Make sure that in all cases where we could not obtain measurement
characteristics the associated fields are set to invalid values.

Note: without this change the "shared" capability of a channel path
for which we could not obtain the measurement characteristics was
incorrectly displayed as 0 (not shared). We will now correctly
report "unknown" in this case.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Reviewed-by: Peter Oberparleiter <oberpar@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

Measurement characteristics are allocated during channel path
registration but not freed during deregistration. Fix this by
embedding these characteristics inside struct channel_path.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Reviewed-by: Peter Oberparleiter <oberpar@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

Currently, on card response failures a combination of card domain and
domain id is recorded in the kernel messages.

According to the message description only the card id will be recorded.
The domain id is not relevant, since the whole card including all domains
is set offline.
Signed-off-by: Ingo Tuchscherer <ingo.tuchscherer@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

Since each iomap_entry handles only one bar of one pci function
(even when disjunct ranges of a bar are mapped) the sanity check
in pci_iomap_range is not needed and can be removed.

Also convert the remaining BUG_ONs to WARN_ONs.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Reviewed-by: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

We receive special notifications from firmware when an error was detected
and a pci function became unusable. Set the error_state accordingly to give
device drivers a hint that they don't need to try error recovery.
Suggested-by: Alexander Schmidt <alexschm@de.ibm.com>
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Reviewed-by: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

Fix the check which bar space we should map to allow available bars only.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Reviewed-by: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

On s390 we need to maintain a mapping between iomem addresses
and arch specific function identifiers. Currently the mapping
table is created as such that we could span the whole iomem
address space. Since we can only map each bar space from each
possible function we have an upper bound for the number of
mapping entries.

This reduces the size of the iomap from 256K to less than 4K
(using the defconfig).
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Reviewed-by: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

Most of the constants defined in pci_io.h depend on each other
and thus can be calculated.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Reviewed-by: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

Provide and use a ZPCI_ADDR macro as the complement of ZPCI_IDX
to get rid of some constants in the code.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Reviewed-by: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

ZPCI_IOMAP_MAX_ENTRIES is off by one. Let's adjust this
for the sake of correctness.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Reviewed-by: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

Commit 3e89e1c5 ("hugetlb: make mm and fs code explicitly non-modular")
moves hugetlb_init() from module_init to subsys_initcall.

The hugetlb_init()->hugetlb_register_node() code accesses "node->dev.kobj"
which is initialized in numa_init_late().

Since numa_init_late() is a device_initcall which is called *after*
subsys_initcall the above mentioned patch breaks NUMA on s390.

So fix this and move numa_init_late() to arch_initcall.

Fixes: 3e89e1c5 ("hugetlb: make mm and fs code explicitly non-modular")
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

Yet another leftover from the 31 bit era. The usual operation
"y = x & PSW_ADDR_INSN" with the PSW_ADDR_INSN mask is a nop for
CONFIG_64BIT.

Therefore remove all usages and hope the code is a bit less confusing.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>

This is a leftover from the 31 bit area. For CONFIG_64BIT the usual
operation "y = x | PSW_ADDR_AMODE" is a nop. Therefore remove all
usages of PSW_ADDR_AMODE and make the code a bit less confusing.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

memblock_alloc() and memblock_alloc_base() will panic on their own if
they can't find free memory. Therefore remove some pointless checks.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Acked-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

Allocating memory with a requested minimum alignment of 1 is wrong
since pg_data_t contains a spinlock which requires an alignment of 4
bytes.

Therefore fix this and ask for an alignment of 8 bytes like it is
guarenteed for all kmalloc requests.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Acked-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

The ssm instruction takes longer that stnsm/stosm as it is often
used to modify DAT and PER. We know that irqsave/irqrestore only
deals with external and I/O interrupts and we know that irqrestore
can transition only from disabled->disabled or disabled->enabled,
so we can use the faster stosm.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>

To improve readability we can use TASK_MAX_SIZE when we just check for the
upper limit.  All places explicitly dealing with 3 vs 4 level pgtables
were left unchanged.
Signed-off-by: Dominik Dingel <dingel@linux.vnet.ibm.com>
Reviewed-By: Sascha Silbe <silbe@linux.vnet.ibm.com>

Pull virtio barrier rework+fixes from Michael Tsirkin:
 "This adds a new kind of barrier, and reworks virtio and xen to use it.

