deliverable/linux.git
8 years agogre: reject GUE and FOU in collect metadata mode
Jiri Benc [Wed, 27 Apr 2016 12:08:01 +0000 (14:08 +0200)] 
gre: reject GUE and FOU in collect metadata mode

The collect metadata mode does not support GUE nor FOU. This might be
implemented later; until then, we should reject such config.

I think this is okay to be changed. It's unlikely anyone has such
configuration (as it doesn't work anyway) and we may need a way to
distinguish whether it's supported or not by the kernel later.

For backwards compatibility with iproute2, it's not possible to just check
the attribute presence (iproute2 always includes the attribute), the actual
value has to be checked, too.

Fixes: 2e15ea390e6f4 ("ip_gre: Add support to collect tunnel metadata.")
Signed-off-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'pegasus-sizes'
David S. Miller [Thu, 28 Apr 2016 21:05:25 +0000 (17:05 -0400)] 
Merge branch 'pegasus-sizes'

Petko Manolov says:

====================
pegasus: correct buffer & packet sizes

As noticed by Lincoln Ramsay <a1291762@gmail.com> some old (usb 1.1) Pegasus
based devices may actually return more bytes than the specified in the datasheet
amount.  That would not be a problem if the allocated space for the SKB was
equal to the parameter passed to usb_fill_bulk_urb().  Some poor bugger (i
really hope it was not me, but 'git blame' is useless in this case, so anyway)
decided to add '+ 8' to the buffer length parameter.  Sometimes the usb transfer
overflows and corrupts the socket structure, leading to kernel panic.

The above doesn't seem to happen for newer (Pegasus2 based) devices which did
help this bug to hide for so long.

The new default is to not include the CRC at the end of each received package.
So far CRC has been ignored which makes no sense to do it in a first place.

The patch is against v4.6-rc5 and was tested on ADM8515 device by transferring
multiple gigabytes of data over a couple of days without any complaints from the
kernel.  Please apply it to whatever net tree you deem fit.

Changes since v1:

 - split the patch in two parts;
 - corrected the subject lines;

Changes since v2:

 - do not append CRC by default (based on a discussion with Johannes Berg);
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agopegasus: fixes reported packet length
Petko Manolov [Wed, 27 Apr 2016 11:24:50 +0000 (14:24 +0300)] 
pegasus: fixes reported packet length

The default Pegasus setup was to append the status and CRC at the end of each
received packet.  The status bits are used to update various stats, but CRC has
been ignored.  The new default is to not append CRC at the end of RX packets.

Signed-off-by: Petko Manolov <petkan@mip-labs.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agopegasus: fixes URB buffer allocation size;
Petko Manolov [Wed, 27 Apr 2016 11:24:49 +0000 (14:24 +0300)] 
pegasus: fixes URB buffer allocation size;

usb_fill_bulk_urb() receives buffer length parameter 8 bytes larger
than what's allocated by alloc_skb(); This seems to be a problem with
older (pegasus usb-1.1) devices, which may silently return more data
than the maximal packet length.

Reported-by: Lincoln Ramsay <a1291762@gmail.com>
Signed-off-by: Petko Manolov <petkan@mip-labs.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'gre-lwt-fixes'
David S. Miller [Thu, 28 Apr 2016 21:02:45 +0000 (17:02 -0400)] 
Merge branch 'gre-lwt-fixes'

Jiri Benc says:

====================
gre: fix lwtunnel support

This patchset fixes a few bugs in ipgre metadata mode implementation.

As an example, in this setup:

ip a a 192.168.1.1/24 dev eth0
ip l a gre1 type gre external
ip l s gre1 up
ip a a 192.168.99.1/24 dev gre1
ip r a 192.168.99.2/32 encap ip dst 192.168.1.2 ttl 10 dev gre1
ping 192.168.99.2

the traffic does not go through before this patchset and does as expected
with it applied.

v3: Back to v1 in order not to break existing users. Dropped patch 3, will
    be fixed in iproute2 instead.
v2: Rejecting invalid configuration, added patch 3, dropped patch for
    ETH_P_TEB (will target net-next).
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agogre: build header correctly for collect metadata tunnels
Jiri Benc [Wed, 27 Apr 2016 09:29:07 +0000 (11:29 +0200)] 
gre: build header correctly for collect metadata tunnels

In ipgre (i.e. not gretap) + collect metadata mode, the skb was assumed to
contain Ethernet header and was encapsulated as ETH_P_TEB. This is not the
case, the interface is ARPHRD_IPGRE and the protocol to be used for
encapsulation is skb->protocol.

Fixes: 2e15ea390e6f4 ("ip_gre: Add support to collect tunnel metadata.")
Signed-off-by: Jiri Benc <jbenc@redhat.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agogre: do not assign header_ops in collect metadata mode
Jiri Benc [Wed, 27 Apr 2016 09:29:06 +0000 (11:29 +0200)] 
gre: do not assign header_ops in collect metadata mode

In ipgre mode (i.e. not gretap) with collect metadata flag set, the tunnel
is incorrectly assumed to be mGRE in NBMA mode (see commit 6a5f44d7a048c).
This is not the case, we're controlling the encapsulation addresses by
lwtunnel metadata. And anyway, assigning dev->header_ops in collect metadata
mode does not make sense.

Although it would be more user firendly to reject requests that specify
both the collect metadata flag and a remote/local IP address, this would
break current users of gretap or introduce ugly code and differences in
handling ipgre and gretap configuration. Keep the current behavior of
remote/local IP address being ignored in such case.

v3: Back to v1, added explanation paragraph.
v2: Reject configuration specifying both remote/local address and collect
    metadata flag.

Fixes: 2e15ea390e6f4 ("ip_gre: Add support to collect tunnel metadata.")
Signed-off-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge tag 'mac80211-for-davem-2016-04-27' of git://git.kernel.org/pub/scm/linux/kerne...
David S. Miller [Thu, 28 Apr 2016 20:55:26 +0000 (16:55 -0400)] 
Merge tag 'mac80211-for-davem-2016-04-27' of git://git./linux/kernel/git/jberg/mac80211

Johannes Berg says:

====================
Just a single fix, for a per-CPU memory leak in a
(root user triggerable) error case.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: phy: at803x: only the AT8030 needs a hardware reset on link change
Timur Tabi [Tue, 26 Apr 2016 17:44:18 +0000 (12:44 -0500)] 
net: phy: at803x: only the AT8030 needs a hardware reset on link change

Commit 13a56b44 ("at803x: Add support for hardware reset") added a
work-around for a hardware bug on the AT8030.  However, the work-around
was being called for all 803x PHYs, even those that don't need it.
Function at803x_link_change_notify() checks to make sure that it only
resets the PHY on the 8030, but it makes more sense to not call that
function at all if it isn't needed.

Signed-off-by: Timur Tabi <timur@codeaurora.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge tag 'batman-adv-fix-for-davem' of git://git.open-mesh.org/linux-merge
David S. Miller [Thu, 28 Apr 2016 20:42:40 +0000 (16:42 -0400)] 
Merge tag 'batman-adv-fix-for-davem' of git://git.open-mesh.org/linux-merge

Antonio Quartulli says:

====================
In this patchset you can find the following fixes:

1) check skb size to avoid reading beyond its border when delivering
   payloads, by Sven Eckelmann
2) initialize last_seen time in neigh_node object to prevent cleanup
   routine from accidentally purge it, by Marek Lindner
3) release "recently added" slave interfaces upon virtual/batman
   interface shutdown, by Sven Eckelmann
4) properly decrease router object reference counter upon routing table
   update, by Sven Eckelmann
5) release queue slots when purging OGM packets of deactivating slave
   interface, by Linus Lüssing

Patch 2 and 3 have no "Fixes:" tag because the offending commits date
back to when batman-adv was not yet officially in the net tree.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agops3_gelic: fix memcpy parameter
Christophe Jaillet [Tue, 26 Apr 2016 02:33:43 +0000 (04:33 +0200)] 
ps3_gelic: fix memcpy parameter

The size allocated for target->hwinfo and the number of bytes copied in it
should be consistent.

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agolan78xx: workaround of forced 100 Full/Half duplex mode error
Woojung Huh [Mon, 25 Apr 2016 22:22:36 +0000 (22:22 +0000)] 
lan78xx: workaround of forced 100 Full/Half duplex mode error

At forced 100 Full & Half duplex mode, chip may fail to set mode correctly
when cable is switched between long(~50+m) and short one.
As workaround, set to 10 before setting to 100 at forced 100 F/H mode.

Signed-off-by: Woojung Huh <woojung.huh@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agolan78xx: fix statistics counter error
Woojung Huh [Mon, 25 Apr 2016 22:22:32 +0000 (22:22 +0000)] 
lan78xx: fix statistics counter error

Fix rx_bytes, tx_bytes and tx_frames error in netdev.stats.
- rx_bytes counted bytes excluding size of struct ethhdr.
- tx_packets didn't count multiple packets in a single urb
- tx_bytes included 8 bytes of extra commands.

Signed-off-by: Woojung Huh <woojung.huh@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dsa: mv88e6xxx: fix uninitialized error return
Colin Ian King [Mon, 25 Apr 2016 22:11:22 +0000 (23:11 +0100)] 
net: dsa: mv88e6xxx: fix uninitialized error return

The error return err is not initialized and there is a possibility
that err is not assigned causing mv88e6xxx_port_bridge_join to
return a garbage error return status. Fix this by initializing err
to 0.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: fix net_gso_ok for new GSO types.
Marcelo Ricardo Leitner [Mon, 25 Apr 2016 18:13:17 +0000 (15:13 -0300)] 
net: fix net_gso_ok for new GSO types.

Fix casting in net_gso_ok. Otherwise the shift on
gso_type << NETIF_F_GSO_SHIFT may hit the 32th bit and make it look like
a INT_MIN, which is then promoted from signed to uint64 which is
0xffffffff80000000, resulting in wrong behavior when it is and'ed with
the feature itself, as in:

This test app:
#include <stdio.h>
#include <stdint.h>

int main(int argc, char **argv)
{
uint64_t feature1;
uint64_t feature2;
int gso_type = 1 << 15;

feature1 = gso_type << 16;
feature2 = (uint64_t)gso_type << 16;
printf("%lx %lx\n", feature1, feature2);

return 0;
}

Gives:
ffffffff80000000 80000000

So that this:
   return (features & feature) == feature;
Actually works on more bits than expected and invalid ones.

Fix is to promote it earlier.

Issue noted while rebasing SCTP GSO patch but posting separetely as
someone else may experience this meanwhile.

Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: ethernet: davinci_emac: Fix devioctl while in fixed link
Neil Armstrong [Mon, 25 Apr 2016 17:41:38 +0000 (19:41 +0200)] 
net: ethernet: davinci_emac: Fix devioctl while in fixed link

When configured in fixed link, the DaVinci emac driver sets the
priv->phydev to NULL and further ioctl calls to the phy_mii_ioctl()
causes the kernel to crash.

Cc: Brian Hutchinson <b.hutchman@gmail.com>
Fixes: 1bb6aa56bb38 ("net: davinci_emac: Add support for fixed-link PHY")
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Reviewed-by: Mugunthan V N <mugunthanvnm@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge tag 'wireless-drivers-for-davem-2016-04-25' of git://git.kernel.org/pub/scm...
David S. Miller [Thu, 28 Apr 2016 18:23:27 +0000 (14:23 -0400)] 
Merge tag 'wireless-drivers-for-davem-2016-04-25' of git://git./linux/kernel/git/kvalo/wireless-drivers

Kalle Valo says:

====================
wireless-drivers fixes for 4.6

ath9k

* fix a couple release old throughput regression on ar9281

iwlwifi

* add new device IDs for 8265
* fix a NULL pointer dereference when paging firmware asserts
* remove a WARNING on gscan capabilities
* fix MODULE_FIRMWARE for 8260
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMAINTAINERS: net: update sfc maintainers
Bert Kenward [Mon, 25 Apr 2016 16:42:12 +0000 (17:42 +0100)] 
MAINTAINERS: net: update sfc maintainers

Add myself and Edward Cree as maintainers.
Remove Shradha Shah, who is on extended leave.

