GitHub/mt8127/android_kernel_alcatel_ttab.git
7 years agousb: gadget: u_ether: remove interrupt throttling
Felipe Balbi [Tue, 1 Nov 2016 11:20:22 +0000 (13:20 +0200)]
usb: gadget: u_ether: remove interrupt throttling

commit fd9afd3cbe404998d732be6cc798f749597c5114 upstream.

According to Dave Miller "the networking stack has a
hard requirement that all SKBs which are transmitted
must have their completion signalled in a fininte
amount of time. This is because, until the SKB is
freed by the driver, it holds onto socket,
netfilter, and other subsystem resources."

In summary, this means that using TX IRQ throttling
for the networking gadgets is, at least, complex and
we should avoid it for the time being.

Reported-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Tested-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Suggested-by: David Miller <davem@davemloft.net>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoUSB: serial: cp210x: fix tiocmget error handling
Johan Hovold [Wed, 19 Oct 2016 13:45:07 +0000 (15:45 +0200)]
USB: serial: cp210x: fix tiocmget error handling

commit de24e0a108bc48062e1c7acaa97014bce32a919f upstream.

The current tiocmget implementation would fail to report errors up the
stack and instead leaked a few bits from the stack as a mask of
modem-status flags.

Fixes: 39a66b8d22a3 ("[PATCH] USB: CP2101 Add support for flow control")
Signed-off-by: Johan Hovold <johan@kernel.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agousb: gadget: function: u_ether: don't starve tx request queue
Felipe Balbi [Tue, 4 Oct 2016 12:14:43 +0000 (15:14 +0300)]
usb: gadget: function: u_ether: don't starve tx request queue

commit 6c83f77278f17a7679001027e9231291c20f0d8a upstream.

If we don't guarantee that we will always get an
interrupt at least when we're queueing our very last
request, we could fall into situation where we queue
every request with 'no_interrupt' set. This will
cause the link to get stuck.

The behavior above has been triggered with g_ether
and dwc3.

Reported-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agousb: misc: legousbtower: Fix NULL pointer deference
Greg Kroah-Hartman [Mon, 19 Sep 2016 18:09:51 +0000 (19:09 +0100)]
usb: misc: legousbtower: Fix NULL pointer deference

commit 2fae9e5a7babada041e2e161699ade2447a01989 upstream.

This patch fixes a NULL pointer dereference caused by a race codition in
the probe function of the legousbtower driver. It re-structures the
probe function to only register the interface after successfully reading
the board's firmware ID.

The probe function does not deregister the usb interface after an error
receiving the devices firmware ID. The device file registered
(/dev/usb/legousbtower%d) may be read/written globally before the probe
function returns. When tower_delete is called in the probe function
(after an r/w has been initiated), core dev structures are deleted while
the file operation functions are still running. If the 0 address is
mappable on the machine, this vulnerability can be used to create a
Local Priviege Escalation exploit via a write-what-where condition by
remapping dev->interrupt_out_buffer in tower_write. A forged USB device
and local program execution would be required for LPE. The USB device
would have to delay the control message in tower_probe and accept
the control urb in tower_open whilst guest code initiated a write to the
device file as tower_delete is called from the error in tower_probe.

This bug has existed since 2003. Patch tested by emulated device.

Reported-by: James Patrick-Evans <james@jmp-e.com>
Tested-by: James Patrick-Evans <james@jmp-e.com>
Signed-off-by: James Patrick-Evans <james@jmp-e.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoUSB: serial: cp210x: fix hardware flow-control disable
Konstantin Shkolnyy [Wed, 4 May 2016 21:56:52 +0000 (16:56 -0500)]
USB: serial: cp210x: fix hardware flow-control disable

commit a377f9e906af4df9071ba8ddba60188cb4013d93 upstream.

A bug in the CRTSCTS handling caused RTS to alternate between

CRTSCTS=0 => "RTS is transmit active signal" and
CRTSCTS=1 => "RTS is used for receive flow control"

instead of

CRTSCTS=0 => "RTS is statically active" and
CRTSCTS=1 => "RTS is used for receive flow control"

This only happened after first having enabled CRTSCTS.

Signed-off-by: Konstantin Shkolnyy <konstantin.shkolnyy@gmail.com>
Fixes: 39a66b8d22a3 ("[PATCH] USB: CP2101 Add support for flow control")
[johan: reword commit message ]
Signed-off-by: Johan Hovold <johan@kernel.org>
[johan: backport to 4.4 ]
Signed-off-by: Johan Hovold <johan@kernel.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agousb: gadget: fsl_qe_udc: signedness bug in qe_get_frame()
Dan Carpenter [Fri, 15 Jul 2016 11:15:47 +0000 (14:15 +0300)]
usb: gadget: fsl_qe_udc: signedness bug in qe_get_frame()

commit f4693b08cc901912a87369c46537b94ed4084ea0 upstream.

We can't assign -EINVAL to a u16.

Fixes: 3948f0e0c999 ('usb: add Freescale QE/CPM USB peripheral controller driver')
Acked-by: Peter Chen <peter.chen@nxp.com>
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoUSB: change bInterval default to 10 ms
Alan Stern [Fri, 16 Sep 2016 14:24:26 +0000 (10:24 -0400)]
USB: change bInterval default to 10 ms

commit 08c5cd37480f59ea39682f4585d92269be6b1424 upstream.

Some full-speed mceusb infrared transceivers contain invalid endpoint
descriptors for their interrupt endpoints, with bInterval set to 0.
In the past they have worked out okay with the mceusb driver, because
the driver sets the bInterval field in the descriptor to 1,
overwriting whatever value may have been there before.  However, this
approach was never sanctioned by the USB core, and in fact it does not
work with xHCI controllers, because they use the bInterval value that
was present when the configuration was installed.

Currently usbcore uses 32 ms as the default interval if the value in
the endpoint descriptor is invalid.  It turns out that these IR
transceivers don't work properly unless the interval is set to 10 ms
or below.  To work around this mceusb problem, this patch changes the
endpoint-descriptor parsing routine, making the default interval value
be 10 ms rather than 32 ms.

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Tested-by: Wade Berrier <wberrier@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agousb: renesas_usbhs: fix clearing the {BRDY,BEMP}STS condition
Yoshihiro Shimoda [Mon, 29 Aug 2016 09:00:38 +0000 (18:00 +0900)]
usb: renesas_usbhs: fix clearing the {BRDY,BEMP}STS condition

commit 519d8bd4b5d3d82c413eac5bb42b106bb4b9ec15 upstream.

The previous driver is possible to stop the transfer wrongly.
For example:
 1) An interrupt happens, but not BRDY interruption.
 2) Read INTSTS0. And than state->intsts0 is not set to BRDY.
 3) BRDY is set to 1 here.
 4) Read BRDYSTS.
 5) Clear the BRDYSTS. And then. the BRDY is cleared wrongly.

Remarks:
 - The INTSTS0.BRDY is read only.
  - If any bits of BRDYSTS are set to 1, the BRDY is set to 1.
  - If BRDYSTS is 0, the BRDY is set to 0.

So, this patch adds condition to avoid such situation. (And about
NRDYSTS, this is not used for now. But, avoiding any side effects,
this patch doesn't touch it.)

Fixes: d5c6a1e024dd ("usb: renesas_usbhs: fixup interrupt status clear method")
Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoUSB: serial: mos7840: fix non-atomic allocation in write path
Alexey Khoroshilov [Thu, 11 Aug 2016 22:05:09 +0000 (01:05 +0300)]
USB: serial: mos7840: fix non-atomic allocation in write path

commit 3b7c7e52efda0d4640060de747768360ba70a7c0 upstream.

There is an allocation with GFP_KERNEL flag in mos7840_write(),
while it may be called from interrupt context.

Follow-up for commit 191252837626 ("USB: kobil_sct: fix non-atomic
allocation in write path")

Found by Linux Driver Verification project (linuxtesting.org).

Signed-off-by: Alexey Khoroshilov <khoroshilov@ispras.ru>
Signed-off-by: Johan Hovold <johan@kernel.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoUSB: serial: mos7720: fix non-atomic allocation in write path
Alexey Khoroshilov [Thu, 11 Aug 2016 22:05:08 +0000 (01:05 +0300)]
USB: serial: mos7720: fix non-atomic allocation in write path

commit 5a5a1d614287a647b36dff3f40c2b0ceabbc83ec upstream.

There is an allocation with GFP_KERNEL flag in mos7720_write(),
while it may be called from interrupt context.

Follow-up for commit 191252837626 ("USB: kobil_sct: fix non-atomic
allocation in write path")

Found by Linux Driver Verification project (linuxtesting.org).

Signed-off-by: Alexey Khoroshilov <khoroshilov@ispras.ru>
Signed-off-by: Johan Hovold <johan@kernel.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoUSB: kobil_sct: fix non-atomic allocation in write path
Johan Hovold [Wed, 29 Oct 2014 08:07:30 +0000 (09:07 +0100)]
USB: kobil_sct: fix non-atomic allocation in write path

commit 191252837626fca0de694c18bb2aa64c118eda89 upstream

Write may be called from interrupt context so make sure to use
GFP_ATOMIC for all allocations in write.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Johan Hovold <johan@kernel.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoUSB: serial: fix memleak in driver-registration error path
Alexey Klimov [Mon, 8 Aug 2016 01:34:46 +0000 (02:34 +0100)]
USB: serial: fix memleak in driver-registration error path

commit 647024a7df36014bbc4479d92d88e6b77c0afcf6 upstream.

udriver struct allocated by kzalloc() will not be freed
if usb_register() and next calls fail. This patch fixes this
by adding one more step with kfree(udriver) in error path.

Signed-off-by: Alexey Klimov <klimov.linux@gmail.com>
Acked-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Johan Hovold <johan@kernel.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agousb: xhci: Fix panic if disconnect
Jim Lin [Tue, 16 Aug 2016 07:18:05 +0000 (10:18 +0300)]
usb: xhci: Fix panic if disconnect

commit 88716a93766b8f095cdef37a8e8f2c93aa233b21 upstream.

After a device is disconnected, xhci_stop_device() will be invoked
in xhci_bus_suspend().
Also the "disconnect" IRQ will have ISR to invoke
xhci_free_virt_device() in this sequence.
xhci_irq -> xhci_handle_event -> handle_cmd_completion ->
xhci_handle_cmd_disable_slot -> xhci_free_virt_device

If xhci->devs[slot_id] has been assigned to NULL in
xhci_free_virt_device(), then virt_dev->eps[i].ring in
xhci_stop_device() may point to an invlid address to cause kernel
panic.

virt_dev = xhci->devs[slot_id];
:
if (virt_dev->eps[i].ring && virt_dev->eps[i].ring->dequeue)

[] Unable to handle kernel paging request at virtual address 00001a68
[] pgd=ffffffc001430000
[] [00001a68] *pgd=000000013c807003, *pud=000000013c807003,
*pmd=000000013c808003, *pte=0000000000000000
[] Internal error: Oops: 96000006 [#1] PREEMPT SMP
[] CPU: 0 PID: 39 Comm: kworker/0:1 Tainted: G     U
[] Workqueue: pm pm_runtime_work
[] task: ffffffc0bc0e0bc0 ti: ffffffc0bc0ec000 task.ti:
ffffffc0bc0ec000
[] PC is at xhci_stop_device.constprop.11+0xb4/0x1a4

This issue is found when running with realtek ethernet device
(0bda:8153).

