HalfSipHash, or hsiphash, is a shortened version of SipHash, which
generates 32-bit outputs using a weaker 64-bit key. It has *much* lower
security margins, and shouldn't be used for anything too sensitive, but
it could be used as a hashtable key function replacement, if the output
is never exposed, and if the security requirement is not too high.
The goal is to make this something that performance-critical jhash users
would be willing to use.
On 64-bit machines, HalfSipHash1-3 is slower than SipHash1-3, so we alias
SipHash1-3 to HalfSipHash1-3 on those systems.
64-bit x86_64:
[ 0.509409] test_siphash: SipHash2-4 cycles: 4049181
[ 0.510650] test_siphash: SipHash1-3 cycles: 2512884
[ 0.512205] test_siphash: HalfSipHash1-3 cycles: 3429920
[ 0.512904] test_siphash: JenkinsHash cycles: 978267
So, we map hsiphash() -> SipHash1-3
32-bit x86:
[ 0.509868] test_siphash: SipHash2-4 cycles: 14812892
[ 0.513601] test_siphash: SipHash1-3 cycles: 9510710
[ 0.515263] test_siphash: HalfSipHash1-3 cycles: 3856157
[ 0.515952] test_siphash: JenkinsHash cycles: 1148567
So, we map hsiphash() -> HalfSipHash1-3
hsiphash() is roughly 3 times slower than jhash(), but comes with a
considerable security improvement.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Reviewed-by: Jean-Philippe Aumasson <jeanphilippe.aumasson@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
SipHash is a 64-bit keyed hash function that is actually a
cryptographically secure PRF, like HMAC. Except SipHash is super fast,
and is meant to be used as a hashtable keyed lookup function, or as a
general PRF for short input use cases, such as sequence numbers or RNG
chaining.
For the first usage:
There are a variety of attacks known as "hashtable poisoning" in which an
attacker forms some data such that the hash of that data will be the
same, and then preceeds to fill up all entries of a hashbucket. This is
a realistic and well-known denial-of-service vector. Currently
hashtables use jhash, which is fast but not secure, and some kind of
rotating key scheme (or none at all, which isn't good). SipHash is meant
as a replacement for jhash in these cases.
There are a modicum of places in the kernel that are vulnerable to
hashtable poisoning attacks, either via userspace vectors or network
vectors, and there's not a reliable mechanism inside the kernel at the
moment to fix it. The first step toward fixing these issues is actually
getting a secure primitive into the kernel for developers to use. Then
we can, bit by bit, port things over to it as deemed appropriate.
While SipHash is extremely fast for a cryptographically secure function,
it is likely a bit slower than the insecure jhash, and so replacements
will be evaluated on a case-by-case basis based on whether or not the
difference in speed is negligible and whether or not the current jhash usage
poses a real security risk.
For the second usage:
A few places in the kernel are using MD5 or SHA1 for creating secure
sequence numbers, syn cookies, port numbers, or fast random numbers.
SipHash is a faster and more fitting, and more secure replacement for MD5
in those situations. Replacing MD5 and SHA1 with SipHash for these uses is
obvious and straight-forward, and so is submitted along with this patch
series. There shouldn't be much of a debate over its efficacy.
Dozens of languages are already using this internally for their hash
tables and PRFs. Some of the BSDs already use this in their kernels.
SipHash is a widely known high-speed solution to a widely known set of
problems, and it's time we catch-up.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Reviewed-by: Jean-Philippe Aumasson <jeanphilippe.aumasson@gmail.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Eric Biggers <ebiggers3@gmail.com>
Cc: David Laight <David.Laight@aculab.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Nothing about the route lookup requires bottom half to be disabled.
Remove the local_bh_disable ... local_bh_enable around ip_route_input.
This appears to be a vestige of days gone by as it has been there
since the beginning of git time.
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The ethtool api {get|set}_settings is deprecated.
We move this driver to new api {get|set}_link_ksettings.
Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The ethtool api {get|set}_settings is deprecated.
We move this driver to new api {get|set}_link_ksettings.
Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The ethtool api {get|set}_settings is deprecated.
We move this driver to new api {get|set}_link_ksettings.
Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The ethtool api {get|set}_settings is deprecated.
We move this driver to new api {get|set}_link_ksettings.
Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The ethtool api {get|set}_settings is deprecated.
We move this driver to new api {get|set}_link_ksettings.
Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
sock_init() call it but not check it's return value,
so change it to void return and add an internal BUG_ON() check.
Signed-off-by: yuan linyu <Linyu.Yuan@alcatel-sbell.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
Willem de Bruijn says:
====================
convert tc_verd to integer bitfields
The skb tc_verd field takes up two bytes but uses far fewer bits.
Convert the remaining use cases to bitfields that fit in existing
holes (depending on config options) and potentially save the two
bytes in struct sk_buff.
This patchset is based on an earlier set by Florian Westphal and its
discussion (http://www.spinics.net/lists/netdev/msg329181.html).
Patches 1 and 2 are low hanging fruit: removing the last traces of
data that are no longer stored in tc_verd.
Patches 3 and 4 convert tc_verd to individual bitfields (5 bits).
Patch 5 reduces TC_AT to a single bitfield,
as AT_STACK is not valid here (unlike in the case of TC_FROM).
Patch 6 changes TC_FROM to two bitfields with clearly defined purpose.
It may be possible to reduce storage further after this initial round.
If tc_skip_classify is set only by IFB, testing skb_iif may suffice.
The L2 header pushing/popping logic can perhaps be shared with
AF_PACKET, which currently not pkt_type for the same purpose.
Changes:
RFC -> v1
- (patch 3): remove no longer needed label in tfc_action_exec
- (patch 5): set tc_at_ingress at the same points as existing
SET_TC_AT calls
Tested ingress mirred + netem + ifb:
ip link set dev ifb0 up
tc qdisc add dev eth0 ingress
tc filter add dev eth0 parent ffff: \
u32 match ip dport 8000 0xffff \
action mirred egress redirect dev ifb0
tc qdisc add dev ifb0 root netem delay 1000ms
nc -u -l 8000 &
ssh $otherhost nc -u $host 8000
Tested egress mirred:
ip link add veth1 type veth peer name veth2
ip link set dev veth1 up
ip link set dev veth2 up
tcpdump -n -i veth2 udp and dst port 8000 &
tc qdisc add dev eth0 root handle 1: prio
tc filter add dev eth0 parent 1:0 \
u32 match ip dport 8000 0xffff \
action mirred egress redirect dev veth1
tc qdisc add dev veth1 root netem delay 1000ms
nc -u $otherhost 8000
Tested ingress mirred:
ip link add veth1 type veth peer name veth2
ip link add veth3 type veth peer name veth4
ip netns add ns0
ip netns add ns1
for i in 1 2 3 4; do \
NS=ns$((${i}%2)); \
ip link set dev veth${i} netns ${NS}; \
ip netns exec ${NS} \
ip addr add dev veth${i} 192.168.1.${i}/24; \
ip netns exec ${NS} \
ip link set dev veth${i} up; \
done
ip netns exec ns0 tc qdisc add dev veth2 ingress
ip netns exec ns0 \
tc filter add dev veth2 parent ffff: \
u32 match ip dport 8000 0xffff \
action mirred ingress redirect dev veth4
ip netns exec ns0 \
tcpdump -n -i veth4 udp and dst port 8000 &
ip netns exec ns1 \
nc -u 192.168.1.2 8000
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
The tc_from field fulfills two roles. It encodes whether a packet was
redirected by an act_mirred device and, if so, whether act_mirred was
called on ingress or egress. Split it into separate fields.
The information is needed by the special IFB loop, where packets are
taken out of the normal path by act_mirred, forwarded to IFB, then
reinjected at their original location (ingress or egress) by IFB.
The IFB device cannot use skb->tc_at_ingress, because that may have
been overwritten as the packet travels from act_mirred to ifb_xmit,
when it passes through tc_classify on the IFB egress path. Cache this
value in skb->tc_from_ingress.
