linux-2.6-microblaze.git
5 years agoRDMA/siw: Fix IPv6 addr_list locking
Bernard Metzler [Wed, 28 Aug 2019 13:03:55 +0000 (15:03 +0200)]
RDMA/siw: Fix IPv6 addr_list locking

Walking the address list of an inet6_dev requires
appropriate locking. Since the called function
siw_listen_address() may sleep, we have to use
rtnl_lock() instead of read_lock_bh().

Also introduces sanity checks if we got a device
from in_dev_get() or in6_dev_get().

Reported-by: Bart Van Assche <bvanassche@acm.org>
Fixes: 6c52fdc244b5 ("rdma/siw: connection management")
Signed-off-by: Bernard Metzler <bmt@zurich.ibm.com>
Link: https://lore.kernel.org/r/20190828130355.22830-1-bmt@zurich.ibm.com
Signed-off-by: Doug Ledford <dledford@redhat.com>
5 years agorbd: restore zeroing past the overlap when reading from parent
Ilya Dryomov [Tue, 27 Aug 2019 14:45:10 +0000 (16:45 +0200)]
rbd: restore zeroing past the overlap when reading from parent

The parent image is read only up to the overlap point, the rest of
the buffer should be zeroed.  This snuck in because as it turns out
the overlap test case has not been triggering this code path for
a while now.

Fixes: a9b67e69949d ("rbd: replace obj_req->tried_parent with obj_req->read_state")
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Jason Dillaman <dillaman@redhat.com>
5 years agolibceph: don't call crypto_free_sync_skcipher() on a NULL tfm
Jia-Ju Bai [Wed, 24 Jul 2019 09:43:06 +0000 (17:43 +0800)]
libceph: don't call crypto_free_sync_skcipher() on a NULL tfm

In set_secret(), key->tfm is assigned to NULL on line 55, and then
ceph_crypto_key_destroy(key) is executed.

ceph_crypto_key_destroy(key)
  crypto_free_sync_skcipher(key->tfm)
    crypto_free_skcipher(&tfm->base);

This happens to work because crypto_sync_skcipher is a trivial wrapper
around crypto_skcipher: &tfm->base is still 0 and crypto_free_skcipher()
handles that.  Let's not rely on the layout of crypto_sync_skcipher.

This bug is found by a static analysis tool STCheck written by us.

Fixes: 69d6302b65a8 ("libceph: Remove VLA usage of skcipher").
Signed-off-by: Jia-Ju Bai <baijiaju1990@gmail.com>
Reviewed-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
5 years agoKVM: arm/arm64: vgic-v2: Handle SGI bits in GICD_I{S,C}PENDR0 as WI
Marc Zyngier [Wed, 28 Aug 2019 10:10:16 +0000 (11:10 +0100)]
KVM: arm/arm64: vgic-v2: Handle SGI bits in GICD_I{S,C}PENDR0 as WI

A guest is not allowed to inject a SGI (or clear its pending state)
by writing to GICD_ISPENDR0 (resp. GICD_ICPENDR0), as these bits are
defined as WI (as per ARM IHI 0048B 4.3.7 and 4.3.8).

Make sure we correctly emulate the architecture.

Fixes: 96b298000db4 ("KVM: arm/arm64: vgic-new: Add PENDING registers handlers")
Cc: stable@vger.kernel.org # 4.7+
Reported-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Signed-off-by: Will Deacon <will@kernel.org>
5 years agotcp: remove empty skb from write queue in error cases
Eric Dumazet [Mon, 26 Aug 2019 16:19:15 +0000 (09:19 -0700)]
tcp: remove empty skb from write queue in error cases

Vladimir Rutsky reported stuck TCP sessions after memory pressure
events. Edge Trigger epoll() user would never receive an EPOLLOUT
notification allowing them to retry a sendmsg().

Jason tested the case of sk_stream_alloc_skb() returning NULL,
but there are other paths that could lead both sendmsg() and sendpage()
to return -1 (EAGAIN), with an empty skb queued on the write queue.

This patch makes sure we remove this empty skb so that
Jason code can detect that the queue is empty, and
call sk->sk_write_space(sk) accordingly.

Fixes: ce5ec440994b ("tcp: ensure epoll edge trigger wakeup when write queue is empty")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Jason Baron <jbaron@akamai.com>
Reported-by: Vladimir Rutsky <rutsky@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/rds: Fix info leak in rds6_inc_info_copy()
Ka-Cheong Poon [Mon, 26 Aug 2019 09:39:12 +0000 (02:39 -0700)]
net/rds: Fix info leak in rds6_inc_info_copy()

The rds6_inc_info_copy() function has a couple struct members which
are leaking stack information.  The ->tos field should hold actual
information and the ->flags field needs to be zeroed out.

Fixes: 3eb450367d08 ("rds: add type of service(tos) infrastructure")
Fixes: b7ff8b1036f0 ("rds: Extend RDS API for IPv6 support")
Reported-by: 黄ID蝴蝶 <butterflyhuangxx@gmail.com>
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Ka-Cheong Poon <ka-cheong.poon@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: fix skb use after free in netpoll
Feng Sun [Mon, 26 Aug 2019 06:46:04 +0000 (14:46 +0800)]
net: fix skb use after free in netpoll

After commit baeababb5b85d5c4e6c917efe2a1504179438d3b
("tun: return NET_XMIT_DROP for dropped packets"),
when tun_net_xmit drop packets, it will free skb and return NET_XMIT_DROP,
netpoll_send_skb_on_dev will run into following use after free cases:
1. retry netpoll_start_xmit with freed skb;
2. queue freed skb in npinfo->txq.
queue_process will also run into use after free case.

hit netpoll_send_skb_on_dev first case with following kernel log:

[  117.864773] kernel BUG at mm/slub.c:306!
[  117.864773] invalid opcode: 0000 [#1] SMP PTI
[  117.864774] CPU: 3 PID: 2627 Comm: loop_printmsg Kdump: loaded Tainted: P           OE     5.3.0-050300rc5-generic #201908182231
[  117.864775] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Ubuntu-1.8.2-1ubuntu1 04/01/2014
[  117.864775] RIP: 0010:kmem_cache_free+0x28d/0x2b0
[  117.864781] Call Trace:
[  117.864781]  ? tun_net_xmit+0x21c/0x460
[  117.864781]  kfree_skbmem+0x4e/0x60
[  117.864782]  kfree_skb+0x3a/0xa0
[  117.864782]  tun_net_xmit+0x21c/0x460
[  117.864782]  netpoll_start_xmit+0x11d/0x1b0
[  117.864788]  netpoll_send_skb_on_dev+0x1b8/0x200
[  117.864789]  __br_forward+0x1b9/0x1e0 [bridge]
[  117.864789]  ? skb_clone+0x53/0xd0
[  117.864790]  ? __skb_clone+0x2e/0x120
[  117.864790]  deliver_clone+0x37/0x50 [bridge]
[  117.864790]  maybe_deliver+0x89/0xc0 [bridge]
[  117.864791]  br_flood+0x6c/0x130 [bridge]
[  117.864791]  br_dev_xmit+0x315/0x3c0 [bridge]
[  117.864792]  netpoll_start_xmit+0x11d/0x1b0
[  117.864792]  netpoll_send_skb_on_dev+0x1b8/0x200
[  117.864792]  netpoll_send_udp+0x2c6/0x3e8
[  117.864793]  write_msg+0xd9/0xf0 [netconsole]
[  117.864793]  console_unlock+0x386/0x4e0
[  117.864793]  vprintk_emit+0x17e/0x280
[  117.864794]  vprintk_default+0x29/0x50
[  117.864794]  vprintk_func+0x4c/0xbc
[  117.864794]  printk+0x58/0x6f
[  117.864795]  loop_fun+0x24/0x41 [printmsg_loop]
[  117.864795]  kthread+0x104/0x140
[  117.864795]  ? 0xffffffffc05b1000
[  117.864796]  ? kthread_park+0x80/0x80
[  117.864796]  ret_from_fork+0x35/0x40

Signed-off-by: Feng Sun <loyou85@gmail.com>
Signed-off-by: Xiaojun Zhao <xiaojunzhao141@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dsa: tag_8021q: Future-proof the reserved fields in the custom VID
Vladimir Oltean [Sun, 25 Aug 2019 18:32:12 +0000 (21:32 +0300)]
net: dsa: tag_8021q: Future-proof the reserved fields in the custom VID

After witnessing the discussion in https://lkml.org/lkml/2019/8/14/151
w.r.t. ioctl extensibility, it became clear that such an issue might
prevent that the 3 RSV bits inside the DSA 802.1Q tag might also suffer
the same fate and be useless for further extension.

So clearly specify that the reserved bits should currently be
transmitted as zero and ignored on receive. The DSA tagger already does
this (and has always did), and is the only known user so far (no
Wireshark dissection plugin, etc). So there should be no incompatibility
to speak of.

Fixes: 0471dd429cea ("net: dsa: tag_8021q: Create a stable binary format")
Signed-off-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoAdd genphy_c45_config_aneg() function to phy-c45.c
Marco Hartmann [Wed, 21 Aug 2019 11:00:46 +0000 (11:00 +0000)]
Add genphy_c45_config_aneg() function to phy-c45.c

Commit 34786005eca3 ("net: phy: prevent PHYs w/o Clause 22 regs from calling
genphy_config_aneg") introduced a check that aborts phy_config_aneg()
if the phy is a C45 phy.
This causes phy_state_machine() to call phy_error() so that the phy
ends up in PHY_HALTED state.

Instead of returning -EOPNOTSUPP, call genphy_c45_config_aneg()
(analogous to the C22 case) so that the state machine can run
correctly.

genphy_c45_config_aneg() closely resembles mv3310_config_aneg()
in drivers/net/phy/marvell10g.c, excluding vendor specific
configurations for 1000BaseT.

Fixes: 22b56e827093 ("net: phy: replace genphy_10g_driver with genphy_c45_driver")

Signed-off-by: Marco Hartmann <marco.hartmann@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agocifs: update internal module number
Steve French [Tue, 27 Aug 2019 22:29:56 +0000 (17:29 -0500)]
cifs: update internal module number

To 2.22

Signed-off-by: Steve French <stfrench@microsoft.com>
5 years agocifs: replace various strncpy with strscpy and similar
Ronnie Sahlberg [Mon, 26 Aug 2019 23:30:14 +0000 (09:30 +1000)]
cifs: replace various strncpy with strscpy and similar

Using strscpy is cleaner, and avoids some problems with
handling maximum length strings.  Linus noticed the
original problem and Aurelien pointed out some additional
problems. Fortunately most of this is SMB1 code (and
in particular the ASCII string handling older, which
is less common).

Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Reviewed-by: Aurelien Aptel <aaptel@suse.com>
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
5 years agonet_sched: fix a NULL pointer deref in ipt action
Cong Wang [Sun, 25 Aug 2019 17:01:32 +0000 (10:01 -0700)]
net_sched: fix a NULL pointer deref in ipt action

The net pointer in struct xt_tgdtor_param is not explicitly
initialized therefore is still NULL when dereferencing it.
So we have to find a way to pass the correct net pointer to
ipt_destroy_target().

The best way I find is just saving the net pointer inside the per
netns struct tcf_idrinfo, which could make this patch smaller.

Fixes: 0c66dc1ea3f0 ("netfilter: conntrack: register hooks in netns when needed by ruleset")
Reported-and-tested-by: itugrok@yahoo.com
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agocifs: Use kzfree() to zero out the password
Dan Carpenter [Tue, 27 Aug 2019 10:59:17 +0000 (13:59 +0300)]
cifs: Use kzfree() to zero out the password

It's safer to zero out the password so that it can never be disclosed.

Fixes: 0c219f5799c7 ("cifs: set domainName when a domain-key is used in multiuser")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
5 years agocifs: set domainName when a domain-key is used in multiuser
Ronnie Sahlberg [Wed, 21 Aug 2019 22:09:50 +0000 (08:09 +1000)]
cifs: set domainName when a domain-key is used in multiuser

RHBZ: 1710429

When we use a domain-key to authenticate using multiuser we must also set
the domainnmame for the new volume as it will be used and passed to the server
in the NTLMSSP Domain-name.

Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
5 years agoMerge tag 'nfs-for-5.3-3' of git://git.linux-nfs.org/projects/trondmy/linux-nfs
Linus Torvalds [Tue, 27 Aug 2019 20:22:57 +0000 (13:22 -0700)]
Merge tag 'nfs-for-5.3-3' of git://git.linux-nfs.org/projects/trondmy/linux-nfs

Pull NFS client bugfixes from Trond Myklebust:
 "Highlights include:

  Stable fixes:

   - Fix a page lock leak in nfs_pageio_resend()

   - Ensure O_DIRECT reports an error if the bytes read/written is 0

   - Don't handle errors if the bind/connect succeeded

   - Revert "NFSv4/flexfiles: Abort I/O early if the layout segment was
     invalidat ed"

  Bugfixes:

   - Don't refresh attributes with mounted-on-file information

   - Fix return values for nfs4_file_open() and nfs_finish_open()

   - Fix pnfs layoutstats reporting of I/O errors

   - Don't use soft RPC calls for pNFS/flexfiles I/O, and don't abort
     for soft I/O errors when the user specifies a hard mount.

   - Various fixes to the error handling in sunrpc

   - Don't report writepage()/writepages() errors twice"

* tag 'nfs-for-5.3-3' of git://git.linux-nfs.org/projects/trondmy/linux-nfs:
  NFS: remove set but not used variable 'mapping'
  NFSv2: Fix write regression
  NFSv2: Fix eof handling
  NFS: Fix writepage(s) error handling to not report errors twice
  NFS: Fix spurious EIO read errors
  pNFS/flexfiles: Don't time out requests on hard mounts
  SUNRPC: Handle connection breakages correctly in call_status()
  Revert "NFSv4/flexfiles: Abort I/O early if the layout segment was invalidated"
  SUNRPC: Handle EADDRINUSE and ENOBUFS correctly
  pNFS/flexfiles: Turn off soft RPC calls
  SUNRPC: Don't handle errors if the bind/connect succeeded
  NFS: On fatal writeback errors, we need to call nfs_inode_remove_request()
  NFS: Fix initialisation of I/O result struct in nfs_pgio_rpcsetup
  NFS: Ensure O_DIRECT reports an error if the bytes read/written is 0
  NFSv4/pnfs: Fix a page lock leak in nfs_pageio_resend()
  NFSv4: Fix return value in nfs_finish_open()
  NFSv4: Fix return values for nfs4_file_open()
  NFS: Don't refresh attributes with mounted-on-file information

5 years agoKVM: x86: Don't update RIP or do single-step on faulting emulation
Sean Christopherson [Fri, 23 Aug 2019 20:55:44 +0000 (13:55 -0700)]
KVM: x86: Don't update RIP or do single-step on faulting emulation

Don't advance RIP or inject a single-step #DB if emulation signals a
fault.  This logic applies to all state updates that are conditional on
clean retirement of the emulation instruction, e.g. updating RFLAGS was
previously handled by commit 38827dbd3fb85 ("KVM: x86: Do not update
EFLAGS on faulting emulation").

Not advancing RIP is likely a nop, i.e. ctxt->eip isn't updated with
ctxt->_eip until emulation "retires" anyways.  Skipping #DB injection
fixes a bug reported by Andy Lutomirski where a #UD on SYSCALL due to
invalid state with EFLAGS.TF=1 would loop indefinitely due to emulation
overwriting the #UD with #DB and thus restarting the bad SYSCALL over
and over.

Cc: Nadav Amit <nadav.amit@gmail.com>
Cc: stable@vger.kernel.org
Reported-by: Andy Lutomirski <luto@kernel.org>
Fixes: 663f4c61b803 ("KVM: x86: handle singlestep during emulation")
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
5 years agoKVM: x86: hyper-v: don't crash on KVM_GET_SUPPORTED_HV_CPUID when kvm_intel.nested...
Vitaly Kuznetsov [Tue, 27 Aug 2019 16:04:02 +0000 (18:04 +0200)]
KVM: x86: hyper-v: don't crash on KVM_GET_SUPPORTED_HV_CPUID when kvm_intel.nested is disabled

If kvm_intel is loaded with nested=0 parameter an attempt to perform
KVM_GET_SUPPORTED_HV_CPUID results in OOPS as nested_get_evmcs_version hook
in kvm_x86_ops is NULL (we assign it in nested_vmx_hardware_setup() and
this only happens in case nested is enabled).

Check that kvm_x86_ops->nested_get_evmcs_version is not NULL before
calling it. With this, we can remove the stub from svm as it is no
longer needed.

Cc: <stable@vger.kernel.org>
Fixes: e2e871ab2f02 ("x86/kvm/hyper-v: Introduce nested_get_evmcs_version() helper")
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
5 years agoMerge tag 'arc-5.3-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupta/arc
Linus Torvalds [Tue, 27 Aug 2019 17:50:27 +0000 (10:50 -0700)]
Merge tag 'arc-5.3-rc7' of git://git./linux/kernel/git/vgupta/arc

Pull ARC updates from Vineet Gupta:

 - support for Edge Triggered IRQs in ARC IDU intc

 - other fixes here and there

* tag 'arc-5.3-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupta/arc:
  arc: prefer __section from compiler_attributes.h
  dt-bindings: IDU-intc: Add support for edge-triggered interrupts
  dt-bindings: IDU-intc: Clean up documentation
  ARCv2: IDU-intc: Add support for edge-triggered interrupts
  ARC: unwind: Mark expected switch fall-throughs
  ARC: [plat-hsdk]: allow to switch between AXI DMAC port configurations
  ARC: fix typo in setup_dma_ops log message
  ARCv2: entry: early return from exception need not clear U & DE bits

5 years agoMerge tag 'mfd-fixes-5.3' of git://git.kernel.org/pub/scm/linux/kernel/git/lee/mfd
Linus Torvalds [Tue, 27 Aug 2019 17:47:01 +0000 (10:47 -0700)]
Merge tag 'mfd-fixes-5.3' of git://git./linux/kernel/git/lee/mfd

Pull MFD fix from Lee Jones:
 "Identify potentially unused functions in rk808 driver when !PM"

* tag 'mfd-fixes-5.3' of git://git.kernel.org/pub/scm/linux/kernel/git/lee/mfd:
  mfd: rk808: Make PM function declaration static
  mfd: rk808: Mark pm functions __maybe_unused

5 years agoMerge tag 'sound-5.3-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai...
Linus Torvalds [Tue, 27 Aug 2019 17:42:03 +0000 (10:42 -0700)]
Merge tag 'sound-5.3-rc7' of git://git./linux/kernel/git/tiwai/sound

Pull sound fixes from Takashi Iwai:
 "A collection of small fixes as usual:

   - More coverage of USB-audio descriptor sanity checks

   - A fix for mute LED regression on Conexant HD-audio codecs

   - A few device-specific fixes and quirks for USB-audio and HD-audio

   - A fix for (die-hard remaining) possible race in sequencer core

   - FireWire oxfw regression fix that was introduced in 5.3-rc1"

* tag 'sound-5.3-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound:
  ALSA: oxfw: fix to handle correct stream for PCM playback
  ALSA: seq: Fix potential concurrent access to the deleted pool
  ALSA: usb-audio: Check mixer unit bitmap yet more strictly
  ALSA: line6: Fix memory leak at line6_init_pcm() error path
  ALSA: usb-audio: Fix invalid NULL check in snd_emuusb_set_samplerate()
  ALSA: hda/ca0132 - Add new SBZ quirk
  ALSA: usb-audio: Add implicit fb quirk for Behringer UFX1604
  ALSA: hda - Fixes inverted Conexant GPIO mic mute led

5 years agodrm/amdgpu: fix GFXOFF on Picasso and Raven2
Aaron Liu [Tue, 27 Aug 2019 14:59:45 +0000 (22:59 +0800)]
drm/amdgpu: fix GFXOFF on Picasso and Raven2

For picasso(adev->pdev->device == 0x15d8)&raven2(adev->rev_id >= 0x8),
firmware is sufficient to support gfxoff.
In commit 98f58ada2d37e, for picasso&raven2,
return directly and cause gfxoff disabled.

Fixes: 98f58ada2d37 ("drm/amdgpu/gfx9: update pg_flags after determining if gfx off is possible")
Reviewed-by: Huang Rui <ray.huang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Aaron Liu <aaron.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
5 years agodrm/amdgpu: Add APTX quirk for Dell Latitude 5495
Kai-Heng Feng [Tue, 27 Aug 2019 09:33:32 +0000 (17:33 +0800)]
drm/amdgpu: Add APTX quirk for Dell Latitude 5495

Needs ATPX rather than _PR3 to really turn off the dGPU. This can save
~5W when dGPU is runtime-suspended.

Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
5 years agodrm/amd/powerplay: correct Vega20 dpm level related settings
Evan Quan [Mon, 19 Aug 2019 05:17:53 +0000 (13:17 +0800)]
drm/amd/powerplay: correct Vega20 dpm level related settings

Correct the settings for auto mode and skip the unnecessary
settings for dcefclk and fclk.

