Non-ND strict packets with a source LLA go through the packet taps
again, while non-ND strict packets with other source addresses do not,
and we can see a clone of those packets on the vrf interface (we should
not). This is due to a series of changes:

Commit 6f12fa77[1] made non-ND strict packets not being pushed again
in the packet taps. This changed with commit 205704c6[2] for those
packets having a source LLA, as they need a lookup with the orig_iif.

The issue now is those packets do not skip the 'vrf_ip6_rcv' function to
the end (as the ones without a source LLA) and go through the check to
call packet taps again. This check was changed by commit 6f12fa77[1]
and do not exclude non-strict packets anymore. Packets matching
'need_strict && !is_ndisc && is_ll_src' are now being sent through the
packet taps again. This can be seen by dumping packets on the vrf
interface.

Fix this by having the same code path for all non-ND strict packets and
selectively lookup with the orig_iif for those with a source LLA. This
has the effect to revert to the pre-205704c6[2] condition, which
should also be easier to maintain.

[1] 6f12fa77 ("vrf: mark skb for multicast or link-local as enslaved to VRF")
[2] 205704c6 ("vrf: packets with lladdr src needs dst at input with orig_iif when needs strict")

Fixes: 205704c6 ("vrf: packets with lladdr src needs dst at input with orig_iif when needs strict")
Cc: Stephen Suryaputra <ssuryaextr@gmail.com>
Reported-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Antoine Tenart <atenart@kernel.org>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Pavel Skripkin says:

====================
net: ethernat: ezchip: bug fixing and code improvments

While manual code reviewing, I found some error in ezchip driver.
Two of them looks very dangerous:
  1. use-after-free in nps_enet_remove
      Accessing netdev private data after free_netdev()

  2. wrong error handling of platform_get_irq()
      It can cause passing negative irq to request_irq()

Also, in 2nd patch I removed redundant check to increase execution
speed and make code more straightforward.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

As documented at drivers/base/platform.c for platform_get_irq:

 * Gets an IRQ for a platform device and prints an error message if finding the
 * IRQ fails. Device drivers should check the return value for errors so as to
 * not pass a negative integer value to the request_irq() APIs.

So, the driver should check that platform_get_irq() return value
is _negative_, not that it's equal to zero, because -ENXIO (return
value from request_irq() if irq was not found) will
pass this check and it leads to passing negative irq to request_irq()

Fixes: 0dd07709 ("NET: Add ezchip ethernet driver")
Signed-off-by: Pavel Skripkin <paskripkin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

err varibale will be set everytime, when code gets
into this path. This check will just slowdown the execution
and that's all.

Fixes: 0dd07709 ("NET: Add ezchip ethernet driver")
Signed-off-by: Pavel Skripkin <paskripkin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

priv is netdev private data, but it is used
after free_netdev(). It can cause use-after-free when accessing priv
pointer. So, fix it by moving free_netdev() after netif_napi_del()
call.

Fixes: 0dd07709 ("NET: Add ezchip ethernet driver")
Signed-off-by: Pavel Skripkin <paskripkin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

static int greth_of_remove(struct platform_device *of_dev)
{
...
	struct greth_private *greth = netdev_priv(ndev);
...
	unregister_netdev(ndev);
	free_netdev(ndev);

	of_iounmap(&of_dev->resource[0], greth->regs, resource_size(&of_dev->resource[0]));
...
}

greth is netdev private data, but it is used
after free_netdev(). It can cause use-after-free when accessing greth
pointer. So, fix it by moving free_netdev() after of_iounmap()
call.

Fixes: d4c41139 ("net: Add Aeroflex Gaisler 10/100/1G Ethernet MAC driver")
Signed-off-by: Pavel Skripkin <paskripkin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Pull networking fixes from Jakub Kicinski:
 "Networking fixes for 5.13-rc7, including fixes from wireless, bpf,
  bluetooth, netfilter and can.