  Plus some fixes here and there"

* tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost: (44 commits)
  checkpatch: add virt barriers
  checkpatch: check for __smp outside barrier.h
  checkpatch.pl: add missing memory barriers
  virtio: make find_vqs() checkpatch.pl-friendly
  virtio_balloon: fix race between migration and ballooning
  virtio_balloon: fix race by fill and leak
  s390: more efficient smp barriers
  s390: use generic memory barriers
  xen/events: use virt_xxx barriers
  xen/io: use virt_xxx barriers
  xenbus: use virt_xxx barriers
  virtio_ring: use virt_store_mb
  sh: move xchg_cmpxchg to a header by itself
  sh: support 1 and 2 byte xchg
  virtio_ring: update weak barriers to use virt_xxx
  Revert "virtio_ring: Update weak barriers to use dma_wmb/rmb"
  asm-generic: implement virt_xxx memory barriers
  x86: define __smp_xxx
  xtensa: define __smp_xxx
  tile: define __smp_xxx
  ...

Pull arch/tile updates from Chris Metcalf:
 "This is a grab bag of changes that includes some NOHZ and
  context-tracking related changes, some debugging improvements,
  JUMP_LABEL support, and some fixes for tilepro allmodconfig support.

  We also remove the now-unused node_has_online_mem() definitions both
  for tile's asm/topology.h as well as in linux/topology.h itself"

* git://git.kernel.org/pub/scm/linux/kernel/git/cmetcalf/linux-tile:
  numa: remove stale node_has_online_mem() define
  arch/tile: move user_exit() to early kernel entry sequence
  tile: fix bug in setting PT_FLAGS_DISABLE_IRQ on kernel entry
  tile: fix tilepro casts for readl, writel, etc
  tile: fix a -Wframe-larger-than warning
  tile: include the syscall number in the backtrace
  MAINTAINERS: add git URL for tile
  arch/tile: adopt prepare_exit_to_usermode() model from x86
  tile/jump_label: add jump label support for TILE-Gx
  tile: define a macro ktext_writable_addr to get writable kernel text address

Pull AVR32 updates from Hans-Christian Noren Egtvedt.

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/egtvedt/linux-avr32:
  mmc: atmel: get rid of struct mci_dma_data
  mmc: atmel-mci: restore dma on AVR32
  avr32: wire up missing syscalls
  avr32: wire up accept4 syscall

Pull btrfs updates from Chris Mason:
 "This has our usual assortment of fixes and cleanups, but the biggest
  change included is Omar Sandoval's free space tree.  It's not the
  default yet, mounting -o space_cache=v2 enables it and sets a readonly
  compat bit.  The tree can actually be deleted and regenerated if there
  are any problems, but it has held up really well in testing so far.

  For very large filesystems (30T+) our existing free space caching code
  can end up taking a huge amount of time during commits.  The new tree
  based code is faster and less work overall to update as the commit
  progresses.

  Omar worked on this during the summer and we'll hammer on it in
  production here at FB over the next few months"

* 'for-linus-4.5' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux-btrfs: (73 commits)
  Btrfs: fix fitrim discarding device area reserved for boot loader's use
  Btrfs: Check metadata redundancy on balance
  btrfs: statfs: report zero available if metadata are exhausted
  btrfs: preallocate path for snapshot creation at ioctl time
  btrfs: allocate root item at snapshot ioctl time
  btrfs: do an allocation earlier during snapshot creation
  btrfs: use smaller type for btrfs_path locks
  btrfs: use smaller type for btrfs_path lowest_level
  btrfs: use smaller type for btrfs_path reada
  btrfs: cleanup, use enum values for btrfs_path reada
  btrfs: constify static arrays
  btrfs: constify remaining structs with function pointers
  btrfs tests: replace whole ops structure for free space tests
  btrfs: use list_for_each_entry* in backref.c
  btrfs: use list_for_each_entry_safe in free-space-cache.c
  btrfs: use list_for_each_entry* in check-integrity.c
  Btrfs: use linux/sizes.h to represent constants
  btrfs: cleanup, remove stray return statements
  btrfs: zero out delayed node upon allocation
  btrfs: pass proper enum type to start_transaction()
  ...