Cc: David S. Miller <davem@davemloft.net>
Cc: Edward Cree <ecree@solarflare.com>
Cc: Shradha Shah <sshah@solarflare.com>
Signed-off-by: Bert Kenward <bkenward@solarflare.com>
Acked-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agosfc: disable RSS when unsupported
Jon Cooper [Mon, 25 Apr 2016 15:51:00 +0000 (16:51 +0100)] 
sfc: disable RSS when unsupported

When certain firmware variants are selected (via the sfboot utility) the
SFC7000 and SFC8000 series NICs don't support RSS. The driver still
tries (and fails) to insert filters with the RSS flag, and the NIC fails
to pass traffic.

When the firmware reports RSS_LIMITED suppress allocating a default RSS
context. The absence of an RSS context is picked up in filter insertion
and RSS flags are discarded.

Signed-off-by: Bert Kenward <bkenward@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomyri10ge: fix sleeping with bh disabled
Stanislaw Gruszka [Mon, 25 Apr 2016 08:59:19 +0000 (10:59 +0200)] 
myri10ge: fix sleeping with bh disabled

napi_disable() can not be called with bh disabled, move locking just
around myri10ge_ss_lock_napi() .

Patches fixes following bug:

[  114.278378] BUG: sleeping function called from invalid context at net/core/dev.c:4383
<snip>
[  114.313712] Call Trace:
[  114.314943]  [<ffffffff817010ce>] dump_stack+0x19/0x1b
[  114.317673]  [<ffffffff810ce7f3>] __might_sleep+0x173/0x230
[  114.320566]  [<ffffffff815b3117>] napi_disable+0x27/0x90
[  114.323254]  [<ffffffffa01e437f>] myri10ge_close+0xbf/0x3f0 [myri10ge]

Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Acked-by: Hyong-Youb Kim <hykim@myri.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoDocumentation: networking: fix spelling mistakes
Eric Engestrom [Mon, 25 Apr 2016 06:36:56 +0000 (07:36 +0100)] 
Documentation: networking: fix spelling mistakes

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agodrm/vmwgfx: Fix order of operation
Sinclair Yeh [Thu, 21 Apr 2016 18:29:31 +0000 (11:29 -0700)] 
drm/vmwgfx: Fix order of operation

mode->hdisplay * (var->bits_per_pixel + 7) gets evaluated before
the division, potentially making the pitch larger than it should
be.

Since the original intention is to do a div-round-up, just use
the macro instead.

Signed-off-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
8 years agodrm/vmwgfx: use vmw_cmd_dx_cid_check for query commands.
Charmaine Lee [Tue, 12 Apr 2016 15:19:08 +0000 (08:19 -0700)] 
drm/vmwgfx: use vmw_cmd_dx_cid_check for query commands.

Instead of calling vmw_cmd_ok, call vmw_cmd_dx_cid_check to
validate the context id for query commands.

Signed-off-by: Charmaine Lee <charmainel@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
8 years agodrm/vmwgfx: Enable SVGA_3D_CMD_DX_SET_PREDICATION
Charmaine Lee [Tue, 12 Apr 2016 15:14:23 +0000 (08:14 -0700)] 
drm/vmwgfx: Enable SVGA_3D_CMD_DX_SET_PREDICATION

Fixes piglit tests nv_conditional_render-* crashes.

Signed-off-by: Charmaine Lee <charmainel@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
8 years agoIB/security: Restrict use of the write() interface
Jason Gunthorpe [Mon, 11 Apr 2016 01:13:13 +0000 (19:13 -0600)] 
IB/security: Restrict use of the write() interface

The drivers/infiniband stack uses write() as a replacement for
bi-directional ioctl().  This is not safe. There are ways to
trigger write calls that result in the return structure that
is normally written to user space being shunted off to user
specified kernel memory instead.

For the immediate repair, detect and deny suspicious accesses to
the write API.

For long term, update the user space libraries and the kernel API
to something that doesn't present the same security vulnerabilities
(likely a structured ioctl() interface).

The impacted uAPI interfaces are generally only available if
hardware from drivers/infiniband is installed in the system.

Reported-by: Jann Horn <jann@thejh.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
[ Expanded check to all known write() entry points ]
Cc: stable@vger.kernel.org
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/hfi1: Use kernel default llseek for ui device
Dean Luick [Fri, 22 Apr 2016 18:17:03 +0000 (11:17 -0700)] 
IB/hfi1: Use kernel default llseek for ui device

The ui device llseek had a mistake with SEEK_END and did
not fully follow seek semantics.  Correct all this by
using a kernel supplied function for fixed size devices.

Cc: Al Viro <viro@ZenIV.linux.org.uk>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/hfi1: Don't attempt to free resources if initialization failed
Mitko Haralanov [Wed, 20 Apr 2016 13:05:36 +0000 (06:05 -0700)] 
IB/hfi1: Don't attempt to free resources if initialization failed

Attempting to free resources which have not been allocated and
initialized properly led to the following kernel backtrace:

    BUG: unable to handle kernel NULL pointer dereference at           (null)
    IP: [<ffffffffa09658fe>] unlock_exp_tids.isra.8+0x2e/0x120 [hfi1]
    PGD 852a43067 PUD 85d4a6067 PMD 0
    Oops: 0000 [#1] SMP
    CPU: 0 PID: 2831 Comm: osu_bw Tainted: G          IO 3.12.18-wfr+ #1
    task: ffff88085b15b540 ti: ffff8808588fe000 task.ti: ffff8808588fe000
    RIP: 0010:[<ffffffffa09658fe>]  [<ffffffffa09658fe>] unlock_exp_tids.isra.8+0x2e/0x120 [hfi1]
    RSP: 0018:ffff8808588ffde0  EFLAGS: 00010282
    RAX: 0000000000000000 RBX: ffff880858a31800 RCX: 0000000000000000
    RDX: ffff88085d971bc0 RSI: ffff880858a318f8 RDI: ffff880858a318c0
    RBP: ffff8808588ffe20 R08: 0000000000000000 R09: 0000000000000000
    R10: ffff88087ffd6f40 R11: 0000000001100348 R12: ffff880852900000
    R13: ffff880858a318c0 R14: 0000000000000000 R15: ffff88085d971be8
    FS:  00007f4674e83740(0000) GS:ffff88087f400000(0000) knlGS:0000000000000000
    CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
    CR2: 0000000000000000 CR3: 000000085c377000 CR4: 00000000001407f0
    Stack:
     ffffffffa0941a71 ffff880858a318f8 ffff88085d971bc0 ffff880858a31800
     ffff880852900000 ffff880858a31800 00000000003ffff7 ffff88085d971bc0
     ffff8808588ffe60 ffffffffa09663fc ffff8808588ffe60 ffff880858a31800
    Call Trace:
     [<ffffffffa0941a71>] ? find_mmu_handler+0x51/0x70 [hfi1]
     [<ffffffffa09663fc>] hfi1_user_exp_rcv_free+0x6c/0x120 [hfi1]
     [<ffffffffa0932809>] hfi1_file_close+0x1a9/0x340 [hfi1]
     [<ffffffff8116c189>] __fput+0xe9/0x270
     [<ffffffff8116c35e>] ____fput+0xe/0x10
     [<ffffffff81065707>] task_work_run+0xa7/0xe0
     [<ffffffff81002969>] do_notify_resume+0x59/0x80
     [<ffffffff814ffc1a>] int_signal+0x12/0x17

This commit re-arranges the context initialization code in a way that
would allow for context event flags to be used to determine whether
the context has been successfully initialized.

In turn, this can be used to skip the resource de-allocation if they
were never allocated in the first place.

Fixes: 3abb33ac6521 ("staging/hfi1: Add TID cache receive init and free funcs")
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com.
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/hfi1: Fix missing lock/unlock in verbs drain callback
Mike Marciniszyn [Wed, 20 Apr 2016 13:05:30 +0000 (06:05 -0700)] 
IB/hfi1: Fix missing lock/unlock in verbs drain callback

The iowait_sdma_drained() callback lacked locking to
protect the qp s_flags field.

This causes the s_flags to be out of sync
on multiple CPUs, potentially corrupting the s_flags.

Fixes: a545f5308b6c ("staging/rdma/hfi: fix CQ completion order issue")
Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Fix send scheduling
Jubin John [Tue, 12 Apr 2016 17:47:00 +0000 (10:47 -0700)] 
IB/rdmavt: Fix send scheduling

call_send is used to determine whether to send immediately or schedule
a send for later. The current logic in rdmavt is inverted and has a
negative impact on the latency of the hfi1 and qib drivers. Fix this
regression by correctly calling send immediately when call_send is set.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/hfi1: Prevent unpinning of wrong pages
Mitko Haralanov [Tue, 12 Apr 2016 17:46:16 +0000 (10:46 -0700)] 
IB/hfi1: Prevent unpinning of wrong pages

The routine used by the SDMA cache to handle already
cached nodes can extend an already existing node.

In its error handling code, the routine will unpin pages
when not all pages of the buffer extension were pinned.

There was a bug in that part of the routine, which would
mistakenly unpin pages from the original set rather than
the newly pinned pages.

This commit fixes that bug by offsetting the page array
to the proper place pointing at the beginning of the newly
pinned pages.

Reviewed-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/hfi1: Fix deadlock caused by locking with wrong scope
Mitko Haralanov [Tue, 12 Apr 2016 17:46:03 +0000 (10:46 -0700)] 
IB/hfi1: Fix deadlock caused by locking with wrong scope

The locking around the interval RB tree is designed to prevent
access to the tree while it's being modified. The locking in its
current form is too overzealous, which is causing a deadlock in
certain cases with the following backtrace:

    Kernel panic - not syncing: Watchdog detected hard LOCKUP on cpu 0
    CPU: 0 PID: 5836 Comm: IMB-MPI1 Tainted: G           O 3.12.18-wfr+ #1
     0000000000000000 ffff88087f206c50 ffffffff814f1caa ffffffff817b53f0
     ffff88087f206cc8 ffffffff814ecd56 0000000000000010 ffff88087f206cd8
     ffff88087f206c78 0000000000000000 0000000000000000 0000000000001662
    Call Trace:
     <NMI>  [<ffffffff814f1caa>] dump_stack+0x45/0x56
     [<ffffffff814ecd56>] panic+0xc2/0x1cb
     [<ffffffff810d4370>] ? restart_watchdog_hrtimer+0x50/0x50
     [<ffffffff810d4432>] watchdog_overflow_callback+0xc2/0xd0
     [<ffffffff81109b4e>] __perf_event_overflow+0x8e/0x2b0
     [<ffffffff8110a714>] perf_event_overflow+0x14/0x20
     [<ffffffff8101c906>] intel_pmu_handle_irq+0x1b6/0x390
     [<ffffffff814f927b>] perf_event_nmi_handler+0x2b/0x50
     [<ffffffff814f8ad8>] nmi_handle.isra.3+0x88/0x180
     [<ffffffff814f8d39>] do_nmi+0x169/0x310
     [<ffffffff814f8177>] end_repeat_nmi+0x1e/0x2e
     [<ffffffff81272600>] ? unmap_single+0x30/0x30
     [<ffffffff814f780d>] ? _raw_spin_lock_irqsave+0x2d/0x40
     [<ffffffff814f780d>] ? _raw_spin_lock_irqsave+0x2d/0x40
     [<ffffffff814f780d>] ? _raw_spin_lock_irqsave+0x2d/0x40
     <<EOE>>  <IRQ>  [<ffffffffa056c4a8>] hfi1_mmu_rb_search+0x38/0x70 [hfi1]
     [<ffffffffa05919cb>] user_sdma_free_request+0xcb/0x120 [hfi1]
     [<ffffffffa0593393>] user_sdma_txreq_cb+0x263/0x350 [hfi1]
     [<ffffffffa057fad7>] ? sdma_txclean+0x27/0x1c0 [hfi1]
     [<ffffffffa0593130>] ? user_sdma_send_pkts+0x1710/0x1710 [hfi1]
     [<ffffffffa057fdd6>] sdma_make_progress+0x166/0x480 [hfi1]
     [<ffffffff810762c9>] ? ttwu_do_wakeup+0x19/0xd0
     [<ffffffffa0581c7e>] sdma_engine_interrupt+0x8e/0x100 [hfi1]
     [<ffffffffa0546bdd>] sdma_interrupt+0x5d/0xa0 [hfi1]
     [<ffffffff81097e57>] handle_irq_event_percpu+0x47/0x1d0
     [<ffffffff81098017>] handle_irq_event+0x37/0x60
     [<ffffffff8109aa5f>] handle_edge_irq+0x6f/0x120
     [<ffffffff810044af>] handle_irq+0xbf/0x150
     [<ffffffff8104c9b7>] ? irq_enter+0x17/0x80
     [<ffffffff8150168d>] do_IRQ+0x4d/0xc0
     [<ffffffff814f7c6a>] common_interrupt+0x6a/0x6a
     <EOI>  [<ffffffff81073524>] ? finish_task_switch+0x54/0xe0
     [<ffffffff814f56c6>] __schedule+0x3b6/0x7e0
     [<ffffffff810763a6>] __cond_resched+0x26/0x30
     [<ffffffff814f5eda>] _cond_resched+0x3a/0x50
     [<ffffffff814f4f82>] down_write+0x12/0x30
     [<ffffffffa0591619>] hfi1_release_user_pages+0x69/0x90 [hfi1]
     [<ffffffffa059173a>] sdma_rb_remove+0x9a/0xc0 [hfi1]
     [<ffffffffa056c00d>] __mmu_rb_remove.isra.5+0x5d/0x70 [hfi1]
     [<ffffffffa056c536>] hfi1_mmu_rb_remove+0x56/0x70 [hfi1]
     [<ffffffffa059427b>] hfi1_user_sdma_process_request+0x74b/0x1160 [hfi1]
     [<ffffffffa055c763>] hfi1_aio_write+0xc3/0x100 [hfi1]
     [<ffffffff8116a14c>] do_sync_readv_writev+0x4c/0x80
     [<ffffffff8116b58b>] do_readv_writev+0xbb/0x230
     [<ffffffff811a9da1>] ? fsnotify+0x241/0x320
     [<ffffffff81073524>] ? finish_task_switch+0x54/0xe0
     [<ffffffff8116b795>] vfs_writev+0x35/0x60
     [<ffffffff8116b8c9>] SyS_writev+0x49/0xc0
     [<ffffffff810cd876>] ? __audit_syscall_exit+0x1f6/0x2a0
     [<ffffffff814ff992>] system_call_fastpath+0x16/0x1b