Signed-off-by: Evan Quan <evan.quan@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Tue, 27 Aug 2019 17:12:48 +0000 (10:12 -0700)]
Merge git://git./linux/kernel/git/netdev/net

Pull networking fixes from David Miller:

 1) Use 32-bit index for tails calls in s390 bpf JIT, from Ilya
    Leoshkevich.

 2) Fix missed EPOLLOUT events in TCP, from Eric Dumazet. Same fix for
    SMC from Jason Baron.

 3) ipv6_mc_may_pull() should return 0 for malformed packets, not
    -EINVAL. From Stefano Brivio.

 4) Don't forget to unpin umem xdp pages in error path of
    xdp_umem_reg(). From Ivan Khoronzhuk.

 5) Fix sta object leak in mac80211, from Johannes Berg.

 6) Fix regression by not configuring PHYLINK on CPU port of bcm_sf2
    switches. From Florian Fainelli.

 7) Revert DMA sync removal from r8169 which was causing regressions on
    some MIPS Loongson platforms. From Heiner Kallweit.

 8) Use after free in flow dissector, from Jakub Sitnicki.

 9) Fix NULL derefs of net devices during ICMP processing across
    collect_md tunnels, from Hangbin Liu.

10) proto_register() memory leaks, from Zhang Lin.

11) Set NLM_F_MULTI flag in multipart netlink messages consistently,
    from John Fastabend.

* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (66 commits)
  r8152: Set memory to all 0xFFs on failed reg reads
  openvswitch: Fix conntrack cache with timeout
  ipv4: mpls: fix mpls_xmit for iptunnel
  nexthop: Fix nexthop_num_path for blackhole nexthops
  net: rds: add service level support in rds-info
  net: route dump netlink NLM_F_MULTI flag missing
  s390/qeth: reject oversized SNMP requests
  sock: fix potential memory leak in proto_register()
  MAINTAINERS: Add phylink keyword to SFF/SFP/SFP+ MODULE SUPPORT
  xfrm/xfrm_policy: fix dst dev null pointer dereference in collect_md mode
  ipv4/icmp: fix rt dst dev null pointer dereference
  openvswitch: Fix log message in ovs conntrack
  bpf: allow narrow loads of some sk_reuseport_md fields with offset > 0
  bpf: fix use after free in prog symbol exposure
  bpf: fix precision tracking in presence of bpf2bpf calls
  flow_dissector: Fix potential use-after-free on BPF_PROG_DETACH
  Revert "r8169: remove not needed call to dma_sync_single_for_device"
  ipv6: propagate ipv6_add_dev's error returns out of ipv6_find_idev
  net/ncsi: Fix the payload copying for the request coming from Netlink
  qed: Add cleanup in qed_slowpath_start()
  ...

5 years agonetfilter: conntrack: make sysctls per-namespace again
Florian Westphal [Tue, 27 Aug 2019 11:24:52 +0000 (13:24 +0200)]
netfilter: conntrack: make sysctls per-namespace again

When I merged the extension sysctl tables with the main one I forgot to
reset them on netns creation.  They currently read/write init_net settings.

Fixes: d912dec12428 ("netfilter: conntrack: merge acct and helper sysctl table with main one")
Fixes: cb2833ed0044 ("netfilter: conntrack: merge ecache and timestamp sysctl tables with main one")
Reported-by: Shmulik Ladkani <shmulik.ladkani@gmail.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
5 years agoKVM: arm/arm64: vgic: Fix potential deadlock when ap_list is long
Heyi Guo [Tue, 27 Aug 2019 11:26:50 +0000 (12:26 +0100)]
KVM: arm/arm64: vgic: Fix potential deadlock when ap_list is long

If the ap_list is longer than 256 entries, merge_final() in list_sort()
will call the comparison callback with the same element twice, causing
a deadlock in vgic_irq_cmp().

Fix it by returning early when irqa == irqb.

Cc: stable@vger.kernel.org # 4.7+
Fixes: 8e4447457965 ("KVM: arm/arm64: vgic-new: Add IRQ sorting")
Signed-off-by: Zenghui Yu <yuzenghui@huawei.com>
Signed-off-by: Heyi Guo <guoheyi@huawei.com>
[maz: massaged commit log and patch, added Fixes and Cc-stable]
Signed-off-by: Marc Zyngier <maz@kernel.org>
Signed-off-by: Will Deacon <will@kernel.org>
5 years agokallsyms: Don't let kallsyms_lookup_size_offset() fail on retrieving the first symbol
Marc Zyngier [Sat, 24 Aug 2019 13:12:31 +0000 (14:12 +0100)]
kallsyms: Don't let kallsyms_lookup_size_offset() fail on retrieving the first symbol

An arm64 kernel configured with

  CONFIG_KPROBES=y
  CONFIG_KALLSYMS=y
  # CONFIG_KALLSYMS_ALL is not set
  CONFIG_KALLSYMS_BASE_RELATIVE=y

reports the following kprobe failure:

  [    0.032677] kprobes: failed to populate blacklist: -22
  [    0.033376] Please take care of using kprobes.

It appears that kprobe fails to retrieve the symbol at address
0xffff000010081000, despite this symbol being in System.map:

  ffff000010081000 T __exception_text_start

This symbol is part of the first group of aliases in the
kallsyms_offsets array (symbol names generated using ugly hacks in
scripts/kallsyms.c):

  kallsyms_offsets:
          .long   0x1000 // do_undefinstr
          .long   0x1000 // efi_header_end
          .long   0x1000 // _stext
          .long   0x1000 // __exception_text_start
          .long   0x12b0 // do_cp15instr

Looking at the implementation of get_symbol_pos(), it returns the
lowest index for aliasing symbols. In this case, it return 0.

But kallsyms_lookup_size_offset() considers 0 as a failure, which
is obviously wrong (there is definitely a valid symbol living there).
In turn, the kprobe blacklisting stops abruptly, hence the original
error.

A CONFIG_KALLSYMS_ALL kernel wouldn't fail as there is always
some random symbols at the beginning of this array, which are never
looked up via kallsyms_lookup_size_offset.

Fix it by considering that get_symbol_pos() is always successful
(which is consistent with the other uses of this function).

Fixes: ffc5089196446 ("[PATCH] Create kallsyms_lookup_size_offset()")
Reviewed-by: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Will Deacon <will@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Signed-off-by: Will Deacon <will@kernel.org>
5 years agoNFS: remove set but not used variable 'mapping'
YueHaibing [Tue, 27 Aug 2019 07:16:36 +0000 (15:16 +0800)]
NFS: remove set but not used variable 'mapping'

Fixes gcc '-Wunused-but-set-variable' warning:

fs/nfs/write.c: In function nfs_page_async_flush:
fs/nfs/write.c:609:24: warning: variable mapping set but not used [-Wunused-but-set-variable]

It is not use since commit aefb623c422e ("NFS: Fix
writepage(s) error handling to not report errors twice")

Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
5 years agoNFSv2: Fix write regression
Trond Myklebust [Tue, 27 Aug 2019 11:03:28 +0000 (07:03 -0400)]
NFSv2: Fix write regression

Ensure we update the write result count on success, since the
RPC call itself does not do so.

Reported-by: Jan Stancek <jstancek@redhat.com>
Reported-by: Naresh Kamboju <naresh.kamboju@linaro.org>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Tested-by: Jan Stancek <jstancek@redhat.com>
5 years agoNFSv2: Fix eof handling
Trond Myklebust [Tue, 27 Aug 2019 00:41:16 +0000 (20:41 -0400)]
NFSv2: Fix eof handling

If we received a reply from the server with a zero length read and
no error, then that implies we are at eof.

Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
5 years agoMerge tag 'kvm-ppc-fixes-5.3-1' of git://git.kernel.org/pub/scm/linux/kernel/git...
Radim Krčmář [Tue, 27 Aug 2019 12:46:17 +0000 (14:46 +0200)]
Merge tag 'kvm-ppc-fixes-5.3-1' of git://git./linux/kernel/git/paulus/powerpc

KVM/PPC fix for 5.3

- Fix bug which could leave locks locked in the host on return
  to a guest.

5 years agomfd: rk808: Make PM function declaration static
Lee Jones [Tue, 27 Aug 2019 06:39:50 +0000 (07:39 +0100)]
mfd: rk808: Make PM function declaration static

Avoids:
  ../drivers/mfd/rk808.c:771:1: warning: symbol 'rk8xx_pm_ops' \
    was not declared. Should it be static?

Fixes: 5752bc4373b2 ("mfd: rk808: Mark pm functions __maybe_unused")
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Lee Jones <lee.jones@linaro.org>
5 years agonetfilter: nf_conntrack_ftp: Fix debug output
Thomas Jarosch [Wed, 21 Aug 2019 14:14:28 +0000 (16:14 +0200)]
netfilter: nf_conntrack_ftp: Fix debug output

The find_pattern() debug output was printing the 'skip' character.
This can be a NULL-byte and messes up further pr_debug() output.

Output without the fix:
kernel: nf_conntrack_ftp: Pattern matches!
kernel: nf_conntrack_ftp: Skipped up to `<7>nf_conntrack_ftp: find_pattern `PORT': dlen = 8
kernel: nf_conntrack_ftp: find_pattern `EPRT': dlen = 8

Output with the fix:
kernel: nf_conntrack_ftp: Pattern matches!
kernel: nf_conntrack_ftp: Skipped up to 0x0 delimiter!
kernel: nf_conntrack_ftp: Match succeeded!
kernel: nf_conntrack_ftp: conntrack_ftp: match `172,17,0,100,200,207' (20 bytes at 4150681645)
kernel: nf_conntrack_ftp: find_pattern `PORT': dlen = 8

Signed-off-by: Thomas Jarosch <thomas.jarosch@intra2net.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
5 years agonetfilter: xt_physdev: Fix spurious error message in physdev_mt_check
Todd Seidelmann [Wed, 21 Aug 2019 15:47:53 +0000 (11:47 -0400)]
netfilter: xt_physdev: Fix spurious error message in physdev_mt_check

Simplify the check in physdev_mt_check() to emit an error message
only when passed an invalid chain (ie, NF_INET_LOCAL_OUT).
This avoids cluttering up the log with errors against valid rules.

For large/heavily modified rulesets, current behavior can quickly
overwhelm the ring buffer, because this function gets called on
every change, regardless of the rule that was changed.

