Both commits 0f17e4c796 ("Add missing
semaphore.h includes") and 4933d07531
("m68k: drivers/input/serio/hp_sdc.c needs <linux/semaphore.h>") added a
We only really need one ;)
Reported-by: Huang Weiyi <weiyi.huang@gmail.com>
Requested-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When decompressing extremely large packets allocating them through
kmalloc is prone to failure. Therefore it's better to use page
frags instead.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch merges the IPv4/IPv6 IPComp implementations since most
of the code is identical. As a result future enhancements will no
longer need to be duplicated.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This fix only affects UBI debugging.
If the the background thread is disabled for debugging purposes,
start it anyway, because otherwise we see tonns of kernel debugging
complaints like this:
INFO: task ubi_bgt0d:26857 blocked for more than 120 seconds.
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
ubi_bgt0d D dd37bf94 0 26857 2
dd37bfcc 00000086 f8e17cea dd37bf94 00000046 00000000 00000000 f5c62430
f5c62430 f5c62590 c2a09c80 f6cbd498 dd8e9cbc 00000296 dd37bfb0 00000296
dd8e9cb8 dd8e9cbc dd37bfcc c0119774 00000000 00000000 c0132e89 f6961560
Call Trace:
[<f8e17cea>] ? ubi_thread+0x0/0x127 [ubi]
[<c0119774>] ? complete+0x43/0x4b
[<c0132e89>] ? kthread+0x0/0x5b
[<f8e17cea>] ? ubi_thread+0x0/0x127 [ubi]
[<c0132eae>] kthread+0x25/0x5b
[<c0132e89>] ? kthread+0x0/0x5b
[<c0104953>] kernel_thread_helper+0x7/0x14
=======================
So start it, and go sleep inside it, instead of creating it and never
start.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
signalfd4, eventfd2, epoll_create1, dup3, pipe2 and inotify_init1
Signed-off-by: Tony Breeds <tony@bakeyournoodle.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
There are only 4 valid name=value pairs for writes to
/proc/ppc64/lparcfg. Current code allocates a buffer to copy
this information in from the user. Since the longest name=value
pair will easily fit into a buffer of 64 characters, simply
put the buffer on the stack instead of allocating the buffer.
Signed-off-by: Nathan Fotenot <nfont@austin.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Update the architecture vector to indicate that Cooperative Memory
Overcommitment is supported if CONFIG_PPC_SMLPAR is set.
Signed-off-by: Nathan Fontenot <nfont@austin.ibm.com>
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Adds support to the ibmvfc driver for collaborative memory overcommit.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Enable the driver to function in a Cooperative Memory Overcommitment (CMO)
environment.
The following changes are made to enable the driver for CMO:
* DMA mapping errors will not result in error messages if entitlement has
been exceeded and resources were not available.
* The driver has a get_desired_dma function defined to function
in a CMO environment. It will indicate how much IO memory it would like
to function.
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Acked by: Brian King <brking@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Acked-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Enable ibmveth for Cooperative Memory Overcommitment (CMO). For this driver
it means calculating a desired amount of IO memory based on the current MTU
and updating this value with the bus when MTU changes occur. Because DMA
mappings can fail, we have added a bounce buffer for temporary cases where
the driver can not map IO memory for the buffer pool.
The following changes are made to enable the driver for CMO:
* DMA mapping errors will not result in error messages if entitlement has
been exceeded and resources were not available.
* DMA mapping errors are handled gracefully, ibmveth_replenish_buffer_pool()
is corrected to check the return from dma_map_single and fail gracefully.
* The driver will have a get_desired_dma function defined to function
in a CMO environment.
* When the MTU is changed, the driver will update the device IO entitlement
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Santiago Leon <santil@us.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Activates larger rx buffer pools when the MTU is changed to a larger
value. This patch de-activates the large rx buffer pools when the MTU
changes to a smaller value.
Signed-off-by: Santiago Leon <santil@us.ibm.com>
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Verify memory entitlement updates can be handled by vio.
Signed-off-by: Nathan Fontenot <nfont@austin.ibm.com>
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
This is a large patch but the normal code path is not affected. For
non-pSeries platforms the code is ifdef'ed out and for non-CMO enabled
pSeries systems this does not affect the normal code path. Devices that
do not perform DMA operations do not need modification with this patch.
The function get_desired_dma was renamed from get_io_entitlement for
clarity.