As evident from the backtrace above, the process was being put to sleep
while holding the lock.

Limiting the scope of the lock only to the RB tree operation fixes the
above error allowing for proper locking and the process being put to
sleep when needed.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/hfi1: Prevent NULL pointer deferences in caching code
Mitko Haralanov [Tue, 12 Apr 2016 17:45:57 +0000 (10:45 -0700)] 
IB/hfi1: Prevent NULL pointer deferences in caching code

There is a potential kernel crash when the MMU notifier calls the
invalidation routines in the hfi1 pinned page caching code for sdma.

The invalidation routine could call the remove callback
for the node, which in turn ends up dereferencing the
current task_struct to get a pointer to the mm_struct.
However, the mm_struct pointer could be NULL resulting in
the following backtrace:

    BUG: unable to handle kernel NULL pointer dereference at 00000000000000a8
    IP: [<ffffffffa041f75a>] sdma_rb_remove+0xaa/0x100 [hfi1]
    15
    task: ffff88085e66e080 ti: ffff88085c244000 task.ti: ffff88085c244000
    RIP: 0010:[<ffffffffa041f75a>]  [<ffffffffa041f75a>] sdma_rb_remove+0xaa/0x100 [hfi1]
    RSP: 0000:ffff88085c245878  EFLAGS: 00010002
    RAX: 0000000000000000 RBX: ffff88105b9bbd40 RCX: ffffea003931a830
    RDX: 0000000000000004 RSI: ffff88105754a9c0 RDI: ffff88105754a9c0
    RBP: ffff88085c245890 R08: ffff88105b9bbd70 R09: 00000000fffffffb
    R10: ffff88105b9bbd58 R11: 0000000000000013 R12: ffff88105754a9c0
    R13: 0000000000000001 R14: 0000000000000001 R15: ffff88105b9bbd40
    FS:  0000000000000000(0000) GS:ffff88107ef40000(0000) knlGS:0000000000000000
    CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
    CR2: 00000000000000a8 CR3: 0000000001a0b000 CR4: 00000000001407e0
    Stack:
     ffff88105b9bbd40 ffff88080ec481a8 ffff88080ec481b8 ffff88085c2458c0
     ffffffffa03fa00e ffff88080ec48190 ffff88080ed9cd00 0000000001024000
     0000000000000000 ffff88085c245920 ffffffffa03fa0e7 0000000000000282
    Call Trace:
     [<ffffffffa03fa00e>] __mmu_rb_remove.isra.5+0x5e/0x70 [hfi1]
     [<ffffffffa03fa0e7>] mmu_notifier_mem_invalidate+0xc7/0xf0 [hfi1]
     [<ffffffffa03fa143>] mmu_notifier_page+0x13/0x20 [hfi1]
     [<ffffffff81156dd0>] __mmu_notifier_invalidate_page+0x50/0x70
     [<ffffffff81140bbb>] try_to_unmap_one+0x20b/0x470
     [<ffffffff81141ee7>] try_to_unmap_anon+0xa7/0x120
     [<ffffffff81141fad>] try_to_unmap+0x4d/0x60
     [<ffffffff8111fd7b>] shrink_page_list+0x2eb/0x9d0
     [<ffffffff81120ab3>] shrink_inactive_list+0x243/0x490
     [<ffffffff81121491>] shrink_lruvec+0x4c1/0x640
     [<ffffffff81121641>] shrink_zone+0x31/0x100
     [<ffffffff81121b0f>] kswapd_shrink_zone.constprop.62+0xef/0x1c0
     [<ffffffff811229e3>] kswapd+0x403/0x7e0
     [<ffffffff811225e0>] ? shrink_all_memory+0xf0/0xf0
     [<ffffffff81068ac0>] kthread+0xc0/0xd0
     [<ffffffff81068a00>] ? insert_kthread_work+0x40/0x40
     [<ffffffff814ff8ec>] ret_from_fork+0x7c/0xb0
     [<ffffffff81068a00>] ? insert_kthread_work+0x40/0x40

To correct this, the mm_struct passed to us by the MMU notifier is
used (which is what should have been done to begin with). This avoids
the broken derefences and ensures that the correct mm_struct is used.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoMAINTAINERS: Update iser/isert maintainer contact info
Sagi Grimberg [Sun, 3 Apr 2016 12:03:12 +0000 (15:03 +0300)] 
MAINTAINERS: Update iser/isert maintainer contact info

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Acked-by: Nicholas Bellinger <nab@linux-iscsi.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/mlx5: Expose correct max_sge_rd limit
Sagi Grimberg [Thu, 31 Mar 2016 16:03:25 +0000 (19:03 +0300)] 
IB/mlx5: Expose correct max_sge_rd limit

mlx5 devices (Connect-IB, ConnectX-4, ConnectX-4-LX) has a limitation
where rdma read work queue entries cannot exceed 512 bytes.
A rdma_read wqe needs to fit in 512 bytes:
- wqe control segment (16 bytes)
- rdma segment (16 bytes)
- scatter elements (16 bytes each)

So max_sge_rd should be: (512 - 16 - 16) / 16 = 30.

Cc: linux-stable@vger.kernel.org
Reported-by: Christoph Hellwig <hch@lst.de>
Tested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagig@grimberg.me>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agommc: sunxi: Disable eMMC HS-DDR (MMC_CAP_1_8V_DDR) for Allwinner A80
Chen-Yu Tsai [Tue, 26 Apr 2016 15:37:26 +0000 (23:37 +0800)] 
mmc: sunxi: Disable eMMC HS-DDR (MMC_CAP_1_8V_DDR) for Allwinner A80

eMMC HS-DDR no longer works on the A80, despite it working when support
for this developed.

Disable it for now.

Signed-off-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
8 years agoperf/x86/intel: Fix incorrect lbr_sel_mask value
Kan Liang [Thu, 21 Apr 2016 09:30:10 +0000 (02:30 -0700)] 
perf/x86/intel: Fix incorrect lbr_sel_mask value

This patch fixes a bug which was introduced by:

 b16a5b52eb90 ("perf/x86: Add option to disable reading branch flags/cycles")

In this patch, lbr_sel_mask is used to mask the lbr_select. But LBR_SEL_MASK
doesn't include the bit for LBR_CALL_STACK. So LBR call stack will never be
set in lbr_select.

This patch corrects the LBR_SEL_MASK by including all valid bits in
LBR_SELECT. Also, the LBR_CALL_STACK bit is different as other bit in
LBR_SELECT. It does not operate in suppress mode, so it needs to be
specially handled in intel_pmu_setup_hw_lbr_filter.

Signed-off-by: Kan Liang <kan.liang@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Link: http://lkml.kernel.org/r/1461231010-4399-1-git-send-email-kan.liang@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
8 years agoperf/x86/intel/pt: Don't die on VMXON
Alexander Shishkin [Tue, 29 Mar 2016 14:43:10 +0000 (17:43 +0300)] 
perf/x86/intel/pt: Don't die on VMXON

Some versions of Intel PT do not support tracing across VMXON, more
specifically, VMXON will clear TraceEn control bit and any attempt to
set it before VMXOFF will throw a #GP, which in the current state of
things will crash the kernel. Namely:

  $ perf record -e intel_pt// kvm -nographic

on such a machine will kill it.

To avoid this, notify the intel_pt driver before VMXON and after
VMXOFF so that it knows when not to enable itself.

Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Gleb Natapov <gleb@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: hpa@zytor.com
Link: http://lkml.kernel.org/r/87oa9dwrfk.fsf@ashishki-desk.ger.corp.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
8 years agoperf/core: Fix perf_event_open() vs. execve() race
Peter Zijlstra [Tue, 26 Apr 2016 09:36:53 +0000 (11:36 +0200)] 
perf/core: Fix perf_event_open() vs. execve() race

Jann reported that the ptrace_may_access() check in
find_lively_task_by_vpid() is racy against exec().

Specifically:

  perf_event_open() execve()

  ptrace_may_access()
commit_creds()
  ... if (get_dumpable() != SUID_DUMP_USER)
  perf_event_exit_task();
  perf_install_in_context()

would result in installing a counter across the creds boundary.

Fix this by wrapping lots of perf_event_open() in cred_guard_mutex.
This should be fine as perf_event_exit_task() is already called with
cred_guard_mutex held, so all perf locks already nest inside it.

Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
8 years agoperf/x86/amd: Set the size of event map array to PERF_COUNT_HW_MAX
Adam Borowski [Wed, 27 Apr 2016 09:35:31 +0000 (11:35 +0200)] 
perf/x86/amd: Set the size of event map array to PERF_COUNT_HW_MAX

The entry for PERF_COUNT_HW_REF_CPU_CYCLES is not used on AMD, but is
referenced by filter_events() which expects undefined events to have a
value of 0.

Found via KASAN:

  UBSAN: Undefined behaviour in arch/x86/events/amd/core.c:132:30
  index 9 is out of range for type 'u64 [9]'
  UBSAN: Undefined behaviour in arch/x86/events/amd/core.c:132:9
  load of address ffffffff81c021c8 with insufficient space for an object of type 'const u64'

Signed-off-by: Adam Borowski <kilobyte@angband.pl>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Link: http://lkml.kernel.org/r/1461749731-30979-1-git-send-email-kilobyte@angband.pl
Signed-off-by: Ingo Molnar <mingo@kernel.org>
8 years agorbd: report unsupported features to syslog
Ilya Dryomov [Wed, 13 Apr 2016 12:15:50 +0000 (14:15 +0200)] 
rbd: report unsupported features to syslog

... instead of just returning an error.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Josh Durgin <jdurgin@redhat.com>
8 years agorbd: fix rbd map vs notify races
Ilya Dryomov [Fri, 15 Apr 2016 14:22:16 +0000 (16:22 +0200)] 
rbd: fix rbd map vs notify races

A while ago, commit 9875201e1049 ("rbd: fix use-after free of
rbd_dev->disk") fixed rbd unmap vs notify race by introducing
an exported wrapper for flushing notifies and sticking it into
do_rbd_remove().