Signed-off-by: Todd Seidelmann <tseidelmann@linode.com>
Acked-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
5 years agorxrpc: Use skb_unshare() rather than skb_cow_data()
David Howells [Tue, 27 Aug 2019 09:13:46 +0000 (10:13 +0100)]
rxrpc: Use skb_unshare() rather than skb_cow_data()

The in-place decryption routines in AF_RXRPC's rxkad security module
currently call skb_cow_data() to make sure the data isn't shared and that
the skb can be written over.  This has a problem, however, as the softirq
handler may be still holding a ref or the Rx ring may be holding multiple
refs when skb_cow_data() is called in rxkad_verify_packet() - and so
skb_shared() returns true and __pskb_pull_tail() dislikes that.  If this
occurs, something like the following report will be generated.

kernel BUG at net/core/skbuff.c:1463!
...
RIP: 0010:pskb_expand_head+0x253/0x2b0
...
Call Trace:
 __pskb_pull_tail+0x49/0x460
 skb_cow_data+0x6f/0x300
 rxkad_verify_packet+0x18b/0xb10 [rxrpc]
 rxrpc_recvmsg_data.isra.11+0x4a8/0xa10 [rxrpc]
 rxrpc_kernel_recv_data+0x126/0x240 [rxrpc]
 afs_extract_data+0x51/0x2d0 [kafs]
 afs_deliver_fs_fetch_data+0x188/0x400 [kafs]
 afs_deliver_to_call+0xac/0x430 [kafs]
 afs_wait_for_call_to_complete+0x22f/0x3d0 [kafs]
 afs_make_call+0x282/0x3f0 [kafs]
 afs_fs_fetch_data+0x164/0x300 [kafs]
 afs_fetch_data+0x54/0x130 [kafs]
 afs_readpages+0x20d/0x340 [kafs]
 read_pages+0x66/0x180
 __do_page_cache_readahead+0x188/0x1a0
 ondemand_readahead+0x17d/0x2e0
 generic_file_read_iter+0x740/0xc10
 __vfs_read+0x145/0x1a0
 vfs_read+0x8c/0x140
 ksys_read+0x4a/0xb0
 do_syscall_64+0x43/0xf0
 entry_SYSCALL_64_after_hwframe+0x44/0xa9

Fix this by using skb_unshare() instead in the input path for DATA packets
that have a security index != 0.  Non-DATA packets don't need in-place
encryption and neither do unencrypted DATA packets.

Fixes: 248f219cb8bc ("rxrpc: Rewrite the data and ack handling code")
Reported-by: Julian Wollrath <jwollrath@web.de>
Signed-off-by: David Howells <dhowells@redhat.com>
5 years agorxrpc: Use the tx-phase skb flag to simplify tracing
David Howells [Mon, 19 Aug 2019 08:25:38 +0000 (09:25 +0100)]
rxrpc: Use the tx-phase skb flag to simplify tracing

Use the previously-added transmit-phase skbuff private flag to simplify the
socket buffer tracing a bit.  Which phase the skbuff comes from can now be
divined from the skb rather than having to be guessed from the call state.

We can also reduce the number of rxrpc_skb_trace values by eliminating the
difference between Tx and Rx in the symbols.

Signed-off-by: David Howells <dhowells@redhat.com>
5 years agorxrpc: Add a private skb flag to indicate transmission-phase skbs
David Howells [Mon, 19 Aug 2019 08:25:37 +0000 (09:25 +0100)]
rxrpc: Add a private skb flag to indicate transmission-phase skbs

Add a flag in the private data on an skbuff to indicate that this is a
transmission-phase buffer rather than a receive-phase buffer.

Signed-off-by: David Howells <dhowells@redhat.com>
5 years agorxrpc: Abstract out rxtx ring cleanup
David Howells [Mon, 19 Aug 2019 08:25:37 +0000 (09:25 +0100)]
rxrpc: Abstract out rxtx ring cleanup

Abstract out rxtx ring cleanup into its own function from its two callers.
This makes it easier to apply the same changes to both.

Signed-off-by: David Howells <dhowells@redhat.com>
5 years agorxrpc: Pass the input handler's data skb reference to the Rx ring
David Howells [Mon, 19 Aug 2019 08:25:36 +0000 (09:25 +0100)]
rxrpc: Pass the input handler's data skb reference to the Rx ring

Pass the reference held on a DATA skb in the rxrpc input handler into the
Rx ring rather than getting an additional ref for this and then dropping
the original ref at the end.

Signed-off-by: David Howells <dhowells@redhat.com>
5 years agorxrpc: Use info in skbuff instead of reparsing a jumbo packet
David Howells [Tue, 27 Aug 2019 08:51:30 +0000 (09:51 +0100)]
rxrpc: Use info in skbuff instead of reparsing a jumbo packet

Use the information now cached in the skbuff private data to avoid the need
to reparse a jumbo packet.  We can find all the subpackets by dead
reckoning, so it's only necessary to note how many there are, whether the
last one is flagged as LAST_PACKET and whether any have the REQUEST_ACK
flag set.

This is necessary as once recvmsg() can see the packet, it can start
modifying it, such as doing in-place decryption.

Fixes: 248f219cb8bc ("rxrpc: Rewrite the data and ack handling code")
Signed-off-by: David Howells <dhowells@redhat.com>
5 years agorxrpc: Improve jumbo packet counting
David Howells [Mon, 19 Aug 2019 08:25:37 +0000 (09:25 +0100)]
rxrpc: Improve jumbo packet counting

Improve the information stored about jumbo packets so that we don't need to
reparse them so much later.

Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Jeffrey Altman <jaltman@auristor.com>
5 years agox86/boot/compressed/64: Fix missing initialization in find_trampoline_placement()
Kirill A. Shutemov [Mon, 26 Aug 2019 13:26:01 +0000 (16:26 +0300)]
x86/boot/compressed/64: Fix missing initialization in find_trampoline_placement()

Gustavo noticed that 'new' can be left uninitialized if 'bios_start'
happens to be less or equal to 'entry->addr + entry->size'.

Initialize the variable at the begin of the iteration to the current value
of 'bios_start'.

Fixes: 0a46fff2f910 ("x86/boot/compressed/64: Fix boot on machines with broken E820 table")
Reported-by: "Gustavo A. R. Silva" <gustavo@embeddedor.com>
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/20190826133326.7cxb4vbmiawffv2r@box
5 years agodrm/i915: Call dma_set_max_seg_size() in i915_driver_hw_probe()
Lyude Paul [Fri, 23 Aug 2019 20:52:51 +0000 (16:52 -0400)]
drm/i915: Call dma_set_max_seg_size() in i915_driver_hw_probe()

Currently, we don't call dma_set_max_seg_size() for i915 because we
intentionally do not limit the segment length that the device supports.
However, this results in a warning being emitted if we try to map
anything larger than SZ_64K on a kernel with CONFIG_DMA_API_DEBUG_SG
enabled:

[    7.751926] DMA-API: i915 0000:00:02.0: mapping sg segment longer
than device claims to support [len=98304] [max=65536]
[    7.751934] WARNING: CPU: 5 PID: 474 at kernel/dma/debug.c:1220
debug_dma_map_sg+0x20f/0x340

This was originally brought up on
https://bugs.freedesktop.org/show_bug.cgi?id=108517 , and the consensus
there was it wasn't really useful to set a limit (and that dma-debug
isn't really all that useful for i915 in the first place). Unfortunately
though, CONFIG_DMA_API_DEBUG_SG is enabled in the debug configs for
various distro kernels. Since a WARN_ON() will disable automatic problem
reporting (and cause any CI with said option enabled to start
complaining), we really should just fix the problem.

Note that as me and Chris Wilson discussed, the other solution for this
would be to make DMA-API not make such assumptions when a driver hasn't
explicitly set a maximum segment size. But, taking a look at the commit
which originally introduced this behavior, commit 78c47830a5cb
("dma-debug: check scatterlist segments"), there is an explicit mention
of this assumption and how it applies to devices with no segment size:

Conversely, devices which are less limited than the rather
conservative defaults, or indeed have no limitations at all
(e.g. GPUs with their own internal MMU), should be encouraged to
set appropriate dma_parms, as they may get more efficient DMA
mapping performance out of it.

So unless there's any concerns (I'm open to discussion!), let's just
follow suite and call dma_set_max_seg_size() with UINT_MAX as our limit
to silence any warnings.

Changes since v3:
* Drop patch for enabling CONFIG_DMA_API_DEBUG_SG in CI. It looks like
  just turning it on causes the kernel to spit out bogus WARN_ONs()
  during some igt tests which would otherwise require teaching igt to
  disable the various DMA-API debugging options causing this. This is
  too much work to be worth it, since DMA-API debugging is useless for
  us. So, we'll just settle with this single patch to squelch WARN_ONs()
  during driver load for users that have CONFIG_DMA_API_DEBUG_SG turned
  on for some reason.
* Move dma_set_max_seg_size() call into i915_driver_hw_probe() - Chris
  Wilson

Signed-off-by: Lyude Paul <lyude@redhat.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: <stable@vger.kernel.org> # v4.18+
Link: https://patchwork.freedesktop.org/patch/msgid/20190823205251.14298-1-lyude@redhat.com
(cherry picked from commit acd674af95d3f627062007429b9c195c6b32361d)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
5 years agodrm/i915/dp: Fix DSC enable code to use cpu_transcoder instead of encoder->type
Manasi Navare [Wed, 21 Aug 2019 21:59:50 +0000 (14:59 -0700)]
drm/i915/dp: Fix DSC enable code to use cpu_transcoder instead of encoder->type

This patch fixes the intel_configure_pps_for_dsc_encoder() function to use
cpu_transcoder instead of encoder->type to select the correct DSC registers
that was wrongly used in the original patch for one DSC register isntance.

