by: Rudi Heitbaum <rudi@heitbaum.com>
Tested-by: Jon Hunter <jonathanh@nvidia.com>
Tested-by: Guenter Roeck <linux@roeck-us.net>
Tested-by: Hulk Robot <hulkrobot@huawei.com>
Tested-by: Salvatore Bonaccorso <carnil@debian.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agocan: j1939: j1939_session_deactivate(): clarify lifetime of session object
Oleksij Rempel [Wed, 14 Jul 2021 11:16:02 +0000 (13:16 +0200)]
can: j1939: j1939_session_deactivate(): clarify lifetime of session object

commit 0c71437dd50dd687c15d8ca80b3b68f10bb21d63 upstream.

The j1939_session_deactivate() is decrementing the session ref-count and
potentially can free() the session. This would cause use-after-free
situation.

However, the code calling j1939_session_deactivate() does always hold
another reference to the session, so that it would not be free()ed in
this code path.

This patch adds a comment to make this clear and a WARN_ON, to ensure
that future changes will not violate this requirement. Further this
patch avoids dereferencing the session pointer as a precaution to avoid
use-after-free if the session is actually free()ed.

Fixes: 9d71dd0c7009 ("can: add support of SAE J1939 protocol")
Link: https://lore.kernel.org/r/20210714111602.24021-1-o.rempel@pengutronix.de
Reported-by: Xiaochen Zou <xzou017@ucr.edu>
Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agoi40e: Add additional info to PHY type error
Lukasz Cieplicki [Mon, 31 May 2021 16:55:49 +0000 (16:55 +0000)]
i40e: Add additional info to PHY type error

commit dc614c46178b0b89bde86ac54fc687a28580d2b7 upstream.

In case of PHY type error occurs, the message was too generic.
Add additional info to PHY type error indicating that it can be
wrong cable connected.

Fixes: 124ed15bf126 ("i40e: Add dual speed module support")
Signed-off-by: Lukasz Cieplicki <lukaszx.cieplicki@intel.com>
Signed-off-by: Michal Maloszewski <michal.maloszewski@intel.com>
Tested-by: Tony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agoRevert "perf map: Fix dso->nsinfo refcounting"
Arnaldo Carvalho de Melo [Fri, 30 Jul 2021 21:26:22 +0000 (18:26 -0300)]
Revert "perf map: Fix dso->nsinfo refcounting"

commit 9bac1bd6e6d36459087a728a968e79e37ebcea1a upstream.

This makes 'perf top' abort in some cases, and the right fix will
involve surgery that is too much to do at this stage, so revert for now
and fix it in the next merge window.

This reverts commit 2d6b74baa7147251c30a46c4996e8cc224aa2dc5.

Cc: Riccardo Mancini <rickyman7@gmail.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Krister Johansen <kjlx@templeofstupid.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agopowerpc/pseries: Fix regression while building external modules
Srikar Dronamraju [Thu, 29 Jul 2021 06:04:49 +0000 (11:34 +0530)]
powerpc/pseries: Fix regression while building external modules

commit 333cf507465fbebb3727f5b53e77538467df312a upstream.

With commit c9f3401313a5 ("powerpc: Always enable queued spinlocks for
64s, disable for others") CONFIG_PPC_QUEUED_SPINLOCKS is always
enabled on ppc64le, external modules that use spinlock APIs are
failing.

  ERROR: modpost: GPL-incompatible module XXX.ko uses GPL-only symbol 'shared_processor'

Before the above commit, modules were able to build without any
issues. Also this problem is not seen on other architectures. This
problem can be workaround if CONFIG_UNINLINE_SPIN_UNLOCK is enabled in
the config. However CONFIG_UNINLINE_SPIN_UNLOCK is not enabled by
default and only enabled in certain conditions like
CONFIG_DEBUG_SPINLOCKS is set in the kernel config.

  #include <linux/module.h>
  spinlock_t spLock;

  static int __init spinlock_test_init(void)
  {
          spin_lock_init(&spLock);
          spin_lock(&spLock);
          spin_unlock(&spLock);
          return 0;
  }

  static void __exit spinlock_test_exit(void)
  {
   printk("spinlock_test unloaded\n");
  }
  module_init(spinlock_test_init);
  module_exit(spinlock_test_exit);

  MODULE_DESCRIPTION ("spinlock_test");
  MODULE_LICENSE ("non-GPL");
  MODULE_AUTHOR ("Srikar Dronamraju");

Given that spin locks are one of the basic facilities for module code,
this effectively makes it impossible to build/load almost any non GPL
modules on ppc64le.

This was first reported at https://github.com/openzfs/zfs/issues/11172

Currently shared_processor is exported as GPL only symbol.
Fix this for parity with other architectures by exposing
shared_processor to non-GPL modules too.