Signed-off-by: Jim Lin <jilin@nvidia.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoUSB: fix typo in wMaxPacketSize validation
Alan Stern [Mon, 22 Aug 2016 20:58:53 +0000 (16:58 -0400)]
USB: fix typo in wMaxPacketSize validation

commit 6c73358c83ce870c0cf32413e5cadb3b9a39c606 upstream.

The maximum value allowed for wMaxPacketSize of a high-speed interrupt
endpoint is 1024 bytes, not 1023.

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Fixes: aed9d65ac327 ("USB: validate wMaxPacketValue entries in endpoint descriptors")
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoUSB: validate wMaxPacketValue entries in endpoint descriptors
Alan Stern [Mon, 1 Aug 2016 19:25:56 +0000 (15:25 -0400)]
USB: validate wMaxPacketValue entries in endpoint descriptors

commit aed9d65ac3278d4febd8665bd7db59ef53e825fe upstream.

Erroneous or malicious endpoint descriptors may have non-zero bits in
reserved positions, or out-of-bounds values.  This patch helps prevent
these from causing problems by bounds-checking the wMaxPacketValue
entries in endpoint descriptors and capping the values at the maximum
allowed.

This issue was first discovered and tests were conducted by Jake Lamberson
<jake.lamberson1@gmail.com>, an intern working for Rosie Hall.

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Reported-by: roswest <roswest@cisco.com>
Tested-by: roswest <roswest@cisco.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[wt: adjusted to 3.10 -- no USB_SPEED_SUPER_PLUS]

Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agousb: dwc3: gadget: increment request->actual once
Felipe Balbi [Fri, 29 Jul 2016 00:17:58 +0000 (03:17 +0300)]
usb: dwc3: gadget: increment request->actual once

commit c7de573471832dff7d31f0c13b0f143d6f017799 upstream.

When using SG lists, we would end up setting
request->actual to:

num_mapped_sgs * (request->length - count)

Let's fix that up by incrementing request->actual
only once.

Reported-by: Brian E Rogers <brian.e.rogers@intel.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agocoredump: fix unfreezable coredumping task
Andrey Ryabinin [Thu, 10 Nov 2016 18:46:38 +0000 (10:46 -0800)]
coredump: fix unfreezable coredumping task

commit 70d78fe7c8b640b5acfad56ad341985b3810998a upstream.

It could be not possible to freeze coredumping task when it waits for
'core_state->startup' completion, because threads are frozen in
get_signal() before they got a chance to complete 'core_state->startup'.

Inability to freeze a task during suspend will cause suspend to fail.
Also CRIU uses cgroup freezer during dump operation.  So with an
unfreezable task the CRIU dump will fail because it waits for a
transition from 'FREEZING' to 'FROZEN' state which will never happen.

Use freezer_do_not_count() to tell freezer to ignore coredumping task
while it waits for core_state->startup completion.

Link: http://lkml.kernel.org/r/1475225434-3753-1-git-send-email-aryabinin@virtuozzo.com
Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Acked-by: Pavel Machek <pavel@ucw.cz>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Tejun Heo <tj@kernel.org>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoswapfile: fix memory corruption via malformed swapfile
Jann Horn [Thu, 10 Nov 2016 18:46:19 +0000 (10:46 -0800)]
swapfile: fix memory corruption via malformed swapfile

commit dd111be69114cc867f8e826284559bfbc1c40e37 upstream.

When root activates a swap partition whose header has the wrong
endianness, nr_badpages elements of badpages are swabbed before
nr_badpages has been checked, leading to a buffer overrun of up to 8GB.

This normally is not a security issue because it can only be exploited
by root (more specifically, a process with CAP_SYS_ADMIN or the ability
to modify a swap file/partition), and such a process can already e.g.
modify swapped-out memory of any other userspace process on the system.

Link: http://lkml.kernel.org/r/1477949533-2509-1-git-send-email-jann@thejh.net
Signed-off-by: Jann Horn <jann@thejh.net>
Acked-by: Kees Cook <keescook@chromium.org>
Acked-by: Jerome Marchand <jmarchan@redhat.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoFix potential infoleak in older kernels
Linus Torvalds [Tue, 8 Nov 2016 10:17:00 +0000 (11:17 +0100)]
Fix potential infoleak in older kernels

Not upstream as it is not needed there.

So a patch something like this might be a safe way to fix the
potential infoleak in older kernels.

THIS IS UNTESTED. It's a very obvious patch, though, so if it compiles
it probably works. It just initializes the output variable with 0 in
the inline asm description, instead of doing it in the exception
handler.

It will generate slightly worse code (a few unnecessary ALU
operations), but it doesn't have any interactions with the exception
handler implementation.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoarc: don't leak bits of kernel stack into coredump
Al Viro [Sat, 10 Sep 2016 20:31:04 +0000 (16:31 -0400)]
arc: don't leak bits of kernel stack into coredump

commit 7798bf2140ebcc36eafec6a4194fffd8d585d471 upstream.

On faulting sigreturn we do get SIGSEGV, all right, but anything
we'd put into pt_regs could end up in the coredump.  And since
__copy_from_user() never zeroed on arc, we'd better bugger off
on its failure without copying random uninitialized bits of
kernel stack into pt_regs...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agofix memory leaks in tracing_buffers_splice_read()
Al Viro [Sat, 17 Sep 2016 22:31:46 +0000 (18:31 -0400)]
fix memory leaks in tracing_buffers_splice_read()

commit 1ae2293dd6d2f5c823cf97e60b70d03631cd622f upstream.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agofix fault_in_multipages_...() on architectures with no-op access_ok()
Al Viro [Tue, 20 Sep 2016 19:07:42 +0000 (20:07 +0100)]
fix fault_in_multipages_...() on architectures with no-op access_ok()

commit e23d4159b109167126e5bcd7f3775c95de7fee47 upstream.

Switching iov_iter fault-in to multipages variants has exposed an old
bug in underlying fault_in_multipages_...(); they break if the range
passed to them wraps around.  Normally access_ok() done by callers will
prevent such (and it's a guaranteed EFAULT - ERR_PTR() values fall into
such a range and they should not point to any valid objects).

However, on architectures where userland and kernel live in different
MMU contexts (e.g. s390) access_ok() is a no-op and on those a range
with a wraparound can reach fault_in_multipages_...().

Since any wraparound means EFAULT there, the fix is trivial - turn
those

    while (uaddr <= end)
    ...
into

    if (unlikely(uaddr > end))
    return -EFAULT;
    do
    ...
    while (uaddr <= end);

Reported-by: Jan Stancek <jstancek@redhat.com>
Tested-by: Jan Stancek <jstancek@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoia64: copy_from_user() should zero the destination on access_ok() failure
Al Viro [Fri, 19 Aug 2016 01:31:41 +0000 (21:31 -0400)]
ia64: copy_from_user() should zero the destination on access_ok() failure

commit a5e541f796f17228793694d64b507f5f57db4cd7 upstream.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoppc32: fix copy_from_user()
Al Viro [Sun, 21 Aug 2016 23:16:26 +0000 (19:16 -0400)]
ppc32: fix copy_from_user()

commit 224264657b8b228f949b42346e09ed8c90136a8e upstream.

should clear on access_ok() failures.  Also remove the useless
range truncation logics.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agosparc32: fix copy_from_user()
Al Viro [Mon, 22 Aug 2016 04:23:07 +0000 (00:23 -0400)]
sparc32: fix copy_from_user()

commit 917400cecb4b52b5cde5417348322bb9c8272fa6 upstream.

Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agomn10300: copy_from_user() should zero on access_ok() failure...
Al Viro [Sat, 20 Aug 2016 20:33:10 +0000 (16:33 -0400)]
mn10300: copy_from_user() should zero on access_ok() failure...

commit ae7cc577ec2a4a6151c9e928fd1f595d953ecef1 upstream.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoopenrisc: fix the fix of copy_from_user()
Guenter Roeck [Sat, 17 Sep 2016 19:57:24 +0000 (12:57 -0700)]
openrisc: fix the fix of copy_from_user()

commit 8e4b72054f554967827e18be1de0e8122e6efc04 upstream.

Since commit acb2505d0119 ("openrisc: fix copy_from_user()"),
copy_from_user() returns the number of bytes requested, not the
number of bytes not copied.

Cc: Al Viro <viro@zeniv.linux.org.uk>
Fixes: acb2505d0119 ("openrisc: fix copy_from_user()")
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoopenrisc: fix copy_from_user()
Al Viro [Sat, 20 Aug 2016 21:05:21 +0000 (17:05 -0400)]
openrisc: fix copy_from_user()

commit acb2505d0119033a80c85ac8d02dccae41271667 upstream.

... that should zero on faults.  Also remove the <censored> helpful
logics wrt range truncation copied from ppc32.  Where it had ever
been needed only in case of copy_from_user() *and* had not been merged
into the mainline until a month after the need had disappeared.
A decade before openrisc went into mainline, I might add...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoparisc: fix copy_from_user()
Al Viro [Sat, 20 Aug 2016 23:03:37 +0000 (19:03 -0400)]
parisc: fix copy_from_user()

commit aace880feea38875fbc919761b77e5732a3659ef upstream.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agometag: copy_from_user() should zero the destination on access_ok() failure
Al Viro [Fri, 19 Aug 2016 02:08:20 +0000 (22:08 -0400)]
metag: copy_from_user() should zero the destination on access_ok() failure

commit 8ae95ed4ae5fc7c3391ed668b2014c9e2079533b upstream.

Acked-by: James Hogan <james.hogan@imgtec.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoalpha: fix copy_from_user()
Al Viro [Wed, 17 Aug 2016 20:02:32 +0000 (16:02 -0400)]
alpha: fix copy_from_user()

commit 2561d309dfd1555e781484af757ed0115035ddb3 upstream.

it should clear the destination even when access_ok() fails.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoasm-generic: make copy_from_user() zero the destination properly
Al Viro [Wed, 17 Aug 2016 20:36:37 +0000 (16:36 -0400)]
asm-generic: make copy_from_user() zero the destination properly

commit 2545e5da080b4839dd859e3b09343a884f6ab0e3 upstream.