Overview
Cooperative Memory Overcommitment (CMO) allows for a set of OS partitions
to be run with less RAM than the aggregate needs of the group of
partitions. The firmware will balance memory between the partitions
and page in/out memory as needed. Based on the number and type of IO
adpaters preset each partition is allocated an amount of memory for
DMA operations and this allocation will be guaranteed to the partition;
this is referred to as the partition's 'entitlement'.
Partitions running in a CMO environment can only have virtual IO devices
present. The VIO bus layer will manage the IO entitlement for the system.
Accounting, at a system and per-device level, is tracked in the VIO bus
code and exposed via sysfs. A set of dma_ops functions are added to
the bus to allow for this accounting.
Bus initialization
At initialization, the bus will calculate the minimum needs of the system
based on providing each device present with a standard minimum entitlement
along with a spare allocation for the bus to handle hotplug events.
If the minimum needs can not be met the system boot will be halted.
Device changes
The significant changes for devices while running under CMO are that the
devices must specify how much dedicated IO entitlement they desire and
must also handle DMA mapping errors that can occur due to constrained
IO memory. The virtual IO drivers are modified to silence errors when
DMA mappings fail for CMO and handle these failures gracefully.
Each devices will be guaranteed a minimum entitlement that can always
be mapped. Devices will specify how much entitlement they desire and
the VIO bus will attempt to provide for this. Devices can change their
desired entitlement level at any point in time to address particular needs
(via vio_cmo_set_dev_desired()), not just at device probe time.
VIO bus changes
The system will have a particular entitlement level available from which
it can provide memory to the devices. The bus defines two pools of memory
within this entitlement, the reserved and excess pools. Each device is
provided with it's own entitlement no less than a system defined minimum
entitlement and no greater than what the device has specified as it's
desired entitlement. The entitlement provided to devices comes from the
reserve pool. The reserve pool can also contain a spare allocation as
large as the system defined minimum entitlement which is used for device
hotplug events. Any entitlement not needed to fulfill the needs of a
reserve pool is placed in the excess pool. Each device is guaranteed
that it can map up to it's entitled level; additional mapping are possible
as long as there is unmapped memory in the excess pool.
Bus probe
As the system starts, each device is given an entitlement equal only
to the system defined minimum entitlement. The reserve pool is equal
to the sum of these entitlements, plus a spare allocation. The VIO bus
also tracks the aggregate desired entitlement of all the devices. If the
system desired entitlement is greater than the size of the reserve pool,
when devices unmap IO memory it will be reserved and a balance operation
will be scheduled for some time in the future.
Entitlement balancing
The balance function tries to fairly distribute entitlement between the
devices in the system with the goal of providing each device with it's
desired amount of entitlement. Devices using more than what would be
ideal will have their entitled set-point adjusted; this will effectively
set a goal for lower IO memory usage as future mappings can fail and
deallocations will trigger a balance operation to distribute the newly
unmapped memory. A fair distribution of entitlement can take several
balance operations to achieve. Entitlement changes and device DLPAR
events will alter the state of CMO and will trigger balance operations.
Hotplug events
The VIO bus allows for changes in system entitlement at run-time via
'vio_cmo_entitlement_update()'. When devices are added the hotplug
device event will be preceded by a system entitlement increase and this
is reversed when devices are removed.
The following changes are made that the VIO bus layer for CMO:
* add IO memory accounting per device structure.
* add IO memory entitlement query function to driver structure.
* during vio bus probe, if CMO is enabled, check that driver has
memory entitlement query function defined. Fail if function not defined.
* fail to register driver if io entitlement function not defined.
* create set of dma_ops at vio level for CMO that will track allocations
and return DMA failures once entitlement is reached. Entitlement will
limited by overall system entitlement. Devices will have a reserved
quantity of memory that is guaranteed, the rest can be used as available.
* expose entitlement, current allocation, desired allocation, and the
allocation error counter for devices to the user through sysfs
* provide mechanism for changing a device's desired entitlement at run time
for devices as an exported function and sysfs tunable
* track any DMA failures for entitled IO memory for each vio device.
* check entitlement against available system entitlement on device add
* track entitlement metrics (high water mark, current usage)
* provide function to reset high water mark
* provide minimum and desired entitlement numbers at a bus level
* provide drivers with a minimum guaranteed entitlement
* balance available entitlement between devices to satisfy their needs
* handle system entitlement changes and device hotplug
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
To support Cooperative Memory Overcommitment (CMO), we need to check
for failure from some of the tce hcalls.