Fixes: 14c73bd344da ("powerpc/vcpu: Assume dedicated processors as non-preempt")
Cc: stable@vger.kernel.org # v5.5+
Reported-by: marc.c.dionne@gmail.com
Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210729060449.292780-1-srikar@linux.vnet.ibm.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agoSMB3: fix readpage for large swap cache
Steve French [Fri, 23 Jul 2021 23:35:15 +0000 (18:35 -0500)]
SMB3: fix readpage for large swap cache

commit f2a26a3cff27dfa456fef386fe5df56dcb4b47b6 upstream.

readpage was calculating the offset of the page incorrectly
for the case of large swapcaches.

    loff_t offset = (loff_t)page->index << PAGE_SHIFT;

As pointed out by Matthew Wilcox, this needs to use
page_file_offset() to calculate the offset instead.
Pages coming from the swap cache have page->index set
to their index within the swapcache, not within the backing
file.  For a sufficiently large swapcache, we could have
overlapping values of page->index within the same backing file.

Suggested by: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: <stable@vger.kernel.org> # v5.7+
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agobpf: Fix pointer arithmetic mask tightening under state pruning
Daniel Borkmann [Fri, 16 Jul 2021 09:18:21 +0000 (09:18 +0000)]
bpf: Fix pointer arithmetic mask tightening under state pruning

commit e042aa532c84d18ff13291d00620502ce7a38dda upstream.

In 7fedb63a8307 ("bpf: Tighten speculative pointer arithmetic mask") we
narrowed the offset mask for unprivileged pointer arithmetic in order to
mitigate a corner case where in the speculative domain it is possible to
advance, for example, the map value pointer by up to value_size-1 out-of-
bounds in order to leak kernel memory via side-channel to user space.

The verifier's state pruning for scalars leaves one corner case open
where in the first verification path R_x holds an unknown scalar with an
aux->alu_limit of e.g. 7, and in a second verification path that same
register R_x, here denoted as R_x', holds an unknown scalar which has
tighter bounds and would thus satisfy range_within(R_x, R_x') as well as
tnum_in(R_x, R_x') for state pruning, yielding an aux->alu_limit of 3:
Given the second path fits the register constraints for pruning, the final
generated mask from aux->alu_limit will remain at 7. While technically
not wrong for the non-speculative domain, it would however be possible
to craft similar cases where the mask would be too wide as in 7fedb63a8307.

One way to fix it is to detect the presence of unknown scalar map pointer
arithmetic and force a deeper search on unknown scalars to ensure that
we do not run into a masking mismatch.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agobpf: verifier: Allocate idmap scratch in verifier env
Lorenz Bauer [Thu, 29 Apr 2021 13:46:56 +0000 (14:46 +0100)]
bpf: verifier: Allocate idmap scratch in verifier env

commit c9e73e3d2b1eb1ea7ff068e05007eec3bd8ef1c9 upstream.

func_states_equal makes a very short lived allocation for idmap,
probably because it's too large to fit on the stack. However the
function is called quite often, leading to a lot of alloc / free
churn. Replace the temporary allocation with dedicated scratch
space in struct bpf_verifier_env.

Signed-off-by: Lorenz Bauer <lmb@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Edward Cree <ecree.xilinx@gmail.com>
Link: https://lore.kernel.org/bpf/20210429134656.122225-4-lmb@cloudflare.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agobpf: Remove superfluous aux sanitation on subprog rejection
Daniel Borkmann [Tue, 29 Jun 2021 09:39:15 +0000 (09:39 +0000)]
bpf: Remove superfluous aux sanitation on subprog rejection

commit 59089a189e3adde4cf85f2ce479738d1ae4c514d upstream.

Follow-up to fe9a5ca7e370 ("bpf: Do not mark insn as seen under speculative
path verification"). The sanitize_insn_aux_data() helper does not serve a
particular purpose in today's code. The original intention for the helper
was that if function-by-function verification fails, a given program would
be cleared from temporary insn_aux_data[], and then its verification would
be re-attempted in the context of the main program a second time.

However, a failure in do_check_subprogs() will skip do_check_main() and
propagate the error to the user instead, thus such situation can never occur.
Given its interaction is not compatible to the Spectre v1 mitigation (due to
comparing aux->seen with env->pass_cnt), just remove sanitize_insn_aux_data()
to avoid future bugs in this area.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agobpf: Fix leakage due to insufficient speculative store bypass mitigation
Daniel Borkmann [Tue, 13 Jul 2021 08:18:31 +0000 (08:18 +0000)]
bpf: Fix leakage due to insufficient speculative store bypass mitigation

[ Upstream commit 2039f26f3aca5b0e419b98f65dd36481337b86ee ]

Spectre v4 gadgets make use of memory disambiguation, which is a set of
techniques that execute memory access instructions, that is, loads and
stores, out of program order; Intel's optimization manual, section 2.4.4.5:

  A load instruction micro-op may depend on a preceding store. Many
  microarchitectures block loads until all preceding store addresses are
  known. The memory disambiguator predicts which loads will not depend on
  any previous stores. When the disambiguator predicts that a load does
  not have such a dependency, the load takes its data from the L1 data
  cache. Eventually, the prediction is verified. If an actual conflict is
  detected, the load and all succeeding instructions are re-executed.

af86ca4e3088 ("bpf: Prevent memory disambiguation attack") tried to mitigate
this attack by sanitizing the memory locations through preemptive "fast"
(low latency) stores of zero prior to the actual "slow" (high latency) store
of a pointer value such that upon dependency misprediction the CPU then
speculatively executes the load of the pointer value and retrieves the zero
value instead of the attacker controlled scalar value previously stored at
that location, meaning, subsequent access in the speculative domain is then
redirected to the "zero page".

The sanitized preemptive store of zero prior to the actual "slow" store is
done through a simple ST instruction based on r10 (frame pointer) with
relative offset to the stack location that the verifier has been tracking
on the original used register for STX, which does not have to be r10. Thus,
there are no memory dependencies for this store, since it's only using r10
and immediate constant of zero; hence af86ca4e3088 /assumed/ a low latency
operation.

However, a recent attack demonstrated that this mitigation is not sufficient
since the preemptive store of zero could also be turned into a "slow" store
and is thus bypassed as well:

  [...]
  // r2 = oob address (e.g. scalar)
  // r7 = pointer to map value
  31: (7b) *(u64 *)(r10 -16) = r2
  // r9 will remain "fast" register, r10 will become "slow" register below
  32: (bf) r9 = r10
  // JIT maps BPF reg to x86 reg:
  //  r9  -> r15 (callee saved)
  //  r10 -> rbp
  // train store forward prediction to break dependency link between both r9
  // and r10 by evicting them from the predictor's LRU table.
  33: (61) r0 = *(u32 *)(r7 +24576)
  34: (63) *(u32 *)(r7 +29696) = r0
  35: (61) r0 = *(u32 *)(r7 +24580)
  36: (63) *(u32 *)(r7 +29700) = r0
  37: (61) r0 = *(u32 *)(r7 +24584)
  38: (63) *(u32 *)(r7 +29704) = r0
  39: (61) r0 = *(u32 *)(r7 +24588)
  40: (63) *(u32 *)(r7 +29708) = r0
  [...]
  543: (61) r0 = *(u32 *)(r7 +25596)
  544: (63) *(u32 *)(r7 +30716) = r0
  // prepare call to bpf_ringbuf_output() helper. the latter will cause rbp
  // to spill to stack memory while r13/r14/r15 (all callee saved regs) remain
  // in hardware registers. rbp becomes slow due to push/pop latency. below is
  // disasm of bpf_ringbuf_output() helper for better visual context:
  //
  // ffffffff8117ee20: 41 54                 push   r12
  // ffffffff8117ee22: 55                    push   rbp
  // ffffffff8117ee23: 53                    push   rbx
  // ffffffff8117ee24: 48 f7 c1 fc ff ff ff  test   rcx,0xfffffffffffffffc
  // ffffffff8117ee2b: 0f 85 af 00 00 00     jne    ffffffff8117eee0 <-- jump taken
  // [...]
  // ffffffff8117eee0: 49 c7 c4 ea ff ff ff  mov    r12,0xffffffffffffffea
  // ffffffff8117eee7: 5b                    pop    rbx
  // ffffffff8117eee8: 5d                    pop    rbp
  // ffffffff8117eee9: 4c 89 e0              mov    rax,r12
  // ffffffff8117eeec: 41 5c                 pop    r12
  // ffffffff8117eeee: c3                    ret
  545: (18) r1 = map[id:4]
  547: (bf) r2 = r7
  548: (b7) r3 = 0
  549: (b7) r4 = 4
  550: (85) call bpf_ringbuf_output#194288
  // instruction 551 inserted by verifier    \
  551: (7a) *(u64 *)(r10 -16) = 0            | /both/ are now slow stores here
  // storing map value pointer r7 at fp-16   | since value of r10 is "slow".
  552: (7b) *(u64 *)(r10 -16) = r7           /
  // following "fast" read to the same memory location, but due to dependency
  // misprediction it will speculatively execute before insn 551/552 completes.
  553: (79) r2 = *(u64 *)(r9 -16)
  // in speculative domain contains attacker controlled r2. in non-speculative
  // domain this contains r7, and thus accesses r7 +0 below.
  554: (71) r3 = *(u8 *)(r2 +0)
  // leak r3

As can be seen, the current speculative store bypass mitigation which the
verifier inserts at line 551 is insufficient since /both/, the write of
the zero sanitation as well as the map value pointer are a high latency
instruction due to prior memory access via push/pop of r10 (rbp) in contrast
to the low latency read in line 553 as r9 (r15) which stays in hardware
registers. Thus, architecturally, fp-16 is r7, however, microarchitecturally,
fp-16 can still be r2.