... in all cases, including the failing access_ok()

Note that some architectures using asm-generic/uaccess.h have
__copy_from_user() not zeroing the tail on failure halfway
through.  This variant works either way.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
[wt: s/might_fault/might_sleep]

Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agomips: copy_from_user() must zero the destination on access_ok() failure
Al Viro [Sat, 20 Aug 2016 20:18:53 +0000 (16:18 -0400)]
mips: copy_from_user() must zero the destination on access_ok() failure

commit e69d700535ac43a18032b3c399c69bf4639e89a2 upstream.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agohexagon: fix strncpy_from_user() error return
Al Viro [Fri, 19 Aug 2016 01:16:49 +0000 (21:16 -0400)]
hexagon: fix strncpy_from_user() error return

commit f35c1e0671728d1c9abc405d05ef548b5fcb2fc4 upstream.

It's -EFAULT, not -1 (and contrary to the comment in there,
__strnlen_user() can return 0 - on faults).

Acked-by: Richard Kuo <rkuo@codeaurora.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agosh: fix copy_from_user()
Al Viro [Mon, 22 Aug 2016 03:39:47 +0000 (23:39 -0400)]
sh: fix copy_from_user()

commit 6e050503a150b2126620c1a1e9b3a368fcd51eac upstream.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoscore: fix copy_from_user() and friends
Al Viro [Mon, 22 Aug 2016 02:30:44 +0000 (22:30 -0400)]
score: fix copy_from_user() and friends

commit b615e3c74621e06cd97f86373ca90d43d6d998aa upstream.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoblackfin: fix copy_from_user()
Al Viro [Fri, 9 Sep 2016 23:16:58 +0000 (19:16 -0400)]
blackfin: fix copy_from_user()

commit 8f035983dd826d7e04f67b28acf8e2f08c347e41 upstream.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agocris: buggered copy_from_user/copy_to_user/clear_user
Al Viro [Thu, 18 Aug 2016 23:34:00 +0000 (19:34 -0400)]
cris: buggered copy_from_user/copy_to_user/clear_user

commit eb47e0293baaa3044022059f1fa9ff474bfe35cb upstream.

* copy_from_user() on access_ok() failure ought to zero the destination
* none of those primitives should skip the access_ok() check in case of
small constant size.

Acked-by: Jesper Nilsson <jesper.nilsson@axis.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agofrv: fix clear_user()
Al Viro [Fri, 19 Aug 2016 00:54:02 +0000 (20:54 -0400)]
frv: fix clear_user()

commit 3b8767a8f00cc6538ba6b1cf0f88502e2fd2eb90 upstream.

It should check access_ok().  Otherwise a bunch of places turn into
trivially exploitable rootholes.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoasm-generic: make get_user() clear the destination on errors
Al Viro [Thu, 18 Aug 2016 03:19:01 +0000 (23:19 -0400)]
asm-generic: make get_user() clear the destination on errors

commit 9ad18b75c2f6e4a78ce204e79f37781f8815c0fa upstream.

both for access_ok() failures and for faults halfway through

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoARC: uaccess: get_user to zero out dest in cause of fault
Vineet Gupta [Fri, 19 Aug 2016 19:10:02 +0000 (12:10 -0700)]
ARC: uaccess: get_user to zero out dest in cause of fault

commit 05d9d0b96e53c52a113fd783c0c97c830c8dc7af upstream.

Al reported potential issue with ARC get_user() as it wasn't clearing
out destination pointer in case of fault due to bad address etc.

Verified using following

| {
|   u32 bogus1 = 0xdeadbeef;
| u64 bogus2 = 0xdead;
| int rc1, rc2;
|
|   pr_info("Orig values %x %llx\n", bogus1, bogus2);
| rc1 = get_user(bogus1, (u32 __user *)0x40000000);
| rc2 = get_user(bogus2, (u64 __user *)0x50000000);
| pr_info("access %d %d, new values %x %llx\n",
| rc1, rc2, bogus1, bogus2);
| }

| [ARCLinux]# insmod /mnt/kernel-module/qtn.ko
| Orig values deadbeef dead
| access -14 -14, new values 0 0

Reported-by: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: linux-snps-arc@lists.infradead.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agos390: get_user() should zero on failure
Al Viro [Mon, 22 Aug 2016 02:00:54 +0000 (22:00 -0400)]
s390: get_user() should zero on failure

commit fd2d2b191fe75825c4c7a6f12f3fef35aaed7dd7 upstream.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoscore: fix __get_user/get_user
Al Viro [Mon, 22 Aug 2016 02:13:39 +0000 (22:13 -0400)]
score: fix __get_user/get_user

commit c2f18fa4cbb3ad92e033a24efa27583978ce9600 upstream.

* should zero on any failure
* __get_user() should use __copy_from_user(), not copy_from_user()

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agosh64: failing __get_user() should zero
Al Viro [Mon, 22 Aug 2016 03:33:47 +0000 (23:33 -0400)]
sh64: failing __get_user() should zero

commit c6852389228df9fb3067f94f3b651de2a7921b36 upstream.

It could be done in exception-handling bits in __get_user_b() et.al.,
but the surgery involved would take more knowledge of sh64 details
than I have or _want_ to have.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agom32r: fix __get_user()
Al Viro [Fri, 9 Sep 2016 23:20:13 +0000 (19:20 -0400)]
m32r: fix __get_user()

commit c90a3bc5061d57e7931a9b7ad14784e1a0ed497d upstream.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agomn10300: failing __get_user() and get_user() should zero
Al Viro [Sat, 20 Aug 2016 20:32:02 +0000 (16:32 -0400)]
mn10300: failing __get_user() and get_user() should zero

commit 43403eabf558d2800b429cd886e996fd555aa542 upstream.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agomicroblaze: fix copy_from_user()
Al Viro [Fri, 9 Sep 2016 23:22:34 +0000 (19:22 -0400)]
microblaze: fix copy_from_user()

commit d0cf385160c12abd109746cad1f13e3b3e8b50b8 upstream.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
[wt: s/might_fault/might_sleep]

Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agomicroblaze: fix __get_user()
Al Viro [Fri, 9 Sep 2016 23:23:33 +0000 (19:23 -0400)]
microblaze: fix __get_user()

commit e98b9e37ae04562d52c96f46b3cf4c2e80222dc1 upstream.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoparisc: Ensure consistent state when switching to kernel stack at syscall entry
John David Anglin [Sat, 29 Oct 2016 03:00:34 +0000 (23:00 -0400)]
parisc: Ensure consistent state when switching to kernel stack at syscall entry

commit 6ed518328d0189e0fdf1bb7c73290d546143ea66 upstream.

We have one critical section in the syscall entry path in which we switch from
the userspace stack to kernel stack. In the event of an external interrupt, the
interrupt code distinguishes between those two states by analyzing the value of
sr7. If sr7 is zero, it uses the kernel stack. Therefore it's important, that
the value of sr7 is in sync with the currently enabled stack.

This patch now disables interrupts while executing the critical section.  This
prevents the interrupt handler to possibly see an inconsistent state which in
the worst case can lead to crashes.

Interestingly, in the syscall exit path interrupts were already disabled in the
critical section which switches back to the userspace stack.

Signed-off-by: John David Anglin <dave.anglin@bell.net>
Signed-off-by: Helge Deller <deller@gmx.de>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agos390/dasd: fix hanging device after clear subchannel
Stefan Haberland [Mon, 8 Aug 2016 12:08:17 +0000 (14:08 +0200)]
s390/dasd: fix hanging device after clear subchannel

commit 9ba333dc55cbb9523553df973adb3024d223e905 upstream.

When a device is in a status where CIO has killed all I/O by itself the
interrupt for a clear request may not contain an irb to determine the
clear function. Instead it contains an error pointer -EIO.
This was ignored by the DASD int_handler leading to a hanging device
waiting for a clear interrupt.

Handle -EIO error pointer correctly for requests that are clear pending and
treat the clear as successful.

Signed-off-by: Stefan Haberland <sth@linux.vnet.ibm.com>
Reviewed-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoavr32: off by one in at32_init_pio()
Dan Carpenter [Wed, 13 Jul 2016 10:08:55 +0000 (13:08 +0300)]
avr32: off by one in at32_init_pio()

commit 55f1cf83d5cf885c75267269729805852039c834 upstream.

The pio_dev[] array has MAX_NR_PIO_DEVICES elements so the > should be
>=.

Fixes: 5f97f7f9400d ('[PATCH] avr32 architecture')
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoavr32: fix 'undefined reference to `___copy_from_user'
Guenter Roeck [Sat, 17 Sep 2016 14:52:49 +0000 (07:52 -0700)]
avr32: fix 'undefined reference to `___copy_from_user'

commit 65c0044ca8d7c7bbccae37f0ff2972f0210e9f41 upstream.

avr32 builds fail with:

arch/avr32/kernel/built-in.o: In function `arch_ptrace':
(.text+0x650): undefined reference to `___copy_from_user'
arch/avr32/kernel/built-in.o:(___ksymtab+___copy_from_user+0x0): undefined
reference to `___copy_from_user'
kernel/built-in.o: In function `proc_doulongvec_ms_jiffies_minmax':
(.text+0x5dd8): undefined reference to `___copy_from_user'
kernel/built-in.o: In function `proc_dointvec_minmax_sysadmin':
sysctl.c:(.text+0x6174): undefined reference to `___copy_from_user'
kernel/built-in.o: In function `ptrace_has_cap':
ptrace.c:(.text+0x69c0): undefined reference to `___copy_from_user'
kernel/built-in.o:ptrace.c:(.text+0x6b90): more undefined references to
`___copy_from_user' follow

Fixes: 8630c32275ba ("avr32: fix copy_from_user()")
Cc: Al Viro <viro@zeniv.linux.org.uk>
Acked-by: Havard Skinnemoen <hskinnemoen@gmail.com>
Acked-by: Hans-Christian Noren Egtvedt <egtvedt@samfundet.no>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoavr32: fix copy_from_user()
Al Viro [Fri, 9 Sep 2016 23:28:23 +0000 (19:28 -0400)]
avr32: fix copy_from_user()

commit 8630c32275bac2de6ffb8aea9d9b11663e7ad28e upstream.

really ugly, but apparently avr32 compilers turns access_ok() into
something so bad that they want it in assembler.  Left that way,
zeroing added in inline wrapper.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agopowerpc/nvram: Fix an incorrect partition merge
Pan Xinhui [Thu, 10 Dec 2015 07:30:02 +0000 (15:30 +0800)]
powerpc/nvram: Fix an incorrect partition merge

commit 11b7e154b132232535befe51c55db048069c8461 upstream.

When we merge two contiguous partitions whose signatures are marked
NVRAM_SIG_FREE, We need update prev's length and checksum, then write it
to nvram, not cur's. So lets fix this mistake now.

Also use memset instead of strncpy to set the partition's name. It's
more readable if we want to fill up with duplicate chars .