  Current release - regressions:

   - mlxsw: spectrum_qdisc: Pass handle, not band number to find_class()
     to fix modifying offloaded qdiscs

   - lantiq: net: fix duplicated skb in rx descriptor ring

   - rtnetlink: fix regression in bridge VLAN configuration, empty info
     is not an error, bot-generated "fix" was not needed

   - libbpf: s/rx/tx/ typo on umem->rx_ring_setup_done to fix umem
     creation

  Current release - new code bugs:

   - ethtool: fix NULL pointer dereference during module EEPROM dump via
     the new netlink API

   - mlx5e: don't update netdev RQs with PTP-RQ, the special purpose
     queue should not be visible to the stack

   - mlx5e: select special PTP queue only for SKBTX_HW_TSTAMP skbs

   - mlx5e: verify dev is present in get devlink port ndo, avoid a panic

  Previous releases - regressions:

   - neighbour: allow NUD_NOARP entries to be force GCed

   - further fixes for fallout from reorg of WiFi locking (staging:
     rtl8723bs, mac80211, cfg80211)

   - skbuff: fix incorrect msg_zerocopy copy notifications

   - mac80211: fix NULL ptr deref for injected rate info

   - Revert "net/mlx5: Arm only EQs with EQEs" it may cause missed IRQs

  Previous releases - always broken:

   - bpf: more speculative execution fixes

   - netfilter: nft_fib_ipv6: skip ipv6 packets from any to link-local

   - udp: fix race between close() and udp_abort() resulting in a panic

   - fix out of bounds when parsing TCP options before packets are
     validated (in netfilter: synproxy, tc: sch_cake and mptcp)

   - mptcp: improve operation under memory pressure, add missing
     wake-ups

   - mptcp: fix double-lock/soft lookup in subflow_error_report()

   - bridge: fix races (null pointer deref and UAF) in vlan tunnel
     egress

   - ena: fix DMA mapping function issues in XDP

   - rds: fix memory leak in rds_recvmsg

  Misc:

   - vrf: allow larger MTUs

   - icmp: don't send out ICMP messages with a source address of 0.0.0.0

   - cdc_ncm: switch to eth%d interface naming"

* tag 'net-5.13-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (139 commits)
  net: ethernet: fix potential use-after-free in ec_bhf_remove
  selftests/net: Add icmp.sh for testing ICMP dummy address responses
  icmp: don't send out ICMP messages with a source address of 0.0.0.0
  net: ll_temac: Avoid ndo_start_xmit returning NETDEV_TX_BUSY
  net: ll_temac: Fix TX BD buffer overwrite
  net: ll_temac: Add memory-barriers for TX BD access
  net: ll_temac: Make sure to free skb when it is completely used
  MAINTAINERS: add Guvenc as SMC maintainer
  bnxt_en: Call bnxt_ethtool_free() in bnxt_init_one() error path
  bnxt_en: Fix TQM fastpath ring backing store computation
  bnxt_en: Rediscover PHY capabilities after firmware reset
  cxgb4: fix wrong shift.
  mac80211: handle various extensible elements correctly
  mac80211: reset profile_periodicity/ema_ap
  cfg80211: avoid double free of PMSR request
  cfg80211: make certificate generation more robust
  mac80211: minstrel_ht: fix sample time check
  net: qed: Fix memcpy() overflow of qed_dcbx_params()
  net: cdc_eem: fix tx fixup skb leak
  net: hamradio: fix memory leak in mkiss_close
  ...

Pull btrfs fix from David Sterba:
 "One more fix, for a space accounting bug in zoned mode. It happens
  when a block group is switched back rw->ro and unusable bytes (due to
  zoned constraints) are subtracted twice.

  It has user visible effects so I consider it important enough for late
  -rc inclusion and backport to stable"

* tag 'for-5.13-rc6-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  btrfs: zoned: fix negative space_info->bytes_readonly

Pull PCI fixes from Bjorn Helgaas:

 - Clear 64-bit flag for host bridge windows below 4GB to fix a resource
   allocation regression added in -rc1 (Punit Agrawal)

 - Fix tegra194 MCFG quirk build regressions added in -rc1 (Jon Hunter)

 - Avoid secondary bus resets on TI KeyStone C667X devices (Antti
   Järvinen)

 - Avoid secondary bus resets on some NVIDIA GPUs (Shanker Donthineni)

 - Work around FLR erratum on Huawei Intelligent NIC VF (Chiqijun)

 - Avoid broken ATS on AMD Navi14 GPU (Evan Quan)

 - Trust Broadcom BCM57414 NIC to isolate functions even though it
   doesn't advertise ACS support (Sriharsha Basavapatna)

 - Work around AMD RS690 BIOSes that don't configure DMA above 4GB
   (Mikel Rychliski)

 - Fix panic during PIO transfer on Aardvark controller (Pali Rohár)