Fixes: 7182414e2530 ("drm/i915/dp: Configure i915 Picture parameter Set registers during DSC enabling")
Cc: Ville Syrjala <ville.syrjala@linux.intel.com>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Cc: <stable@vger.kernel.org> # v5.0+
Signed-off-by: Manasi Navare <manasi.d.navare@intel.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190821215950.24223-1-manasi.d.navare@intel.com
(cherry picked from commit d4c61c4a16decd8ace8660f22c81609a539fccba)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
5 years agodrm/i915: Don't deballoon unused ggtt drm_mm_node in linux guest
Xiong Zhang [Tue, 20 Aug 2019 05:46:17 +0000 (13:46 +0800)]
drm/i915: Don't deballoon unused ggtt drm_mm_node in linux guest

The following call trace may exist in linux guest dmesg when guest i915
driver is unloaded.
[   90.776610] [drm:vgt_deballoon_space.isra.0 [i915]] deballoon space: range [0x0 - 0x0] 0 KiB.
[   90.776621] BUG: unable to handle kernel NULL pointer dereference at 00000000000000c0
[   90.776691] IP: drm_mm_remove_node+0x4d/0x320 [drm]
[   90.776718] PGD 800000012c7d0067 P4D 800000012c7d0067 PUD 138e4c067 PMD 0
[   90.777091] task: ffff9adab60f2f00 task.stack: ffffaf39c0fe0000
[   90.777142] RIP: 0010:drm_mm_remove_node+0x4d/0x320 [drm]
[   90.777573] Call Trace:
[   90.777653]  intel_vgt_deballoon+0x4c/0x60 [i915]
[   90.777729]  i915_ggtt_cleanup_hw+0x121/0x190 [i915]
[   90.777792]  i915_driver_unload+0x145/0x180 [i915]
[   90.777856]  i915_pci_remove+0x15/0x20 [i915]
[   90.777890]  pci_device_remove+0x3b/0xc0
[   90.777916]  device_release_driver_internal+0x157/0x220
[   90.777945]  driver_detach+0x39/0x70
[   90.777967]  bus_remove_driver+0x51/0xd0
[   90.777990]  pci_unregister_driver+0x23/0x90
[   90.778019]  SyS_delete_module+0x1da/0x240
[   90.778045]  entry_SYSCALL_64_fastpath+0x24/0x87
[   90.778072] RIP: 0033:0x7f34312af067
[   90.778092] RSP: 002b:00007ffdea3da0d8 EFLAGS: 00000206
[   90.778297] RIP: drm_mm_remove_node+0x4d/0x320 [drm] RSP: ffffaf39c0fe3dc0
[   90.778344] ---[ end trace f4b1bc8305fc59dd ]---

Four drm_mm_node are used to reserve guest ggtt space, but some of them
may be skipped and not initialised due to space constraints in
intel_vgt_balloon(). If drm_mm_remove_node() is called with
uninitialized drm_mm_node, the above call trace occurs.

This patch check drm_mm_node's validity before calling
drm_mm_remove_node().

Fixes: ff8f797557c7("drm/i915: return the correct usable aperture size under gvt environment")
Cc: stable@vger.kernel.org
Signed-off-by: Xiong Zhang <xiong.y.zhang@intel.com>
Acked-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/1566279978-9659-1-git-send-email-xiong.y.zhang@intel.com
(cherry picked from commit 4776f3529d6b1e47f02904ad1d264d25ea22b27b)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
5 years agodrm/i915: Do not create a new max_bpc prop for MST connectors
Ville Syrjälä [Tue, 20 Aug 2019 16:16:57 +0000 (19:16 +0300)]
drm/i915: Do not create a new max_bpc prop for MST connectors

We're not allowed to create new properties after device registration
so for MST connectors we need to either create the max_bpc property
earlier, or we reuse one we already have. Let's do the latter apporach
since the corresponding SST connector already has the prop and its
min/max are correct also for the MST connector.

The problem was highlighted by commit 4f5368b5541a ("drm/kms:
Catch mode_object lifetime errors") which results in the following
spew:
[ 1330.878941] WARNING: CPU: 2 PID: 1554 at drivers/gpu/drm/drm_mode_object.c:45 __drm_mode_object_add+0xa0/0xb0 [drm]
...
[ 1330.879008] Call Trace:
[ 1330.879023]  drm_property_create+0xba/0x180 [drm]
[ 1330.879036]  drm_property_create_range+0x15/0x30 [drm]
[ 1330.879048]  drm_connector_attach_max_bpc_property+0x62/0x80 [drm]
[ 1330.879086]  intel_dp_add_mst_connector+0x11f/0x140 [i915]
[ 1330.879094]  drm_dp_add_port.isra.20+0x20b/0x440 [drm_kms_helper]
...

Cc: stable@vger.kernel.org
Cc: Lyude Paul <lyude@redhat.com>
Cc: sunpeng.li@amd.com
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Sean Paul <sean@poorly.run>
Fixes: 5ca0ef8a56b8 ("drm/i915: Add max_bpc property for DP MST")
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190820161657.9658-1-ville.syrjala@linux.intel.com
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
(cherry picked from commit 1b9bd09630d4db4827cc04d358a41a16a6bc2cb0)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
5 years agomfd: rk808: Mark pm functions __maybe_unused
Arnd Bergmann [Mon, 8 Jul 2019 12:53:02 +0000 (14:53 +0200)]
mfd: rk808: Mark pm functions __maybe_unused

The newly added suspend/resume functions are only used if CONFIG_PM
is enabled:

drivers/mfd/rk808.c:752:12: error: 'rk8xx_resume' defined but not used [-Werror=unused-function]
drivers/mfd/rk808.c:732:12: error: 'rk8xx_suspend' defined but not used [-Werror=unused-function]

Mark them as __maybe_unused so the compiler can silently drop them
when they are not needed.

Fixes: 586c1b4125b3 ("mfd: rk808: Add RK817 and RK809 support")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Lee Jones <lee.jones@linaro.org>
5 years agophy: marvell: phy-mvebu-cp110-comphy: rename instances of DLT
Matt Pelland [Thu, 1 Aug 2019 19:50:59 +0000 (15:50 -0400)]
phy: marvell: phy-mvebu-cp110-comphy: rename instances of DLT

The documentation for Marvell's cp110 phy refers to these
registers/register regions as DTL control, DTL frequency loop enable,
etc. This patch aligns the relevant code for these accordingly.

Signed-off-by: Matt Pelland <mpelland@starry.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agophy: marvell: phy-mvebu-cp110-comphy: implement RXAUI support
Matt Pelland [Thu, 1 Aug 2019 19:50:58 +0000 (15:50 -0400)]
phy: marvell: phy-mvebu-cp110-comphy: implement RXAUI support

Marvell's cp110 phy supports RXAUI on lanes 2, 3, 4, and 5 when
connected to port zero. When used in this mode, lanes operate in pairs
of two (2 and 3, 4 and 5).

Signed-off-by: Matt Pelland <mpelland@starry.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agodt-bindings: pci: add PHY properties to Armada 7K/8K controller bindings
Miquel Raynal [Wed, 31 Jul 2019 12:21:21 +0000 (14:21 +0200)]
dt-bindings: pci: add PHY properties to Armada 7K/8K controller bindings

Armada CP110 PCIe controller can have from one to four PHYs for
configuring SERDES lanes (PCIe x1, PCIe x2 or PCIe x4). Describe the
phys and phy-names properties in the bindings.

Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agodt-bindings: phy: Add Marvell COMPHY clocks
Miquel Raynal [Wed, 31 Jul 2019 12:21:20 +0000 (14:21 +0200)]
dt-bindings: phy: Add Marvell COMPHY clocks

Marvell CP110 COMPHY block is fed by 3 clocks. Describe each of them in the
bindings.

Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agophy: mvebu-cp110-comphy: Update comment about powering off all lanes at boot
Miquel Raynal [Wed, 31 Jul 2019 12:21:19 +0000 (14:21 +0200)]
phy: mvebu-cp110-comphy: Update comment about powering off all lanes at boot

Now that all COMPHY modes are supported by the driver, update the
comment stating that mvebu_comphy_power_off() should be called for
each lane. This is still wrong because for compatibility reasons, it
might break users running an old firmware (the driver only uses SMC
calls for SATA, USB and PCIe configuration, there is no code in Linux
to fallback on in these cases.

Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agophy: mvebu-cp110-comphy: Add PCIe support
Grzegorz Jaszczyk [Wed, 31 Jul 2019 12:21:18 +0000 (14:21 +0200)]
phy: mvebu-cp110-comphy: Add PCIe support

Add PCIe support by filling the COMPHY modes table.

Also add a new macro to generate the right value for the firmware
depending on the width (PCI x1, x2, x4, etc). The width will be passed
by the core as the "submode" argument of the ->set_mode() callback. If
this argument is zero, default to x1 mode.

Signed-off-by: Grzegorz Jaszczyk <jaz@semihalf.com>
[miquel.raynal@bootlin.com: adapt the content to the mainline driver]
Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agophy: mvebu-cp110-comphy: Cosmetic change in a helper
Miquel Raynal [Wed, 31 Jul 2019 12:21:17 +0000 (14:21 +0200)]
phy: mvebu-cp110-comphy: Cosmetic change in a helper

Before adding more logic, simplify a bit the writing of the
mvebu_comphy_get_mode() helper by using a pointer instead of
referencing a configuration with the entire table name.

Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agophy: mvebu-cp110-comphy: Add SATA support
Grzegorz Jaszczyk [Wed, 31 Jul 2019 12:21:16 +0000 (14:21 +0200)]
phy: mvebu-cp110-comphy: Add SATA support

Add the corresponding entries in the COMPHY modes table.

SATA support does not need any additional care.

Signed-off-by: Grzegorz Jaszczyk <jaz@semihalf.com>
[miquel.raynal@bootlin.com: adapt the content to the mainline driver]
Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agophy: mvebu-cp110-comphy: Add USB3 host/device support
Grzegorz Jaszczyk [Wed, 31 Jul 2019 12:21:15 +0000 (14:21 +0200)]
phy: mvebu-cp110-comphy: Add USB3 host/device support

Add USB3 host/device support by adding the right entries in the COMPHY
modes table. A new macro is created to instantiate a "generic" mode
ie. not an Ethernet one. This macro will be re-used when adding SATA
support.

Signed-off-by: Grzegorz Jaszczyk <jaz@semihalf.com>
[miquel.raynal@bootlin.com: adapt the content to the mainline driver]
Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agophy: mvebu-cp110-comphy: Allow non-Ethernet modes to be configured
Miquel Raynal [Wed, 31 Jul 2019 12:21:14 +0000 (14:21 +0200)]
phy: mvebu-cp110-comphy: Allow non-Ethernet modes to be configured

The COMPHY can configure the SERDES lanes in several non-Ethernet
modes: SATA, USB3, PCIe. Drop the condition limiting the driver to
Ethernet modes only before adding support for more.

Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agophy: mvebu-cp110-comphy: Rename the macro handling only Ethernet modes
Miquel Raynal [Wed, 31 Jul 2019 12:21:13 +0000 (14:21 +0200)]
phy: mvebu-cp110-comphy: Rename the macro handling only Ethernet modes

Before adding support for other PHY modes (not Ethernet ones), let's
rename the MVEBU_COMPHY_CONF macro to a more specific (and shorter)
appellation.

Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agophy: mvebu-cp110-comphy: Add RXAUI support
Grzegorz Jaszczyk [Wed, 31 Jul 2019 12:21:12 +0000 (14:21 +0200)]
phy: mvebu-cp110-comphy: Add RXAUI support

Add support for RXAUI mode by adding an entry in the COMPHY modes list.

There is no user for this mode yet so we can enforce an up-to-date
firmware and return an error otherwise without breaking anywone.

