The majority of SMP platforms handle their IPIs through do_IRQ()
which calls irq_{enter/exit}(). When a call function IPI is received,
smp_call_function_interrupt() is called which also calls
irq_{enter,exit}(), meaning irq_count is raised twice.

When tick broadcasting is used (which is implemented via a call
function IPI), this incorrectly causes all CPU idle time on the core
receiving broadcast ticks to be accounted as time spent servicing
IRQs, as account_process_tick() will account as such if irq_count is
greater than 1. This results in 100% CPU usage being reported on a
core which receives its ticks via broadcast.

This patch removes the SMP smp_call_function_interrupt() wrapper which
calls irq_{enter,exit}(). Platforms which handle their IPIs through
do_IRQ() now call generic_smp_call_function_interrupt() directly to
avoid incrementing irq_count a second time. Platforms which don't
(loongson, sgi-ip27, sibyte) call generic_smp_call_function_interrupt()
wrapped in irq_{enter,exit}().
Signed-off-by: NAlex Smith <alex.smith@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/10770/Signed-off-by: NRalf Baechle <ralf@linux-mips.org>

71eeedcf (MIPS: Lemote 2F: Fix build caused
by recent mass rename.) only fixed one instance of this issue in arch/mips
but missed a 2nd one in drivers/cpufreq/loongson2_cpufreq.c.

[ralf@linux-mips.org: dropped the one segment for the already fixed
instance and changed the other avoiding an include <path.h> without a /
because that's generally is a bad idea.]
Signed-off-by: NRalf Baechle <ralf@linux-mips.org>
Acked-by: NViresh Kumar <viresh.kumar@linaro.org>
Patchwork: https://patchwork.linux-mips.org/patch/10659/

Ted Ts'o reports that his Lenovo T540p ThinkPad crashes at boot if
attached to the docking station.  This is a regression that he was able
to bisect to commit 8c7b5ccb: "drm/i915: Use atomic helpers for
computing changed flags:"

The reason seems to be the new call to drm_atomic_helper_check_modeset()
added to intel_modeset_compute_config(), which in turn calls
update_connector_routing(), and somehow ends up picking a NULL crtc for
the connector state, causing the subsequent drm_crtc_index() to OOPS.

Daniel Vetter says that the fundamental issue seems to be confusion in
the encoder selection, and this isn't the right fix, but while he chases
down the proper fix, this at least avoids the NULL pointer dereference
and makes Ted's docking station work again.
Reported-bisected-and-tested-by: NTheodore Ts'o <tytso@mit.edu>
Cc: Daniel Vetter <daniel.vetter@intel.com>
Cc: Mani Nikula <jani.nikula@linux.intel.com>
Cc: Dave Airlie <airlied@gmail.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Commit 50649ab1 ("stmmac: drop driver from stmmac platform code")
was a bit overzealous in removing code and dropped the MODULE_*
macro's that are still needed since stmmac_platform can be a module.
Fix this by putting the macro's remvoed in 50649ab1 back.

This fixes the following errors when used as a module:
  stmmac_platform: module license 'unspecified' taints kernel.
  Disabling lock debugging due to kernel taint
  stmmac_platform: Unknown symbol devm_kmalloc (err 0)
  stmmac_platform: Unknown symbol stmmac_suspend (err 0)
  stmmac_platform: Unknown symbol platform_get_irq_byname (err 0)
  stmmac_platform: Unknown symbol stmmac_dvr_remove (err 0)
  stmmac_platform: Unknown symbol platform_get_resource (err 0)
  stmmac_platform: Unknown symbol of_get_phy_mode (err 0)
  stmmac_platform: Unknown symbol of_property_read_u32_array (err 0)
  stmmac_platform: Unknown symbol of_alias_get_id (err 0)
  stmmac_platform: Unknown symbol stmmac_resume (err 0)
  stmmac_platform: Unknown symbol stmmac_dvr_probe (err 0)

Fixes: 50649ab1 ("stmmac: drop driver from stmmac platform code")
Reported-by: NIgor Gnatenko <i.gnatenko.brain@gmail.com>
Signed-off-by: NJoachim Eastwood <manabian@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The wol_en flag is 0 by default anyway, and we have the
following inconsistency: a MAGIC packet wol capable eth
interface is registered as a wake-up source but unable
to wake-up the system as wol_en is 0 (wake-on flag set to 'd').
Calling set_wakeup_enable() at netdev open is just redundant
because wol_en is 0 by default.
Let only ethtool call set_wakeup_enable() for now.