Initial thoughts to address this issue was to track spilled pointer loads
from stack and enforce their load via LDX through r10 as well so that /both/
the preemptive store of zero /as well as/ the load use the /same/ register
such that a dependency is created between the store and load. However, this
option is not sufficient either since it can be bypassed as well under
speculation. An updated attack with pointer spill/fills now _all_ based on
r10 would look as follows:

  [...]
  // r2 = oob address (e.g. scalar)
  // r7 = pointer to map value
  [...]
  // longer store forward prediction training sequence than before.
  2062: (61) r0 = *(u32 *)(r7 +25588)
  2063: (63) *(u32 *)(r7 +30708) = r0
  2064: (61) r0 = *(u32 *)(r7 +25592)
  2065: (63) *(u32 *)(r7 +30712) = r0
  2066: (61) r0 = *(u32 *)(r7 +25596)
  2067: (63) *(u32 *)(r7 +30716) = r0
  // store the speculative load address (scalar) this time after the store
  // forward prediction training.
  2068: (7b) *(u64 *)(r10 -16) = r2
  // preoccupy the CPU store port by running sequence of dummy stores.
  2069: (63) *(u32 *)(r7 +29696) = r0
  2070: (63) *(u32 *)(r7 +29700) = r0
  2071: (63) *(u32 *)(r7 +29704) = r0
  2072: (63) *(u32 *)(r7 +29708) = r0
  2073: (63) *(u32 *)(r7 +29712) = r0
  2074: (63) *(u32 *)(r7 +29716) = r0
  2075: (63) *(u32 *)(r7 +29720) = r0
  2076: (63) *(u32 *)(r7 +29724) = r0
  2077: (63) *(u32 *)(r7 +29728) = r0
  2078: (63) *(u32 *)(r7 +29732) = r0
  2079: (63) *(u32 *)(r7 +29736) = r0
  2080: (63) *(u32 *)(r7 +29740) = r0
  2081: (63) *(u32 *)(r7 +29744) = r0
  2082: (63) *(u32 *)(r7 +29748) = r0
  2083: (63) *(u32 *)(r7 +29752) = r0
  2084: (63) *(u32 *)(r7 +29756) = r0
  2085: (63) *(u32 *)(r7 +29760) = r0
  2086: (63) *(u32 *)(r7 +29764) = r0
  2087: (63) *(u32 *)(r7 +29768) = r0
  2088: (63) *(u32 *)(r7 +29772) = r0
  2089: (63) *(u32 *)(r7 +29776) = r0
  2090: (63) *(u32 *)(r7 +29780) = r0
  2091: (63) *(u32 *)(r7 +29784) = r0
  2092: (63) *(u32 *)(r7 +29788) = r0
  2093: (63) *(u32 *)(r7 +29792) = r0
  2094: (63) *(u32 *)(r7 +29796) = r0
  2095: (63) *(u32 *)(r7 +29800) = r0
  2096: (63) *(u32 *)(r7 +29804) = r0
  2097: (63) *(u32 *)(r7 +29808) = r0
  2098: (63) *(u32 *)(r7 +29812) = r0
  // overwrite scalar with dummy pointer; same as before, also including the
  // sanitation store with 0 from the current mitigation by the verifier.
  2099: (7a) *(u64 *)(r10 -16) = 0         | /both/ are now slow stores here
  2100: (7b) *(u64 *)(r10 -16) = r7        | since store unit is still busy.
  // load from stack intended to bypass stores.
  2101: (79) r2 = *(u64 *)(r10 -16)
  2102: (71) r3 = *(u8 *)(r2 +0)
  // leak r3
  [...]

Looking at the CPU microarchitecture, the scheduler might issue loads (such
as seen in line 2101) before stores (line 2099,2100) because the load execution
units become available while the store execution unit is still busy with the
sequence of dummy stores (line 2069-2098). And so the load may use the prior
stored scalar from r2 at address r10 -16 for speculation. The updated attack
may work less reliable on CPU microarchitectures where loads and stores share
execution resources.

This concludes that the sanitizing with zero stores from af86ca4e3088 ("bpf:
Prevent memory disambiguation attack") is insufficient. Moreover, the detection
of stack reuse from af86ca4e3088 where previously data (STACK_MISC) has been
written to a given stack slot where a pointer value is now to be stored does
not have sufficient coverage as precondition for the mitigation either; for
several reasons outlined as follows:

 1) Stack content from prior program runs could still be preserved and is
    therefore not "random", best example is to split a speculative store
    bypass attack between tail calls, program A would prepare and store the
    oob address at a given stack slot and then tail call into program B which
    does the "slow" store of a pointer to the stack with subsequent "fast"
    read. From program B PoV such stack slot type is STACK_INVALID, and
    therefore also must be subject to mitigation.