* tag 'pci-v5.13-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci:
  PCI: aardvark: Fix kernel panic during PIO transfer
  PCI: Add AMD RS690 quirk to enable 64-bit DMA
  PCI: Add ACS quirk for Broadcom BCM57414 NIC
  PCI: Mark AMD Navi14 GPU ATS as broken
  PCI: Work around Huawei Intelligent NIC VF FLR erratum
  PCI: Mark some NVIDIA GPUs to avoid bus reset
  PCI: Mark TI C667X to avoid bus reset
  PCI: tegra194: Fix MCFG quirk build regressions
  PCI: of: Clear 64-bit flag for non-prefetchable memory below 4GB

If a task is killed during a page fault, it does not currently call
sb_end_pagefault(), which means that the filesystem cannot be frozen
at any time thereafter.  This may be reported by lockdep like this:

====================================
WARNING: fsstress/10757 still has locks held!
5.13.0-rc4-build4+ #91 Not tainted
------------------------------------
1 lock held by fsstress/10757:
 #0: ffff888104eac530
 (
sb_pagefaults

as filesystem freezing is modelled as a lock.

Fix this by removing all the direct returns from within the function,
and using 'ret' to indicate whether we were interrupted or successful.

Fixes: 1cf7a151 ("afs: Implement shared-writeable mmap")
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: David Howells <dhowells@redhat.com>
cc: linux-afs@lists.infradead.org
Link: https://lore.kernel.org/r/20210616154900.1958373-1-willy@infradead.org/Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

static void ec_bhf_remove(struct pci_dev *dev)
{
...
	struct ec_bhf_priv *priv = netdev_priv(net_dev);

	unregister_netdev(net_dev);
	free_netdev(net_dev);

	pci_iounmap(dev, priv->dma_io);
	pci_iounmap(dev, priv->io);
...
}

priv is netdev private data, but it is used
after free_netdev(). It can cause use-after-free when accessing priv
pointer. So, fix it by moving free_netdev() after pci_iounmap()
calls.

Fixes: 6af55ff5 ("Driver for Beckhoff CX5020 EtherCAT master module.")
Signed-off-by: Pavel Skripkin <paskripkin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge tag 'mac80211-for-net-2021-06-18' of git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211

Johannes Berg says:

====================
A couple of straggler fixes:
 * a minstrel HT sample check fix
 * peer measurement could double-free on races
 * certificate file generation at build time could
   sometimes hang
 * some parameters weren't reset between connections
   in mac80211
 * some extensible elements were treated as non-
   extensible, possibly causuing bad connections
   (or failures) if the AP adds data
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

This adds a new icmp.sh selftest for testing that the kernel will respond
correctly with an ICMP unreachable message with the dummy (192.0.0.8)
source address when there are no IPv4 addresses configured to use as source
addresses.
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

When constructing ICMP response messages, the kernel will try to pick a
suitable source address for the outgoing packet. However, if no IPv4
addresses are configured on the system at all, this will fail and we end up
producing an ICMP message with a source address of 0.0.0.0. This can happen
on a box routing IPv4 traffic via v6 nexthops, for instance.

Since 0.0.0.0 is not generally routable on the internet, there's a good
chance that such ICMP messages will never make it back to the sender of the
original packet that the ICMP message was sent in response to. This, in
turn, can create connectivity and PMTUd problems for senders. Fortunately,
RFC7600 reserves a dummy address to be used as a source for ICMP
messages (192.0.0.8/32), so let's teach the kernel to substitute that
address as a last resort if the regular source address selection procedure
fails.

Below is a quick example reproducing this issue with network namespaces:

ip netns add ns0
ip l add type veth peer netns ns0
ip l set dev veth0 up
ip a add 10.0.0.1/24 dev veth0
ip a add fc00:dead:cafe:42::1/64 dev veth0
ip r add 10.1.0.0/24 via inet6 fc00:dead:cafe:42::2
ip -n ns0 l set dev veth0 up
ip -n ns0 a add fc00:dead:cafe:42::2/64 dev veth0
ip -n ns0 r add 10.0.0.0/24 via inet6 fc00:dead:cafe:42::1
ip netns exec ns0 sysctl -w net.ipv4.icmp_ratelimit=0
ip netns exec ns0 sysctl -w net.ipv4.ip_forward=1
tcpdump -tpni veth0 -c 2 icmp &
ping -w 1 10.1.0.1 > /dev/null
tcpdump: verbose output suppressed, use -v[v]... for full protocol decode
listening on veth0, link-type EN10MB (Ethernet), snapshot length 262144 bytes
IP 10.0.0.1 > 10.1.0.1: ICMP echo request, id 29, seq 1, length 64
IP 0.0.0.0 > 10.0.0.1: ICMP net 10.1.0.1 unreachable, length 92
2 packets captured
2 packets received by filter
0 packets dropped by kernel