A similar problem exists on the rbd map path, though: the watch is
registered in rbd_dev_image_probe(), while the disk is set up quite
a few steps later, in rbd_dev_device_setup().  Nothing prevents
a notify from coming in and crashing on a NULL rbd_dev->disk:

    BUG: unable to handle kernel NULL pointer dereference at 0000000000000050
    Call Trace:
     [<ffffffffa0508344>] rbd_watch_cb+0x34/0x180 [rbd]
     [<ffffffffa04bd290>] do_event_work+0x40/0xb0 [libceph]
     [<ffffffff8109d5db>] process_one_work+0x17b/0x470
     [<ffffffff8109e3ab>] worker_thread+0x11b/0x400
     [<ffffffff8109e290>] ? rescuer_thread+0x400/0x400
     [<ffffffff810a5acf>] kthread+0xcf/0xe0
     [<ffffffff810b41b3>] ? finish_task_switch+0x53/0x170
     [<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
     [<ffffffff81645dd8>] ret_from_fork+0x58/0x90
     [<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
    RIP  [<ffffffffa050828a>] rbd_dev_refresh+0xfa/0x180 [rbd]

If an error occurs during rbd map, we have to error out, potentially
tearing down a watch.  Just like on rbd unmap, notifies have to be
flushed, otherwise rbd_watch_cb() may end up trying to read in the
image header after rbd_dev_image_release() has run:

    Assertion failure in rbd_dev_header_info() at line 4722:

     rbd_assert(rbd_image_format_valid(rbd_dev->image_format));

    Call Trace:
     [<ffffffff81cccee0>] ? rbd_parent_request_create+0x150/0x150
     [<ffffffff81cd4e59>] rbd_dev_refresh+0x59/0x390
     [<ffffffff81cd5229>] rbd_watch_cb+0x69/0x290
     [<ffffffff81fde9bf>] do_event_work+0x10f/0x1c0
     [<ffffffff81107799>] process_one_work+0x689/0x1a80
     [<ffffffff811076f7>] ? process_one_work+0x5e7/0x1a80
     [<ffffffff81132065>] ? finish_task_switch+0x225/0x640
     [<ffffffff81107110>] ? pwq_dec_nr_in_flight+0x2b0/0x2b0
     [<ffffffff81108c69>] worker_thread+0xd9/0x1320
     [<ffffffff81108b90>] ? process_one_work+0x1a80/0x1a80
     [<ffffffff8111b02d>] kthread+0x21d/0x2e0
     [<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
     [<ffffffff82022802>] ret_from_fork+0x22/0x40
     [<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
    RIP  [<ffffffff81ccd8f9>] rbd_dev_header_info+0xa19/0x1e30

To fix this, a) check if RBD_DEV_FLAG_EXISTS is set before calling
revalidate_disk(), b) move ceph_osdc_flush_notifies() call into
rbd_dev_header_unwatch_sync() to cover rbd map error paths and c) turn
header read-in into a critical section.  The latter also happens to
take care of rbd map foo@bar vs rbd snap rm foo@bar race.

Fixes: http://tracker.ceph.com/issues/15490
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Josh Durgin <jdurgin@redhat.com>
8 years agox86/apic: Handle zero vector gracefully in clear_vector_irq()
Keith Busch [Wed, 27 Apr 2016 20:22:32 +0000 (14:22 -0600)] 
x86/apic: Handle zero vector gracefully in clear_vector_irq()

If x86_vector_alloc_irq() fails x86_vector_free_irqs() is invoked to cleanup
the already allocated vectors. This subsequently calls clear_vector_irq().

The failed irq has no vector assigned, which triggers the BUG_ON(!vector) in
clear_vector_irq().

We cannot suppress the call to x86_vector_free_irqs() for the failed
interrupt, because the other data related to this irq must be cleaned up as
well. So calling clear_vector_irq() with vector == 0 is legitimate.

Remove the BUG_ON and return if vector is zero,

[ tglx: Massaged changelog ]

Fixes: b5dc8e6c21e7 "x86/irq: Use hierarchical irqdomain to manage CPU interrupt vectors"
Signed-off-by: Keith Busch <keith.busch@intel.com>
Cc: stable@vger.kernel.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
8 years agothermal: use %d to print S32 parameters
Leo Yan [Tue, 29 Mar 2016 11:24:15 +0000 (19:24 +0800)] 
thermal: use %d to print S32 parameters

Power allocator's parameters are S32 type, so use %d to print them.

Acked-by: Javi Merino <javi.merino@arm.com>
Signed-off-by: Leo Yan <leo.yan@linaro.org>
Signed-off-by: Eduardo Valentin <edubezval@gmail.com>
8 years agothermal: hisilicon: increase temperature resolution
Leo Yan [Tue, 29 Mar 2016 11:23:32 +0000 (19:23 +0800)] 
thermal: hisilicon: increase temperature resolution

When calculate temperature, old code firstly do division and then
convert to "millicelsius" unit. This will lose resolution and only can
read back temperature with "Celsius" unit.

So firstly scale step value to "millicelsius" and then do division, so
finally we can increase resolution for temperature value. Also refine
the calculation from temperature value to step value.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Signed-off-by: Eduardo Valentin <edubezval@gmail.com>
8 years agosparc64: Fix bootup regressions on some Kconfig combinations.
David S. Miller [Wed, 27 Apr 2016 21:27:37 +0000 (17:27 -0400)] 
sparc64: Fix bootup regressions on some Kconfig combinations.

The system call tracing bug fix mentioned in the Fixes tag
below increased the amount of assembler code in the sequence
of assembler files included by head_64.S

This caused to total set of code to exceed 0x4000 bytes in
size, which overflows the expression in head_64.S that works
to place swapper_tsb at address 0x408000.

When this is violated, the TSB is not properly aligned, and
also the trap table is not aligned properly either.  All of
this together results in failed boots.

So, do two things:

1) Simplify some code by using ba,a instead of ba/nop to get
   those bytes back.

2) Add a linker script assertion to make sure that if this
   happens again the build will fail.

Fixes: 1a40b95374f6 ("sparc: Fix system call tracing register handling.")
Reported-by: Meelis Roos <mroos@linux.ee>
Reported-by: Joerg Abraham <joerg.abraham@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'bnxt_en-fixes'
David S. Miller [Wed, 27 Apr 2016 20:18:46 +0000 (16:18 -0400)] 
Merge branch 'bnxt_en-fixes'

Michael Chan says:

====================
bnxt_en: Bug fixes for net.

Only use MSIX on VF, and fix rx page buffers on architectures with
PAGE_SIZE >= 64K.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobnxt_en: Divide a page into 32K buffers for the aggregation ring if necessary.
Michael Chan [Mon, 25 Apr 2016 06:30:51 +0000 (02:30 -0400)] 
bnxt_en: Divide a page into 32K buffers for the aggregation ring if necessary.

If PAGE_SIZE is bigger than BNXT_RX_PAGE_SIZE, that means the native CPU
page is bigger than the maximum length of the RX BD.  Divide the page
into multiple 32K buffers for the aggregation ring.

Add an offset field in the bnxt_sw_rx_agg_bd struct to keep track of the
page offset of each buffer.  Since each page can be referenced by multiple
buffer entries, call get_page() as needed to get the proper reference
count.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobnxt_en: Limit RX BD pages to be no bigger than 32K.
Michael Chan [Mon, 25 Apr 2016 06:30:50 +0000 (02:30 -0400)] 
bnxt_en: Limit RX BD pages to be no bigger than 32K.

The RX BD length field of this device is 16-bit, so the largest buffer
size is 65535.  For LRO and GRO, we allocate native CPU pages for the
aggregation ring buffers.  It won't work if the native CPU page size is
64K or bigger.

We fix this by defining BNXT_RX_PAGE_SIZE to be native CPU page size
up to 32K.  Replace PAGE_SIZE with BNXT_RX_PAGE_SIZE in all appropriate
places related to the rx aggregation ring logic.

The next patch will add additional logic to divide the page into 32K
chunks for aggrgation ring buffers if PAGE_SIZE is bigger than
BNXT_RX_PAGE_SIZE.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobnxt_en: Don't fallback to INTA on VF.
Michael Chan [Mon, 25 Apr 2016 06:30:49 +0000 (02:30 -0400)] 
bnxt_en: Don't fallback to INTA on VF.

Only MSI-X can be used on a VF.  The driver should fail initialization
if it cannot successfully enable MSI-X.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'for-4.6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq
Linus Torvalds [Wed, 27 Apr 2016 19:03:59 +0000 (12:03 -0700)] 
Merge branch 'for-4.6-fixes' of git://git./linux/kernel/git/tj/wq

Pull workqueue fix from Tejun Heo:
 "So, it turns out we had a silly bug in the most fundamental part of
  workqueue for a very long time.  AFAICS, this dates back to pre-git
  era and has quite likely been there from the time workqueue was first
  introduced.

  A work item uses its PENDING bit to synchronize multiple queuers.
  Anyone who wins the PENDING bit owns the pending state of the work
  item.  Whether a queuer wins or loses the race, one thing should be
  guaranteed - there will soon be at least one execution of the work
  item - where "after" means that the execution instance would be able
  to see all the changes that the queuer has made prior to the queueing
  attempt.

  Unfortunately, we were missing a smp_mb() after clearing PENDING for
  execution, so nothing guaranteed visibility of the changes that a
  queueing loser has made, which manifested as a reproducible blk-mq
  stall.

  Lots of kudos to Roman for debugging the problem.  The patch for
  -stable is the minimal one.  For v3.7, Peter is working on a patch to
  make the code path slightly more efficient and less fragile"

* 'for-4.6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq:
  workqueue: fix ghost PENDING flag while doing MQ IO

8 years agoMerge branch 'for-4.6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj...
Linus Torvalds [Wed, 27 Apr 2016 18:41:14 +0000 (11:41 -0700)] 
Merge branch 'for-4.6-fixes' of git://git./linux/kernel/git/tj/cgroup

Pull cgroup fixes from Tejun Heo:
 "Two patches to fix a deadlock which can be easily triggered if memcg
  charge moving is used.

  This bug was introduced while converting threadgroup locking to a
  global percpu_rwsem and is caused by cgroup controller task migration
  path depending on the ability to create new kthreads.  cpuset had a
  similar issue which was fixed by performing heavy-lifting operations
  asynchronous to task migration.  The two patches fix the same issue in
  memcg in a similar way.  The first patch makes the mechanism generic
  and the second relocates memcg charge moving outside the migration
  path.

  Given that we don't want to perform heavy operations while
  writelocking threadgroup lock anyway, moving them out of the way is a
  desirable solution.  One thing to note is that the problem was
  difficult to debug because lockdep couldn't figure out the deadlock
  condition.  Looking into how to improve that"

* 'for-4.6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup:
  memcg: relocate charge moving from ->attach to ->post_attach
  cgroup, cpuset: replace cpuset_post_attach_flush() with cgroup_subsys->post_attach callback

8 years agoMerge branch 'i2c/for-current' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa...
Linus Torvalds [Wed, 27 Apr 2016 18:34:45 +0000 (11:34 -0700)] 
Merge branch 'i2c/for-current' of git://git./linux/kernel/git/wsa/linux

Pull i2c fixes from Wolfram Sang:
 "I2C has one buildfix, one ABBA deadlock fix, and three simple 'add ID'
  patches"

* 'i2c/for-current' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
  i2c: exynos5: Fix possible ABBA deadlock by keeping I2C clock prepared
  i2c: cpm: Fix build break due to incompatible pointer types
  i2c: ismt: Add Intel DNV PCI ID
  i2c: xlp9xx: add support for Broadcom Vulcan
  i2c: rk3x: add support for rk3228

8 years agoMerge tag 'arc-4.6-rc6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupta/arc
Linus Torvalds [Wed, 27 Apr 2016 16:46:21 +0000 (09:46 -0700)] 
Merge tag 'arc-4.6-rc6-fixes' of git://git./linux/kernel/git/vgupta/arc

Pull ARC fixes from Vineet Gupta:

 - lockdep now works for ARCv2 builds

 - enable DT reserved-memory binding (for forthcoming HDMI driver)

* tag 'arc-4.6-rc6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupta/arc:
  ARC: add support for reserved memory defined by device tree
  ARC: support generic per-device coherent dma mem
  Documentation: dt: arc: fix spelling mistakes
  ARCv2: Enable LOCKDEP

8 years agoMerge tag 'nios2-v4.6-fix' of git://git.kernel.org/pub/scm/linux/kernel/git/lftan...
Linus Torvalds [Wed, 27 Apr 2016 16:33:24 +0000 (09:33 -0700)] 
Merge tag 'nios2-v4.6-fix' of git://git./linux/kernel/git/lftan/nios2

Pull arch/nios2 fix from Ley Foon Tan:
 "memset: use the right constraint modifier for the %4 output operand"

* tag 'nios2-v4.6-fix' of git://git.kernel.org/pub/scm/linux/kernel/git/lftan/nios2:
  nios2: memset: use the right constraint modifier for the %4 output operand

8 years agodrm/amdgpu: disable vm interrupts with vm_fault_stop=2
Flora Cui [Mon, 25 Apr 2016 08:06:17 +0000 (16:06 +0800)] 
drm/amdgpu: disable vm interrupts with vm_fault_stop=2

V2: disable all vm interrupts in late_init()

Signed-off-by: Flora Cui <Flora.Cui@amd.com>
Reviewed-by: Ken Wang <Qingqing.Wang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
8 years agodrm/amdgpu: print a message if ATPX dGPU power control is missing
Alex Deucher [Mon, 25 Apr 2016 17:14:47 +0000 (13:14 -0400)] 
drm/amdgpu: print a message if ATPX dGPU power control is missing

It will help identify problematic boards.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
8 years agoRevert "drm/amdgpu: disable runtime pm on PX laptops without dGPU power control"
Alex Deucher [Mon, 25 Apr 2016 17:12:18 +0000 (13:12 -0400)] 
Revert "drm/amdgpu: disable runtime pm on PX laptops without dGPU power control"

This reverts commit bedf2a65c1aa8fb29ba8527fd00c0f68ec1f55f1.