Pull more networking fixes from David Miller:

 1) Fix brcmfmac build with older gcc, from Arend van Spriel.

 2) IRQ values unintentionally truncated to u8 in mlx5 driver, from
    Doron Tsur.

 3) Fix build warnings wrt tcp cgroup changes, from Geert Uytterhoeven.

 4) Limit deep recursion in ovs stack, from Hannes Frederic Sowa.

 5) at803x phy driver bug fixes from, Martin Blumenstingl.

 6) Fix TSO handling in hns driver, from Daode Huang

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (22 commits)
  ovs: limit ovs recursions in ovs_execute_actions to not corrupt stack
  team: Replace rcu_read_lock with a mutex in team_vlan_rx_kill_vid
  net: hns: bug fix about hisilicon TSO BD mode
  brcmfmac: fix BRCMF_FW_NVRAM_DEF macro for older gcc compilers
  net: phy: at803x: Add the interrupt register bit definitions
  net: phy: at803x: Clean up duplicate register definitions
  net: phy: at803x: Allow specifying the RGMII RX clock delay via phy mode
  net: phy: at803x: Don't set gbit features for the AR8030 phy
  arm64: bpf: add extra pass to handle faulty codegen
  arm64: insn: remove BUG_ON from codegen
  sctp: the temp asoc's transports should not be hashed/unhashed
  net/mlx5_core: Fix trimming down IRQ number
  tcp_memcontrol: Forward declare cgroup_subsys and mem_cgroup stucts
  batman-adv: Drop immediate orig_node free function
  batman-adv: Drop immediate batadv_hard_iface free function
  batman-adv: Drop immediate neigh_ifinfo free function
  batman-adv: Drop immediate batadv_hardif_neigh_node free function
  batman-adv: Drop immediate batadv_neigh_node free function
  batman-adv: Drop immediate batadv_orig_ifinfo free function
  batman-adv: Avoid recursive call_rcu for batadv_nc_node
  ...

Pull IDE updates from David Miller:
 "Just a few small changes this merge window, marking ops const, printf
  string type fixes, etc"

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/ide:
  drivers/ide: make ide-scan-pci.c driver explicitly non-modular
  ide: constify ide_dma_ops structures
  ide: silence some underflow warnings

Pull RTC updates from Alexandre Belloni:
 "Core:
   - fix module reference count in rtc-proc
   - Replace simple_strtoul by kstrtoul

  New driver:
   - Epson RX8010SJ

  Subsystem wide cleanups:
   - use %ph for short hex dumps
   - constify *_chip_ops structures

  Drivers:
   - abx80x: Microcrystal rv1805 support, alarm support
   - cmos: prevent kernel warning on IRQ flags mismatch
   - s5m: various cleanups
   - rv8803: rx8900 compatibility, small error path fix
   - sunxi: various cleanups
   - lpc32xx: remove irq > NR_IRQS check from probe()
   - imxdi: fix spelling mistake in warning message
   - ds1685: don't try to micromanage sysfs output size
   - da9063: avoid writing undefined data to rtc
   - gemini: Remove unnecessary platform_set_drvdata()
   - efi: add efi_procfs in efi_rtc_ops
   - pcf8523: refuse to write dates later than 2099"

* tag 'rtc-4.5' of git://git.kernel.org/pub/scm/linux/kernel/git/abelloni/linux: (24 commits)
  rtc: cmos: prevent kernel warning on IRQ flags mismatch
  rtc: rtc-ds2404: constify ds2404_chip_ops structures
  rtc: s5m: Make register configuration per S2MPS device to remove exceptions
  rtc: s5m: Add separate field for storing auto-cleared mask in register config
  rtc: s5m: Cleanup by removing useless 'rtc' prefix from fields
  rtc: Replace simple_strtoul by kstrtoul
  rtc: abx80x: add alarm support
  rtc: abx80x: Add Microcrystal rv1805 support
  rtc: v3020: constify v3020_chip_ops structures
  rtc: rv8803: Extend compatibility with the rx8900
  rtc: rv8803: fix handling return value of i2c_smbus_read_byte_data
  rtc: Add Epson RX8010SJ RTC driver
  rtc: lpc32xx: remove irq > NR_IRQS check from probe()
  rtc: imxdi: fix spelling mistake in warning message
  rtc: ds1685: don't try to micromanage sysfs output size
  rtc: use %ph for short hex dumps
  rtc: da9063: avoid writing undefined data to rtc
  rtc: sunxi: use of_device_get_match_data
  rtc: sunxi: constify the data_year_param structure
  rtc: sunxi: fix signedness issues
  ...