With this patch the above capture changes to:
IP 10.0.0.1 > 10.1.0.1: ICMP echo request, id 31127, seq 1, length 64
IP 192.0.0.8 > 10.0.0.1: ICMP net 10.1.0.1 unreachable, length 92

Fixes: 1da177e4 ("Linux-2.6.12-rc2")
Reported-by: Juliusz Chroboczek <jch@irif.fr>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

As documented in Documentation/networking/driver.rst, the ndo_start_xmit
method must not return NETDEV_TX_BUSY under any normal circumstances, and
as recommended, we simply stop the tx queue in advance, when there is a
risk that the next xmit would cause a NETDEV_TX_BUSY return.
Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Just as the initial check, we need to ensure num_frag+1 buffers available,
as that is the number of buffers we are going to use.

This fixes a buffer overflow, which might be seen during heavy network
load. Complete lockup of TEMAC was reproducible within about 10 minutes of
a particular load.

Fixes: 84823ff8 ("net: ll_temac: Fix race condition causing TX hang")
Cc: stable@vger.kernel.org # v5.4+
Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add a couple of memory-barriers to ensure correct ordering of read/write
access to TX BDs.

In xmit_done, we should ensure that reading the additional BD fields are
only done after STS_CTRL_APP0_CMPLT bit is set.

When xmit_done marks the BD as free by setting APP0=0, we need to ensure
that the other BD fields are reset first, so we avoid racing with the xmit
path, which writes to the same fields.

Finally, making sure to read APP0 of next BD after the current BD, ensures
that we see all available buffers.
Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

With the skb pointer piggy-backed on the TX BD, we have a simple and
efficient way to free the skb buffer when the frame has been transmitted.
But in order to avoid freeing the skb while there are still fragments from
the skb in use, we need to piggy-back on the TX BD of the skb, not the
first.

Without this, we are doing use-after-free on the DMA side, when the first
BD of a multi TX BD packet is seen as completed in xmit_done, and the
remaining BDs are still being processed.

Cc: stable@vger.kernel.org # v5.4+
Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add Guvenc as maintainer for Shared Memory Communications (SMC)
Sockets.

Cc: Julian Wiedmann <jwi@linux.ibm.com>
Acked-by: Guvenc Gulce <guvenc@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Michael Chan says:

====================
bnxt_en: Bug fixes

This patchset includes 3 small bug fixes to reinitialize PHY capabilities
after firmware reset, setup the chip's internal TQM fastpath ring
backing memory properly for RoCE traffic, and to free ethtool related
memory if driver probe fails.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

bnxt_ethtool_init() may have allocated some memory and we need to
call bnxt_ethtool_free() to properly unwind if bnxt_init_one()
fails.

Fixes: 7c380918 ("bnxt_en: Refactor bnxt_init_one() and turn on TPA support on 57500 chips.")
Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

TQM fastpath ring needs to be sized to store both the requester
and responder side of RoCE QPs in TQM for supporting bi-directional
tests.  Fix bnxt_alloc_ctx_mem() to multiply the RoCE QPs by a factor of
2 when computing the number of entries for TQM fastpath ring.  This
fixes an RX pipeline stall issue when running bi-directional max
RoCE QP tests.

Fixes: c7dd7ab4 ("bnxt_en: Improve TQM ring context memory sizing formulas.")
Signed-off-by: Rukhsana Ansari <rukhsana.ansari@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

There is a missing bnxt_probe_phy() call in bnxt_fw_init_one() to
rediscover the PHY capabilities after a firmware reset.  This can cause
some PHY related functionalities to fail after a firmware reset.  For
example, in multi-host, the ability for any host to configure the PHY
settings may be lost after a firmware reset.