The bflock is obviously obsoleted, its utility has been corroded
over time.  The bitfield flags used today in gianfar are accessed
only on the init/ config path, with no real possibility of
concurrency - nothing that would justify smth. like bflock.
Signed-off-by: NClaudiu Manoil <claudiu.manoil@freescale.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If we disable NAPI in the first place we can mask the device's
interrupts (and halt it) without fearing that imask may be
concurrently accessed from interrupt context, so there's
no need to do local_irq_save() around gfar_halt_nodisable().
lock_rx_qs()/unlock_tx_qs() are just obsoleted and potentially
buggy routines.  The txlock is currently used in the driver only
to manage TX congestion, it has nothing to do with halting the
device.  With these changes, the TX processing is stopped before
gfar_halt().

Compact gfar_halt() is used instead of gfar_halt_nodisable(),
as it disables Rx/TX DMA h/w blocks and the Rx/TX h/w queues.
gfar_start() re-enables all these blocks on resume.  Enabling
the magic-packet mode remains the same, note that the RX block
is re-enabled just before entering sleep mode.

Add IRQF_NO_SUSPEND flag for the error interrupt line, to signal
that the interrupt line must remain active during sleep in order
to wake the system by magic packet (MAG) reception interrupt.
(On some systems the MAG interrupt did trigger w/o this flag
as well, but on others it didn't.)

Without these fixes, when suspended during fair Tx traffic the
interface occasionally failed to be woken up by magic packet.
Signed-off-by: NClaudiu Manoil <claudiu.manoil@freescale.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

CC      drivers/net/ethernet/freescale/gianfar.o
drivers/net/ethernet/freescale/gianfar.c:568:13: warning: 'lock_tx_qs'
defined but not used [-Wunused-function]
 static void lock_tx_qs(struct gfar_private *priv)
             ^
drivers/net/ethernet/freescale/gianfar.c:576:13: warning: 'unlock_tx_qs'
defined but not used [-Wunused-function]
 static void unlock_tx_qs(struct gfar_private *priv)
             ^
Reported-by: NScott Wood <scottwood@freescale.com>
Signed-off-by: NClaudiu Manoil <claudiu.manoil@freescale.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This reverts commit b9855f03.
The patch break existing DMA usage case. For example, audio SOC
dmaengine never release channel and cause virt-dma to cache too
much memory in descriptor to exhaust system memory.
Signed-off-by: NVinod Koul <vinod.koul@intel.com>

Commit 6f166312 ("dmaengine: mv_xor: add support for a38x command
in descriptor mode") introduced the support for a feature that
appeared in Armada 38x: specifying the operation to be performed in a
per-descriptor basis rather than globally per channel.

However, when doing so, it changed the function mv_chan_set_mode() to
use:

  if (IS_ENABLED(__BIG_ENDIAN))

instead of:

  #if defined(__BIG_ENDIAN)

While IS_ENABLED() is perfectly fine for CONFIG_* symbols, it is not
for other symbols such as __BIG_ENDIAN that is provided directly by
the compiler. Consequently, the commit broke support for big-endian,
as the XOR_DESCRIPTOR_SWAP flag was not set in the XOR channel
configuration register.