Fixes: fa2b4e54d41f ("powerpc/nvram: Improve partition removal")
Signed-off-by: Pan Xinhui <xinhui.pan@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agopowerpc/64: Fix incorrect return value from __copy_tofrom_user
Paul Mackerras [Tue, 11 Oct 2016 11:25:47 +0000 (22:25 +1100)]
powerpc/64: Fix incorrect return value from __copy_tofrom_user

commit 1a34439e5a0b2235e43f96816dbb15ee1154f656 upstream.

Debugging a data corruption issue with virtio-net/vhost-net led to
the observation that __copy_tofrom_user was occasionally returning
a value 16 larger than it should.  Since the return value from
__copy_tofrom_user is the number of bytes not copied, this means
that __copy_tofrom_user can occasionally return a value larger
than the number of bytes it was asked to copy.  In turn this can
cause higher-level copy functions such as copy_page_to_iter_iovec
to corrupt memory by copying data into the wrong memory locations.

It turns out that the failing case involves a fault on the store
at label 79, and at that point the first unmodified byte of the
destination is at R3 + 16.  Consequently the exception handler
for that store needs to add 16 to R3 before using it to work out
how many bytes were not copied, but in this one case it was not
adding the offset to R3.  To fix it, this moves the label 179 to
the point where we add 16 to R3.  I have checked manually all the
exception handlers for the loads and stores in this code and the
rest of them are correct (it would be excellent to have an
automated test of all the exception cases).

This bug has been present since this code was initially
committed in May 2002 to Linux version 2.5.20.

Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agopowerpc/powernv: Use CPU-endian PEST in pnv_pci_dump_p7ioc_diag_data()
Gavin Shan [Tue, 2 Aug 2016 04:10:32 +0000 (14:10 +1000)]
powerpc/powernv: Use CPU-endian PEST in pnv_pci_dump_p7ioc_diag_data()

commit 5adaf8629b193f185ca5a1665b9e777a0579f518 upstream.

This fixes the warnings reported from sparse:

  pci.c:312:33: warning: restricted __be64 degrades to integer
  pci.c:313:33: warning: restricted __be64 degrades to integer

Fixes: cee72d5bb489 ("powerpc/powernv: Display diag data on p7ioc EEH errors")
Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agopowerpc/vdso64: Use double word compare on pointers
Anton Blanchard [Sun, 25 Sep 2016 07:16:53 +0000 (17:16 +1000)]
powerpc/vdso64: Use double word compare on pointers

commit 5045ea37377ce8cca6890d32b127ad6770e6dce5 upstream.

__kernel_get_syscall_map() and __kernel_clock_getres() use cmpli to
check if the passed in pointer is non zero. cmpli maps to a 32 bit
compare on binutils, so we ignore the top 32 bits.

A simple test case can be created by passing in a bogus pointer with
the bottom 32 bits clear. Using a clk_id that is handled by the VDSO,
then one that is handled by the kernel shows the problem:

  printf("%d\n", clock_getres(CLOCK_REALTIME, (void *)0x100000000));
  printf("%d\n", clock_getres(CLOCK_BOOTTIME, (void *)0x100000000));

And we get:

  0
  -1

The bigger issue is if we pass a valid pointer with the bottom 32 bits
clear, in this case we will return success but won't write any data
to the pointer.

I stumbled across this issue because the LLVM integrated assembler
doesn't accept cmpli with 3 arguments. Fix this by converting them to
cmpldi.

Fixes: a7f290dad32e ("[PATCH] powerpc: Merge vdso's and add vdso support to 32 bits kernel")
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agopowerpc/mm: Don't alias user region to other regions below PAGE_OFFSET
Paul Mackerras [Fri, 2 Sep 2016 11:47:59 +0000 (21:47 +1000)]
powerpc/mm: Don't alias user region to other regions below PAGE_OFFSET

commit f077aaf0754bcba0fffdbd925bc12f09cd1e38aa upstream.

In commit c60ac5693c47 ("powerpc: Update kernel VSID range", 2013-03-13)
we lost a check on the region number (the top four bits of the effective
address) for addresses below PAGE_OFFSET.  That commit replaced a check
that the top 18 bits were all zero with a check that bits 46 - 59 were
zero (performed for all addresses, not just user addresses).

This means that userspace can access an address like 0x1000_0xxx_xxxx_xxxx
and we will insert a valid SLB entry for it.  The VSID used will be the
same as if the top 4 bits were 0, but the page size will be some random
value obtained by indexing beyond the end of the mm_ctx_high_slices_psize
array in the paca.  If that page size is the same as would be used for
region 0, then userspace just has an alias of the region 0 space.  If the
page size is different, then no HPTE will be found for the access, and
the process will get a SIGSEGV (since hash_page_mm() will refuse to create
a HPTE for the bogus address).

The access beyond the end of the mm_ctx_high_slices_psize can be at most
5.5MB past the array, and so will be in RAM somewhere.  Since the access
is a load performed in real mode, it won't fault or crash the kernel.
At most this bug could perhaps leak a little bit of information about
blocks of 32 bytes of memory located at offsets of i * 512kB past the
paca->mm_ctx_high_slices_psize array, for 1 <= i <= 11.

Fixes: c60ac5693c47 ("powerpc: Update kernel VSID range")
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoMIPS: ptrace: Fix regs_return_value for kernel context
Marcin Nowakowski [Wed, 12 Oct 2016 07:32:56 +0000 (09:32 +0200)]
MIPS: ptrace: Fix regs_return_value for kernel context

commit 74f1077b5b783e7bf4fa3007cefdc8dbd6c07518 upstream.

Currently regs_return_value always negates reg[2] if it determines
the syscall has failed, but when called in kernel context this check is
invalid and may result in returning a wrong value.

This fixes errors reported by CONFIG_KPROBES_SANITY_TEST

Fixes: d7e7528bcd45 ("Audit: push audit success and retcode into arch ptrace.h")
Signed-off-by: Marcin Nowakowski <marcin.nowakowski@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/14381/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoMIPS: Malta: Fix IOCU disable switch read for MIPS64
Paul Burton [Fri, 2 Sep 2016 15:07:10 +0000 (16:07 +0100)]
MIPS: Malta: Fix IOCU disable switch read for MIPS64

commit 305723ab439e14debc1d339aa04e835d488b8253 upstream.

Malta boards used with CPU emulators feature a switch to disable use of
an IOCU. Software has to check this switch & ignore any present IOCU if
the switch is closed. The read used to do this was unsafe for 64 bit
kernels, as it simply casted the address 0xbf403000 to a pointer &
dereferenced it. Whilst in a 32 bit kernel this would access kseg1, in a
64 bit kernel this attempts to access xuseg & results in an address
error exception.

Fix by accessing a correctly formed ckseg1 address generated using the
CKSEG1ADDR macro.

Whilst modifying this code, define the name of the register and the bit
we care about within it, which indicates whether PCI DMA is routed to
the IOCU or straight to DRAM. The code previously checked that bit 0 was
also set, but the least significant 7 bits of the CONFIG_GEN0 register
contain the value of the MReqInfo signal provided to the IOCU OCP bus,
so singling out bit 0 makes little sense & that part of the check is
dropped.

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Fixes: b6d92b4a6bdb ("MIPS: Add option to disable software I/O coherency.")
Cc: Matt Redfearn <matt.redfearn@imgtec.com>
Cc: Masahiro Yamada <yamada.masahiro@socionext.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/14187/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoarm64: debug: avoid resetting stepping state machine when TIF_SINGLESTEP
Will Deacon [Fri, 26 Aug 2016 10:36:39 +0000 (11:36 +0100)]
arm64: debug: avoid resetting stepping state machine when TIF_SINGLESTEP

commit 3a402a709500c5a3faca2111668c33d96555e35a upstream.

When TIF_SINGLESTEP is set for a task, the single-step state machine is
enabled and we must take care not to reset it to the active-not-pending
state if it is already in the active-pending state.

Unfortunately, that's exactly what user_enable_single_step does, by
unconditionally setting the SS bit in the SPSR for the current task.
This causes failures in the GDB testsuite, where GDB ends up missing
expected step traps if the instruction being stepped generates another
trap, e.g. PTRACE_EVENT_FORK from an SVC instruction.

This patch fixes the problem by preserving the current state of the
stepping state machine when TIF_SINGLESTEP is set on the current thread.

Cc: <stable@vger.kernel.org>
Reported-by: Yao Qi <yao.qi@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoarm64: spinlocks: implement smp_mb__before_spinlock() as smp_mb()
Will Deacon [Mon, 5 Sep 2016 10:56:05 +0000 (11:56 +0100)]
arm64: spinlocks: implement smp_mb__before_spinlock() as smp_mb()

commit 872c63fbf9e153146b07f0cece4da0d70b283eeb upstream.

smp_mb__before_spinlock() is intended to upgrade a spin_lock() operation
to a full barrier, such that prior stores are ordered with respect to
loads and stores occuring inside the critical section.

Unfortunately, the core code defines the barrier as smp_wmb(), which
is insufficient to provide the required ordering guarantees when used in
conjunction with our load-acquire-based spinlock implementation.

This patch overrides the arm64 definition of smp_mb__before_spinlock()
to map to a full smp_mb().

Cc: Peter Zijlstra <peterz@infradead.org>
Reported-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoarm64: Define AT_VECTOR_SIZE_ARCH for ARCH_DLINFO
James Hogan [Mon, 25 Jul 2016 15:59:52 +0000 (16:59 +0100)]
arm64: Define AT_VECTOR_SIZE_ARCH for ARCH_DLINFO

commit 3146bc64d12377a74dbda12b96ea32da3774ae07 upstream.

AT_VECTOR_SIZE_ARCH should be defined with the maximum number of
NEW_AUX_ENT entries that ARCH_DLINFO can contain, but it wasn't defined
for arm64 at all even though ARCH_DLINFO will contain one NEW_AUX_ENT
for the VDSO address.

This shouldn't be a problem as AT_VECTOR_SIZE_BASE includes space for
AT_BASE_PLATFORM which arm64 doesn't use, but lets define it now and add
the comment above ARCH_DLINFO as found in several other architectures to
remind future modifiers of ARCH_DLINFO to keep AT_VECTOR_SIZE_ARCH up to
date.

Fixes: f668cd1673aa ("arm64: ELF definitions")
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: linux-arm-kernel@lists.infradead.org
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoarm64: avoid returning from bad_mode
Mark Rutland [Mon, 23 Jan 2017 18:28:50 +0000 (18:28 +0000)]
arm64: avoid returning from bad_mode

commit 7d9e8f71b989230bc613d121ca38507d34ada849 upstream.

Generally, taking an unexpected exception should be a fatal event, and
bad_mode is intended to cater for this. However, it should be possible
to contain unexpected synchronous exceptions from EL0 without bringing
the kernel down, by sending a SIGILL to the task.

We tried to apply this approach in commit 9955ac47f4ba1c95 ("arm64:
don't kill the kernel on a bad esr from el0"), by sending a signal for
any bad_mode call resulting from an EL0 exception.