Fixes: ec5d31e3 ("bnxt_en: Handle firmware reset status during IF_UP.")
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

While fixing coverity warning, commit dd2c7967 introduced typo in
shift value. Fix that.
Signed-off-by: Pavel Machek (CIP) <pavel@denx.de>
Fixes: dd2c7967 ("cxgb4: Fix unintentional sign extension issues")
Signed-off-by: David S. Miller <davem@davemloft.net>

Pull ARC fixes from Vineet Gupta:

 - ARCv2 userspace ABI not populating a few registers

 - Unbork CONFIG_HARDENED_USERCOPY for ARC

* tag 'arc-5.13-rc7-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupta/arc:
  ARC: fix CONFIG_HARDENED_USERCOPY
  ARCv2: save ABI registers across signal handling

Pull tracing fixes from Steven Rostedt:

 - Have recordmcount check for valid st_shndx otherwise some archs may
   have invalid references for the mcount location.

 - Two fixes done for mapping pids to task names. Traces were not
   showing the names of tasks when they should have.

 - Fix to trace_clock_global() to prevent it from going backwards

* tag 'trace-v5.13-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace:
  tracing: Do no increment trace_clock_global() by one
  tracing: Do not stop recording comms if the trace file is being read
  tracing: Do not stop recording cmdlines when tracing is off
  recordmcount: Correct st_shndx handling

Pull printk fixup from Petr Mladek:
 "Fix misplaced EXPORT_SYMBOL(vsprintf)"

* tag 'printk-for-5.13-fixup' of git://git.kernel.org/pub/scm/linux/kernel/git/printk/linux:
  printk: Move EXPORT_SYMBOL() closer to vprintk definition

Pull power management fix from Rafael Wysocki:
 "Remove recently added frequency invariance support from the CPPC
  cpufreq driver, because it has turned out to be problematic and it
  cannot be fixed properly on time for 5.13 (Viresh Kumar)"

* tag 'pm-5.13-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  Revert "cpufreq: CPPC: Add support for frequency invariance"

Pull USB fixes from Greg KH:
 "Here are three small USB fixes for reported problems for 5.13-rc7.
  They include:

   - disable autosuspend for a cypress USB hub

   - fix the battery charger detection for the chipidea driver

   - fix a kernel panic in the dwc3 driver due to a previous change in
     5.13-rc1.

  All have been in linux-next with no reported problems"

* tag 'usb-5.13-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb:
  usb: core: hub: Disable autosuspend for Cypress CY7C65632
  usb: chipidea: imx: Fix Battery Charger 1.2 CDP detection
  usb: dwc3: core: fix kernel panic when do reboot

Pull drm fixes from Dave Airlie:
 "Not much happening in fixes land this week only one PR for two amdgpu
  powergating fixes was waiting for me, maybe something will show up
  over the weekend, maybe not.

  amdgpu:

   - GFX9 and 10 powergating fixes"

* tag 'drm-fixes-2021-06-18' of git://anongit.freedesktop.org/drm/drm:
  drm/amdgpu/gfx10: enlarge CP_MEC_DOORBELL_RANGE_UPPER to cover full doorbell.
  drm/amdgpu/gfx9: fix the doorbell missing when in CGPG issue.

Trying to start a new PIO transfer by writing value 0 in PIO_START register
when previous transfer has not yet completed (which is indicated by value 1
in PIO_START) causes an External Abort on CPU, which results in kernel
panic:

    SError Interrupt on CPU0, code 0xbf000002 -- SError
    Kernel panic - not syncing: Asynchronous SError Interrupt

To prevent kernel panic, it is required to reject a new PIO transfer when
previous one has not finished yet.

If previous PIO transfer is not finished yet, the kernel may issue a new
PIO request only if the previous PIO transfer timed out.

In the past the root cause of this issue was incorrectly identified (as it
often happens during link retraining or after link down event) and special
hack was implemented in Trusted Firmware to catch all SError events in EL3,
to ignore errors with code 0xbf000002 and not forwarding any other errors
to kernel and instead throw panic from EL3 Trusted Firmware handler.

Links to discussion and patches about this issue:
https://git.trustedfirmware.org/TF-A/trusted-firmware-a.git/commit/?id=3c7dcdac5c50
https://lore.kernel.org/linux-pci/20190316161243.29517-1-repk@triplefau.lt/
https://lore.kernel.org/linux-pci/971be151d24312cc533989a64bd454b4@www.loen.fr/
https://review.trustedfirmware.org/c/TF-A/trusted-firmware-a/+/1541

But the real cause was the fact that during link retraining or after link
down event the PIO transfer may take longer time, up to the 1.44s until it
times out. This increased probability that a new PIO transfer would be
issued by kernel while previous one has not finished yet.