The primarily visible effect was some nasty warnings and failures
appearing during the self-test of the XOR unit:

[    1.197368] mv_xor d0060900.xor: error on chan 0. intr cause 0x00000082
[    1.197393] mv_xor d0060900.xor: config       0x00008440
[    1.197410] mv_xor d0060900.xor: activation   0x00000000
[    1.197427] mv_xor d0060900.xor: intr cause   0x00000082
[    1.197443] mv_xor d0060900.xor: intr mask    0x000003f7
[    1.197460] mv_xor d0060900.xor: error cause  0x00000000
[    1.197477] mv_xor d0060900.xor: error addr   0x00000000
[    1.197491] ------------[ cut here ]------------
[    1.197513] WARNING: CPU: 0 PID: 1 at ../drivers/dma/mv_xor.c:664 mv_xor_interrupt_handler+0x14c/0x170()

See also:

  http://storage.kernelci.org/next/next-20150617/arm-mvebu_v7_defconfig+CONFIG_CPU_BIG_ENDIAN=y/lab-khilman/boot-armada-xp-openblocks-ax3-4.txtSigned-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Fixes: 6f166312 ("dmaengine: mv_xor: add support for a38x command in descriptor mode")
Reviewed-by: NMaxime Ripard <maxime.ripard@free-electrons.com>
Signed-off-by: NVinod Koul <vinod.koul@intel.com>

There is an overlap in dma ring cmd csr region due to sharing of ethernet
ring cmd csr region. This patch fix the resource overlapping by mapping
the entire dma ring cmd csr region.
Signed-off-by: NRameshwar Prasad Sahu <rsahu@apm.com>
Signed-off-by: NVinod Koul <vinod.koul@intel.com>

This patch adds the missing update of the transfer data width in
at_xdmac_prep_slave_sg().

Indeed, for each item in the scatter-gather list, we check whether the
transfer length is aligned with the data width provided by
dmaengine_slave_config(). If so, we directly use this data width for the
current part of the transfer we are preparing. Otherwise, the data width
is reduced to 8 bits (1 byte). Of course, the actual number of register
accesses must also be updated to match the new data width.

So one chunk was missing in the original patch (see Fixes tag below): the
number of register accesses was correctly set to (len >> fixed_dwidth) in
mbr_ubc but the real data width was not updated in mbr_cfg. Since mbr_cfg
may change for each part of the scatter-gather transfer this also explains
why the original patch used the Descriptor View 2 instead of the
Descriptor View 1.

Let's take the example of a DMA transfer to write 8bit data into an Atmel
USART with FIFOs. When FIFOs are enabled in the USART, its Transmit
Holding Register (THR) works in multidata mode, that is to say that up to
4 8bit data can be written into the THR in a single 32bit access and it is
still possible to write only one data with a 8bit access. To take
advantage of this new feature, the DMA driver was modified to allow
multiple dwidths when doing slave transfers.
For instance, when the total length is 22 bytes, the USART driver splits
the transfer into 2 parts:

First part: 20 bytes transferred through 5 32bit writes into THR
Second part: 2 bytes transferred though 2 8bit writes into THR

For the second part, the data width was first set to 4_BYTES by the USART
driver thanks to dmaengine_slave_config() then at_xdmac_prep_slave_sg()
reduces this data width to 1_BYTE because the 2 byte length is not aligned
with the original 4_BYTES data width. Since the data width is modified,
the actual number of writes into THR must be set accordingly.
Signed-off-by: NCyrille Pitchen <cyrille.pitchen@atmel.com>
Fixes: 6d3a7d9e ("dmaengine: at_xdmac: allow muliple dwidths when doing slave transfers")
Cc: stable@vger.kernel.org #4.0 and later
Acked-by: NNicolas Ferre <nicolas.ferre@atmel.com>
Acked-by: NLudovic Desroches <ludovic.desroches@atmel.com>
Signed-off-by: NVinod Koul <vinod.koul@intel.com>

As claimed by the programmer datasheet and confirmed by the IP designer,
the Block Transfer Size (BTSIZE) bitfield of the Channel x Control A
Register (CTRLAx) always refers to a number of Source Width (SRC_WIDTH)
transfers.

Both the SRC_WIDTH and BTSIZE bitfields can be extacted from the CTRLAx
register to compute the DMA residue. So the 'tx_width' field is useless
and can be removed from the struct at_desc.