Pull fbdev updates from Tomi Valkeinen:
 "Summary:

   - pxafb: device-tree support
   - An unsafe kernel parameter 'lockless_register_fb' for debugging
     problems happening while inside the console lock
   - Small miscellaneous fixes & cleanups
   - omapdss: add writeback support functions
   - Separation of omapfb and omapdrm (see below)

  About the separation of omapfb and omapdrm, see

    http://permalink.gmane.org/gmane.comp.video.dri.devel/143151

  for longer story.  The short version:

  omapfb and omapdrm have shared low level drivers (omapdss and panel
  drivers), making further development of omapdrm difficult.  After
  these patches omapfb and omapdrm have their own versions of the
  drivers, which are more or less direct copies for now but will diverge
  soon.

  This also means that omapfb (everything under drivers/video/fbdev/omap2/)
  is now in maintenance mode, and all new development will be done for
  omapdrm (drivers/gpu/drm/omapdrm/)"

* tag 'fbdev-4.5' of git://git.kernel.org/pub/scm/linux/kernel/git/tomba/linux: (49 commits)
  video: fbdev: pxafb: fix out of memory error path
  drm/omap: make omapdrm select OMAP2_DSS
  drm/omap: move omapdss & displays under omapdrm
  omapfb: move vrfb into omapfb
  omapfb: take omapfb's private omapdss into use
  omapfb/displays: change CONFIG_DISPLAY_* to CONFIG_FB_OMAP2_*
  omapfb/dss: change CONFIG_OMAP* to CONFIG_FB_OMAP*
  omapdss: remove CONFIG_OMAP2_DSS_VENC from omapdss.h
  omapfb: copy omapdss & displays for omapfb
  omapfb: allow compilation only if DRM_OMAP is disabled
  fbdev: omap2: panel-dpi: simplify gpio setting
  fbdev: omap2: panel-dpi: in .disable first disable backlight then display
  OMAPDSS: DSS: fix a warning message
  video: omapdss: delete unneeded of_node_put
  OMAPDSS: DISPC: Remove boolean comparisons
  OMAPDSS: DSI: cleanup DSI_IRQ_ERROR_MASK define
  OMAPDSS: remove extra out == NULL checks
  OMAPDSS: change internal dispc functions to static
  OMAPDSS: make a two dss feat funcs internal to omapdss
  OMAPDSS: remove extra EXPORT_SYMBOLs
  ...

This isn't used anywhere, so delete it.

Looks like the last usage (in x86-specific code) was removed by Tejun
in 2011 in commit bd6709a9 ("x86, NUMA: Make 32bit use common NUMA
init path").
Signed-off-by: Chris Metcalf <cmetcalf@ezchip.com>

This ensures that we always notify context tracking that we
have exited from user space no matter how we enter the kernel.
It is similar to how arm64 handles context tracking, for example.

This allows the removal of all the exception_enter() calls that
were added in commit 49e4e156 ("tile: support CONTEXT_TRACKING and
thus NOHZ_FULL").
Signed-off-by: Chris Metcalf <cmetcalf@ezchip.com>

This flag value is saved in ptregs and used to decide whether
to disable irqs when returning from the kernel.  Commit 1168df528fe4
("tile: don't assume user privilege is zero") performed a bad
merge from some KVM-enabled code that had not yet been upstreamed.