However, this also applies to other unexpected exceptions, such as
SError and FIQ. The entry paths for these exceptions branch to bad_mode
without configuring the link register, and have no kernel_exit. Thus, if
we take one of these exceptions from EL0, bad_mode will eventually
return to the original user link register value.

This patch fixes this by introducing a new bad_el0_sync handler to cater
for the recoverable case, and restoring bad_mode to its original state,
whereby it calls panic() and never returns. The recoverable case
branches to bad_el0_sync with a bl, and returns to userspace via the
usual ret_to_user mechanism.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Fixes: 9955ac47f4ba1c95 ("arm64: don't kill the kernel on a bad esr from el0")
Reported-by: Mark Salter <msalter@redhat.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: stable@vger.kernel.org
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoARM: sa1111: fix pcmcia suspend/resume
Russell King [Tue, 6 Sep 2016 13:34:05 +0000 (14:34 +0100)]
ARM: sa1111: fix pcmcia suspend/resume

commit 06dfe5cc0cc684e735cb0232fdb756d30780b05d upstream.

SA1111 PCMCIA was broken when PCMCIA switched to using dev_pm_ops for
the PCMCIA socket class.  PCMCIA used to handle suspend/resume via the
socket hosting device, which happened at normal device suspend/resume
time.

However, the referenced commit changed this: much of the resume now
happens much earlier, in the noirq resume handler of dev_pm_ops.

However, on SA1111, the PCMCIA device is not accessible as the SA1111
has not been resumed at _noirq time.  It's slightly worse than that,
because the SA1111 has already been put to sleep at _noirq time, so
suspend doesn't work properly.

Fix this by converting the core SA1111 code to use dev_pm_ops as well,
and performing its own suspend/resume at noirq time.

This fixes these errors in the kernel log:

pcmcia_socket pcmcia_socket0: time out after reset
pcmcia_socket pcmcia_socket1: time out after reset

and the resulting lack of PCMCIA cards after a S2RAM cycle.

Fixes: d7646f7632549 ("pcmcia: use dev_pm_ops for class pcmcia_socket_class")
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoARM: sa1100: clear reset status prior to reboot
Russell King [Fri, 19 Aug 2016 15:34:45 +0000 (16:34 +0100)]
ARM: sa1100: clear reset status prior to reboot

commit da60626e7d02a4f385cae80e450afc8b07035368 upstream.

Clear the current reset status prior to rebooting the platform.  This
adds the bit missing from 04fef228fb00 ("[ARM] pxa: introduce
reset_status and clear_reset_status for driver's usage").

Fixes: 04fef228fb00 ("[ARM] pxa: introduce reset_status and clear_reset_status for driver's usage")
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoARM: 8618/1: decompressor: reset ttbcr fields to use TTBR0 on ARMv7
Srinivas Ramana [Fri, 30 Sep 2016 14:03:31 +0000 (15:03 +0100)]
ARM: 8618/1: decompressor: reset ttbcr fields to use TTBR0 on ARMv7

commit 117e5e9c4cfcb7628f08de074fbfefec1bb678b7 upstream.

If the bootloader uses the long descriptor format and jumps to
kernel decompressor code, TTBCR may not be in a right state.
Before enabling the MMU, it is required to clear the TTBCR.PD0
field to use TTBR0 for translation table walks.

The commit dbece45894d3a ("ARM: 7501/1: decompressor:
reset ttbcr for VMSA ARMv7 cores") does the reset of TTBCR.N, but
doesn't consider all the bits for the size of TTBCR.N.

Clear TTBCR.PD0 field and reset all the three bits of TTBCR.N to
indicate the use of TTBR0 and the correct base address width.

Fixes: dbece45894d3 ("ARM: 7501/1: decompressor: reset ttbcr for VMSA ARMv7 cores")
Acked-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Srinivas Ramana <sramana@codeaurora.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoARM: 8616/1: dt: Respect property size when parsing CPUs
Robin Murphy [Mon, 26 Sep 2016 15:50:55 +0000 (16:50 +0100)]
ARM: 8616/1: dt: Respect property size when parsing CPUs

commit ba6dea4f7cedb4b1c17e36f4087675d817c2e24b upstream.

Whilst MPIDR values themselves are less than 32 bits, it is still
perfectly valid for a DT to have #address-cells > 1 in the CPUs node,
resulting in the "reg" property having leading zero cell(s). In that
situation, the big-endian nature of the data conspires with the current
behaviour of only reading the first cell to cause the kernel to think
all CPUs have ID 0, and become resoundingly unhappy as a consequence.

Take the full property length into account when parsing CPUs so as to
be correct under any circumstances.

Cc: Russell King <linux@armlinux.org.uk>
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoiommu/amd: Free domain id when free a domain of struct dma_ops_domain
Baoquan He [Thu, 15 Sep 2016 08:50:52 +0000 (16:50 +0800)]
iommu/amd: Free domain id when free a domain of struct dma_ops_domain

commit c3db901c54466a9c135d1e6e95fec452e8a42666 upstream.

The current code missed freeing domain id when free a domain of
struct dma_ops_domain.

Signed-off-by: Baoquan He <bhe@redhat.com>
Fixes: ec487d1a110a ('x86, AMD IOMMU: add domain allocation and deallocation functions')
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoiommu/amd: Update Alias-DTE in update_device_table()
Joerg Roedel [Tue, 26 Jul 2016 13:18:54 +0000 (15:18 +0200)]
iommu/amd: Update Alias-DTE in update_device_table()

commit 3254de6bf74fe94c197c9f819fe62a3a3c36f073 upstream.

Not doing so might cause IO-Page-Faults when a device uses
an alias request-id and the alias-dte is left in a lower
page-mode which does not cover the address allocated from
the iova-allocator.

Fixes: 492667dacc0a ('x86/amd-iommu: Remove amd_iommu_pd_table')
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agox86/um: reuse asm-generic/barrier.h
Michael S. Tsirkin [Mon, 21 Dec 2015 07:22:18 +0000 (09:22 +0200)]
x86/um: reuse asm-generic/barrier.h

commit 577f183acc88645eae116326cc2203dc88ea730c upstream.

On x86/um CONFIG_SMP is never defined.  As a result, several macros
match the asm-generic variant exactly. Drop the local definitions and
pull in asm-generic/barrier.h instead.

This is in preparation to refactoring this code area.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Richard Weinberger <richard@nod.at>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agox86/build: Build compressed x86 kernels as PIE
H.J. Lu [Thu, 17 Mar 2016 03:04:35 +0000 (20:04 -0700)]
x86/build: Build compressed x86 kernels as PIE

commit 6d92bc9d483aa1751755a66fee8fb39dffb088c0 upstream.

The 32-bit x86 assembler in binutils 2.26 will generate R_386_GOT32X
relocation to get the symbol address in PIC.  When the compressed x86
kernel isn't built as PIC, the linker optimizes R_386_GOT32X relocations
to their fixed symbol addresses.  However, when the compressed x86
kernel is loaded at a different address, it leads to the following
load failure:

  Failed to allocate space for phdrs

during the decompression stage.

If the compressed x86 kernel is relocatable at run-time, it should be
compiled with -fPIE, instead of -fPIC, if possible and should be built as
Position Independent Executable (PIE) so that linker won't optimize
R_386_GOT32X relocation to its fixed symbol address.

Older linkers generate R_386_32 relocations against locally defined
symbols, _bss, _ebss, _got and _egot, in PIE.  It isn't wrong, just less
optimal than R_386_RELATIVE.  But the x86 kernel fails to properly handle
R_386_32 relocations when relocating the kernel.  To generate
R_386_RELATIVE relocations, we mark _bss, _ebss, _got and _egot as
hidden in both 32-bit and 64-bit x86 kernels.

To build a 64-bit compressed x86 kernel as PIE, we need to disable the
relocation overflow check to avoid relocation overflow errors. We do
this with a new linker command-line option, -z noreloc-overflow, which
got added recently:

 commit 4c10bbaa0912742322f10d9d5bb630ba4e15dfa7
 Author: H.J. Lu <hjl.tools@gmail.com>
 Date:   Tue Mar 15 11:07:06 2016 -0700

    Add -z noreloc-overflow option to x86-64 ld

    Add -z noreloc-overflow command-line option to the x86-64 ELF linker to
    disable relocation overflow check.  This can be used to avoid relocation
    overflow check if there will be no dynamic relocation overflow at
    run-time.

The 64-bit compressed x86 kernel is built as PIE only if the linker supports
-z noreloc-overflow.  So far 64-bit relocatable compressed x86 kernel
boots fine even when it is built as a normal executable.

Signed-off-by: H.J. Lu <hjl.tools@gmail.com>
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: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
[ Edited the changelog and comments. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agox86/paravirt: Do not trace _paravirt_ident_*() functions
Steven Rostedt [Wed, 25 May 2016 17:47:26 +0000 (13:47 -0400)]
x86/paravirt: Do not trace _paravirt_ident_*() functions

commit 15301a570754c7af60335d094dd2d1808b0641a5 upstream.

Å\81ukasz Daniluk reported that on a RHEL kernel that his machine would lock up
after enabling function tracer. I asked him to bisect the functions within
available_filter_functions, which he did and it came down to three:

  _paravirt_nop(), _paravirt_ident_32() and _paravirt_ident_64()

It was found that this is only an issue when noreplace-paravirt is added
to the kernel command line.

This means that those functions are most likely called within critical
sections of the funtion tracer, and must not be traced.