Before this patch, atc_prep_slave_sg() was not consistent: BTSIZE was
correctly initialized according to the SRC_WIDTH but 'tx_width' was always
set to reg_width, which was incorrect for MEM_TO_DEV transfers. It led to
bad DMA residue when 'tx_width' != SRC_WIDTH.

Also the 'tx_width' field was mostly set only in the first and last
descriptors. Depending on the kind of DMA transfer, this field remained
uninitialized for intermediate descriptors. The accurate DMA residue was
computed only when the currently processed descriptor was the first or the
last of the chain. This algorithm was a little bit odd. An accurate DMA
residue can always be computed using the SRC_WIDTH and BTSIZE bitfields
in the CTRLAx register.

Finally, the test to check whether the currently processed descriptor is
the last of the chain was wrong: for cyclic transfer, last_desc->lli.dscr
is NOT equal to zero, since set_desc_eol() is never called, but logically
equal to first_desc->txd.phys. This bug has a side effect on the
drivers/tty/serial/atmel_serial.c driver, which uses cyclic DMA transfer
to receive data. Since the DMA residue was wrong each time the DMA
transfer reaches the second (and last) period of the transfer, no more
data were received by the USART driver till the cyclic DMA transfer loops
back to the first period.
Signed-off-by: NCyrille Pitchen <cyrille.pitchen@atmel.com>
Acked-by: NTorsten Fleischer <torfl6749@gmail.com>
Tested-by: NJirí Prchal <jiri.prchal@aksignal.cz>
Acked-by: NNicolas Ferre <nicolas.ferre@atmel.com>
Signed-off-by: NVinod Koul <vinod.koul@intel.com>

When using descriptor view 2 or higher, we don't write the configuration
into AT_XDMAC_CC register because this configuration will be fetch from
the descriptor. Unfortunately, the PROT bit is not updated with this
method, we have to do it manually before enabling the channel.
Signed-off-by: NLudovic Desroches <ludovic.desroches@atmel.com>
Signed-off-by: NVinod Koul <vinod.koul@intel.com>

With the grouping of multi-function devices a non-ATS
capable device might also end up in the same domain as an
IOMMUv2 capable device.
So handle this situation gracefully and don't consider it a
bug anymore.
Tested-by: NOded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: NJoerg Roedel <jroedel@suse.de>

The device reset is necessary if the hw becomes abnormal and stops
transmitting packets.
Signed-off-by: NHayes Wang <hayeswang@realtek.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add rtl8152_pre_reset() and rtl8152_post_reset() which are used when
calling usb_reset_device(). The two functions could reduce the time
of reset when calling usb_reset_device() after probe().
Signed-off-by: NHayes Wang <hayeswang@realtek.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fix a memory leak with scsi-mq triggered by commands with large data
transfer length.

__sg_alloc_table() sets both table->nents and table->orig_nents to the
same value.  When the scatterlist is DMA-mapped, table->nents is
overwritten with the (possibly smaller) size of the DMA-mapped
scatterlist, while table->orig_nents retains the original size of the
allocated scatterlist.  scsi_free_sgtable() should therefore check
orig_nents instead of nents, and all code that initializes sdb->table
without calling __sg_alloc_table() should set both nents and orig_nents.

Fixes: d285203c ("scsi: add support for a blk-mq based I/O path.")
Cc: <stable@vger.kernel.org> # 3.17+
Signed-off-by: NTony Battersby <tonyb@cybernetics.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NEwan D. Milne <emilne@redhat.com>
Signed-off-by: NJames Bottomley <JBottomley@Odin.com>

Fixes another signed / unsigned array indexing bug in the ipr driver.
Currently, when hrrq_index wraps, it becomes a negative number. We
do the modulo, but still have a negative number, so we end up indexing
backwards in the array. Given where the hrrq array is located in memory,
we probably won't actually reference memory we don't own, but nonetheless
ipr is still looking at data within struct ipr_ioa_cfg and interpreting it as
struct ipr_hrr_queue data, so bad things could certainly happen.