 2) The STACK_SPILL must not be coupled to register_is_const(&stack->spilled_ptr)
    condition, for example, the previous content of that memory location could
    also be a pointer to map or map value. Without the fix, a speculative
    store bypass is not mitigated in such precondition and can then lead to
    a type confusion in the speculative domain leaking kernel memory near
    these pointer types.

While brainstorming on various alternative mitigation possibilities, we also
stumbled upon a retrospective from Chrome developers [0]:

  [...] For variant 4, we implemented a mitigation to zero the unused memory
  of the heap prior to allocation, which cost about 1% when done concurrently
  and 4% for scavenging. Variant 4 defeats everything we could think of. We
  explored more mitigations for variant 4 but the threat proved to be more
  pervasive and dangerous than we anticipated. For example, stack slots used
  by the register allocator in the optimizing compiler could be subject to
  type confusion, leading to pointer crafting. Mitigating type confusion for
  stack slots alone would have required a complete redesign of the backend of
  the optimizing compiler, perhaps man years of work, without a guarantee of
  completeness. [...]

From BPF side, the problem space is reduced, however, options are rather
limited. One idea that has been explored was to xor-obfuscate pointer spills
to the BPF stack:

  [...]
  // preoccupy the CPU store port by running sequence of dummy stores.
  [...]
  2106: (63) *(u32 *)(r7 +29796) = r0
  2107: (63) *(u32 *)(r7 +29800) = r0
  2108: (63) *(u32 *)(r7 +29804) = r0
  2109: (63) *(u32 *)(r7 +29808) = r0
  2110: (63) *(u32 *)(r7 +29812) = r0
  // overwrite scalar with dummy pointer; xored with random 'secret' value
  // of 943576462 before store ...
  2111: (b4) w11 = 943576462
  2112: (af) r11 ^= r7
  2113: (7b) *(u64 *)(r10 -16) = r11
  2114: (79) r11 = *(u64 *)(r10 -16)
  2115: (b4) w2 = 943576462
  2116: (af) r2 ^= r11
  // ... and restored with the same 'secret' value with the help of AX reg.
  2117: (71) r3 = *(u8 *)(r2 +0)
  [...]

While the above would not prevent speculation, it would make data leakage
infeasible by directing it to random locations. In order to be effective
and prevent type confusion under speculation, such random secret would have
to be regenerated for each store. The additional complexity involved for a
tracking mechanism that prevents jumps such that restoring spilled pointers
would not get corrupted is not worth the gain for unprivileged. Hence, the
fix in here eventually opted for emitting a non-public BPF_ST | BPF_NOSPEC
instruction which the x86 JIT translates into a lfence opcode. Inserting the
latter in between the store and load instruction is one of the mitigations
options [1]. The x86 instruction manual notes:

  [...] An LFENCE that follows an instruction that stores to memory might
  complete before the data being stored have become globally visible. [...]

The latter meaning that the preceding store instruction finished execution
and the store is at minimum guaranteed to be in the CPU's store queue, but
it's not guaranteed to be in that CPU's L1 cache at that point (globally
visible). The latter would only be guaranteed via sfence. So the load which
is guaranteed to execute after the lfence for that local CPU would have to
rely on store-to-load forwarding. [2], in section 2.3 on store buffers says:

  [...] For every store operation that is added to the ROB, an entry is
  allocated in the store buffer. This entry requires both the virtual and
  physical address of the target. Only if there is no free entry in the store
  buffer, the frontend stalls until there is an empty slot available in the
  store buffer again. Otherwise, the CPU can immediately continue adding
  subsequent instructions to the ROB and execute them out of order. On Intel
  CPUs, the store buffer has up to 56 entries. [...]

One small upside on the fix is that it lifts constraints from af86ca4e3088
where the sanitize_stack_off relative to r10 must be the same when coming
from different paths. The BPF_ST | BPF_NOSPEC gets emitted after a BPF_STX
or BPF_ST instruction. This happens either when we store a pointer or data
value to the BPF stack for the first time, or upon later pointer spills.
The former needs to be enforced since otherwise stale stack data could be
leaked under speculation as outlined earlier. For non-x86 JITs the BPF_ST |
BPF_NOSPEC mapping is currently optimized away, but others could emit a
speculation barrier as well if necessary. For real-world unprivileged
programs e.g. generated by LLVM, pointer spill/fill is only generated upon
register pressure and LLVM only tries to do that for pointers which are not
used often. The program main impact will be the initial BPF_ST | BPF_NOSPEC
sanitation for the STACK_INVALID case when the first write to a stack slot
occurs e.g. upon map lookup. In future we might refine ways to mitigate
the latter cost.