After applying this change into the kernel, it is possible to revert the
mentioned TF-A hack and SError events do not have to be caught in TF-A EL3.

Link: https://lore.kernel.org/r/20210608203655.31228-1-pali@kernel.orgSigned-off-by: Pali Rohár <pali@kernel.org>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Marek Behún <kabel@kernel.org>
Cc: stable@vger.kernel.org # 7fbcb5da ("PCI: aardvark: Don't rely on jiffies while holding spinlock")

Although the AMD RS690 chipset has 64-bit DMA support, BIOS implementations
sometimes fail to configure the memory limit registers correctly.

The Acer F690GVM mainboard uses this chipset and a Marvell 88E8056 NIC. The
sky2 driver programs the NIC to use 64-bit DMA, which will not work:

  sky2 0000:02:00.0: error interrupt status=0x8
  sky2 0000:02:00.0 eth0: tx timeout
  sky2 0000:02:00.0 eth0: transmit ring 0 .. 22 report=0 done=0

Other drivers required by this mainboard either don't support 64-bit DMA,
or have it disabled using driver specific quirks. For example, the ahci
driver has quirks to enable or disable 64-bit DMA depending on the BIOS
version (see ahci_sb600_enable_64bit() in ahci.c). This ahci quirk matches
against the SB600 SATA controller, but the real issue is almost certainly
with the RS690 PCI host that it was commonly attached to.

To avoid this issue in all drivers with 64-bit DMA support, fix the
configuration of the PCI host. If the kernel is aware of physical memory
above 4GB, but the BIOS never configured the PCI host with this
information, update the registers with our values.

[bhelgaas: drop PCI_DEVICE_ID_ATI_RS690 definition]
Link: https://lore.kernel.org/r/20210611214823.4898-1-mikel@mikelr.comSigned-off-by: Mikel Rychliski <mikel@mikelr.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>

The Broadcom BCM57414 NIC may be a multi-function device.  While it does
not advertise an ACS capability, peer-to-peer transactions are not possible
between the individual functions, so it is safe to treat them as fully
isolated.

Add an ACS quirk for this device so the functions can be in independent
IOMMU groups and attached individually to userspace applications using
VFIO.

[bhelgaas: commit log]
Link: https://lore.kernel.org/r/1621645997-16251-1-git-send-email-michael.chan@broadcom.comSigned-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Cc: stable@vger.kernel.org

Observed unexpected GPU hang during runpm stress test on 0x7341 rev 0x00.
Further debugging shows broken ATS is related.

Disable ATS on this part.  Similar issues on other devices:

  a2da5d8c ("PCI: Mark AMD Raven iGPU ATS as broken in some platforms")
  45beb31d ("PCI: Mark AMD Navi10 GPU rev 0x00 ATS as broken")
  5e89cd30 ("PCI: Mark AMD Navi14 GPU rev 0xc5 ATS as broken")
Suggested-by: Alex Deucher <alexander.deucher@amd.com>
Link: https://lore.kernel.org/r/20210602021255.939090-1-evan.quan@amd.comSigned-off-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Krzysztof Wilczyński <kw@linux.com>
Cc: stable@vger.kernel.org

pcie_flr() starts a Function Level Reset (FLR), waits 100ms (the maximum
time allowed for FLR completion by PCIe r5.0, sec 6.6.2), and waits for the
FLR to complete.  It assumes the FLR is complete when a config read returns
valid data.

When we do an FLR on several Huawei Intelligent NIC VFs at the same time,
firmware on the NIC processes them serially.  The VF may respond to config
reads before the firmware has completed its reset processing.  If we bind a
driver to the VF (e.g., by assigning the VF to a virtual machine) in the
interval between the successful config read and completion of the firmware
reset processing, the NIC VF driver may fail to load.

Prevent this driver failure by waiting for the NIC firmware to complete its
reset processing.  Not all NIC firmware supports this feature.

[bhelgaas: commit log]
Link: https://support.huawei.com/enterprise/en/doc/EDOC1100063073/87950645/vm-oss-occasionally-fail-to-load-the-in200-driver-when-the-vf-performs-flr
Link: https://lore.kernel.org/r/20210414132301.1793-1-chiqijun@huawei.comSigned-off-by: Chiqijun <chiqijun@huawei.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Cc: stable@vger.kernel.org

Some NVIDIA GPU devices do not work with SBR.  Triggering SBR leaves the
device inoperable for the current system boot. It requires a system
hard-reboot to get the GPU device back to normal operating condition
post-SBR. For the affected devices, enable NO_BUS_RESET quirk to avoid the
issue.