These changes for the pseries platform affect the powerpc architecture;
patches for the other affected platforms are included in this patch.
pSeries platform IOMMU code changes:
* platform TCE functions must handle H_NOT_ENOUGH_RESOURCES errors and
return an error.
Architecture IOMMU code changes:
* Calls to ppc_md.tce_build need to check return values and return
DMA_MAPPING_ERROR for transient errors.
Architecture changes:
* struct machdep_calls for tce_build*_pSeriesLP functions need to change
to indicate failure.
* all other platforms will need updates to iommu functions to match the new
calling semantics; they will return 0 on success. The other platforms
default configs have been built, but no further testing was performed.
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Acked-by: Olof Johansson <olof@lixom.net>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
With the addition of Cooperative Memory Overcommitment (CMO) support
for IBM Power Systems, two fields have been added to the VPA to report
paging statistics. Add support in lparcfg to report them to userspace.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Adds a collaborative memory manager, which acts as a simple balloon driver
for System p machines that support cooperative memory overcommitment
(CMO).
Adds a platform configuration option for CMO called PPC_SMLPAR.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Newer versions of firmware support page states, which are used by the
collaborative memory manager (future patch) to "loan" pages to the
hypervisor for use by other partitions.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
For Cooperative Memory Overcommitment (CMO), set the FW_FEATURE_CMO
flag in powerpc_firmware_features from the rtas ibm,get-system-parameters
table prior to calling iommu_init_early_pSeries.
With this, any CMO specific functionality can be controlled by checking:
firmware_has_feature(FW_FEATURE_CMO)
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Split the retrieval of processor entitlement data returned in the H_GET_PPP
hcall into its own helper routine.
Signed-off-by: Nathan Fontenot <nfont@austin.ibm.com>
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Update /proc/ppc64/lparcfg to display Cooperative Memory
Overcommitment statistics as reported by the H_GET_MPP hcall. This
also updates the lparcfg interface to allow setting memory entitlement
and weight.
Signed-off-by: Nathan Fontenot <nfont@austin.ibm.com>
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Split the retrieval and setting of processor entitlement and weight into
helper routines. This also removes the printing of the raw values
returned from h_get_ppp, the values are already parsed and printed.
Signed-off-by: Nathan Fontenot <nfont@austin.ibm.com>
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Remove the extraneous error reporting used when a hcall made from lparcfg fails.
Signed-off-by: Nathan Fontenot <nfont@austin.ibm.com>
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
My previous patch to fix compilation with binutils-2.17 causes
a "file truncated" build error from ld with binutils 2.15 (and
possibly older), and a warning with 2.16 and 2.17.
This fixes it.
Signed-off-by: Segher Boessenkool <segher@kernel.crashing.org>
Acked-by: Chuck Meade <chuckmeade@mindspring.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
At the moment the fixed mapping is by default strongly ordered (the
iommu_fixed=weak boot option must be used to make the fixed mapping weakly
ordered). If we're on a setup where the southbridge is being used in
endpoint mode (triblade and CAB boards) the default should be a weakly
ordered fixed mapping.
This adds a check so that if a node of type pcie-endpoint can be found in
the device tree the fixed mapping is set to be weak by default (but can be
overridden using iommu_fixed=strong).
Signed-off-by: Mark Nelson <markn@au1.ibm.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
This patch implements support for HW based watchpoint via the
DBSR_DAC (Data Address Compare) facility of the BookE processors.
It does so by interfacing with the existing DABR breakpoint code
and adding the necessary bits and pieces for the new bits to
be properly set or cleared
Signed-off-by: Luis Machado <luisgpm@br.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
A struct sysdev_attribute * parameter was added to the show routine by
commit 4a0b2b4dbe "sysdev: Pass the
attribute to the low level sysdev show/store function".
This eliminates a warning:
arch/powerpc/kernel/sysfs.c:538: warning: initialization from incompatible pointer type
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Stash the first platform string matched by identify_cpu() in
powerpc_base_platform, and supply that to the ELF loader for the value
of AT_BASE_PLATFORM.