Signed-off-by: Grzegorz Jaszczyk <jaz@semihalf.com>
[miquel.raynal@bootlin.com: adapt the content to the mainline driver]
Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agophy: mvebu-cp110-comphy: List already supported Ethernet modes
Miquel Raynal [Wed, 31 Jul 2019 12:21:11 +0000 (14:21 +0200)]
phy: mvebu-cp110-comphy: List already supported Ethernet modes

Currently, the driver supports setting lanes to 1000BASEX, 2500BASEX,
10GKR. Complete the COMPHY modes list by adding two (already
supported) cases for lane 4.

Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agophy: mvebu-cp110-comphy: Add SMC call support
Grzegorz Jaszczyk [Wed, 31 Jul 2019 12:21:10 +0000 (14:21 +0200)]
phy: mvebu-cp110-comphy: Add SMC call support

Keep the exact same list of supported configurations but first try to
use the firmware's implementation. If it fails, try the legacy method:
Linux implementation.

Signed-off-by: Grzegorz Jaszczyk <jaz@semihalf.com>
[miquel.raynal@bootlin.com: adapt the content to the mainline driver]
Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Tested-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Tested-by: Grzegorz Jaszczyk <jaz@semihalf.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agophy: mvebu-cp110-comphy: Explicitly initialize the lane submode
Miquel Raynal [Wed, 31 Jul 2019 12:21:09 +0000 (14:21 +0200)]
phy: mvebu-cp110-comphy: Explicitly initialize the lane submode

Explicitly set the lane submode (enum) to a known invalid value.

Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agophy: mvebu-cp110-comphy: Add clocks support
Miquel Raynal [Wed, 31 Jul 2019 12:21:08 +0000 (14:21 +0200)]
phy: mvebu-cp110-comphy: Add clocks support

There is no public clock tree that implies such dependencies between
the MG/MG-core/AXI clocks and the COMPHY IP but accessing the COMPHY
registers while one of the three clocks are disabled stalls the CPU.

This happens if, for instance, the COMPHY driver probe is deferred
(eg. the USB Vbus regulator driver is not yet visible). The MVPP2
driver which also needs these clocks (among others) will
prepare/enable the clocks, then be deferred, and disable/unprepare
them. Next COMPHY lane to be configured would produce an infinite
stall.

Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Tested-by: Grzegorz Jaszczyk <jaz@semihalf.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agophy-rockchip-inno-hdmi: Fix RK3328_TERM_RESISTOR_CALIB_SPEED_7_0's third value
Nathan Chancellor [Wed, 7 Aug 2019 19:23:05 +0000 (12:23 -0700)]
phy-rockchip-inno-hdmi: Fix RK3328_TERM_RESISTOR_CALIB_SPEED_7_0's third value

After commit "linux/bits.h: Add compile time sanity check of GENMASK
inputs" [1], arm64 defconfig builds started failing:

In file included from ../include/linux/bits.h:22,
                 from ../include/linux/bitops.h:5,
                 from ../include/linux/kernel.h:12,
                 from ../include/linux/clk.h:13,
                 from ../drivers/phy/rockchip/phy-rockchip-inno-hdmi.c:9:
../drivers/phy/rockchip/phy-rockchip-inno-hdmi.c: In function 'inno_hdmi_phy_rk3328_power_on':
../include/linux/build_bug.h:16:45: error: negative width in bit-field '<anonymous>'
   16 | #define BUILD_BUG_ON_ZERO(e) (sizeof(struct { int:(-!!(e)); }))
      |                                             ^
../include/linux/bits.h:24:18: note: in expansion of macro 'BUILD_BUG_ON_ZERO'
   24 |  ((unsigned long)BUILD_BUG_ON_ZERO(__builtin_choose_expr( \
      |                  ^~~~~~~~~~~~~~~~~
../include/linux/bits.h:39:3: note: in expansion of macro 'GENMASK_INPUT_CHECK'
   39 |  (GENMASK_INPUT_CHECK(high, low) + __GENMASK(high, low))
      |   ^~~~~~~~~~~~~~~~~~~
../drivers/phy/rockchip/phy-rockchip-inno-hdmi.c:24:42: note: in expansion of macro 'GENMASK'
   24 | #define UPDATE(x, h, l)  (((x) << (l)) & GENMASK((h), (l)))
      |                                          ^~~~~~~
../drivers/phy/rockchip/phy-rockchip-inno-hdmi.c:201:50: note: in expansion of macro 'UPDATE'
  201 | #define RK3328_TERM_RESISTOR_CALIB_SPEED_7_0(x)  UPDATE(x, 7, 9)
      |                                                  ^~~~~~
../drivers/phy/rockchip/phy-rockchip-inno-hdmi.c:1046:26: note: in expansion of macro 'RK3328_TERM_RESISTOR_CALIB_SPEED_7_0'
 1046 |   inno_write(inno, 0xc6, RK3328_TERM_RESISTOR_CALIB_SPEED_7_0(v));
      |                          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

As pointed out by Robin and Guenter, inno_write's val argument is an
8-bit value so having a mask larger than that doesn't make sense. This
also matches the rest of the *_7_0 macros in this driver.

[1]: https://lore.kernel.org/lkml/20190801230358.4193-2-rikard.falkeborn@gmail.com/

Reported-by: Andrzej Hajda <a.hajda@samsung.com>
Reported-by: Guenter Roeck <linux@roeck-us.net>
Reported-by: kernelci.org bot <bot@kernelci.org>
Reported-by: Naresh Kamboju <naresh.kamboju@linaro.org>
Suggested-by: Guenter Roeck <linux@roeck-us.net>
Suggested-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
Reviewed-by: Heiko Stuebner <heiko@sntech.de>
Tested-by: Heiko Stuebner <heiko@sntech.de>
5 years agoKVM: PPC: Book3S: Fix incorrect guest-to-user-translation error handling
Alexey Kardashevskiy [Mon, 26 Aug 2019 04:55:20 +0000 (14:55 +1000)]
KVM: PPC: Book3S: Fix incorrect guest-to-user-translation error handling

H_PUT_TCE_INDIRECT handlers receive a page with up to 512 TCEs from
a guest. Although we verify correctness of TCEs before we do anything
with the existing tables, there is a small window when a check in
kvmppc_tce_validate might pass and right after that the guest alters
the page of TCEs, causing an early exit from the handler and leaving
srcu_read_lock(&vcpu->kvm->srcu) (virtual mode) or lock_rmap(rmap)
(real mode) locked.

This fixes the bug by jumping to the common exit code with an appropriate
unlock.

Cc: stable@vger.kernel.org # v4.11+
Fixes: 121f80ba68f1 ("KVM: PPC: VFIO: Add in-kernel acceleration for VFIO")
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years agodrm/powerplay: Fix Vega20 power reading again
Kent Russell [Fri, 23 Aug 2019 13:13:18 +0000 (09:13 -0400)]
drm/powerplay: Fix Vega20 power reading again

For the 40.46 SMU release, they changed CurrSocketPower to
AverageSocketPower, but this was changed back in 40.47 so just check if
it's 40.46 and make the appropriate change

Tested with 40.45, 40.46 and 40.47 successfully

Signed-off-by: Kent Russell <kent.russell@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 years agodrm/powerplay: Fix Vega20 Average Power value v4
Kent Russell [Thu, 22 Aug 2019 12:17:40 +0000 (08:17 -0400)]
drm/powerplay: Fix Vega20 Average Power value v4

The SMU changed reading from CurrSocketPower to AverageSocketPower, so
reflect this accordingly. This fixes the issue where Average Power
Consumption was being reported as 0 from SMU 40.46-onward

v2: Fixed headline prefix
v3: Add check for SMU version for proper compatibility
v4: Style fix

Signed-off-by: Kent Russell <kent.russell@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 years agobpf: handle 32-bit zext during constant blinding
Naveen N. Rao [Wed, 21 Aug 2019 19:23:58 +0000 (00:53 +0530)]
bpf: handle 32-bit zext during constant blinding

Since BPF constant blinding is performed after the verifier pass, the
ALU32 instructions inserted for doubleword immediate loads don't have a
corresponding zext instruction. This is causing a kernel oops on powerpc
and can be reproduced by running 'test_cgroup_storage' with
bpf_jit_harden=2.

Fix this by emitting BPF_ZEXT during constant blinding if
prog->aux->verifier_zext is set.

Fixes: a4b1d3c1ddf6cb ("bpf: verifier: insert zero extension according to analysis result")
Reported-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Reviewed-by: Jiong Wang <jiong.wang@netronome.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agonfp: bpf: fix latency bug when updating stack index register
Jiong Wang [Sat, 24 Aug 2019 02:00:28 +0000 (19:00 -0700)]
nfp: bpf: fix latency bug when updating stack index register

NFP is using Local Memory to model stack. LM_addr could be used as base of
a 16 32-bit word region of Local Memory. Then, if the stack offset is
beyond the current region, the local index needs to be updated. The update
needs at least three cycles to take effect, therefore the sequence normally
looks like:

  local_csr_wr[ActLMAddr3, gprB_5]
  nop
  nop
  nop

If the local index switch happens on a narrow loads, then the instruction
preparing value to zero high 32-bit of the destination register could be
counted as one cycle, the sequence then could be something like:

  local_csr_wr[ActLMAddr3, gprB_5]
  nop
  nop
  immed[gprB_5, 0]

However, we have zero extension optimization that zeroing high 32-bit could
be eliminated, therefore above IMMED insn won't be available for which case
the first sequence needs to be generated.

Fixes: 0b4de1ff19bf ("nfp: bpf: eliminate zero extension code-gen")
Signed-off-by: Jiong Wang <jiong.wang@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agodrm/amdgpu: fix dma_fence_wait without reference
Christian König [Fri, 16 Aug 2019 12:56:35 +0000 (14:56 +0200)]
drm/amdgpu: fix dma_fence_wait without reference

We need to grab a reference to the fence we wait for.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 years agoNFS: Fix writepage(s) error handling to not report errors twice
Trond Myklebust [Sat, 24 Aug 2019 14:39:00 +0000 (10:39 -0400)]
NFS: Fix writepage(s) error handling to not report errors twice

If writepage()/writepages() saw an error, but handled it without
reporting it, we should not be re-reporting that error on exit.

Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
5 years agoNFS: Fix spurious EIO read errors
Trond Myklebust [Thu, 15 Aug 2019 16:26:05 +0000 (12:26 -0400)]
NFS: Fix spurious EIO read errors

If the client attempts to read a page, but the read fails due to some
spurious error (e.g. an ACCESS error or a timeout, ...) then we need
to allow other processes to retry.
Also try to report errors correctly when doing a synchronous readpage.

Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
5 years agopNFS/flexfiles: Don't time out requests on hard mounts
Trond Myklebust [Fri, 23 Aug 2019 12:37:50 +0000 (08:37 -0400)]
pNFS/flexfiles: Don't time out requests on hard mounts

If the mount is hard, we should ignore the 'io_maxretrans' module
parameter so that we always keep retrying.

Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
5 years agoSUNRPC: Handle connection breakages correctly in call_status()
Trond Myklebust [Fri, 16 Aug 2019 12:58:48 +0000 (08:58 -0400)]
SUNRPC: Handle connection breakages correctly in call_status()

If the connection breaks while we're waiting for a reply from the
server, then we want to immediately try to reconnect.

Fixes: ec6017d90359 ("SUNRPC fix regression in umount of a secure mount")
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
5 years agoRevert "NFSv4/flexfiles: Abort I/O early if the layout segment was invalidated"
Trond Myklebust [Fri, 16 Aug 2019 12:37:26 +0000 (08:37 -0400)]
Revert "NFSv4/flexfiles: Abort I/O early if the layout segment was invalidated"

This reverts commit a79f194aa4879e9baad118c3f8bb2ca24dbef765.
The mechanism for aborting I/O is racy, since we are not guaranteed that
the request is asleep while we're changing both task->tk_status and
task->tk_action.

Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Cc: stable@vger.kernel.org # v5.1
5 years agoSUNRPC: Handle EADDRINUSE and ENOBUFS correctly
Trond Myklebust [Fri, 16 Aug 2019 02:55:19 +0000 (22:55 -0400)]
SUNRPC: Handle EADDRINUSE and ENOBUFS correctly

If a connect or bind attempt returns EADDRINUSE, that means we want to
retry with a different port. It is not a fatal connection error.
Similarly, ENOBUFS is not fatal, but just indicates a memory allocation
issue. Retry after a short delay.

Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
5 years agopNFS/flexfiles: Turn off soft RPC calls
Trond Myklebust [Fri, 16 Aug 2019 00:18:48 +0000 (20:18 -0400)]
pNFS/flexfiles: Turn off soft RPC calls

The pNFS/flexfiles I/O requests are sent with the SOFTCONN flag set, so
they automatically time out if the connection breaks. It should
therefore not be necessary to have the soft flag set in addition.

Fixes: 5f01d9539496 ("nfs41: create NFSv3 DS connection if specified")
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
5 years agoSUNRPC: Don't handle errors if the bind/connect succeeded
Trond Myklebust [Thu, 15 Aug 2019 21:26:17 +0000 (17:26 -0400)]
SUNRPC: Don't handle errors if the bind/connect succeeded

Don't handle errors in call_bind_status()/call_connect_status()
if it turns out that a previous call caused it to succeed.

Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Cc: stable@vger.kernel.org # v5.1+
5 years agox86/apic: Include the LDR when clearing out APIC registers
Bandan Das [Mon, 26 Aug 2019 10:15:13 +0000 (06:15 -0400)]
x86/apic: Include the LDR when clearing out APIC registers

Although APIC initialization will typically clear out the LDR before
setting it, the APIC cleanup code should reset the LDR.

This was discovered with a 32-bit KVM guest jumping into a kdump
kernel. The stale bits in the LDR triggered a bug in the KVM APIC
implementation which caused the destination mapping for VCPUs to be
corrupted.

Note that this isn't intended to paper over the KVM APIC bug. The kernel
has to clear the LDR when resetting the APIC registers except when X2APIC
is enabled.

This lacks a Fixes tag because missing to clear LDR goes way back into pre
git history.

[ tglx: Made x2apic_enabled a function call as required ]

Signed-off-by: Bandan Das <bsd@redhat.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/20190826101513.5080-3-bsd@redhat.com
5 years agox86/apic: Do not initialize LDR and DFR for bigsmp
Bandan Das [Mon, 26 Aug 2019 10:15:12 +0000 (06:15 -0400)]
x86/apic: Do not initialize LDR and DFR for bigsmp

Legacy apic init uses bigsmp for smp systems with 8 and more CPUs. The
bigsmp APIC implementation uses physical destination mode, but it
nevertheless initializes LDR and DFR. The LDR even ends up incorrectly with
multiple bit being set.

This does not cause a functional problem because LDR and DFR are ignored
when physical destination mode is active, but it triggered a problem on a
32-bit KVM guest which jumps into a kdump kernel.

The multiple bits set unearthed a bug in the KVM APIC implementation. The
code which creates the logical destination map for VCPUs ignores the
disabled state of the APIC and ends up overwriting an existing valid entry
and as a result, APIC calibration hangs in the guest during kdump
initialization.

Remove the bogus LDR/DFR initialization.

This is not intended to work around the KVM APIC bug. The LDR/DFR
ininitalization is wrong on its own.

The issue goes back into the pre git history. The fixes tag is the commit
in the bitkeeper import which introduced bigsmp support in 2003.

  git://git.kernel.org/pub/scm/linux/kernel/git/tglx/history.git

Fixes: db7b9e9f26b8 ("[PATCH] Clustered APIC setup for >8 CPU systems")
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Bandan Das <bsd@redhat.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/20190826101513.5080-2-bsd@redhat.com
5 years agoarc: prefer __section from compiler_attributes.h
Nick Desaulniers [Mon, 12 Aug 2019 21:50:35 +0000 (14:50 -0700)]
arc: prefer __section from compiler_attributes.h

Reported-by: Sedat Dilek <sedat.dilek@gmail.com>
Suggested-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
5 years agodt-bindings: IDU-intc: Add support for edge-triggered interrupts
Mischa Jonker [Wed, 24 Jul 2019 12:04:36 +0000 (14:04 +0200)]
dt-bindings: IDU-intc: Add support for edge-triggered interrupts

This updates the documentation for supporting an optional extra interrupt
cell to specify edge vs level triggered.

Signed-off-by: Mischa Jonker <mischa.jonker@synopsys.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
5 years agodt-bindings: IDU-intc: Clean up documentation
Mischa Jonker [Wed, 24 Jul 2019 12:04:35 +0000 (14:04 +0200)]
dt-bindings: IDU-intc: Clean up documentation

* Some lines exceeded 80 characters.
* Clarified statement about AUX register interface

Signed-off-by: Mischa Jonker <mischa.jonker@synopsys.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
5 years agoARCv2: IDU-intc: Add support for edge-triggered interrupts
Mischa Jonker [Wed, 24 Jul 2019 12:04:34 +0000 (14:04 +0200)]
ARCv2: IDU-intc: Add support for edge-triggered interrupts

This adds support for an optional extra interrupt cell to specify edge
vs level triggered. It is backward compatible with dts files with only
one cell, and will default to level-triggered in such a case.

Note that I had to make a change to idu_irq_set_affinity as well, as
this function was setting the interrupt type to "level" unconditionally,
since this was the only type supported previously.

Signed-off-by: Mischa Jonker <mischa.jonker@synopsys.com>
Reviewed-by: Vineet Gupta <vgupta@synopsys.com>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
5 years agoALSA: oxfw: fix to handle correct stream for PCM playback
Takashi Sakamoto [Mon, 26 Aug 2019 13:55:15 +0000 (22:55 +0900)]
ALSA: oxfw: fix to handle correct stream for PCM playback

When userspace application calls ioctl(2) to configure hardware for PCM
playback substream, ALSA OXFW driver handles incoming AMDTP stream.
In this case, outgoing AMDTP stream should be handled.

This commit fixes the bug for v5.3-rc kernel.

Fixes: 4f380d007052 ("ALSA: oxfw: configure packet format in pcm.hw_params callback")
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
5 years agouprobes/x86: Fix detection of 32-bit user mode
Sebastian Mayr [Sun, 28 Jul 2019 15:26:17 +0000 (17:26 +0200)]
uprobes/x86: Fix detection of 32-bit user mode

32-bit processes running on a 64-bit kernel are not always detected
correctly, causing the process to crash when uretprobes are installed.

The reason for the crash is that in_ia32_syscall() is used to determine the
process's mode, which only works correctly when called from a syscall.

In the case of uretprobes, however, the function is called from a exception
and always returns 'false' on a 64-bit kernel. In consequence this leads to
corruption of the process's return address.

Fix this by using user_64bit_mode() instead of in_ia32_syscall(), which
is correct in any situation.

[ tglx: Add a comment and the following historical info ]

This should have been detected by the rename which happened in commit

  abfb9498ee13 ("x86/entry: Rename is_{ia32,x32}_task() to in_{ia32,x32}_syscall()")

which states in the changelog:

    The is_ia32_task()/is_x32_task() function names are a big misnomer: they
    suggests that the compat-ness of a system call is a task property, which
    is not true, the compatness of a system call purely depends on how it
    was invoked through the system call layer.
    .....

and then it went and blindly renamed every call site.

Sadly enough this was already mentioned here:

   8faaed1b9f50 ("uprobes/x86: Introduce sizeof_long(), cleanup adjust_ret_addr() and
arch_uretprobe_hijack_return_addr()")

where the changelog says:

    TODO: is_ia32_task() is not what we actually want, TS_COMPAT does
    not necessarily mean 32bit. Fortunately syscall-like insns can't be
    probed so it actually works, but it would be better to rename and
    use is_ia32_frame().

and goes all the way back to:

    0326f5a94dde ("uprobes/core: Handle breakpoint and singlestep exceptions")

Oh well. 7+ years until someone actually tried a uretprobe on a 32bit
process on a 64bit kernel....

Fixes: 0326f5a94dde ("uprobes/core: Handle breakpoint and singlestep exceptions")
Signed-off-by: Sebastian Mayr <me@sam.st>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Dmitry Safonov <dsafonov@virtuozzo.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/20190728152617.7308-1-me@sam.st
5 years agophy: qcom-qmp: Correct ready status, again
Bjorn Andersson [Tue, 6 Aug 2019 00:42:56 +0000 (17:42 -0700)]
phy: qcom-qmp: Correct ready status, again

Despite extensive testing of commit 885bd765963b ("phy: qcom-qmp: Correct
READY_STATUS poll break condition") I failed to conclude that the
PHYSTATUS bit of the PCS_STATUS register used in PCIe and USB3 falls as
the PHY gets ready. Similar to the prior bug with UFS the code will
generally get past the check before the transition and thereby
"succeed".

Correct the name of the register used PCIe and USB3 PHYs, replace
mask_pcs_ready with a constant expression depending on the type of the
PHY and check for the appropriate ready state.