The only issue with the old code is that we will read the interrupt
mask in more conditions than we need to (e.g., coming from user
space when user space has the Interrupt Critical Section bit set, or
coming from a guest kernel), which is a slow multi-cycle operation.
This change saves those few cycles in the common case.
Signed-off-by: Chris Metcalf <cmetcalf@ezchip.com>

Missing parentheses could cause an argument of the form
"integer + pointer" to get cast to "(long)integer + pointer"
and remain a pointer type, causing compiler warnings.
Signed-off-by: Chris Metcalf <cmetcalf@ezchip.com>

The warning occurs in setup.c, where it is known that it can't be
a problem, but it's still a good idea to silence the warning.
The onstack array is converted from an s32 to a u8, which still
is plenty of range for the values being managed there.
Signed-off-by: Chris Metcalf <cmetcalf@ezchip.com>

This information is easily available in the backtrace data and can
be helpful when trying to figure out the backtrace, particularly
if we're early in kernel entry or late in kernel exit.
Signed-off-by: Chris Metcalf <cmetcalf@ezchip.com>

Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Chris Metcalf <cmetcalf@ezchip.com>

This change is a prerequisite change for TASK_ISOLATION but also
stands on its own for readability and maintainability.  The existing
tile do_work_pending() was called in a loop from assembly on
the slow path; this change moves the loop into C code as well.
For the x86 version see commit c5c46f59 ("x86/entry: Add new,
comprehensible entry and exit handlers written in C").

This change exposes a pre-existing bug on the older tilepro platform;
the singlestep processing is done last, but on tilepro (unlike tilegx)
we enable interrupts while doing that processing, so we could in
theory miss a signal or other asynchronous event.  A future change
could fix this by breaking the singlestep work into a "prepare"
step done in the main loop, and a "trigger" step done after exiting
the loop.  Since this change is intended as purely a restructuring
change, we call out the bug explicitly now, but don't yet fix it.
Signed-off-by: Chris Metcalf <cmetcalf@ezchip.com>

The Kconfig for this support is currently:

config IDEPCI_PCIBUS_ORDER
        bool "Probe IDE PCI devices in the PCI bus order (DEPRECATED)"

...meaning that it currently is not being built as a module by anyone.
Lets change the initcall to be the equivalent device_initcall, so that
when reading the driver code, there is no doubt it is builtin-only.

Unlike other similar changes, we leave the module.h header to be
included since this code interacts with other drivers and needs to
know what a struct module is.

Cc: "David S. Miller" <davem@davemloft.net>
Cc: linux-ide@vger.kernel.org
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The ide_dma_ops structures are never modified, so declare these as const,
as is already done for the others.

Done with the help of Coccinelle.
Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>

Back in the day we used to just say this code was root only so it was
ok that the bounds checking was sloppy.  These days it annoys static
checkers so we fix it.

In the original code "c > INT_MAX" was never true since "c" was an int.
I am not sure what was intended so I left it alone.  But because I made
"c" unsigned it means we don't have a warning any more.

The second warning is that we cap "i" but allow negatives leading to an
underflow of the ide_disks_chs[] array.  The third set of warnings is
because these values come from the user and we cap most of the upper
bounds but allow negative values.  Negative cylinders doesn't make
sense.

drivers/ide/ide.c:262 ide_set_disk_chs() warn: impossible condition '(c > ((~0 >> 1))) => (s32min-s32max > s32max)'
drivers/ide/ide.c:270 ide_set_disk_chs() warn: check 'ide_disks_chs[i]' for negative offsets 'i' = s32min.  extra = 's32min-19'
drivers/ide/ide.c:271 ide_set_disk_chs() warn: no lower bound on 'h'
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

It was seen that defective configurations of openvswitch could overwrite
the STACK_END_MAGIC and cause a hard crash of the kernel because of too
many recursions within ovs.

This problem arises due to the high stack usage of openvswitch. The rest
of the kernel is fine with the current limit of 10 (RECURSION_LIMIT).

We use the already existing recursion counter in ovs_execute_actions to
implement an upper bound of 5 recursions.

Cc: Pravin Shelar <pshelar@ovn.org>
Cc: Simon Horman <simon.horman@netronome.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Simon Horman <simon.horman@netronome.com>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

We can't be within an RCU read-side critical section when deleting
VLANs, as underlying drivers might sleep during the hardware operation.
Therefore, replace the RCU critical section with a mutex. This is
consistent with team_vlan_rx_add_vid.

Fixes: 3d249d4c ("net: introduce ethernet teaming device")
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The current upstreaming code fails to set the tso_mode register
when initilizes, when processes large size packets, the default 4 bd is
not enough, so this patch initilizes it and set the default value to 8 bds
Signed-off-by: Daode Huang <huangdaode@hisilicon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>