See the radeon revert for an extended description.

Cc: stable@vger.kernel.org
8 years agodrm/radeon: fix vertical bars appear on monitor (v2)
Vitaly Prosyak [Thu, 14 Apr 2016 17:34:03 +0000 (13:34 -0400)] 
drm/radeon: fix vertical bars appear on monitor (v2)

When crtc/timing is disabled on boot the dig block
should be stopped in order ignore timing from crtc,
reset the steering fifo otherwise we get display
corruption or hung in dp sst mode.

v2: agd: fix coding style

Signed-off-by: Vitaly Prosyak <vitaly.prosyak@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
8 years agodrm/ttm: fix kref count mess in ttm_bo_move_to_lru_tail
Flora Cui [Wed, 20 Apr 2016 02:23:47 +0000 (10:23 +0800)] 
drm/ttm: fix kref count mess in ttm_bo_move_to_lru_tail

Fixes the following scenario:

1. Page table bo allocated in vram and linked to man->lru.
   tbo->list_kref.refcount=2
2. Page table bo is swapped out and removed from man->lru.
   tbo->list_kref.refcount=1
3. Command submission from userspace.  Page table bo is moved
   to vram.  ttm_bo_move_to_lru_tail() link it to man->lru and
   don't increase the kref count.

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Flora Cui <Flora.Cui@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
8 years agoMerge tag 'platform-drivers-x86-v4.6-3' of git://git.infradead.org/users/dvhart/linux...
Linus Torvalds [Wed, 27 Apr 2016 15:57:11 +0000 (08:57 -0700)] 
Merge tag 'platform-drivers-x86-v4.6-3' of git://git.infradead.org/users/dvhart/linux-platform-drivers-x86

Pull x86 platform driver fix from Darren Hart:
 "Fix regression caused by hotkey enabling value in toshiba_acpi"

* tag 'platform-drivers-x86-v4.6-3' of git://git.infradead.org/users/dvhart/linux-platform-drivers-x86:
  toshiba_acpi: Fix regression caused by hotkey enabling value

8 years agoMerge tag 'asoc-fix-v4.6-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/brooni...
Takashi Iwai [Wed, 27 Apr 2016 15:30:49 +0000 (17:30 +0200)] 
Merge tag 'asoc-fix-v4.6-rc5' of git://git./linux/kernel/git/broonie/sound into for-linus

ASoC: Fixes for v4.6

This is a fairly large collection of fixes but almost all driver
specific ones, especially to the new Intel drivers which have had a lot
of recent development.  The one core fix is a change to the debugfs code
to avoid crashes in some relatively unusual configurations.

8 years agoARC: add support for reserved memory defined by device tree
Alexey Brodkin [Tue, 26 Apr 2016 16:29:34 +0000 (19:29 +0300)] 
ARC: add support for reserved memory defined by device tree

Enable reserved memory initialization from device tree.

Signed-off-by: Alexey Brodkin <abrodkin@synopsys.com>
Cc: Grant Likely <grant.likely@linaro.org>
Cc: Marek Szyprowski <m.szyprowski@samsung.com>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
8 years agoARC: support generic per-device coherent dma mem
Alexey Brodkin [Tue, 26 Apr 2016 16:29:33 +0000 (19:29 +0300)] 
ARC: support generic per-device coherent dma mem

Signed-off-by: Alexey Brodkin <abrodkin@synopsys.com>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
8 years agonios2: memset: use the right constraint modifier for the %4 output operand
Romain Perier [Tue, 19 Apr 2016 10:17:32 +0000 (18:17 +0800)] 
nios2: memset: use the right constraint modifier for the %4 output operand

Depending on the size of the area to be memset'ed, the nios2 memset implementation
either uses a naive loop (for buffers smaller or equal than 8 bytes) or a more optimized
implementation (for buffers larger than 8 bytes). This implementation does 4-byte stores
rather than 1-byte stores to speed up memset.

However, we discovered that on our nios2 platform, memset() was not properly setting the
buffer to the expected value. A memset of 0xff would not set the entire buffer to 0xff, but to:

0xff 0x00 0xff 0x00 0xff 0x00 0xff 0x00 ...

Which is obviously incorrect. Our investigation has revealed that the problem lies in the
incorrect constraints used in the inline assembly.

The following piece of assembly, from the nios2 memset implementation, is supposed to
create a 4-byte value that repeats 4 times the 1-byte pattern passed as memset argument:

/* fill8 %3, %5 (c & 0xff) */
"       slli    %4, %5, 8\n"
"       or      %4, %4, %5\n"
"       slli    %3, %4, 16\n"
"       or      %3, %3, %4\n"

However, depending on the compiler and optimization level, this code might be compiled as:

34: 280a923a  slli r5,r5,8
38: 294ab03a  or r5,r5,r5
3c: 2808943a  slli r4,r5,16
40: 2148b03a  or r4,r4,r5

This is wrong because r5 gets used both for %5 and %4, which leads to the final pattern
stored in r4 to be 0xff00ff00 rather than the expected 0xffffffff.

%4 is defined with the "=r" constraint, i.e as an output operand. However, as explained in
http://www.ethernut.de/en/documents/arm-inline-asm.html, this does not prevent gcc from
using the same register for an output operand (%4) and input operand (%5). By using the
constraint modifier '&', we indicate that the register should be used for output only. With this
change, we get the following assembly output:

34: 2810923a  slli r8,r5,8
38: 4150b03a  or r8,r8,r5
3c: 400e943a  slli r7,r8,16
40: 3a0eb03a  or r7,r7,r8

Which correctly produces the 0xffffffff pattern when 0xff is passed as the memset() pattern.

It is worth mentioning the observed consequence of this bug: we were hitting the kernel
BUG() in mm/bootmem.c:__free() that verifies when marking a page as free that it was
previously marked as occupied (i.e that the bit was set to 1). The entire bootmem bitmap is
set to 0xff bit via a memset() during the bootmem initialization. The bootmem_free() call right
after the initialization was finding some bits to be set to 0, which didn't make sense since the
bitmap has just been memset'ed to 0xff. Except that due to the bug explained above, the
bitmap was in fact initialized to 0xff00ff00.

Thanks to Marek Vasut for his help and feedback.

Signed-off-by: Romain Perier <romain.perier@free-electrons.com>
Acked-by: Marek Vasut <marex@denx.de>
Acked-by: Ley Foon Tan <lftan@altera.com>
8 years agomac80211: fix statistics leak if dev_alloc_name() fails
Johannes Berg [Tue, 26 Apr 2016 11:47:08 +0000 (13:47 +0200)] 
mac80211: fix statistics leak if dev_alloc_name() fails

In the case that dev_alloc_name() fails, e.g. because the name was
given by the user and already exists, we need to clean up properly
and free the per-CPU statistics. Fix that.

Cc: stable@vger.kernel.org
Fixes: 5a490510ba5f ("mac80211: use per-CPU TX/RX statistics")
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
8 years agos390/sclp_ctl: fix potential information leak with /dev/sclp
Martin Schwidefsky [Mon, 25 Apr 2016 15:54:28 +0000 (17:54 +0200)] 
s390/sclp_ctl: fix potential information leak with /dev/sclp

The sclp_ctl_ioctl_sccb function uses two copy_from_user calls to
retrieve the sclp request from user space. The first copy_from_user
fetches the length of the request which is stored in the first two
bytes of the request. The second copy_from_user gets the complete
sclp request, but this copies the length field a second time.
A malicious user may have changed the length in the meantime.

Reported-by: Pengfei Wang <wpengfeinudt@gmail.com>
Reviewed-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
8 years agopowerpc: wire up preadv2 and pwritev2 syscalls
Rui Salvaterra [Tue, 19 Apr 2016 12:23:36 +0000 (13:23 +0100)] 
powerpc: wire up preadv2 and pwritev2 syscalls

Wire up preadv2/pwritev2 in the same way as preadv/pwritev. Fixes two
build warnings on ppc64.

mpe: Lightly tested with fio (slightly hacked to add the syscall
wrappers):

  fio-4217  [009] ....  1304.635300: sys_preadv2(fd: 3, vec:
  10025821de0, vlen: 1, pos_l: 6253000, pos_h: 0, flags: 1)
  fio-4217  [009] ....  1304.635474: sys_preadv2 -> 0x1000

Signed-off-by: Rui Salvaterra <rsalvaterra@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agocxl: Poll for outstanding IRQs when detaching a context
Michael Neuling [Fri, 22 Apr 2016 04:57:49 +0000 (14:57 +1000)] 
cxl: Poll for outstanding IRQs when detaching a context

When detaching contexts, we may still have interrupts in the system
which are yet to be delivered to any CPU and be acked in the PSL.
This can result in a subsequent unrelated process getting an spurious
IRQ or an interrupt for a non-existent context.

This polls the PSL to ensure that the PSL is clear of IRQs for the
detached context, before removing the context from the idr.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Tested-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Acked-by: Ian Munsie <imunsie@au1.ibm.com>
Tested-by: Vaibhav Jain <vaibhav@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agocxl: Keep IRQ mappings on context teardown
Michael Neuling [Fri, 22 Apr 2016 04:57:48 +0000 (14:57 +1000)] 
cxl: Keep IRQ mappings on context teardown

Keep IRQ mappings on context teardown.  This won't leak IRQs as if we
allocate the mapping again, the generic code will give the same
mapping used last time.

Doing this works around a race in the generic code. Masking the
interrupt introduces a race which can crash the kernel or result in
IRQ that is never EOIed. The lost of EOI results in all subsequent
mappings to the same HW IRQ never receiving an interrupt.

We've seen this race with cxl test cases which are doing heavy context
startup and teardown at the same time as heavy interrupt load.

A fix to the generic code is being investigated also.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Cc: stable@vger.kernel.org # 3.8
Tested-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Acked-by: Ian Munsie <imunsie@au1.ibm.com>
Tested-by: Vaibhav Jain <vaibhav@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
8 years agodrm/virtio: send vblank event after crtc updates
Gustavo Padovan [Thu, 14 Apr 2016 17:58:54 +0000 (10:58 -0700)] 
drm/virtio: send vblank event after crtc updates

virtio_gpu was failing to send vblank events when using the atomic IOCTL
with the DRM_MODE_PAGE_FLIP_EVENT flag set. This patch fixes each and
enables atomic pageflips updates.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Cc: stable@vger.kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
8 years agodrm/dp/mst: Restore primary hub guid on resume
Lyude [Wed, 13 Apr 2016 20:50:18 +0000 (16:50 -0400)] 
drm/dp/mst: Restore primary hub guid on resume

Some hubs are forgetful, and end up forgetting whatever GUID we set
previously after we do a suspend/resume cycle. This can lead to
hotplugging breaking (along with probably other things) since the hub
will start sending connection notifications with the wrong GUID. As
such, we need to check on resume whether or not the GUID the hub is
giving us is valid.