In newer kenels _paravirt_nop() is defined within gcc asm(), and is no
longer an issue.  But both _paravirt_ident_{32,64}() causes the
following splat when they are traced:

 mm/pgtable-generic.c:33: bad pmd ffff8800d2435150(0000000001d00054)
 mm/pgtable-generic.c:33: bad pmd ffff8800d3624190(0000000001d00070)
 mm/pgtable-generic.c:33: bad pmd ffff8800d36a5110(0000000001d00054)
 mm/pgtable-generic.c:33: bad pmd ffff880118eb1450(0000000001d00054)
 NMI watchdog: BUG: soft lockup - CPU#2 stuck for 22s! [systemd-journal:469]
 Modules linked in: e1000e
 CPU: 2 PID: 469 Comm: systemd-journal Not tainted 4.6.0-rc4-test+ #513
 Hardware name: Hewlett-Packard HP Compaq Pro 6300 SFF/339A, BIOS K01 v02.05 05/07/2012
 task: ffff880118f740c0 ti: ffff8800d4aec000 task.ti: ffff8800d4aec000
 RIP: 0010:[<ffffffff81134148>]  [<ffffffff81134148>] queued_spin_lock_slowpath+0x118/0x1a0
 RSP: 0018:ffff8800d4aefb90  EFLAGS: 00000246
 RAX: 0000000000000000 RBX: 0000000000000000 RCX: ffff88011eb16d40
 RDX: ffffffff82485760 RSI: 000000001f288820 RDI: ffffea0000008030
 RBP: ffff8800d4aefb90 R08: 00000000000c0000 R09: 0000000000000000
 R10: ffffffff821c8e0e R11: 0000000000000000 R12: ffff880000200fb8
 R13: 00007f7a4e3f7000 R14: ffffea000303f600 R15: ffff8800d4b562e0
 FS:  00007f7a4e3d7840(0000) GS:ffff88011eb00000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 00007f7a4e3f7000 CR3: 00000000d3e71000 CR4: 00000000001406e0
 Call Trace:
   _raw_spin_lock+0x27/0x30
   handle_pte_fault+0x13db/0x16b0
   handle_mm_fault+0x312/0x670
   __do_page_fault+0x1b1/0x4e0
   do_page_fault+0x22/0x30
   page_fault+0x28/0x30
   __vfs_read+0x28/0xe0
   vfs_read+0x86/0x130
   SyS_read+0x46/0xa0
   entry_SYSCALL_64_fastpath+0x1e/0xa8
 Code: 12 48 c1 ea 0c 83 e8 01 83 e2 30 48 98 48 81 c2 40 6d 01 00 48 03 14 c5 80 6a 5d 82 48 89 0a 8b 41 08 85 c0 75 09 f3 90 8b 41 08 <85> c0 74 f7 4c 8b 09 4d 85 c9 74 08 41 0f 18 09 eb 02 f3 90 8b

Reported-by: Å\81ukasz Daniluk <lukasz.daniluk@intel.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agox86/mm/pat, /dev/mem: Remove superfluous error message
Jiri Kosina [Fri, 8 Jul 2016 09:38:28 +0000 (11:38 +0200)]
x86/mm/pat, /dev/mem: Remove superfluous error message

commit 39380b80d72723282f0ea1d1bbf2294eae45013e upstream.

Currently it's possible for broken (or malicious) userspace to flood a
kernel log indefinitely with messages a-la

Program dmidecode tried to access /dev/mem between f0000->100000

because range_is_allowed() is case of CONFIG_STRICT_DEVMEM being turned on
dumps this information each and every time devmem_is_allowed() fails.

Reportedly userspace that is able to trigger contignuous flow of these
messages exists.

It would be possible to rate limit this message, but that'd have a
questionable value; the administrator wouldn't get information about all
the failing accessess, so then the information would be both superfluous
and incomplete at the same time :)

Returning EPERM (which is what is actually happening) is enough indication
for userspace what has happened; no need to log this particular error as
some sort of special condition.

Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@kernel.org>
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: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Luis R. Rodriguez <mcgrof@suse.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Toshi Kani <toshi.kani@hp.com>
Link: http://lkml.kernel.org/r/alpine.LNX.2.00.1607081137020.24757@cbobk.fhfr.pm
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agox86/apic: Do not init irq remapping if ioapic is disabled
Wanpeng Li [Tue, 23 Aug 2016 12:07:19 +0000 (20:07 +0800)]
x86/apic: Do not init irq remapping if ioapic is disabled

commit 2e63ad4bd5dd583871e6602f9d398b9322d358d9 upstream.

native_smp_prepare_cpus
  -> default_setup_apic_routing
    -> enable_IR_x2apic
      -> irq_remapping_prepare
        -> intel_prepare_irq_remapping
          -> intel_setup_irq_remapping

So IR table is setup even if "noapic" boot parameter is added. As a result we
crash later when the interrupt affinity is set due to a half initialized
remapping infrastructure.

Prevent remap initialization when IOAPIC is disabled.

Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Joerg Roedel <joro@8bytes.org>
Link: http://lkml.kernel.org/r/1471954039-3942-1-git-send-email-wanpeng.li@hotmail.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agox86/mm: Disable preemption during CR3 read+write
Sebastian Andrzej Siewior [Fri, 5 Aug 2016 13:37:39 +0000 (15:37 +0200)]
x86/mm: Disable preemption during CR3 read+write

commit 5cf0791da5c162ebc14b01eb01631cfa7ed4fa6e upstream.

There's a subtle preemption race on UP kernels:

Usually current->mm (and therefore mm->pgd) stays the same during the
lifetime of a task so it does not matter if a task gets preempted during
the read and write of the CR3.

But then, there is this scenario on x86-UP:

TaskA is in do_exit() and exit_mm() sets current->mm = NULL followed by:

 -> mmput()
 -> exit_mmap()
 -> tlb_finish_mmu()
 -> tlb_flush_mmu()
 -> tlb_flush_mmu_tlbonly()
 -> tlb_flush()
 -> flush_tlb_mm_range()
 -> __flush_tlb_up()
 -> __flush_tlb()
 ->  __native_flush_tlb()

At this point current->mm is NULL but current->active_mm still points to
the "old" mm.

Let's preempt taskA _after_ native_read_cr3() by taskB. TaskB has its
own mm so CR3 has changed.

Now preempt back to taskA. TaskA has no ->mm set so it borrows taskB's
mm and so CR3 remains unchanged. Once taskA gets active it continues
where it was interrupted and that means it writes its old CR3 value
back. Everything is fine because userland won't need its memory
anymore.

Now the fun part:

Let's preempt taskA one more time and get back to taskB. This
time switch_mm() won't do a thing because oldmm (->active_mm)
is the same as mm (as per context_switch()). So we remain
with a bad CR3 / PGD and return to userland.

The next thing that happens is handle_mm_fault() with an address for
the execution of its code in userland. handle_mm_fault() realizes that
it has a PTE with proper rights so it returns doing nothing. But the
CPU looks at the wrong PGD and insists that something is wrong and
faults again. And again. And one more timeâ\80¦

This pagefault circle continues until the scheduler gets tired of it and
puts another task on the CPU. It gets little difficult if the task is a
RT task with a high priority. The system will either freeze or it gets
fixed by the software watchdog thread which usually runs at RT-max prio.
But waiting for the watchdog will increase the latency of the RT task
which is no good.

Fix this by disabling preemption across the critical code section.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Borislav Petkov <bp@suse.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-mm@kvack.org
Link: http://lkml.kernel.org/r/1470404259-26290-1-git-send-email-bigeasy@linutronix.de
[ Prettified the changelog. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agox86/traps: Ignore high word of regs->cs in early_idt_handler_common
Andy Lutomirski [Thu, 1 Dec 2016 17:26:42 +0000 (09:26 -0800)]
x86/traps: Ignore high word of regs->cs in early_idt_handler_common

This is a backport of:
commit fc0e81b2bea0ebceb71889b61d2240856141c9ee upstream

On the 80486 DX, it seems that some exceptions may leave garbage in
the high bits of CS.  This causes sporadic failures in which
early_fixup_exception() refuses to fix up an exception.

As far as I can tell, this has been buggy for a long time, but the
problem seems to have been exacerbated by commits:

  1e02ce4cccdc ("x86: Store a per-cpu shadow copy of CR4")
  e1bfc11c5a6f ("x86/init: Fix cr4_init_shadow() on CR4-less machines")

This appears to have broken for as long as we've had early
exception handling.

[ This backport should apply to kernels from 3.4 - 4.5. ]

Fixes: 4c5023a3fa2e ("x86-32: Handle exception table entries during early boot")
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: stable@vger.kernel.org
Reported-by: Matthew Whitehead <tedheadster@gmail.com>
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agox86/xen: fix upper bound of pmd loop in xen_cleanhighmap()
Juergen Gross [Thu, 23 Jun 2016 05:12:27 +0000 (07:12 +0200)]
x86/xen: fix upper bound of pmd loop in xen_cleanhighmap()

commit 1cf38741308c64d08553602b3374fb39224eeb5a upstream.

xen_cleanhighmap() is operating on level2_kernel_pgt only. The upper
bound of the loop setting non-kernel-image entries to zero should not
exceed the size of level2_kernel_pgt.

Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoxen-pciback: Add name prefix to global 'permissive' variable
Ben Hutchings [Sun, 12 Apr 2015 23:26:35 +0000 (00:26 +0100)]
xen-pciback: Add name prefix to global 'permissive' variable

commit 8014bcc86ef112eab9ee1db312dba4e6b608cf89 upstream.

The variable for the 'permissive' module parameter used to be static
but was recently changed to be extern.  This puts it in the kernel
global namespace if the driver is built-in, so its name should begin
with a prefix identifying the driver.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Fixes: af6fc858a35b ("xen-pciback: limit guest control of command register")
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoxen/pciback: Don't allow MSI-X ops if PCI_COMMAND_MEMORY is not set.
Konrad Rzeszutek Wilk [Mon, 2 Nov 2015 23:13:27 +0000 (18:13 -0500)]
xen/pciback: Don't allow MSI-X ops if PCI_COMMAND_MEMORY is not set.

commit 408fb0e5aa7fda0059db282ff58c3b2a4278baa0 upstream.

commit f598282f51 ("PCI: Fix the NIU MSI-X problem in a better way")
teaches us that dealing with MSI-X can be troublesome.

Further checks in the MSI-X architecture shows that if the
PCI_COMMAND_MEMORY bit is turned of in the PCI_COMMAND we
may not be able to access the BAR (since they are memory regions).

Since the MSI-X tables are located in there.. that can lead
to us causing PCIe errors. Inhibit us performing any
operation on the MSI-X unless the MEMORY bit is set.

Note that Xen hypervisor with:
"x86/MSI-X: access MSI-X table only after having enabled MSI-X"
will return:
xen_pciback: 0000:0a:00.1: error -6 enabling MSI-X for guest 3!

When the generic MSI code tries to setup the PIRQ without
MEMORY bit set. Which means with later versions of Xen
(4.6) this patch is not neccessary.

This is part of XSA-157

Reviewed-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoxen/pciback: For XEN_PCI_OP_disable_msi[|x] only disable if device has MSI(X) enabled.
Konrad Rzeszutek Wilk [Wed, 1 Apr 2015 14:49:47 +0000 (10:49 -0400)]
xen/pciback: For XEN_PCI_OP_disable_msi[|x] only disable if device has MSI(X) enabled.

commit 7cfb905b9638982862f0331b36ccaaca5d383b49 upstream.

Otherwise just continue on, returning the same values as
previously (return of 0, and op->result has the PIRQ value).

This does not change the behavior of XEN_PCI_OP_disable_msi[|x].

The pci_disable_msi or pci_disable_msix have the checks for
msi_enabled or msix_enabled so they will error out immediately.

However the guest can still call these operations and cause
us to disable the 'ack_intr'. That means the backend IRQ handler
for the legacy interrupt will not respond to interrupts anymore.

This will lead to (if the device is causing an interrupt storm)
for the Linux generic code to disable the interrupt line.

Naturally this will only happen if the device in question
is plugged in on the motherboard on shared level interrupt GSI.