Each ipr adapter has anywhere from 1 to 16 HRRQs. By default, we use 2 on new
adapters.  Let's take an example:

Assume ioa_cfg->hrrq_index=0x7fffffffe and ioa_cfg->hrrq_num=4:

The atomic_add_return will then return -1. We mod this with 3 and get -2, add
one and get -1 for an array index.

On adapters which support more than a single HRRQ, we dedicate HRRQ to adapter
initialization and error interrupts so that we can optimize the other queues
for fast path I/O. So all normal I/O uses HRRQ 1-15. So we want to spread the
I/O requests across those HRRQs.

With the default module parameter settings, this bug won't hit, only when
someone sets the ipr.number_of_msix parameter to a value larger than 3 is when
bad things start to happen.

Cc: <stable@vger.kernel.org>
Tested-by: NWen Xiong <wenxiong@linux.vnet.ibm.com>
Reviewed-by: NWen Xiong <wenxiong@linux.vnet.ibm.com>
Reviewed-by: NGabriel Krisman Bertazi <krisman@linux.vnet.ibm.com>
Signed-off-by: NBrian King <brking@linux.vnet.ibm.com>
Reviewed-by: NMartin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: NJames Bottomley <JBottomley@Odin.com>

When ipr's internal driver trace was changed to an atomic, a signed/unsigned
bug slipped in which results in us indexing backwards in our memory buffer
writing on memory that does not belong to us. This patch fixes this by removing
the modulo and instead just mask off the low bits.

Cc: <stable@vger.kernel.org>
Tested-by: NWen Xiong <wenxiong@linux.vnet.ibm.com>
Reviewed-by: NWen Xiong <wenxiong@linux.vnet.ibm.com>
Reviewed-by: NGabriel Krisman Bertazi <krisman@linux.vnet.ibm.com>
Signed-off-by: NBrian King <brking@linux.vnet.ibm.com>
Reviewed-by: NMartin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: NJames Bottomley <JBottomley@Odin.com>

Make sure we have the host lock held when calling scsi_report_bus_reset. Fixes
a crash seen as the __devices list in the scsi host was changing as we were
iterating through it.

Cc: <stable@vger.kernel.org>
Reviewed-by: NWen Xiong <wenxiong@linux.vnet.ibm.com>
Reviewed-by: NGabriel Krisman Bertazi <krisman@linux.vnet.ibm.com>
Signed-off-by: NBrian King <brking@linux.vnet.ibm.com>
Reviewed-by: NMartin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: NJames Bottomley <JBottomley@Odin.com>

Even though it is documented how to specifiy efi parameters, it is
possible to cause a kernel panic due to a dereference of a NULL pointer when
parsing such parameters if "efi" alone is given:

PANIC: early exception 0e rip 10:ffffffff812fb361 error 0 cr2 0
[ 0.000000] CPU: 0 PID: 0 Comm: swapper Not tainted 4.2.0-rc1+ #450
[ 0.000000]  ffffffff81fe20a9 ffffffff81e03d50 ffffffff8184bb0f 00000000000003f8
[ 0.000000]  0000000000000000 ffffffff81e03e08 ffffffff81f371a1 64656c62616e6520
[ 0.000000]  0000000000000069 000000000000005f 0000000000000000 0000000000000000
[ 0.000000] Call Trace:
[ 0.000000]  [<ffffffff8184bb0f>] dump_stack+0x45/0x57
[ 0.000000]  [<ffffffff81f371a1>] early_idt_handler_common+0x81/0xae
[ 0.000000]  [<ffffffff812fb361>] ? parse_option_str+0x11/0x90
[ 0.000000]  [<ffffffff81f4dd69>] arch_parse_efi_cmdline+0x15/0x42
[ 0.000000]  [<ffffffff81f376e1>] do_early_param+0x50/0x8a
[ 0.000000]  [<ffffffff8106b1b3>] parse_args+0x1e3/0x400
[ 0.000000]  [<ffffffff81f37a43>] parse_early_options+0x24/0x28
[ 0.000000]  [<ffffffff81f37691>] ? loglevel+0x31/0x31
[ 0.000000]  [<ffffffff81f37a78>] parse_early_param+0x31/0x3d
[ 0.000000]  [<ffffffff81f3ae98>] setup_arch+0x2de/0xc08
[ 0.000000]  [<ffffffff8109629a>] ? vprintk_default+0x1a/0x20
[ 0.000000]  [<ffffffff81f37b20>] start_kernel+0x90/0x423
[ 0.000000]  [<ffffffff81f37495>] x86_64_start_reservations+0x2a/0x2c
[ 0.000000]  [<ffffffff81f37582>] x86_64_start_kernel+0xeb/0xef
[ 0.000000] RIP 0xffffffff81ba2efc