Signed-off-by: Lyude <cpaul@redhat.com>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1460580618-7421-1-git-send-email-cpaul@redhat.com
Cc: stable@vger.kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
8 years agodrm/dp/mst: Get validated port ref in drm_dp_update_payload_part1()
cpaul@redhat.com [Fri, 22 Apr 2016 20:08:46 +0000 (16:08 -0400)] 
drm/dp/mst: Get validated port ref in drm_dp_update_payload_part1()

We can thank KASAN for finding this, otherwise I probably would have spent
hours on it. This fixes a somewhat harder to trigger kernel panic, occuring
while enabling MST where the port we were currently updating the payload on
would have all of it's refs dropped before we finished what we were doing:

==================================================================
BUG: KASAN: use-after-free in drm_dp_update_payload_part1+0xb3f/0xdb0 [drm_kms_helper] at addr ffff8800d29de018
Read of size 4 by task Xorg/973
=============================================================================
BUG kmalloc-2048 (Tainted: G    B   W      ): kasan: bad access detected
-----------------------------------------------------------------------------

INFO: Allocated in drm_dp_add_port+0x1aa/0x1ed0 [drm_kms_helper] age=16477 cpu=0 pid=2175
___slab_alloc+0x472/0x490
__slab_alloc+0x20/0x40
kmem_cache_alloc_trace+0x151/0x190
drm_dp_add_port+0x1aa/0x1ed0 [drm_kms_helper]
drm_dp_send_link_address+0x526/0x960 [drm_kms_helper]
drm_dp_check_and_send_link_address+0x1ac/0x210 [drm_kms_helper]
drm_dp_mst_link_probe_work+0x77/0xd0 [drm_kms_helper]
process_one_work+0x562/0x1350
worker_thread+0xd9/0x1390
kthread+0x1c5/0x260
ret_from_fork+0x22/0x40
INFO: Freed in drm_dp_free_mst_port+0x50/0x60 [drm_kms_helper] age=7521 cpu=0 pid=2175
__slab_free+0x17f/0x2d0
kfree+0x169/0x180
drm_dp_free_mst_port+0x50/0x60 [drm_kms_helper]
drm_dp_destroy_connector_work+0x2b8/0x490 [drm_kms_helper]
process_one_work+0x562/0x1350
worker_thread+0xd9/0x1390
kthread+0x1c5/0x260
ret_from_fork+0x22/0x40

which on this T460s, would eventually lead to kernel panics in somewhat
random places later in intel_mst_enable_dp() if we got lucky enough.

Signed-off-by: Lyude <cpaul@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
8 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Tue, 26 Apr 2016 23:25:51 +0000 (16:25 -0700)] 
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Handle v4/v6 mixed sockets properly in soreuseport, from Craig
    Gallak.

 2) Bug fixes for the new macsec facility (missing kmalloc NULL checks,
    missing locking around netdev list traversal, etc.) from Sabrina
    Dubroca.

 3) Fix handling of host routes on ifdown in ipv6, from David Ahern.

 4) Fix double-fdput in bpf verifier.  From Jann Horn.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (31 commits)
  bpf: fix double-fdput in replace_map_fd_with_map_ptr()
  net: ipv6: Delete host routes on an ifdown
  Revert "ipv6: Revert optional address flusing on ifdown."
  net/mlx4_en: fix spurious timestamping callbacks
  net: dummy: remove note about being Y by default
  cxgbi: fix uninitialized flowi6
  ipv6: Revert optional address flusing on ifdown.
  ipv4/fib: don't warn when primary address is missing if in_dev is dead
  net/mlx5: Add pci shutdown callback
  net/mlx5_core: Remove static from local variable
  net/mlx5e: Use vport MTU rather than physical port MTU
  net/mlx5e: Fix minimum MTU
  net/mlx5e: Device's mtu field is u16 and not int
  net/mlx5_core: Add ConnectX-5 to list of supported devices
  net/mlx5e: Fix MLX5E_100BASE_T define
  net/mlx5_core: Fix soft lockup in steering error flow
  qlcnic: Update version to 5.3.64
  net: stmmac: socfpga: Remove re-registration of reset controller
  macsec: fix netlink attribute validation
  macsec: add missing macsec prefix in uapi
  ...

8 years agoMerge branch 'drm-etnaviv-fixes' of git://git.pengutronix.de:/git/lst/linux into...
Dave Airlie [Tue, 26 Apr 2016 23:19:06 +0000 (09:19 +1000)] 
Merge branch 'drm-etnaviv-fixes' of git://git.pengutronix.de:/git/lst/linux into drm-fixes

just a single fix to not move the GPU linear window on cores where it
might lead to inconsistent views of the memory by different engines in
the core, thus breaking relocs and possibly causing other fun.

* 'drm-etnaviv-fixes' of git://git.pengutronix.de:/git/lst/linux:
  drm/etnaviv: don't move linear memory window on 3D cores without MC2.0

8 years agoMerge tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm...
Linus Torvalds [Tue, 26 Apr 2016 23:17:01 +0000 (16:17 -0700)] 
Merge tag 'fixes-for-linus' of git://git./linux/kernel/git/arm/arm-soc

Pull ARM SoC fixes from Arnd Bergmann:
 "Here are the latest bug fixes for ARM SoCs, mostly addressing recent
  regressions.  Changes are across several platforms, so I'm listing
  every change separately here.

  Regressions since 4.5:

   - A correction of the psci firmware DT binding, to prevent users from
     relying on unintended semantics

   - Actually getting the newly merged clock driver for some OMAP
     platforms to work

   - A revert of patches for the Qualcomm BAM, these need to be reworked
     for 4.7 to avoid breaking boards other than the one they were
     intended for

   - A correction for the I2C device nodes on the Socionext Uniphier
     platform

   - i.MX SDHCI was broken for non-DT platforms due to a change with the
     setting of the DMA mask

   - A revert of a patch that accidentally added a nonexisting clock on
     the Rensas "Porter" board

   - A couple of OMAP fixes that are all related to suspend after the
     power domain changes for dra7

   - On Mediatek, revert part of the power domain initialization changes
     that broke mt8173-evb

  Fixes for older bugs:

   - Workaround for an "external abort" in the omap34xx suspend/resume
     code.

   - The USB1/eSATA should not be listed as an excon device on
     am57xx-beagle-x15 (broken since v4.0)

   - A v4.5 regression in the TI AM33xx and AM43XX DT specifying
     incorrect DMA request lines for the GPMC

   - The jiffies calibration on Renesas platforms was incorrect for some
     modern CPU cores.

   - A hardware errata woraround for clockdomains on TI DRA7"

* tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc:
  drivers: firmware: psci: unify enable-method binding on ARM {64,32}-bit systems
  arm64: dts: uniphier: fix I2C nodes of PH1-LD20
  ARM: shmobile: timer: Fix preset_lpj leading to too short delays
  Revert "ARM: dts: porter: Enable SCIF_CLK frequency and pins"
  ARM: dts: r8a7791: Don't disable referenced optional clocks
  Revert "ARM: OMAP: Catch callers of revision information prior to it being populated"
  ARM: OMAP3: Fix external abort on 36xx waking from off mode idle
  ARM: dts: am57xx-beagle-x15: remove extcon_usb1
  ARM: dts: am437x: Fix GPMC dma properties
  ARM: dts: am33xx: Fix GPMC dma properties
  Revert "soc: mediatek: SCPSYS: Fix double enabling of regulators"
  ARM: mach-imx: sdhci-esdhc-imx: initialize DMA mask
  ARM: DRA7: clockdomain: Implement timer workaround for errata i874
  ARM: OMAP: Catch callers of revision information prior to it being populated
  ARM: dts: dra7: Correct clock tree for sys_32k_ck
  ARM: OMAP: DRA7: Provide proper class to omap2_set_globals_tap
  ARM: OMAP: DRA7: wakeupgen: Skip SAR save for wakeupgen
  Revert "dts: msm8974: Add dma channels for blsp2_i2c1 node"
  Revert "dts: msm8974: Add blsp2_bam dma node"
  ARM: dts: Add clocks for dm814x ADPLL

8 years agodevpts: more pty driver interface cleanups
Linus Torvalds [Tue, 26 Apr 2016 03:04:08 +0000 (20:04 -0700)] 
devpts: more pty driver interface cleanups

This is more prep-work for the upcoming pty changes.  Still just code
cleanup with no actual semantic changes.

This removes a bunch pointless complexity by just having the slave pty
side remember the dentry associated with the devpts slave rather than
the inode.  That allows us to remove all the "look up the dentry" code
for when we want to remove it again.

Together with moving the tty pointer from "inode->i_private" to
"dentry->d_fsdata" and getting rid of pointless inode locking, this
removes about 30 lines of code.  Not only is the end result smaller,
it's simpler and easier to understand.

The old code, for example, depended on the d_find_alias() to not just
find the dentry, but also to check that it is still hashed, which in
turn validated the tty pointer in the inode.

That is a _very_ roundabout way to say "invalidate the cached tty
pointer when the dentry is removed".

The new code just does

dentry->d_fsdata = NULL;

in devpts_pty_kill() instead, invalidating the tty pointer rather more
directly and obviously.  Don't do something complex and subtle when the
obvious straightforward approach will do.

The rest of the patch (ie apart from code deletion and the above tty
pointer clearing) is just switching the calling convention to pass the
dentry or file pointer around instead of the inode.

Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Peter Anvin <hpa@zytor.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Peter Hurley <peter@hurleysoftware.com>
Cc: Serge Hallyn <serge.hallyn@ubuntu.com>
Cc: Willy Tarreau <w@1wt.eu>
Cc: Aurelien Jarno <aurelien@aurel32.net>
Cc: Alan Cox <gnomes@lxorguk.ukuu.org.uk>
Cc: Jann Horn <jann@thejh.net>
Cc: Greg KH <greg@kroah.com>
Cc: Jiri Slaby <jslaby@suse.com>
Cc: Florian Weimer <fw@deneb.enyo.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agobpf: fix double-fdput in replace_map_fd_with_map_ptr()
Jann Horn [Tue, 26 Apr 2016 20:26:26 +0000 (22:26 +0200)] 
bpf: fix double-fdput in replace_map_fd_with_map_ptr()

When bpf(BPF_PROG_LOAD, ...) was invoked with a BPF program whose bytecode
references a non-map file descriptor as a map file descriptor, the error
handling code called fdput() twice instead of once (in __bpf_map_get() and
in replace_map_fd_with_map_ptr()). If the file descriptor table of the
current task is shared, this causes f_count to be decremented too much,
allowing the struct file to be freed while it is still in use
(use-after-free). This can be exploited to gain root privileges by an
unprivileged user.

This bug was introduced in
commit 0246e64d9a5f ("bpf: handle pseudo BPF_LD_IMM64 insn"), but is only
exploitable since
commit 1be7f75d1668 ("bpf: enable non-root eBPF programs") because
previously, CAP_SYS_ADMIN was required to reach the vulnerable code.

(posted publicly according to request by maintainer)

Signed-off-by: Jann Horn <jannh@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge remote-tracking branches 'asoc/fix/rt5640' and 'asoc/fix/wm8962' into asoc...
Mark Brown [Tue, 26 Apr 2016 18:25:18 +0000 (19:25 +0100)] 
Merge remote-tracking branches 'asoc/fix/rt5640' and 'asoc/fix/wm8962' into asoc-linus

8 years agoMerge remote-tracking branches 'asoc/fix/arizona', 'asoc/fix/cs35l32', 'asoc/fix...
Mark Brown [Tue, 26 Apr 2016 18:25:15 +0000 (19:25 +0100)] 
Merge remote-tracking branches 'asoc/fix/arizona', 'asoc/fix/cs35l32', 'asoc/fix/hdac', 'asoc/fix/nau8825' and 'asoc/fix/rt5616' into asoc-linus

8 years agoMerge remote-tracking branch 'asoc/fix/intel' into asoc-linus
Mark Brown [Tue, 26 Apr 2016 18:25:14 +0000 (19:25 +0100)] 
Merge remote-tracking branch 'asoc/fix/intel' into asoc-linus

8 years agoMerge remote-tracking branch 'asoc/fix/dapm' into asoc-linus
Mark Brown [Tue, 26 Apr 2016 18:25:13 +0000 (19:25 +0100)] 
Merge remote-tracking branch 'asoc/fix/dapm' into asoc-linus

8 years agoRevert "x86/mm/32: Set NX in __supported_pte_mask before enabling paging"
Andy Lutomirski [Tue, 26 Apr 2016 15:52:44 +0000 (08:52 -0700)] 
Revert "x86/mm/32: Set NX in __supported_pte_mask before enabling paging"

This reverts commit 320d25b6a05f8b73c23fc21025d2906ecdd2d4fc.