  [0] https://arxiv.org/pdf/1902.05178.pdf
  [1] https://msrc-blog.microsoft.com/2018/05/21/analysis-and-mitigation-of-speculative-store-bypass-cve-2018-3639/
  [2] https://arxiv.org/pdf/1905.05725.pdf

Fixes: af86ca4e3088 ("bpf: Prevent memory disambiguation attack")
Fixes: f7cf25b2026d ("bpf: track spill/fill of constants")
Co-developed-by: Piotr Krysiuk <piotras@gmail.com>
Co-developed-by: Benedict Schlueter <benedict.schlueter@rub.de>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Piotr Krysiuk <piotras@gmail.com>
Signed-off-by: Benedict Schlueter <benedict.schlueter@rub.de>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agobpf: Introduce BPF nospec instruction for mitigating Spectre v4
Daniel Borkmann [Tue, 13 Jul 2021 08:18:31 +0000 (08:18 +0000)]
bpf: Introduce BPF nospec instruction for mitigating Spectre v4

[ Upstream commit f5e81d1117501546b7be050c5fbafa6efd2c722c ]

In case of JITs, each of the JIT backends compiles the BPF nospec instruction
/either/ to a machine instruction which emits a speculation barrier /or/ to
/no/ machine instruction in case the underlying architecture is not affected
by Speculative Store Bypass or has different mitigations in place already.

This covers both x86 and (implicitly) arm64: In case of x86, we use 'lfence'
instruction for mitigation. In case of arm64, we rely on the firmware mitigation
as controlled via the ssbd kernel parameter. Whenever the mitigation is enabled,
it works for all of the kernel code with no need to provide any additional
instructions here (hence only comment in arm64 JIT). Other archs can follow
as needed. The BPF nospec instruction is specifically targeting Spectre v4
since i) we don't use a serialization barrier for the Spectre v1 case, and
ii) mitigation instructions for v1 and v4 might be different on some archs.

The BPF nospec is required for a future commit, where the BPF verifier does
annotate intermediate BPF programs with speculation barriers.

Co-developed-by: Piotr Krysiuk <piotras@gmail.com>
Co-developed-by: Benedict Schlueter <benedict.schlueter@rub.de>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Piotr Krysiuk <piotras@gmail.com>
Signed-off-by: Benedict Schlueter <benedict.schlueter@rub.de>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agocan: hi311x: fix a signedness bug in hi3110_cmd()
Dan Carpenter [Thu, 29 Jul 2021 14:12:46 +0000 (17:12 +0300)]
can: hi311x: fix a signedness bug in hi3110_cmd()

[ Upstream commit f6b3c7848e66e9046c8a79a5b88fd03461cc252b ]

The hi3110_cmd() is supposed to return zero on success and negative
error codes on failure, but it was accidentally declared as a u8 when
it needs to be an int type.

Fixes: 57e83fb9b746 ("can: hi311x: Add Holt HI-311x CAN driver")
Link: https://lore.kernel.org/r/20210729141246.GA1267@kili
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agosis900: Fix missing pci_disable_device() in probe and remove
Wang Hai [Wed, 28 Jul 2021 12:11:07 +0000 (20:11 +0800)]
sis900: Fix missing pci_disable_device() in probe and remove

[ Upstream commit 89fb62fde3b226f99b7015280cf132e2a7438edf ]

Replace pci_enable_device() with pcim_enable_device(),
pci_disable_device() and pci_release_regions() will be
called in release automatically.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Wang Hai <wanghai38@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agotulip: windbond-840: Fix missing pci_disable_device() in probe and remove
Wang Hai [Wed, 28 Jul 2021 07:43:13 +0000 (15:43 +0800)]
tulip: windbond-840: Fix missing pci_disable_device() in probe and remove

[ Upstream commit 76a16be07b209a3f507c72abe823bd3af1c8661a ]

Replace pci_enable_device() with pcim_enable_device(),
pci_disable_device() and pci_release_regions() will be
called in release automatically.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Wang Hai <wanghai38@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agosctp: fix return value check in __sctp_rcv_asconf_lookup
Marcelo Ricardo Leitner [Wed, 28 Jul 2021 02:40:54 +0000 (23:40 -0300)]
sctp: fix return value check in __sctp_rcv_asconf_lookup

[ Upstream commit 557fb5862c9272ad9b21407afe1da8acfd9b53eb ]

As Ben Hutchings noticed, this check should have been inverted: the call
returns true in case of success.