Signed-off-by: Nathan Lynch <ntl@pobox.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Some IBM POWER-based platforms have the ability to run in a
mode which mostly appears to the OS as a different processor from the
actual hardware. For example, a Power6 system may appear to be a
Power5+, which makes the AT_PLATFORM value "power5+". This means that
programs are restricted to the ISA supported by Power5+;
Power6-specific instructions are treated as illegal.
However, some applications (virtual machines, optimized libraries) can
benefit from knowledge of the underlying CPU model. A new aux vector
entry, AT_BASE_PLATFORM, will denote the actual hardware. For
example, on a Power6 system in Power5+ compatibility mode, AT_PLATFORM
will be "power5+" and AT_BASE_PLATFORM will be "power6". The idea is
that AT_PLATFORM indicates the instruction set supported, while
AT_BASE_PLATFORM indicates the underlying microarchitecture.
If the architecture has defined ELF_BASE_PLATFORM, copy that value to
the user stack in the same manner as ELF_PLATFORM.
Signed-off-by: Nathan Lynch <ntl@pobox.com>
Acked-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
* git://git.kernel.org/pub/scm/linux/kernel/git/rusty/linux-2.6-for-linus:
virtio: Add transport feature handling stub for virtio_ring.
virtio: Rename set_features to finalize_features
virtio: Formally reserve bits 28-31 to be 'transport' features.
s390: use virtio_console for KVM on s390
virtio: console as a config option
virtio_console: use virtqueue notification for hvc_console
hvc_console: rework setup to replace irq functions with callbacks
virtio_blk: check for hardsector size from host
virtio: Use bus_type probe and remove methods
virtio: don't always force a notification when ring is full
virtio: clarify that ABI is usable by any implementations
virtio: Recycle unused recv buffer pages for large skbs in net driver
virtio net: Allow receiving SG packets
virtio net: Add ethtool ops for SG/GSO
virtio: fix virtio_net xmit of freed skb bug
Obvious misc patch been in my queue (& linux-next) for over a cycle.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
To prepare for virtio_ring transport feature bits, hook in a call in
all the users to manipulate them. This currently just clears all the
bits, since it doesn't understand any features.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Rather than explicitly handing the features to the lower-level, we just
hand the virtio_device and have it set the features. This make it clear
that it has the chance to manipulate the features of the device at this
point (and that all feature negotiation is already done).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We assign feature bits as required, but it makes sense to reserve some
for the particular transport, rather than the particular device.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This patch enables virtio_console as the default console on kvm for
s390. We currently use the same notify hack as lguest for early
console output. I will try to address this for lguest and s390 later.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
I also added a small Kconfig change that allows the user to specify the
virtio console in menuconfig.
(Fixes to export symbols from Stephen Rothwell <sfr@canb.auug.org.au>)
(Fixes for CONFIG_VIRTIO_CONSOLE=y vs CONFIG_VIRTIO=m from Christian himself)
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
This patch exploits the new notifier callbacks of the hvc_console. We can
use the virtio callbacks instead of the polling code.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This patch tries to change hvc_console to not use request_irq/free_irq if
the backend does not use irqs. This allows virtio_console to use hvc_console
without having a linker reference to request_irq/free_irq.
In addition, together with patch 2/3 it improves the performance for virtio
console input. (an earlier version of this patch was tested by Yajin on lguest)
The irq specific code is moved to hvc_irq.c and selected by the drivers that
use irqs (System p, System i, XEN).
I replaced "int irq" with the opaque "int data". The request_irq and
free_irq calls are replaced with notifier_add and notifier_del. I have also
changed the code a bit to call the notifier_add and notifier_del inside the
spinlock area as the callbacks are found via hp->ops.
Changes since last version:
o remove ifdef
o reintroduce "irq_requested" as "notified"
o cleanups, sparse..
I did not move the timer based polling into a separate polling scheme. I
played with several variants, but it seems we need to sleep/schedule in
a thread even for irq based consoles, as there are throttleing and buffer
size constraints.
I also kept hvc_struct defined in hvc_console.h so that hvc_irq.c can access
the irq_requested element.
Feedback is appreciated. virtio_console is currently the only available console
for kvm on s390. I plan to push this change as soon as all affected parties
agree on it. I would love to get test results from System p, Xen etc.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Currently virtio_blk assumes a 512 byte hard sector size. This can cause
trouble / performance issues if the backing has a different block size
(like a file on an ext3 file system formatted with 4k block size or a dasd).