This is part of XSA-157

Reviewed-by: David Vrabel <david.vrabel@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoxen/pciback: Do not install an IRQ handler for MSI interrupts.
Konrad Rzeszutek Wilk [Mon, 2 Nov 2015 22:24:08 +0000 (17:24 -0500)]
xen/pciback: Do not install an IRQ handler for MSI interrupts.

commit a396f3a210c3a61e94d6b87ec05a75d0be2a60d0 upstream.

Otherwise an guest can subvert the generic MSI code to trigger
an BUG_ON condition during MSI interrupt freeing:

 for (i = 0; i < entry->nvec_used; i++)
        BUG_ON(irq_has_action(entry->irq + i));

Xen PCI backed installs an IRQ handler (request_irq) for
the dev->irq whenever the guest writes PCI_COMMAND_MEMORY
(or PCI_COMMAND_IO) to the PCI_COMMAND register. This is
done in case the device has legacy interrupts the GSI line
is shared by the backend devices.

To subvert the backend the guest needs to make the backend
to change the dev->irq from the GSI to the MSI interrupt line,
make the backend allocate an interrupt handler, and then command
the backend to free the MSI interrupt and hit the BUG_ON.

Since the backend only calls 'request_irq' when the guest
writes to the PCI_COMMAND register the guest needs to call
XEN_PCI_OP_enable_msi before any other operation. This will
cause the generic MSI code to setup an MSI entry and
populate dev->irq with the new PIRQ value.

Then the guest can write to PCI_COMMAND PCI_COMMAND_MEMORY
and cause the backend to setup an IRQ handler for dev->irq
(which instead of the GSI value has the MSI pirq). See
'xen_pcibk_control_isr'.

Then the guest disables the MSI: XEN_PCI_OP_disable_msi
which ends up triggering the BUG_ON condition in 'free_msi_irqs'
as there is an IRQ handler for the entry->irq (dev->irq).

Note that this cannot be done using MSI-X as the generic
code does not over-write dev->irq with the MSI-X PIRQ values.

The patch inhibits setting up the IRQ handler if MSI or
MSI-X (for symmetry reasons) code had been called successfully.

P.S.
Xen PCIBack when it sets up the device for the guest consumption
ends up writting 0 to the PCI_COMMAND (see xen_pcibk_reset_device).
XSA-120 addendum patch removed that - however when upstreaming said
addendum we found that it caused issues with qemu upstream. That
has now been fixed in qemu upstream.

This is part of XSA-157

Reviewed-by: David Vrabel <david.vrabel@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoxen/pciback: Return error on XEN_PCI_OP_enable_msix when device has MSI or MSI-X...
Konrad Rzeszutek Wilk [Mon, 2 Nov 2015 23:07:44 +0000 (18:07 -0500)]
xen/pciback: Return error on XEN_PCI_OP_enable_msix when device has MSI or MSI-X enabled

commit 5e0ce1455c09dd61d029b8ad45d82e1ac0b6c4c9 upstream.

The guest sequence of:

  a) XEN_PCI_OP_enable_msix
  b) XEN_PCI_OP_enable_msix

results in hitting an NULL pointer due to using freed pointers.

The device passed in the guest MUST have MSI-X capability.

The a) constructs and SysFS representation of MSI and MSI groups.
The b) adds a second set of them but adding in to SysFS fails (duplicate entry).
'populate_msi_sysfs' frees the newly allocated msi_irq_groups (note that
in a) pdev->msi_irq_groups is still set) and also free's ALL of the
MSI-X entries of the device (the ones allocated in step a) and b)).

The unwind code: 'free_msi_irqs' deletes all the entries and tries to
delete the pdev->msi_irq_groups (which hasn't been set to NULL).
However the pointers in the SysFS are already freed and we hit an
NULL pointer further on when 'strlen' is attempted on a freed pointer.

The patch adds a simple check in the XEN_PCI_OP_enable_msix to guard
against that. The check for msi_enabled is not stricly neccessary.

This is part of XSA-157

Reviewed-by: David Vrabel <david.vrabel@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoxen/pciback: Return error on XEN_PCI_OP_enable_msi when device has MSI or MSI-X enabled
Konrad Rzeszutek Wilk [Fri, 3 Apr 2015 15:08:22 +0000 (11:08 -0400)]
xen/pciback: Return error on XEN_PCI_OP_enable_msi when device has MSI or MSI-X enabled

commit 56441f3c8e5bd45aab10dd9f8c505dd4bec03b0d upstream.

The guest sequence of:

 a) XEN_PCI_OP_enable_msi
 b) XEN_PCI_OP_enable_msi
 c) XEN_PCI_OP_disable_msi

results in hitting an BUG_ON condition in the msi.c code.

The MSI code uses an dev->msi_list to which it adds MSI entries.
Under the above conditions an BUG_ON() can be hit. The device
passed in the guest MUST have MSI capability.

The a) adds the entry to the dev->msi_list and sets msi_enabled.
The b) adds a second entry but adding in to SysFS fails (duplicate entry)
and deletes all of the entries from msi_list and returns (with msi_enabled
is still set).  c) pci_disable_msi passes the msi_enabled checks and hits:

BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));

and blows up.

The patch adds a simple check in the XEN_PCI_OP_enable_msi to guard
against that. The check for msix_enabled is not stricly neccessary.

This is part of XSA-157.

Reviewed-by: David Vrabel <david.vrabel@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoxen/pciback: Save the number of MSI-X entries to be copied later.
Konrad Rzeszutek Wilk [Thu, 11 Feb 2016 21:10:24 +0000 (16:10 -0500)]
xen/pciback: Save the number of MSI-X entries to be copied later.

commit d159457b84395927b5a52adb72f748dd089ad5e5 upstream.

Commit 8135cf8b092723dbfcc611fe6fdcb3a36c9951c5 (xen/pciback: Save
xen_pci_op commands before processing it) broke enabling MSI-X because
it would never copy the resulting vectors into the response.  The
number of vectors requested was being overwritten by the return value
(typically zero for success).

Save the number of vectors before processing the op, so the correct
number of vectors are copied afterwards.

Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: <stable@vger.kernel.org>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoxen/pciback: Save xen_pci_op commands before processing it
Konrad Rzeszutek Wilk [Mon, 16 Nov 2015 17:40:48 +0000 (12:40 -0500)]
xen/pciback: Save xen_pci_op commands before processing it

commit 8135cf8b092723dbfcc611fe6fdcb3a36c9951c5 upstream.

Double fetch vulnerabilities that happen when a variable is
fetched twice from shared memory but a security check is only
performed the first time.

The xen_pcibk_do_op function performs a switch statements on the op->cmd
value which is stored in shared memory. Interestingly this can result
in a double fetch vulnerability depending on the performed compiler
optimization.

This patch fixes it by saving the xen_pci_op command before
processing it. We also use 'barrier' to make sure that the
compiler does not perform any optimization.

This is part of XSA155.

Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Jan Beulich <JBeulich@suse.com>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: "Jan Beulich" <JBeulich@suse.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoxen-blkback: only read request operation from shared ring once
Roger Pau Monné [Tue, 3 Nov 2015 16:34:09 +0000 (16:34 +0000)]
xen-blkback: only read request operation from shared ring once

commit 1f13d75ccb806260079e0679d55d9253e370ec8a upstream.

A compiler may load a switch statement value multiple times, which could
be bad when the value is in memory shared with the frontend.

When converting a non-native request to a native one, ensure that
src->operation is only loaded once by using READ_ONCE().

This is part of XSA155.

Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: "Jan Beulich" <JBeulich@suse.com>
[wt: s/READ_ONCE/ACCESS_ONCE for 3.10]
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoxen-netback: use RING_COPY_REQUEST() throughout
David Vrabel [Fri, 30 Oct 2015 15:17:06 +0000 (15:17 +0000)]
xen-netback: use RING_COPY_REQUEST() throughout

commit 68a33bfd8403e4e22847165d149823a2e0e67c9c upstream.

Instead of open-coding memcpy()s and directly accessing Tx and Rx
requests, use the new RING_COPY_REQUEST() that ensures the local copy
is correct.

This is more than is strictly necessary for guest Rx requests since
only the id and gref fields are used and it is harmless if the
frontend modifies these.

This is part of XSA155.

Reviewed-by: Wei Liu <wei.liu2@citrix.com>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
[wt: adjustments for 3.10 : netbk_rx_meta instead of struct xenvif_rx_meta]

Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoxen-netback: don't use last request to determine minimum Tx credit
David Vrabel [Fri, 30 Oct 2015 15:16:01 +0000 (15:16 +0000)]
xen-netback: don't use last request to determine minimum Tx credit

commit 0f589967a73f1f30ab4ac4dd9ce0bb399b4d6357 upstream.

The last from guest transmitted request gives no indication about the
minimum amount of credit that the guest might need to send a packet
since the last packet might have been a small one.

Instead allow for the worst case 128 KiB packet.

This is part of XSA155.

Reviewed-by: Wei Liu <wei.liu2@citrix.com>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoxen: Add RING_COPY_REQUEST()
David Vrabel [Fri, 30 Oct 2015 14:58:08 +0000 (14:58 +0000)]
xen: Add RING_COPY_REQUEST()

commit 454d5d882c7e412b840e3c99010fe81a9862f6fb upstream.

Using RING_GET_REQUEST() on a shared ring is easy to use incorrectly
(i.e., by not considering that the other end may alter the data in the
shared ring while it is being inspected).  Safe usage of a request
generally requires taking a local copy.

Provide a RING_COPY_REQUEST() macro to use instead of
RING_GET_REQUEST() and an open-coded memcpy().  This takes care of
ensuring that the copy is done correctly regardless of any possible
compiler optimizations.

Use a volatile source to prevent the compiler from reordering or
omitting the copy.

This is part of XSA155.

Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agox86/mm/xen: Suppress hugetlbfs in PV guests
Jan Beulich [Thu, 21 Apr 2016 06:27:04 +0000 (00:27 -0600)]
x86/mm/xen: Suppress hugetlbfs in PV guests

commit 103f6112f253017d7062cd74d17f4a514ed4485c upstream.