Cc: stable@vger.kernel.org
Cc: Vivek Gautam <vivek.gautam@codeaurora.org>
Cc: Evan Green <evgreen@chromium.org>
Cc: Niklas Cassel <niklas.cassel@linaro.org>
Reported-by: Marc Gonzalez <marc.w.gonzalez@free.fr>
Fixes: 885bd765963b ("phy: qcom-qmp: Correct READY_STATUS poll break condition")
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Tested-by: Marc Gonzalez <marc.w.gonzalez@free.fr>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agophy: qualcomm: phy-qcom-qmp: Add of_node_put() before return
Nishka Dasgupta [Thu, 8 Aug 2019 07:29:37 +0000 (12:59 +0530)]
phy: qualcomm: phy-qcom-qmp: Add of_node_put() before return

Each iteration of for_each_available_child_of_node() puts the previous
node, but in the case of a return from the middle of the loop, there is
no put, thus causing a memory leak. Hence create a new label,
err_node_put, that puts the previous node (child) before returning the
required value. Also include the statement pm_runtime_disable() under
this label in order to avoid repetition among mid-loop return
conditions. Edit the mid-loop return statements to instead go to this
new label err_node_put.
Issue found with Coccinelle.

Signed-off-by: Nishka Dasgupta <nishkadg.linux@gmail.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agophy: renesas: rcar-gen3-usb2: Disable clearing VBUS in over-current
Yoshihiro Shimoda [Tue, 6 Aug 2019 08:51:19 +0000 (17:51 +0900)]
phy: renesas: rcar-gen3-usb2: Disable clearing VBUS in over-current

The hardware manual should be revised, but the initial value of
VBCTRL.OCCLREN is set to 1 actually. If the bit is set, the hardware
clears VBCTRL.VBOUT and ADPCTRL.DRVVBUS registers automatically
when the hardware detects over-current signal from a USB power switch.
However, since the hardware doesn't have any registers which
indicates over-current, the driver cannot handle it at all. So, if
"is_otg_channel" hardware detects over-current, since ADPCTRL.DRVVBUS
register is cleared automatically, the channel cannot be used after
that.

To resolve this behavior, this patch sets the VBCTRL.OCCLREN to 0
to keep ADPCTRL.DRVVBUS even if the "is_otg_channel" hardware
detects over-current. (We assume a USB power switch itself protects
over-current and turns the VBUS off.)

This patch is inspired by a BSP patch from Kazuya Mizuguchi.

Fixes: 1114e2d31731 ("phy: rcar-gen3-usb2: change the mode to OTG on the combined channel")
Cc: <stable@vger.kernel.org> # v4.5+
Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agophy: ti: am654-serdes: fix an use-after-free in serdes_am654_clk_register()
Wen Yang [Mon, 8 Jul 2019 06:19:05 +0000 (14:19 +0800)]
phy: ti: am654-serdes: fix an use-after-free in serdes_am654_clk_register()

The regmap_node variable is still being used in the syscon_node_to_regmap()
call after the of_node_put() call, which may result in use-after-free.

Fixes: 71e2f5c5c224 ("phy: ti: Add a new SERDES driver for TI's AM654x SoC")
Signed-off-by: Wen Yang <wen.yang99@zte.com.cn>
Cc: Kishon Vijay Abraham I <kishon@ti.com>
Cc: Roger Quadros <rogerq@ti.com>
Reviewed-by: Roger Quadros <rogerq@ti.com>
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
5 years agox86/apic: Fix arch_dynirq_lower_bound() bug for DT enabled machines
Thomas Gleixner [Wed, 21 Aug 2019 13:16:31 +0000 (15:16 +0200)]
x86/apic: Fix arch_dynirq_lower_bound() bug for DT enabled machines

Rahul Tanwar reported the following bug on DT systems:

> 'ioapic_dynirq_base' contains the virtual IRQ base number. Presently, it is
> updated to the end of hardware IRQ numbers but this is done only when IOAPIC
> configuration type is IOAPIC_DOMAIN_LEGACY or IOAPIC_DOMAIN_STRICT. There is
> a third type IOAPIC_DOMAIN_DYNAMIC which applies when IOAPIC configuration
> comes from devicetree.
>
> See dtb_add_ioapic() in arch/x86/kernel/devicetree.c
>
> In case of IOAPIC_DOMAIN_DYNAMIC (DT/OF based system), 'ioapic_dynirq_base'
> remains to zero initialized value. This means that for OF based systems,
> virtual IRQ base will get set to zero.

Such systems will very likely not even boot.

For DT enabled machines ioapic_dynirq_base is irrelevant and not
updated, so simply map the IRQ base 1:1 instead.

Reported-by: Rahul Tanwar <rahul.tanwar@linux.intel.com>
Tested-by: Rahul Tanwar <rahul.tanwar@linux.intel.com>
Tested-by: Andy Shevchenko <andriy.shevchenko@intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: alan@linux.intel.com
Cc: bp@alien8.de
Cc: cheol.yong.kim@intel.com
Cc: qi-ming.wu@intel.com
Cc: rahul.tanwar@intel.com
Cc: rppt@linux.ibm.com
Cc: tony.luck@intel.com
Link: http://lkml.kernel.org/r/20190821081330.1187-1-rahul.tanwar@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
5 years agoACPI / property: Add two new Thunderbolt property GUIDs to the list
Mika Westerberg [Tue, 14 May 2019 13:56:21 +0000 (16:56 +0300)]
ACPI / property: Add two new Thunderbolt property GUIDs to the list

Ice Lake Thunderbolt controller includes two new device property
compatible properties that we need to be able to extract in the driver
so add them to the growing array of GUIDs.

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
5 years agothunderbolt: Add support for Intel Ice Lake
Mika Westerberg [Tue, 16 Jan 2018 20:19:00 +0000 (22:19 +0200)]
thunderbolt: Add support for Intel Ice Lake

The Thunderbolt controller is integrated into the Ice Lake CPU itself
and requires special flows to power it on and off using force power bit
in NHI VSEC registers. Runtime PM (RTD3) and Sx flows also differ from
the discrete solutions. Now the firmware notifies the driver whether
RTD3 entry or exit are possible. The driver is responsible of sending
Go2Sx command through link controller mailbox when system enters Sx
states (suspend-to-mem/disk). Rest of the ICM firwmare flows follow
Titan Ridge.

Signed-off-by: Raanan Avargil <raanan.avargil@intel.com>
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Yehezkel Bernat <YehezkelShB@gmail.com>
Tested-by: Mario Limonciello <mario.limonciello@dell.com>
5 years agothunderbolt: Expose active parts of NVM even if upgrade is not supported
Mika Westerberg [Tue, 5 Feb 2019 09:51:40 +0000 (12:51 +0300)]
thunderbolt: Expose active parts of NVM even if upgrade is not supported

Ice Lake Thunderbolt controller NVM firmware is part of the BIOS image
which means it is not writable through the DMA port anymore. However, we
can still read it so we can keep nvm_version and active parts of NVM.
This way users still can find out the active NVM version and other
potentially useful information directly from Linux.

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Yehezkel Bernat <YehezkelShB@gmail.com>
Tested-by: Mario Limonciello <mario.limonciello@dell.com>
5 years agothunderbolt: Hide switch attributes that are not set
Mika Westerberg [Tue, 11 Sep 2018 12:34:23 +0000 (15:34 +0300)]
thunderbolt: Hide switch attributes that are not set

Thunderbolt host routers may not always contain DROM that includes
device identification information. This is mostly needed for Ice Lake
systems but some Falcon Ridge controllers on PCs also do not have DROM.

In that case hide the identification attributes.

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Yehezkel Bernat <YehezkelShB@gmail.com>
Tested-by: Mario Limonciello <mario.limonciello@dell.com>
5 years agothunderbolt: Do not fail adding switch if some port is not implemented
Mika Westerberg [Wed, 4 Jul 2018 05:50:01 +0000 (08:50 +0300)]
thunderbolt: Do not fail adding switch if some port is not implemented

There are two ways to mark a port as unimplemented. Typical way is to
return port type as TB_TYPE_INACTIVE when its config space is read.
Alternatively if the port is not physically present (such as ports 10
and 11 in ICL) reading from port config space returns
TB_CFG_ERROR_INVALID_CONFIG_SPACE instead. Currently the driver bails
out from adding the switch if it receives any error during port
inititialization which is wrong.

Handle this properly and just leave the port as TB_TYPE_INACTIVE before
continuing to the next port.

This also allows us to get rid of special casing for Light Ridge port 5
in eeprom.c.

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Yehezkel Bernat <YehezkelShB@gmail.com>
Tested-by: Mario Limonciello <mario.limonciello@dell.com>
5 years agothunderbolt: Use 32-bit writes when writing ring producer/consumer
Mika Westerberg [Wed, 4 Jul 2018 05:46:07 +0000 (08:46 +0300)]
thunderbolt: Use 32-bit writes when writing ring producer/consumer

The register access should be using 32-bit reads/writes according to the
datasheet. With the previous generation hardware 16-bit writes have been
working but starting with ICL this is not the case anymore so fix
producer/consumer register update to use correct width register address.

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Yehezkel Bernat <YehezkelShB@gmail.com>
Tested-by: Mario Limonciello <mario.limonciello@dell.com>
5 years agothunderbolt: Move NVM upgrade support flag to struct icm
Mika Westerberg [Tue, 16 Jan 2018 20:30:40 +0000 (22:30 +0200)]
thunderbolt: Move NVM upgrade support flag to struct icm

This is depends on the controller and on the platform/CPU we are
running. Move it to struct icm so we can set it per controller.

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Yehezkel Bernat <YehezkelShB@gmail.com>
Tested-by: Mario Limonciello <mario.limonciello@dell.com>
5 years agothunderbolt: Correct path indices for PCIe tunnel
Mika Westerberg [Tue, 11 Jun 2019 16:31:26 +0000 (19:31 +0300)]
thunderbolt: Correct path indices for PCIe tunnel

PCIe tunnel path indices got mixed up when we added support for tunnels
between switches that are not adjacent. This did not affect the
functionality as it is just an index but fix it now nevertheless to make
the code easier to understand.

Reported-by: Rajmohan Mani <rajmohan.mani@intel.com>
Fixes: 8c7acaaf020f ("thunderbolt: Extend tunnel creation to more than 2 adjacent switches")
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Yehezkel Bernat <YehezkelShB@gmail.com>
5 years agor8152: Set memory to all 0xFFs on failed reg reads
Prashant Malani [Sat, 24 Aug 2019 08:36:19 +0000 (01:36 -0700)]
r8152: Set memory to all 0xFFs on failed reg reads

get_registers() blindly copies the memory written to by the
usb_control_msg() call even if the underlying urb failed.

This could lead to junk register values being read by the driver, since
some indirect callers of get_registers() ignore the return values. One
example is:
  ocp_read_dword() ignores the return value of generic_ocp_read(), which
  calls get_registers().

So, emulate PCI "Master Abort" behavior by setting the buffer to all
0xFFs when usb_control_msg() fails.

This patch is copied from the r8152 driver (v2.12.0) published by
Realtek (www.realtek.com).

Signed-off-by: Prashant Malani <pmalani@chromium.org>
Acked-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>