Lets add a feature flag that tells the guest to use a different hard sector
size than 512 byte.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Hook up to the probe() and remove() methods in bus_type
rather than device_driver. The latter has been preferred
since 2.6.16.
Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We force notification when the ring is full, even if the host has
indicated it doesn't want to know. This seemed like a good idea at
the time: if we fill the transmit ring, we should tell the host
immediately.
Unfortunately this logic also applies to the receiving ring, which is
refilled constantly. We should introduce real notification thesholds
to replace this logic. Meanwhile, removing the logic altogether breaks
the heuristics which KVM uses, so we use a hack: only notify if there are
outgoing parts of the new buffer.
Here are the number of exits with lguest's crappy network implementation:
Before:
network xmit 7859051 recv 236420
After:
network xmit 7858610 recv 118136
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We want others to implement and use virtio, so it makes sense to BSD
license the non-__KERNEL__ parts of the headers to make this crystal
clear.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Acked-by: Mark McLoughlin <markmc@redhat.com>
Acked-by: Ryan Harper <ryanh@us.ibm.com>
Acked-by: Eric Van Hensbergen <ericvh@gmail.com>
Acked-by: Anthony Liguori <aliguori@us.ibm.com>
If we hack the virtio_net driver to always allocate full-sized (64k+)
skbuffs, the driver slows down (lguest numbers):
Time to receive 1GB (small buffers): 10.85 seconds
Time to receive 1GB (64k+ buffers): 24.75 seconds
Of course, large buffers use up more space in the ring, so we increase
that from 128 to 2048:
Time to receive 1GB (64k+ buffers, 2k ring): 16.61 seconds
If we recycle pages rather than using alloc_page/free_page:
Time to receive 1GB (64k+ buffers, 2k ring, recycle pages): 10.81 seconds
This demonstrates that with efficient allocation, we don't need to
have a separate "small buffer" queue.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Finally this patch lets virtio_net receive GSO packets in addition
to sending them. This can definitely be optimised for the non-GSO
case. For comparison the Xen approach stores one page in each skb
and uses subsequent skb's pages to construct an SG skb instead of
preallocating the maximum amount of pages per skb.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au> (added feature bits)
This patch adds some basic ethtool operations to virtio_net so
I could test SG without GSO (which was really useful because TSO
turned out to be buggy :)
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au> (remove MTU setting)
On Mon, 2008-05-26 at 17:42 +1000, Rusty Russell wrote:
> If we fail to transmit a packet, we assume the queue is full and put
> the skb into last_xmit_skb. However, if more space frees up before we
> xmit it, we loop, and the result can be transmitting the same skb twice.
>
> Fix is simple: set skb to NULL if we've used it in some way, and check
> before sending.
...
> diff -r 564237b31993 drivers/net/virtio_net.c
> --- a/drivers/net/virtio_net.c Mon May 19 12:22:00 2008 +1000
> +++ b/drivers/net/virtio_net.c Mon May 19 12:24:58 2008 +1000
> @@ -287,21 +287,25 @@ again:
> free_old_xmit_skbs(vi);
>
> /* If we has a buffer left over from last time, send it now. */
> - if (vi->last_xmit_skb) {
> + if (unlikely(vi->last_xmit_skb)) {
> if (xmit_skb(vi, vi->last_xmit_skb) != 0) {
> /* Drop this skb: we only queue one. */
> vi->dev->stats.tx_dropped++;
> kfree_skb(skb);
> + skb = NULL;
> goto stop_queue;
> }
> vi->last_xmit_skb = NULL;
With this, may drop an skb and then later in the function discover that
we could have sent it after all. Poor wee skb :)
How about the incremental patch below?
Cheers,
Mark.
Subject: [PATCH] virtio_net: Delay dropping tx skbs
Currently we drop the skb in start_xmit() if we have a
queued buffer and fail to transmit it.
However, if we delay dropping it until we've stopped the
queue and enabled the tx notification callback, then there
is a chance space might become available for it.
Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This fixes the following compile error caused by commit
f9247273cb ("UFS: add const to parser
token table"):
CC fs/nfs/nfsroot.o
/home/bunk/linux/kernel-2.6/git/linux-2.6/fs/nfs/nfsroot.c:130: error: tokens causes a section type conflict
make[3]: *** [fs/nfs/nfsroot.o] Error 1
Signed-off-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