This issue will be fixed in the next generation of hardware.

Link: https://lore.kernel.org/r/20210608054857.18963-8-ameynarkhede03@gmail.comSigned-off-by: Shanker Donthineni <sdonthineni@nvidia.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Sinan Kaya <okaya@kernel.org>
Cc: stable@vger.kernel.org

Some TI KeyStone C667X devices do not support bus/hot reset.  The PCIESS
automatically disables LTSSM when Secondary Bus Reset is received and
device stops working.  Prevent bus reset for these devices.  With this
change, the device can be assigned to VMs with VFIO, but it will leak state
between VMs.

Reference: https://e2e.ti.com/support/processors/f/791/t/954382
Link: https://lore.kernel.org/r/20210315102606.17153-1-antti.jarvinen@gmail.comSigned-off-by: Antti Järvinen <antti.jarvinen@gmail.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Kishon Vijay Abraham I <kishon@ti.com>
Cc: stable@vger.kernel.org

7f100744 ("PCI: tegra: Add Tegra194 MCFG quirks for ECAM errata")
caused a few build regressions:

  - 7f100744 removed the Makefile rule for CONFIG_PCIE_TEGRA194, so
    pcie-tegra.c can no longer be built as a module.  Restore that rule.

  - 7f100744 added "#ifdef CONFIG_PCIE_TEGRA194" around the native
    driver, but that's only set when the driver is built-in (for a module,
    CONFIG_PCIE_TEGRA194_MODULE is defined).

    The ACPI quirk is completely independent of the rest of the native
    driver, so move the quirk to its own file and remove the #ifdef in the
    native driver.

  - 7f100744 added symbols that are always defined but used only when
    CONFIG_PCIEASPM, which causes warnings when CONFIG_PCIEASPM is not set:

      drivers/pci/controller/dwc/pcie-tegra194.c:259:18: warning: ‘event_cntr_data_offset’ defined but not used [-Wunused-const-variable=]
      drivers/pci/controller/dwc/pcie-tegra194.c:250:18: warning: ‘event_cntr_ctrl_offset’ defined but not used [-Wunused-const-variable=]
      drivers/pci/controller/dwc/pcie-tegra194.c:243:27: warning: ‘pcie_gen_freq’ defined but not used [-Wunused-const-variable=]

Fixes: 7f100744 ("PCI: tegra: Add Tegra194 MCFG quirks for ECAM errata")
Link: https://lore.kernel.org/r/20210610064134.336781-1-jonathanh@nvidia.comSigned-off-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Thierry Reding <treding@nvidia.com>

Alexandru and Qu reported this resource allocation failure on ROCKPro64 v2
and ROCK Pi 4B, both based on the RK3399:

  pci_bus 0000:00: root bus resource [mem 0xfa000000-0xfbdfffff 64bit]
  pci 0000:00:00.0: PCI bridge to [bus 01]
  pci 0000:00:00.0: BAR 14: no space for [mem size 0x00100000]
  pci 0000:01:00.0: reg 0x10: [mem 0x00000000-0x00003fff 64bit]

"BAR 14" is the PCI bridge's 32-bit non-prefetchable window, and our PCI
allocation code isn't smart enough to allocate it in a host bridge window
marked as 64-bit, even though this should work fine.

A DT host bridge description includes the windows from the CPU address
space to the PCI bus space.  On a few architectures (microblaze, powerpc,
sparc), the DT may also describe PCI devices themselves, including their
BARs.

Before 9d57e61b ("of/pci: Add IORESOURCE_MEM_64 to resource flags for
64-bit memory addresses"), of_bus_pci_get_flags() ignored the fact that
some DT addresses described 64-bit windows and BARs.  That was a problem
because the virtio virtual NIC has a 32-bit BAR and a 64-bit BAR, and the
driver couldn't distinguish them.

9d57e61b set IORESOURCE_MEM_64 for those 64-bit DT ranges, which fixed
the virtio driver.  But it also set IORESOURCE_MEM_64 for host bridge
windows, which exposed the fact that the PCI allocator isn't smart enough
to put 32-bit resources in those 64-bit windows.