This change was problematic for a couple of reasons:

1. It missed a some entry points (Xen things and 64-bit native).

2. The entry it changed can be executed more than once.  This isn't
   really a problem, but it conflated per-cpu state setup and global
   state setup.

3. It broke 64-bit non-NX.  64-bit non-NX worked the other way around from
   32-bit -- __supported_pte_mask had NX set initially and was *cleared*
   in x86_configure_nx.  With the patch applied, it never got cleared.

Reported-and-tested-by: Meelis Roos <mroos@linux.ee>
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/59bd15f7f4b56b633a611b7f70876c6d2ad01a98.1461685884.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
8 years agoRDMA/iw_cxgb4: Fix bar2 virt addr calculation for T4 chips
Hariprasad S [Tue, 5 Apr 2016 04:53:48 +0000 (10:23 +0530)] 
RDMA/iw_cxgb4: Fix bar2 virt addr calculation for T4 chips

For T4, kernel mode qps don't use the user doorbell. User mode qps during
flow control db ringing are forced into kernel, where user doorbell is
treated as kernel doorbell and proper bar2 offset in bar2 virtual space is
calculated, which incase of T4 is a bogus address, causing a kernel panic
due to illegal write during doorbell ringing.
In case of T4, kernel mode qp bar2 virtual address should be 0. Added T4
check during bar2 virtual address calculation to return 0. Fixed Bar2
range checks based on bar2 physical address.

The below oops will be fixed

  <1>BUG: unable to handle kernel paging request at 000000000002aa08
  <1>IP: [<ffffffffa011d800>] c4iw_uld_control+0x4e0/0x880 [iw_cxgb4]
  <4>PGD 1416a8067 PUD 15bf35067 PMD 0
  <4>Oops: 0002 [#1] SMP
  <4>last sysfs file:
  /sys/devices/pci0000:00/0000:00:03.0/0000:02:00.4/infiniband/cxgb4_0/node_guid
  <4>CPU 5
  <4>Modules linked in: rdma_ucm rdma_cm ib_cm ib_sa ib_mad ib_uverbs
  ip6table_filter ip6_tables ebtable_nat ebtables ipt_MASQUERADE
  iptable_nat nf_nat nf_conntrack_ipv4 nf_defrag_ipv4 xt_state nf_conntrack
  ipt_REJECT xt_CHECKSUM iptable_mangle iptable_filter ip_tables bridge autofs4
  target_core_iblock target_core_file target_core_pscsi target_core_mod
  configfs bnx2fc cnic uio fcoe libfcoe libfc scsi_transport_fc scsi_tgt 8021q
  garp stp llc cpufreq_ondemand acpi_cpufreq freq_table mperf vhost_net macvtap
  macvlan tun kvm uinput microcode iTCO_wdt iTCO_vendor_support sg joydev
  serio_raw i2c_i801 i2c_core lpc_ich mfd_core e1000e ptp pps_core ioatdma dca
  i7core_edac edac_core shpchp ext3 jbd mbcache sd_mod crc_t10dif pata_acpi
  ata_generic ata_piix iw_cxgb4 iw_cm ib_core ib_addr cxgb4 ipv6 dm_mirror
  dm_region_hash dm_log dm_mod [last unloaded: scsi_wait_scan]
  <4>
  Supermicro X8ST3/X8ST3
  <4>RIP: 0010:[<ffffffffa011d800>]  [<ffffffffa011d800>]
  c4iw_uld_control+0x4e0/0x880 [iw_cxgb4]
  <4>RSP: 0000:ffff880155a03db0  EFLAGS: 00010006
  <4>RAX: 000000000000001d RBX: ffff88013ae5fc00 RCX: ffff880155adb180
  <4>RDX: 000000000002aa00 RSI: 0000000000000001 RDI: ffff88013ae5fdf8
  <4>RBP: ffff880155a03e10 R08: 0000000000000000 R09: 0000000000000001
  <4>R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000000
  <4>R13: 000000000000001d R14: ffff880156414ab0 R15: ffffe8ffffc05b88
  <4>FS:  0000000000000000(0000) GS:ffff8800282a0000(0000) knlGS:0000000000000000
  <4>CS:  0010 DS: 0018 ES: 0018 CR0: 000000008005003b
  <4>CR2: 000000000002aa08 CR3: 000000015bd0e000 CR4: 00000000000007e0
  <4>DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
  <4>DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
  <4>Process cxgb4 (pid: 394, threadinfo ffff880155a00000, task ffff880156414ab0)
  <4>Stack:
  <4> ffff880156415068 ffff880155adb180 ffff880155a03df0 ffffffffa00a344b
  <4><d> 00000000000003e8 ffff880155920000 0000000000000004 ffff880155920000
  <4><d> ffff88015592d438 ffffffffa00a3860 ffff880155a03fd8 ffffe8ffffc05b88
  <4>Call Trace:
  <4> [<ffffffffa00a344b>] ? enable_txq_db+0x2b/0x80 [cxgb4]
  <4> [<ffffffffa00a3860>] ? process_db_full+0x0/0xa0 [cxgb4]
  <4> [<ffffffffa00a38a6>] process_db_full+0x46/0xa0 [cxgb4]
  <4> [<ffffffff8109fda0>] worker_thread+0x170/0x2a0
  <4> [<ffffffff810a6aa0>] ? autoremove_wake_function+0x0/0x40
  <4> [<ffffffff8109fc30>] ? worker_thread+0x0/0x2a0
  <4> [<ffffffff810a660e>] kthread+0x9e/0xc0
  <4> [<ffffffff8100c28a>] child_rip+0xa/0x20
  <4> [<ffffffff810a6570>] ? kthread+0x0/0xc0
  <4> [<ffffffff8100c280>] ? child_rip+0x0/0x20
  <4>Code: e9 ba 00 00 00 66 0f 1f 44 00 00 44 8b 05 29 07 02 00 45 85 c0 0f 85
  71 02 00 00 8b 83 70 01 00 00 45 0f b7 ed c1 e0 0f 44 09 e8 <89> 42 08 0f ae f8
  66 c7 83 82 01 00 00 00 00 44 0f b7 ab dc 01
  <1>RIP  [<ffffffffa011d800>] c4iw_uld_control+0x4e0/0x880 [iw_cxgb4]
  <4> RSP <ffff880155a03db0>
  <4>CR2: 000000000002aa08`

Based on original work by Bharat Potnuri <bharat@chelsio.com>

Fixes: 74217d4c6a4fb0d8 ("iw_cxgb4: support for bar2 qid densities exceeding the page size")
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Reviewed-by: Leon Romanovsky <leon@leon.nu>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoiw_cxgb4: handle draining an idle qp
Steve Wise [Tue, 12 Apr 2016 13:55:03 +0000 (06:55 -0700)] 
iw_cxgb4: handle draining an idle qp

In c4iw_drain_sq/rq(), if the particular queue is already empty
then don't block.

Fixes: ce4af14d94aa ('iw_cxgb4: add queue drain functions')
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoiw_cxgb3: initialize ibdev.iwcm->ifname for port mapping
Steve Wise [Tue, 12 Apr 2016 13:55:01 +0000 (06:55 -0700)] 
iw_cxgb3: initialize ibdev.iwcm->ifname for port mapping

The IWCM uses ibdev.iwcm->ifname for registration with the iwarp
port map daemon.  But iw_cxgb3 did not initialize this field which
causes intermittent registration failures based on the contents of the
uninitialized memory.

Fixes: c1340e8aa628 ("iw_cxgb3: support for iWARP port mapping")
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoiw_cxgb4: initialize ibdev.iwcm->ifname for port mapping
Steve Wise [Tue, 12 Apr 2016 13:54:54 +0000 (06:54 -0700)] 
iw_cxgb4: initialize ibdev.iwcm->ifname for port mapping

The IWCM uses ibdev.iwcm->ifname for registration with the iwarp
port map daemon.  But iw_cxgb4 did not initialize this field which
causes intermittent registration failures based on the contents of the
uninitialized memory.

Fixes: 170003c894d9 ("iw_cxgb4: remove port mapper related code")
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/core: Don't drain non-existent rq queue-pair
Sagi Grimberg [Tue, 26 Apr 2016 14:55:38 +0000 (17:55 +0300)] 
IB/core: Don't drain non-existent rq queue-pair

The drain_rq function expects a normal receive qp to drain.  A qp can
only have either a normal rq or an srq.  If there is an srq, there
is no rq to drain.  Until the API supports draining SRQs, simply
skip draining the rq when the qp has an srq attached.

Fixes: 765d67748bcf ("IB: new common API for draining queues")
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agonet: ipv6: Delete host routes on an ifdown
David Ahern [Fri, 22 Apr 2016 03:56:12 +0000 (20:56 -0700)] 
net: ipv6: Delete host routes on an ifdown

It was a simple idea -- save IPv6 configured addresses on a link down
so that IPv6 behaves similar to IPv4. As always the devil is in the
details and the IPv6 stack as too many behavioral differences from IPv4
making the simple idea more complicated than it needs to be.

The current implementation for keeping IPv6 addresses can panic or spit
out a warning in one of many paths:

1. IPv6 route gets an IPv4 route as its 'next' which causes a panic in
   rt6_fill_node while handling a route dump request.

2. rt->dst.obsolete is set to DST_OBSOLETE_DEAD hitting the WARN_ON in
   fib6_del

3. Panic in fib6_purge_rt because rt6i_ref count is not 1.

The root cause of all these is references related to the host route for
an address that is retained.

So, this patch deletes the host route every time the ifdown loop runs.
Since the host route is deleted and will be re-generated an up there is
no longer a need for the l3mdev fix up. On the 'admin up' side move
addrconf_permanent_addr into the NETDEV_UP event handling so that it
runs only once versus on UP and CHANGE events.

All of the current panics and warnings appear to be related to
addresses on the loopback device, but given the catastrophic nature when
a bug is triggered this patch takes the conservative approach and evicts
all host routes rather than trying to determine when it can be re-used
and when it can not. That can be a later optimizaton if desired.

Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoRevert "ipv6: Revert optional address flusing on ifdown."
David S. Miller [Tue, 26 Apr 2016 15:47:41 +0000 (11:47 -0400)] 
Revert "ipv6: Revert optional address flusing on ifdown."

This reverts commit 841645b5f2dfceac69b78fcd0c9050868d41ea61.

Ok, this puts the feature back.  I've decided to apply David A.'s
bug fix and run with that rather than make everyone wait another
whole release for this feature.

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoworkqueue: fix ghost PENDING flag while doing MQ IO
Roman Pen [Tue, 26 Apr 2016 11:15:35 +0000 (13:15 +0200)] 
workqueue: fix ghost PENDING flag while doing MQ IO

The bug in a workqueue leads to a stalled IO request in MQ ctx->rq_list
with the following backtrace:

[  601.347452] INFO: task kworker/u129:5:1636 blocked for more than 120 seconds.
[  601.347574]       Tainted: G           O    4.4.5-1-storage+ #6
[  601.347651] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[  601.348142] kworker/u129:5  D ffff880803077988     0  1636      2 0x00000000
[  601.348519] Workqueue: ibnbd_server_fileio_wq ibnbd_dev_file_submit_io_worker [ibnbd_server]
[  601.348999]  ffff880803077988 ffff88080466b900 ffff8808033f9c80 ffff880803078000
[  601.349662]  ffff880807c95000 7fffffffffffffff ffffffff815b0920 ffff880803077ad0
[  601.350333]  ffff8808030779a0 ffffffff815b01d5 0000000000000000 ffff880803077a38
[  601.350965] Call Trace:
[  601.351203]  [<ffffffff815b0920>] ? bit_wait+0x60/0x60
[  601.351444]  [<ffffffff815b01d5>] schedule+0x35/0x80
[  601.351709]  [<ffffffff815b2dd2>] schedule_timeout+0x192/0x230
[  601.351958]  [<ffffffff812d43f7>] ? blk_flush_plug_list+0xc7/0x220
[  601.352208]  [<ffffffff810bd737>] ? ktime_get+0x37/0xa0
[  601.352446]  [<ffffffff815b0920>] ? bit_wait+0x60/0x60
[  601.352688]  [<ffffffff815af784>] io_schedule_timeout+0xa4/0x110
[  601.352951]  [<ffffffff815b3a4e>] ? _raw_spin_unlock_irqrestore+0xe/0x10
[  601.353196]  [<ffffffff815b093b>] bit_wait_io+0x1b/0x70
[  601.353440]  [<ffffffff815b056d>] __wait_on_bit+0x5d/0x90
[  601.353689]  [<ffffffff81127bd0>] wait_on_page_bit+0xc0/0xd0
[  601.353958]  [<ffffffff81096db0>] ? autoremove_wake_function+0x40/0x40
[  601.354200]  [<ffffffff81127cc4>] __filemap_fdatawait_range+0xe4/0x140
[  601.354441]  [<ffffffff81127d34>] filemap_fdatawait_range+0x14/0x30
[  601.354688]  [<ffffffff81129a9f>] filemap_write_and_wait_range+0x3f/0x70
[  601.354932]  [<ffffffff811ced3b>] blkdev_fsync+0x1b/0x50
[  601.355193]  [<ffffffff811c82d9>] vfs_fsync_range+0x49/0xa0
[  601.355432]  [<ffffffff811cf45a>] blkdev_write_iter+0xca/0x100
[  601.355679]  [<ffffffff81197b1a>] __vfs_write+0xaa/0xe0
[  601.355925]  [<ffffffff81198379>] vfs_write+0xa9/0x1a0
[  601.356164]  [<ffffffff811c59d8>] kernel_write+0x38/0x50

The underlying device is a null_blk, with default parameters:

  queue_mode    = MQ
  submit_queues = 1

Verification that nullb0 has something inflight:

root@pserver8:~# cat /sys/block/nullb0/inflight
       0        1
root@pserver8:~# find /sys/block/nullb0/mq/0/cpu* -name rq_list -print -exec cat {} \;
...
/sys/block/nullb0/mq/0/cpu2/rq_list
CTX pending:
        ffff8838038e2400
...