That field is valid only if a packet arriving at ifb_xmit came from
act_mirred. Other packets can be crafted to reach ifb_xmit. These
must be dropped. Set tc_redirected on redirection and drop all packets
that do not have this bit set.
Both fields are set only on cloned skbs in tc actions, so original
packet sources do not have to clear the bit when reusing packets
(notably, pktgen and octeon).
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Field tc_at is used only within tc actions to distinguish ingress from
egress processing. A single bit is sufficient for this purpose.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Extract the remaining two fields from tc_verd and remove the __u16
completely. TC_AT and TC_FROM are converted to equivalent two-bit
integer fields tc_at and tc_from. Where possible, use existing
helper skb_at_tc_ingress when reading tc_at. Introduce helper
skb_reset_tc to clear fields.
Not documenting tc_from and tc_at, because they will be replaced
with single bit fields in follow-on patches.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Packets sent by the IFB device skip subsequent tc classification.
A single bit governs this state. Move it out of tc_verd in
anticipation of removing that __u16 completely.
The new bitfield tc_skip_classify temporarily uses one bit of a
hole, until tc_verd is removed completely in a follow-up patch.
Remove the bit hole comment. It could be 2, 3, 4 or 5 bits long.
With that many options, little value in documenting it.
Introduce a helper function to deduplicate the logic in the two
sites that check this bit.
The field tc_skip_classify is set only in IFB on skbs cloned in
act_mirred, so original packet sources do not have to clear the
bit when reusing packets (notably, pktgen and octeon).
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This field is no longer kept in tc_verd. Remove it from the global
definition of that struct.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove the last reference to tc_verd's munge and redirect ttl bits.
These fields are no longer used.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
While it is useful to know which MDIO device is being registered, demote
the dev_info() to a dev_dbg().
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
In dev_get_stats() the statistic structure storage has already been
zeroed. Therefore network drivers do not need to call memset() again.
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The network device operation for reading statistics is only called
in one place, and it ignores the return value. Having a structure
return value is potentially confusing because some future driver could
incorrectly assume that the return value was used.
Fix all drivers with ndo_get_stats64 to have a void function.
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jeff Kirsher says:
====================
100GbE Intel Wired LAN Driver Updates 2017-01-08
This series contains updates to fm10k only.
Ngai-Mint changes the driver to use the MAC pointer in the fm10k_mac_info
structure for fm10k_get_host_state_generic(). Fixed a race condition
where the mailbox interrupt request bits can be cleared before being
handled causing certain mailbox messages from the PF to be untreated
and the PF will enter in some inactive state.
Jake removes the typecast of u8 to char, and the extra variable that was
created for the typecast. Bumps the driver version. Added back the
receive descriptor timestamp value so that applications built on top
of the IES API can function properly. Cleaned up the debug statistics
flag, since debug statistics were removed and the flag was missed in
the removal.
Scott limits the DMA sync for CPU to the actual length of the packet,
instead of the entire buffer, since the DMA sync occurs every time a
packet is received.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
ipmr_get_route has 1 caller and the nowait arg is 0. Remove the arg and
simplify ipmr_get_route accordingly.
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Because every call to octeon_flush_iq() has a hardcoded 1 for the
pending_thresh argument, simplify that function by removing that argument.
This avoids one atomic read as well.
Signed-off-by: Derek Chickles <derek.chickles@cavium.com>
Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: Satanand Burla <satananda.burla@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The debug statistics were removed due to complications with the ethtool
statistics API which are not possible to resolve without a new
statistics interface. The flag was left behind, but we no longer need
it.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This was accidentally removed when we defeatured the full 1588 Clock
support. We need to report the Rx descriptor timestamp value so that
applications built on top of the IES API can function properly.
Additionally, remove the FM10K_FLAG_RX_TS_ENABLED, as it is not used now
that 1588 functionality has been removed.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
On packet RX, we perform a dma sync for cpu before passing the
packet up. Here we limit that sync to the actual length of the
incoming packet, rather than always syncing the entire buffer.