This panic is not reproducible with "efi=" as this will result in a non-NULL
zero-length string.

Thus, verify that the pointer to the parameter string is not NULL. This is
consistent with other parameter-parsing functions which check for NULL pointers.
Signed-off-by: NRicardo Neri <ricardo.neri-calderon@linux.intel.com>
Cc: Dave Young <dyoung@redhat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: NMatt Fleming <matt.fleming@intel.com>

The old style of memory interleaving swizzled upto the end of the
first even bank of memory, and then used the remainder as unswizzled on
the unpaired bank - i.e. swizzling is not constant for all memory. This
causes problems when we try to migrate memory and so the kernel prevents
migration at all when we detect L-shaped inconsistent swizzling.
However, this issue also extends to userspace who try to manually detile
into memory as the swizzling for an individual page is unknown (it
depends on its physical address only known to the kernel), userspace
cannot correctly swizzle.

Note that this is a new attempt for the previously merged one,
reverted in

commit d82c0ba6
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Tue Jul 14 12:29:27 2015 +0200

    Revert "drm/i915: Declare the swizzling unknown for L-shaped configurations"

This is cc: stable since we need it to fix up troubles with wc cpu
mmaps that userspace recently started to use widely.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91105Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: stable@vger.kernel.org
[danvet: Add note about previous (failed attempt).]
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

If the device does not support the aliasing ppgtt, we must translate
user bind requests (PIN_USER) from LOCAL_BIND to a GLOBAL_BIND. However,
since this is device specific we cannot do this conveniently in the
upper layers and so must manage the vma->bound flags in the backend.

Partial revert of commit 75d04a37 [4.2-rc1]
Author: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Date:   Tue Apr 28 17:56:17 2015 +0300

    drm/i915/gtt: Allocate va range only if vma is not bound

Note this was spotted by Daniel originally, but we dropped the ball in
getting the fix in before the bug going wild. Sorry all.

Reported-by: Vincent Legoll vincent.legoll@gmail.com
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91133
References: https://bugs.freedesktop.org/show_bug.cgi?id=90224Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Michel Thierry <michel.thierry@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Some AMD systems also have non-PCI devices which can do DMA.
Those can't be handled by the AMD IOMMU, as the hardware can
only handle PCI. These devices would end up with no dma_ops,
as neither the per-device nor the global dma_ops will get
set. SWIOTLB provides global dma_ops when it is active, so
make sure there are global dma_ops too when swiotlb is
disabled.
Signed-off-by: NJoerg Roedel <jroedel@suse.de>

In passthrough mode (iommu=pt) all devices are identity
mapped. If a device does not support 64bit DMA it might
still need remapping. Make sure swiotlb is initialized to
provide this remapping.
Signed-off-by: NJoerg Roedel <jroedel@suse.de>

Since devices with IOMMUv2 functionality might be in the
same group as devices without it, allow those devices in
IOMMUv2 domains too.
Otherwise attaching the group with the IOMMUv2 device to the
domain will fail.
Signed-off-by: NJoerg Roedel <jroedel@suse.de>

Remove the AMD IOMMU driver implementation for passthrough
mode and rely on the new iommu core features for that.
Signed-off-by: NJoerg Roedel <jroedel@suse.de>