Reported-by: Ben Hutchings <ben@decadent.org.uk>
Fixes: 0c5dc070ff3d ("sctp: validate from_addr_param return")
Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Reviewed-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agonet/mlx5e: Fix nullptr in mlx5e_hairpin_get_mdev()
Dima Chumak [Mon, 26 Apr 2021 12:16:26 +0000 (15:16 +0300)]
net/mlx5e: Fix nullptr in mlx5e_hairpin_get_mdev()

[ Upstream commit b1c2f6312c5005c928a72e668bf305a589d828d4 ]

The result of __dev_get_by_index() is not checked for NULL and then gets
dereferenced immediately.

Also, __dev_get_by_index() must be called while holding either RTNL lock
or @dev_base_lock, which isn't satisfied by mlx5e_hairpin_get_mdev() or
its callers. This makes the underlying hlist_for_each_entry() loop not
safe, and can have adverse effects in itself.

Fix by using dev_get_by_index() and handling nullptr return value when
ifindex device is not found. Update mlx5e_hairpin_get_mdev() callers to
check for possible PTR_ERR() result.

Fixes: 77ab67b7f0f9 ("net/mlx5e: Basic setup of hairpin object")
Addresses-Coverity: ("Dereference null return value")
Signed-off-by: Dima Chumak <dchumak@nvidia.com>
Reviewed-by: Vlad Buslov <vladbu@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agonet/mlx5: Fix flow table chaining
Maor Gottlieb [Mon, 26 Jul 2021 06:20:14 +0000 (09:20 +0300)]
net/mlx5: Fix flow table chaining

[ Upstream commit 8b54874ef1617185048029a3083d510569e93751 ]

Fix a bug when flow table is created in priority that already
has other flow tables as shown in the below diagram.
If the new flow table (FT-B) has the lowest level in the priority,
we need to connect the flow tables from the previous priority (p0)
to this new table. In addition when this flow table is destroyed
(FT-B), we need to connect the flow tables from the previous
priority (p0) to the next level flow table (FT-C) in the same
priority of the destroyed table (if exists).

                       ---------
                       |root_ns|
                       ---------
                            |
            --------------------------------
            |               |              |
       ----------      ----------      ---------
       |p(prio)-x|     |   p-y  |      |   p-n |
       ----------      ----------      ---------
            |               |
     ----------------  ------------------
     |ns(e.g bypass)|  |ns(e.g. kernel) |
     ----------------  ------------------
            |            |           |
-------        ------       ----
        |  p0 |        | p1 |       |p2|
        -------        ------       ----
           |             |    \
        --------       ------- ------
        | FT-A |       |FT-B | |FT-C|
        --------       ------- ------

Fixes: f90edfd279f3 ("net/mlx5_core: Connect flow tables")
Signed-off-by: Maor Gottlieb <maorg@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoskmsg: Make sk_psock_destroy() static
Cong Wang [Wed, 27 Jan 2021 22:15:01 +0000 (14:15 -0800)]
skmsg: Make sk_psock_destroy() static

[ Upstream commit 8063e184e49011f6f3f34f6c358dc8a83890bb5b ]

sk_psock_destroy() is a RCU callback, I can't see any reason why
it could be used outside.

Signed-off-by: Cong Wang <cong.wang@bytedance.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jakub Sitnicki <jakub@cloudflare.com>
Cc: Lorenz Bauer <lmb@cloudflare.com>
Link: https://lore.kernel.org/bpf/20210127221501.46866-1-xiyou.wangcong@gmail.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agodrm/msm/dp: Initialize the INTF_CONFIG register
Bjorn Andersson [Thu, 22 Jul 2021 02:44:34 +0000 (19:44 -0700)]
drm/msm/dp: Initialize the INTF_CONFIG register

[ Upstream commit f9a39932fa54b6421e751ada7a285da809146421 ]

Some bootloaders set the widebus enable bit in the INTF_CONFIG register,
but configuration of widebus isn't yet supported ensure that the
register has a known value, with widebus disabled.

Fixes: c943b4948b58 ("drm/msm/dp: add displayPort driver support")
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Reviewed-by: Stephen Boyd <swboyd@chromium.org>
Link: https://lore.kernel.org/r/20210722024434.3313167-1-bjorn.andersson@linaro.org
Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agodrm/msm/dpu: Fix sm8250_mdp register length
Robert Foss [Mon, 28 Jun 2021 08:50:33 +0000 (10:50 +0200)]
drm/msm/dpu: Fix sm8250_mdp register length

[ Upstream commit b910a0206b59eb90ea8ff76d146f4c3156da61e9 ]

The downstream dts lists this value as 0x494, and not
0x45c.