During debug it became clear that stalled request is always inserted in
the rq_list from the following path:

   save_stack_trace_tsk + 34
   blk_mq_insert_requests + 231
   blk_mq_flush_plug_list + 281
   blk_flush_plug_list + 199
   wait_on_page_bit + 192
   __filemap_fdatawait_range + 228
   filemap_fdatawait_range + 20
   filemap_write_and_wait_range + 63
   blkdev_fsync + 27
   vfs_fsync_range + 73
   blkdev_write_iter + 202
   __vfs_write + 170
   vfs_write + 169
   kernel_write + 56

So blk_flush_plug_list() was called with from_schedule == true.

If from_schedule is true, that means that finally blk_mq_insert_requests()
offloads execution of __blk_mq_run_hw_queue() and uses kblockd workqueue,
i.e. it calls kblockd_schedule_delayed_work_on().

That means, that we race with another CPU, which is about to execute
__blk_mq_run_hw_queue() work.

Further debugging shows the following traces from different CPUs:

  CPU#0                                  CPU#1
  ----------------------------------     -------------------------------
  reqeust A inserted
  STORE hctx->ctx_map[0] bit marked
  kblockd_schedule...() returns 1
  <schedule to kblockd workqueue>
                                         request B inserted
                                         STORE hctx->ctx_map[1] bit marked
                                         kblockd_schedule...() returns 0
  *** WORK PENDING bit is cleared ***
  flush_busy_ctxs() is executed, but
  bit 1, set by CPU#1, is not observed

As a result request B pended forever.

This behaviour can be explained by speculative LOAD of hctx->ctx_map on
CPU#0, which is reordered with clear of PENDING bit and executed _before_
actual STORE of bit 1 on CPU#1.

The proper fix is an explicit full barrier <mfence>, which guarantees
that clear of PENDING bit is to be executed before all possible
speculative LOADS or STORES inside actual work function.

Signed-off-by: Roman Pen <roman.penyaev@profitbricks.com>
Cc: Gioh Kim <gi-oh.kim@profitbricks.com>
Cc: Michael Wang <yun.wang@profitbricks.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: linux-block@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Cc: stable@vger.kernel.org
Signed-off-by: Tejun Heo <tj@kernel.org>
8 years agodrivers: firmware: psci: unify enable-method binding on ARM {64,32}-bit systems
Sudeep Holla [Fri, 22 Apr 2016 15:18:02 +0000 (16:18 +0100)] 
drivers: firmware: psci: unify enable-method binding on ARM {64,32}-bit systems

Currently ARM CPUs DT bindings allows different enable-method value for
PSCI based systems. On ARM 64-bit this property is required and must be
"psci" while on ARM 32-bit systems this property is optional and must
be "arm,psci" if present.

However, "arm,psci" has always been the compatible string for the PSCI
node, and was never intended to be the enable-method. So this is a bug
in the binding and not a deliberate attempt at specifying 32-bit
differently.

This is problematic if 32-bit OS is run on 64-bit system which has
"psci" as enable-method rather than the expected "arm,psci".

So let's unify the value into "psci" and remove support for "arm,psci"
before it finds any users.

Reported-by: Soby Mathew <Soby.Mathew@arm.com>
Cc: Rob Herring <robh+dt@kernel.org>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
8 years agoALSA: hda - Update BCLK also at hotplug for i915 HSW/BDW
Takashi Iwai [Thu, 21 Apr 2016 14:39:17 +0000 (16:39 +0200)] 
ALSA: hda - Update BCLK also at hotplug for i915 HSW/BDW

The recent bug report suggests that BCLK setup for i915 HSW/BDW needs
to be updated at each HDMI hotplug, not only at initialization and
resume.  That is, we need to update HSW_EM4 and HSW_EM5 registers at
ELD notification, too.  Otherwise the HDMI audio may be out of sync
and played in a wrong pitch.

However, the HDA codec driver has no access to the controller
registers, and currently the code managing these registers is in
hda_intel.c, i.e. local to the controller driver.  For allowing the
explicit BCLK update from the codec driver, as in this patch, the
former haswell_set_bclk() in hda_intel.c is moved to hdac_i915.c and
exposed as snd_hdac_i915_set_bclk().  This is called from both the HDA
controller driver and intel_pin_eld_notify() in HDMI codec driver.

Along with this change, snd_hdac_get_display_clk() gets dropped as
it's no longer used.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91410
Cc: <stable@vger.kernel.org> # v4.5+
Signed-off-by: Takashi Iwai <tiwai@suse.de>
8 years agoALSA: hda - Add dock support for ThinkPad X260
Conrad Kostecki [Tue, 26 Apr 2016 08:08:10 +0000 (10:08 +0200)] 
ALSA: hda - Add dock support for ThinkPad X260

Fixes audio output on a ThinkPad X260, when using Lenovo CES 2013
docking station series (basic, pro, ultra).

Signed-off-by: Conrad Kostecki <ck+linuxkernel@bl4ckb0x.de>
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
8 years agonet/mlx4_en: fix spurious timestamping callbacks
Eric Dumazet [Sat, 23 Apr 2016 18:35:46 +0000 (11:35 -0700)] 
net/mlx4_en: fix spurious timestamping callbacks

When multiple skb are TX-completed in a row, we might incorrectly keep
a timestamp of a prior skb and cause extra work.

Fixes: ec693d47010e8 ("net/mlx4_en: Add HW timestamping (TS) support")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Reviewed-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dummy: remove note about being Y by default
Ivan Babrou [Sat, 23 Apr 2016 12:58:03 +0000 (12:58 +0000)] 
net: dummy: remove note about being Y by default

Signed-off-by: Ivan Babrou <ivan@cloudflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMD: make bio mergeable
Shaohua Li [Mon, 25 Apr 2016 23:52:38 +0000 (16:52 -0700)] 
MD: make bio mergeable

blk_queue_split marks bio unmergeable, which makes sense for normal bio.
But if dispatching the bio to underlayer disk, the blk_queue_split
checks are invalid, hence it's possible the bio becomes mergeable.

In the reported bug, this bug causes trim against raid0 performance slash
https://bugzilla.kernel.org/show_bug.cgi?id=117051

Reported-and-tested-by: Park Ju Hyung <qkrwngud825@gmail.com>
Fixes: 6ac45aeb6bca(block: avoid to merge splitted bio)
Cc: stable@vger.kernel.org (v4.3+)
Cc: Ming Lei <ming.lei@canonical.com>
Cc: Neil Brown <neilb@suse.de>
Reviewed-by: Jens Axboe <axboe@fb.com>
Signed-off-by: Shaohua Li <shli@fb.com>
8 years agocpufreq: intel_pstate: Fix processing for turbo activation ratio
Srinivas Pandruvada [Sat, 23 Apr 2016 02:53:59 +0000 (19:53 -0700)] 
cpufreq: intel_pstate: Fix processing for turbo activation ratio

When the config TDP level is not nominal (level = 0), the MSR values for
reading level 1 and level 2 ratios contain power in low 14 bits and actual
ratio bits are at bits [23:16]. The current processing for level 1 and
level 2 is wrong as there is no shift done to get actual ratio.

Fixes: 6a35fc2d6c22 (cpufreq: intel_pstate: get P1 from TAR when available)
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Cc: 4.4+ <stable@vger.kernel.org> # 4.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
8 years agocxgbi: fix uninitialized flowi6
Jiri Benc [Fri, 22 Apr 2016 11:09:13 +0000 (13:09 +0200)] 
cxgbi: fix uninitialized flowi6

ip6_route_output looks into different fields in the passed flowi6 structure,
yet cxgbi passes garbage in nearly all those fields. Zero the structure out
first.

Fixes: fc8d0590d9142 ("libcxgbi: Add ipv6 api to driver")
Signed-off-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomemcg: relocate charge moving from ->attach to ->post_attach
Tejun Heo [Thu, 21 Apr 2016 23:09:02 +0000 (19:09 -0400)] 
memcg: relocate charge moving from ->attach to ->post_attach

Hello,

So, this ended up a lot simpler than I originally expected.  I tested
it lightly and it seems to work fine.  Petr, can you please test these
two patches w/o the lru drain drop patch and see whether the problem
is gone?

Thanks.
------ 8< ------
If charge moving is used, memcg performs relabeling of the affected
pages from its ->attach callback which is called under both
cgroup_threadgroup_rwsem and thus can't create new kthreads.  This is
fragile as various operations may depend on workqueues making forward
progress which relies on the ability to create new kthreads.

There's no reason to perform charge moving from ->attach which is deep
in the task migration path.  Move it to ->post_attach which is called
after the actual migration is finished and cgroup_threadgroup_rwsem is
dropped.

* move_charge_struct->mm is added and ->can_attach is now responsible
  for pinning and recording the target mm.  mem_cgroup_clear_mc() is
  updated accordingly.  This also simplifies mem_cgroup_move_task().

* mem_cgroup_move_task() is now called from ->post_attach instead of
  ->attach.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@kernel.org>
Debugged-and-tested-by: Petr Mladek <pmladek@suse.com>
Reported-by: Cyril Hrubis <chrubis@suse.cz>
Reported-by: Johannes Weiner <hannes@cmpxchg.org>
Fixes: 1ed1328792ff ("sched, cgroup: replace signal_struct->group_rwsem with a global percpu_rwsem")
Cc: <stable@vger.kernel.org> # 4.4+
8 years agocgroup, cpuset: replace cpuset_post_attach_flush() with cgroup_subsys->post_attach...
Tejun Heo [Thu, 21 Apr 2016 23:06:48 +0000 (19:06 -0400)] 
cgroup, cpuset: replace cpuset_post_attach_flush() with cgroup_subsys->post_attach callback

Since e93ad19d0564 ("cpuset: make mm migration asynchronous"), cpuset
kicks off asynchronous NUMA node migration if necessary during task
migration and flushes it from cpuset_post_attach_flush() which is
called at the end of __cgroup_procs_write().  This is to avoid
performing migration with cgroup_threadgroup_rwsem write-locked which
can lead to deadlock through dependency on kworker creation.

memcg has a similar issue with charge moving, so let's convert it to
an official callback rather than the current one-off cpuset specific
function.  This patch adds cgroup_subsys->post_attach callback and
makes cpuset register cpuset_post_attach_flush() as its ->post_attach.

The conversion is mostly one-to-one except that the new callback is
called under cgroup_mutex.  This is to guarantee that no other
migration operations are started before ->post_attach callbacks are
finished.  cgroup_mutex is one of the outermost mutex in the system
and has never been and shouldn't be a problem.  We can add specialized
synchronization around __cgroup_procs_write() but I don't think
there's any noticeable benefit.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: <stable@vger.kernel.org> # 4.4+ prerequisite for the next patch
This page took 0.06261 seconds and 5 git commands to generate.