Since the conversion to default domains the
iommu_attach_device function only works for devices with
their own group. But this isn't always true for current
IOMMUv2 capable devices, so use iommu_attach_group instead.
Signed-off-by: NJoerg Roedel <jroedel@suse.de>

Use proper typecasting while performing byte-by-byte copy
Signed-off-by: NShahed Shaikh <shahed.shaikh@qlogic.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Cortina phy does not have kernel driver and we don't attach
device with phy layer for intefaces like XFI, XLAUI etc,
Hence check for interface type before calling disconnect.
Signed-off-by: NThanneeru Srinivasulu <tsrinivasulu@caviumnetworks.com>
Signed-off-by: NAleksey Makarov <aleksey.makarov@caviumnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NSunil Goutham <sgoutham@cavium.com>
Signed-off-by: NAleksey Makarov <aleksey.makarov@caviumnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This fixes a crash when changing rss with multiple traffic flows.

While interface teardown, disable tx queues after all NAPI threads
are done. If done otherwise tx queues might be woken up inside NAPI
if any CQE_TX are processed.
Signed-off-by: NSunil Goutham <sgoutham@cavium.com>
Signed-off-by: NAleksey Makarov <aleksey.makarov@caviumnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If a txq (SQ) remains in stopped state after this timeout its
considered as stuck and interface is reinited.
Signed-off-by: NSunil Goutham <sgoutham@cavium.com>
Signed-off-by: NAleksey Makarov <aleksey.makarov@caviumnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Previously TXQ is wakedup whenever napi is executed
and irrespective of if any CQE_TX are processed or not.
Added 'txq_stop' and 'txq_wake' counters to aid in debugging
if there are any future issues.
Signed-off-by: NSunil Goutham <sgoutham@cavium.com>
Signed-off-by: NAleksey Makarov <aleksey.makarov@caviumnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Suppressing standard alloc_pages() warnings. Some kernel configs limit
alloc size and the network driver may fail. Do not drop a kernel
warning in this case, instead just drop a oneliner that the network
driver could not be loaded since the buffer could not be allocated.
Signed-off-by: NSunil Goutham <sgoutham@cavium.com>
Signed-off-by: NAleksey Makarov <aleksey.makarov@caviumnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fixing TSO packages not being counted.
Signed-off-by: NSunil Goutham <sgoutham@cavium.com>
Signed-off-by: NAleksey Makarov <aleksey.makarov@caviumnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fix for memory leak when changing queue/channel count via ethtool
Signed-off-by: NSunil Goutham <sgoutham@cavium.com>
Signed-off-by: NAleksey Makarov <aleksey.makarov@caviumnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

With earlier configured value sufficient number of CQEs are not
being reserved for transmitted packets. Hence under heavy incoming
traffic load, receive notifications will take away most of the CQ
thus transmit notifications will be lost resulting in tx skbs not
being freed.

Finally SQ will be full and it will be stopped, watchdog timer
will kick in. After this fix receive notifications will not take
morethan half of CQ reserving the rest for transmit notifications.

Also changed CQ & SQ sizes from 16k to 4k.
This is also due to the receive notifications taking first half of
CQ under heavy load and time taken by NAPI to clear transmit notifications
will increase with higher queue sizes. Again results in SQ being stopped.
Signed-off-by: NSunil Goutham <sgoutham@cavium.com>
Signed-off-by: NAleksey Makarov <aleksey.makarov@caviumnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fixed 'tso_hdrs' memory not being freed properly.
Also fixed SQ skbuff maintenance issues.
Signed-off-by: NSunil Goutham <sgoutham@cavium.com>
Signed-off-by: NAleksey Makarov <aleksey.makarov@caviumnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Switching back to LDD transactions from LDWB.

While transmitting packets out with LDWB transactions
data integrity issues are seen very frequently.
hence switching back to LDD.
Signed-off-by: NSunil Goutham <sgoutham@cavium.com>
Signed-off-by: NRobert Richter <rrichter@cavium.com>
Signed-off-by: NAleksey Makarov <aleksey.makarov@caviumnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>