Clear IORESOURCE_MEM_64 from host bridge windows since we don't need that
information.
Suggested-by: Bjorn Helgaas <bhelgaas@google.com>
Fixes: 9d57e61b ("of/pci: Add IORESOURCE_MEM_64 to resource flags for 64-bit memory addresses")
Link: https://lore.kernel.org/r/20210614230457.752811-1-punitagrawal@gmail.com
Reported-at: https://lore.kernel.org/lkml/7a1e2ebc-f7d8-8431-d844-41a9c36a8911@arm.com/
Reported-at: https://lore.kernel.org/lkml/YMyTUv7Jsd89PGci@m4/T/#uReported-by: Alexandru Elisei <alexandru.elisei@arm.com>
Reported-by: Qu Wenruo <wqu@suse.com>
Tested-by: Alexandru Elisei <alexandru.elisei@arm.com>
Tested-by: Domenico Andreoli <domenico.andreoli@linux.com>
Signed-off-by: Punit Agrawal <punitagrawal@gmail.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Acked-by: Ard Biesheuvel <ardb@kernel.org>

The trace_clock_global() tries to make sure the events between CPUs is
somewhat in order. A global value is used and updated by the latest read
of a clock. If one CPU is ahead by a little, and is read by another CPU, a
lock is taken, and if the timestamp of the other CPU is behind, it will
simply use the other CPUs timestamp.

The lock is also only taken with a "trylock" due to tracing, and strange
recursions can happen. The lock is not taken at all in NMI context.

In the case where the lock is not able to be taken, the non synced
timestamp is returned. But it will not be less than the saved global
timestamp.

The problem arises because when the time goes "backwards" the time
returned is the saved timestamp plus 1. If the lock is not taken, and the
plus one to the timestamp is returned, there's a small race that can cause
the time to go backwards!

	CPU0				CPU1
	----				----
				trace_clock_global() {
				    ts = clock() [ 1000 ]
				    trylock(clock_lock) [ success ]
				    global_ts = ts; [ 1000 ]

				    <interrupted by NMI>
 trace_clock_global() {
    ts = clock() [ 999 ]
    if (ts < global_ts)
	ts = global_ts + 1 [ 1001 ]

    trylock(clock_lock) [ fail ]

    return ts [ 1001]
 }
				    unlock(clock_lock);
				    return ts; [ 1000 ]
				}

 trace_clock_global() {
    ts = clock() [ 1000 ]
    if (ts < global_ts) [ false 1000 == 1000 ]

    trylock(clock_lock) [ success ]
    global_ts = ts; [ 1000 ]
    unlock(clock_lock)

    return ts; [ 1000 ]
 }

The above case shows to reads of trace_clock_global() on the same CPU, but
the second read returns one less than the first read. That is, time when
backwards, and this is not what is allowed by trace_clock_global().

This was triggered by heavy tracing and the ring buffer checker that tests
for the clock going backwards:

 Ring buffer clock went backwards: 20613921464 -> 20613921463
 ------------[ cut here ]------------
 WARNING: CPU: 2 PID: 0 at kernel/trace/ring_buffer.c:3412 check_buffer+0x1b9/0x1c0
 Modules linked in:
 [..]
 [CPU: 2]TIME DOES NOT MATCH expected:20620711698 actual:20620711697 delta:6790234 before:20613921463 after:20613921463
   [20613915818] PAGE TIME STAMP
   [20613915818] delta:0
   [20613915819] delta:1
   [20613916035] delta:216
   [20613916465] delta:430
   [20613916575] delta:110
   [20613916749] delta:174
   [20613917248] delta:499
   [20613917333] delta:85
   [20613917775] delta:442
   [20613917921] delta:146
   [20613918321] delta:400
   [20613918568] delta:247
   [20613918768] delta:200
   [20613919306] delta:538
   [20613919353] delta:47
   [20613919980] delta:627
   [20613920296] delta:316
   [20613920571] delta:275
   [20613920862] delta:291
   [20613921152] delta:290
   [20613921464] delta:312
   [20613921464] delta:0 TIME EXTEND
   [20613921464] delta:0

This happened more than once, and always for an off by one result. It also
started happening after commit aafe104a was added.

Cc: stable@vger.kernel.org
Fixes: aafe104a ("tracing: Restructure trace_clock_global() to never block")
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>