Huge pages are not normally available to PV guests. Not suppressing
hugetlbfs use results in an endless loop of page faults when user mode
code tries to access a hugetlbfs mapped area (since the hypervisor
denies such PTEs to be created, but error indications can't be
propagated out of xen_set_pte_at(), just like for various of its
siblings), and - once killed in an oops like this:

  kernel BUG at .../fs/hugetlbfs/inode.c:428!
  invalid opcode: 0000 [#1] SMP
  ...
  RIP: e030:[<ffffffff811c333b>]  [<ffffffff811c333b>] remove_inode_hugepages+0x25b/0x320
  ...
  Call Trace:
   [<ffffffff811c3415>] hugetlbfs_evict_inode+0x15/0x40
   [<ffffffff81167b3d>] evict+0xbd/0x1b0
   [<ffffffff8116514a>] __dentry_kill+0x19a/0x1f0
   [<ffffffff81165b0e>] dput+0x1fe/0x220
   [<ffffffff81150535>] __fput+0x155/0x200
   [<ffffffff81079fc0>] task_work_run+0x60/0xa0
   [<ffffffff81063510>] do_exit+0x160/0x400
   [<ffffffff810637eb>] do_group_exit+0x3b/0xa0
   [<ffffffff8106e8bd>] get_signal+0x1ed/0x470
   [<ffffffff8100f854>] do_signal+0x14/0x110
   [<ffffffff810030e9>] prepare_exit_to_usermode+0xe9/0xf0
   [<ffffffff814178a5>] retint_user+0x8/0x13

This is CVE-2016-3961 / XSA-174.

Reported-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Juergen Gross <JGross@suse.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Luis R. Rodriguez <mcgrof@suse.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Toshi Kani <toshi.kani@hp.com>
Cc: xen-devel <xen-devel@lists.xenproject.org>
Link: http://lkml.kernel.org/r/57188ED802000078000E431C@prv-mh.provo.novell.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoppp: defer netns reference release for ppp channel
WANG Cong [Wed, 6 Jul 2016 05:12:36 +0000 (22:12 -0700)]
ppp: defer netns reference release for ppp channel

commit 205e1e255c479f3fd77446415706463b282f94e4 upstream

Matt reported that we have a NULL pointer dereference
in ppp_pernet() from ppp_connect_channel(),
i.e. pch->chan_net is NULL.

This is due to that a parallel ppp_unregister_channel()
could happen while we are in ppp_connect_channel(), during
which pch->chan_net set to NULL. Since we need a reference
to net per channel, it makes sense to sync the refcnt
with the life time of the channel, therefore we should
release this reference when we destroy it.

Fixes: 1f461dcdd296 ("ppp: take reference on channels netns")
Reported-by: Matt Bennett <Matt.Bennett@alliedtelesis.co.nz>
Cc: Paul Mackerras <paulus@samba.org>
Cc: linux-ppp@vger.kernel.org
Cc: Guillaume Nault <g.nault@alphalink.fr>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoPM / devfreq: Fix incorrect type issue.
Xiaolong Ye [Fri, 11 Sep 2015 03:05:23 +0000 (11:05 +0800)]
PM / devfreq: Fix incorrect type issue.

commit 5f25f066f75a67835abb5e400471a27abd09395b upstream

time_in_state in struct devfreq is defined as unsigned long, so
devm_kzalloc should use sizeof(unsigned long) as argument instead
of sizeof(unsigned int), otherwise it will cause unexpected result
in 64bit system.

Signed-off-by: Xiaolong Ye <yexl@marvell.com>
Signed-off-by: Kevin Liu <kliu5@marvell.com>
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoKVM: Disable irq while unregistering user notifier
Ignacio Alvarado [Fri, 4 Nov 2016 19:15:55 +0000 (12:15 -0700)]
KVM: Disable irq while unregistering user notifier

commit 1650b4ebc99da4c137bfbfc531be4a2405f951dd upstream.

Function user_notifier_unregister should be called only once for each
registered user notifier.

Function kvm_arch_hardware_disable can be executed from an IPI context
which could cause a race condition with a VCPU returning to user mode
and attempting to unregister the notifier.

Signed-off-by: Ignacio Alvarado <ikalvarado@google.com>
Fixes: 18863bdd60f8 ("KVM: x86 shared msr infrastructure")
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Radim KrÄ\8dmáÅ\99 <rkrcmar@redhat.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoKVM: x86: fix missed SRCU usage in kvm_lapic_set_vapic_addr
Paolo Bonzini [Thu, 17 Nov 2016 14:55:46 +0000 (15:55 +0100)]
KVM: x86: fix missed SRCU usage in kvm_lapic_set_vapic_addr

commit 7301d6abaea926d685832f7e1f0c37dd206b01f4 upstream.

Reported by syzkaller:

    [ INFO: suspicious RCU usage. ]
    4.9.0-rc4+ #47 Not tainted
    -------------------------------
    ./include/linux/kvm_host.h:536 suspicious rcu_dereference_check() usage!

    stack backtrace:
    CPU: 1 PID: 6679 Comm: syz-executor Not tainted 4.9.0-rc4+ #47
    Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011
     ffff880039e2f6d0 ffffffff81c2e46b ffff88003e3a5b40 0000000000000000
     0000000000000001 ffffffff83215600 ffff880039e2f700 ffffffff81334ea9
     ffffc9000730b000 0000000000000004 ffff88003c4f8420 ffff88003d3f8000
    Call Trace:
     [<     inline     >] __dump_stack lib/dump_stack.c:15
     [<ffffffff81c2e46b>] dump_stack+0xb3/0x118 lib/dump_stack.c:51
     [<ffffffff81334ea9>] lockdep_rcu_suspicious+0x139/0x180 kernel/locking/lockdep.c:4445
     [<     inline     >] __kvm_memslots include/linux/kvm_host.h:534
     [<     inline     >] kvm_memslots include/linux/kvm_host.h:541
     [<ffffffff8105d6ae>] kvm_gfn_to_hva_cache_init+0xa1e/0xce0 virt/kvm/kvm_main.c:1941
     [<ffffffff8112685d>] kvm_lapic_set_vapic_addr+0xed/0x140 arch/x86/kvm/lapic.c:2217

Reported-by: Dmitry Vyukov <dvyukov@google.com>
Fixes: fda4e2e85589191b123d31cdc21fd33ee70f50fd
Cc: Andrew Honig <ahonig@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Radim KrÄ\8dmáÅ\99 <rkrcmar@redhat.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoKVM: x86: fix wbinvd_dirty_mask use-after-free
Ido Yariv [Fri, 21 Oct 2016 16:39:57 +0000 (12:39 -0400)]
KVM: x86: fix wbinvd_dirty_mask use-after-free

commit bd768e146624cbec7122ed15dead8daa137d909d upstream.

vcpu->arch.wbinvd_dirty_mask may still be used after freeing it,
corrupting memory. For example, the following call trace may set a bit
in an already freed cpu mask:
    kvm_arch_vcpu_load
    vcpu_load
    vmx_free_vcpu_nested
    vmx_free_vcpu
    kvm_arch_vcpu_free

Fix this by deferring freeing of wbinvd_dirty_mask.

Signed-off-by: Ido Yariv <ido@wizery.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Radim KrÄ\8dmáÅ\99 <rkrcmar@redhat.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoKVM: MIPS: Make ERET handle ERL before EXL
James Hogan [Tue, 25 Oct 2016 15:11:11 +0000 (16:11 +0100)]
KVM: MIPS: Make ERET handle ERL before EXL

commit ede5f3e7b54a4347be4d8525269eae50902bd7cd upstream.

The ERET instruction to return from exception is used for returning from
exception level (Status.EXL) and error level (Status.ERL). If both bits
are set however we should be returning from ERL first, as ERL can
interrupt EXL, for example when an NMI is taken. KVM however checks EXL
first.

Fix the order of the checks to match the pseudocode in the instruction
set manual.

Fixes: e685c689f3a8 ("KVM/MIPS32: Privileged instruction/target branch emulation.")
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim KrÄ\8dmáÅ\99" <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoKVM: nVMX: postpone VMCS changes on MSR_IA32_APICBASE write
Radim KrÄ\8dmáÅ\99 [Mon, 8 Aug 2016 18:16:23 +0000 (20:16 +0200)]
KVM: nVMX: postpone VMCS changes on MSR_IA32_APICBASE write

commit dccbfcf52cebb8963246eba5b177b77f26b34da0 upstream.

If vmcs12 does not intercept APIC_BASE writes, then KVM will handle the
write with vmcs02 as the current VMCS.
This will incorrectly apply modifications intended for vmcs01 to vmcs02
and L2 can use it to gain access to L0's x2APIC registers by disabling
virtualized x2APIC while using msr bitmap that assumes enabled.

Postpone execution of vmx_set_virtual_x2apic_mode until vmcs01 is the
current VMCS.  An alternative solution would temporarily make vmcs01 the
current VMCS, but it requires more care.

Fixes: 8d14695f9542 ("x86, apicv: add virtual x2apic support")
Reported-by: Jim Mattson <jmattson@google.com>
Reviewed-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Radim KrÄ\8dmáÅ\99 <rkrcmar@redhat.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoKVM: MIPS: Drop other CPU ASIDs on guest MMU changes
James Hogan [Wed, 9 Nov 2016 14:46:24 +0000 (14:46 +0000)]
KVM: MIPS: Drop other CPU ASIDs on guest MMU changes

commit 91e4f1b6073dd680d86cdb7e42d7cccca9db39d8 upstream.

When a guest TLB entry is replaced by TLBWI or TLBWR, we only invalidate
TLB entries on the local CPU. This doesn't work correctly on an SMP host
when the guest is migrated to a different physical CPU, as it could pick
up stale TLB mappings from the last time the vCPU ran on that physical
CPU.

Therefore invalidate both user and kernel host ASIDs on other CPUs,
which will cause new ASIDs to be generated when it next runs on those
CPUs.

We're careful only to do this if the TLB entry was already valid, and
only for the kernel ASID where the virtual address it mapped is outside
of the guest user address range.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim KrÄ\8dmáÅ\99" <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Cc: <stable@vger.kernel.org> # 3.10.x-
Cc: Jiri Slaby <jslaby@suse.cz>
[james.hogan@imgtec.com: Backport to 3.10..3.16]
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
7 years agoKVM: MIPS: Precalculate MMIO load resume PC
James Hogan [Wed, 9 Nov 2016 16:13:50 +0000 (16:13 +0000)]
KVM: MIPS: Precalculate MMIO load resume PC

commit e1e575f6b026734be3b1f075e780e91ab08ca541 upstream.

The advancing of the PC when completing an MMIO load is done before
re-entering the guest, i.e. before restoring the guest ASID. However if
the load is in a branch delay slot it may need to access guest code to
read the prior branch instruction. This isn't safe in TLB mapped code at
the moment, nor in the future when we'll access unmapped guest segments
using direct user accessors too, as it could read the branch from host
user memory instead.

Therefore calculate the resume PC in advance while we're still in the
right context and save it in the new vcpu->arch.io_pc (replacing the no
longer needed vcpu->arch.pending_load_cause), and restore it on MMIO
completion.

Fixes: e685c689f3a8 ("KVM/MIPS32: Privileged instruction/target branch emulation.")
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim KrÄ\8dmáÅ\99 <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Cc: <stable@vger.kernel.org> # 3.10.x-3.16.x: 5f508c43a764: MIPS: KVM: Fix unused variable build warning
Cc: <stable@vger.kernel.org> # 3.10.x-3.16.x
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
[james.hogan@imgtec.com: Backport to 3.10..3.16]
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>