Fixes: af776a3e1c30 ("drm/msm/dpu: add SM8250 to hw catalog")
Signed-off-by: Robert Foss <robert.foss@linaro.org>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
Reviewed-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@somainline.org>
Link: https://lore.kernel.org/r/20210628085033.9905-1-robert.foss@linaro.org
Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agonet: llc: fix skb_over_panic
Pavel Skripkin [Sat, 24 Jul 2021 21:11:59 +0000 (00:11 +0300)]
net: llc: fix skb_over_panic

[ Upstream commit c7c9d2102c9c098916ab9e0ab248006107d00d6c ]

Syzbot reported skb_over_panic() in llc_pdu_init_as_xid_cmd(). The
problem was in wrong LCC header manipulations.

Syzbot's reproducer tries to send XID packet. llc_ui_sendmsg() is
doing following steps:

1. skb allocation with size = len + header size
len is passed from userpace and header size
is 3 since addr->sllc_xid is set.

2. skb_reserve() for header_len = 3
3. filling all other space with memcpy_from_msg()

Ok, at this moment we have fully loaded skb, only headers needs to be
filled.

Then code comes to llc_sap_action_send_xid_c(). This function pushes 3
bytes for LLC PDU header and initializes it. Then comes
llc_pdu_init_as_xid_cmd(). It initalizes next 3 bytes *AFTER* LLC PDU
header and call skb_push(skb, 3). This looks wrong for 2 reasons:

1. Bytes rigth after LLC header are user data, so this function
   was overwriting payload.

2. skb_push(skb, 3) call can cause skb_over_panic() since
   all free space was filled in llc_ui_sendmsg(). (This can
   happen is user passed 686 len: 686 + 14 (eth header) + 3 (LLC
   header) = 703. SKB_DATA_ALIGN(703) = 704)

So, in this patch I added 2 new private constansts: LLC_PDU_TYPE_U_XID
and LLC_PDU_LEN_U_XID. LLC_PDU_LEN_U_XID is used to correctly reserve
header size to handle LLC + XID case. LLC_PDU_TYPE_U_XID is used by
llc_pdu_header_init() function to push 6 bytes instead of 3. And finally
I removed skb_push() call from llc_pdu_init_as_xid_cmd().

This changes should not affect other parts of LLC, since after
all steps we just transmit buffer.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Reported-and-tested-by: syzbot+5e5a981ad7cc54c4b2b4@syzkaller.appspotmail.com
Signed-off-by: Pavel Skripkin <paskripkin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoKVM: x86: Check the right feature bit for MSR_KVM_ASYNC_PF_ACK access
Vitaly Kuznetsov [Thu, 22 Jul 2021 12:30:18 +0000 (14:30 +0200)]
KVM: x86: Check the right feature bit for MSR_KVM_ASYNC_PF_ACK access

[ Upstream commit 0a31df6823232516f61f174907e444f710941dfe ]

MSR_KVM_ASYNC_PF_ACK MSR is part of interrupt based asynchronous page fault
interface and not the original (deprecated) KVM_FEATURE_ASYNC_PF. This is
stated in Documentation/virt/kvm/msr.rst.

Fixes: 66570e966dd9 ("kvm: x86: only provide PV features if enabled in guest's CPUID")
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Reviewed-by: Oliver Upton <oupton@google.com>
Message-Id: <20210722123018.260035-1-vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agomlx4: Fix missing error code in mlx4_load_one()
Jiapeng Chong [Fri, 23 Jul 2021 10:36:09 +0000 (18:36 +0800)]
mlx4: Fix missing error code in mlx4_load_one()

[ Upstream commit 7e4960b3d66d7248b23de3251118147812b42da2 ]

The error code is missing in this code scenario, add the error code
'-EINVAL' to the return value 'err'.

Eliminate the follow smatch warning:

drivers/net/ethernet/mellanox/mlx4/main.c:3538 mlx4_load_one() warn:
missing error code 'err'.

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Fixes: 7ae0e400cd93 ("net/mlx4_core: Flexible (asymmetric) allocation of EQs and MSI-X vectors for PF/VFs")
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
3 years agoocteontx2-pf: Fix interface down flag on error
Geetha sowjanya [Sun, 25 Jul 2021 07:59:03 +0000 (13:29 +0530)]
octeontx2-pf: Fix interface down flag on error

[ Upstream commit 69f0aeb13bb548e2d5710a350116e03f0273302e ]

In the existing code while changing the number of TX/RX
queues using ethtool the PF/VF interface resources are
freed and reallocated (otx2_stop and otx2_open is called)
if the device is in running state. If any resource allocation
fails in otx2_open, driver free already allocated resources
and return. But again, when the number of queues changes
as the device state still running oxt2_stop is called.
In which we try to free already freed resources leading
to driver crash.
This patch fixes the issue by setting the INTF_DOWN flag on
error and free the resources in otx2_stop only if the flag is
not set.

Fixes: 50fe6c02e5ad ("octeontx2-pf: Register and handle link notifications")
Signed-off-by: Geetha sowjanya <gakula@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <Sunil.Goutham@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>