Signed-off-by: Scott Peterson <scott.d.peterson@intel.com>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Partially revert commit 5e93cbadd3 ("fm10k: Reset mailbox global
interrupts", 2016-06-07)
The register bits related to this commit are now solely being handled by
the IES API. Recent changes in the IES API will allow an automatic
recovery from improper handling of these bits.
Signed-off-by: Ngai-Mint Kwan <ngai-mint.kwan@intel.com>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Multiple IES API resets can cause a race condition where the mailbox
interrupt request bits can be cleared before being handled. This can
leave certain mailbox messages from the PF to be untreated and the PF
will enter in some inactive state. If this situation occurs, the IES API
will initiate a mailbox version reset which, then, trigger a mailbox
state change. Once this mailbox transition occurs (from OPEN to CONNECT
state), a request for reset will be returned.
This ensures that PF will undergo a reset whenever IES API encounters an
unknown global mailbox interrupt event or whenever the IES API
terminates.
Signed-off-by: Ngai-Mint Kwan <ngai-mint.kwan@intel.com>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
We don't need to typecast a u8 * into a char *, so just remove the extra
variable.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Since a pointer "mac" to fm10k_mac_info structure exists, use it to
access the contents of its members.
Signed-off-by: Ngai-Mint Kwan <ngai-mint.kwan@intel.com>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Isolate the HWMON support in DSA in its own file. Currently only the
legacy DSA code is concerned.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Murali Karicheri says:
====================
netcp: enhancements and minor fixes
This series is for net-next. This propagates enhancements and minor
bug fixes from internal version of the driver to keep the upstream
in sync. Please review and apply if this looks good.
Tested on all of K2HK/E/L boards with nfs rootfs.
Test logs below
K2HK-EVM: http://pastebin.ubuntu.com/23754106/
k2L-EVM: http://pastebin.ubuntu.com/23754143/
K2E-EVM: http://pastebin.ubuntu.com/23754159/
History:
v1 - dropped 1/10 amd 2/10 of v0 based on comments from Rob as
it needs more work before submission
v0 - Initial version
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
For NetCP NU Switch ALE, some of the mask bits are different than
defaults used in the driver. Add a new macro DEFINE_ALE_FIELD1 that use
a configurable mask bits and use it in the driver. These bits are set to
correct values by using the new variables added to cpsw_ale structure
and re-used in the macros. The parameter nu_switch_ale is configured by
the caller driver to indicate the ALE is for that switch and is used in
the ALE driver to do customization as needed.
Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: Sekhar Nori <nsekhar@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
ALE h/w on newer version of NetCP (K2E/L/G) does provide a ALE_STATUS
register for the size of the ALE Table implemented in h/w. Currently
for example we set ALE Table size to 1024 for NetCP ALE on
K2E even though the ALE Status/Documentation shows it has 8192 entries.
So take advantage of this register to read the size of ALE table supported
and use that value in the driver for the newer version of NetCP ALE.
For NetCP lite, ALE Table size is much less (64) and indicated by a size
of zero in ALE_STATUS. So use that as a default for now. While at it,
also fix the ale table size on 10G switch to 2048 per User guide
http://www.ti.com/lit/ug/spruhj5/spruhj5.pdf
Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: Sekhar Nori <nsekhar@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In NU Ethernet switch used on some of the Keystone SoCs, there is
separate UNKNOWNVLAN register for membership, unreg mcast flood, reg
mcast flood and force untag egress bits in ALE. So control for these
fields require different address offset, shift and size of field.
As this ALE has the same version number as ALE in CPSW found on other
SoCs, customization based on version number is not possible. So
use a configuration parameter, nu_switch_ale, to identify the ALE
ALE found in NU Switch. Different treatment is needed for NU Switch
ALE due to difference in the ale table bits, separate unknown vlan
registers etc. The register information available in ale_controls,
needs to be updated to support the netcp NU switch h/w. So it is not
constant array any more since it needs to be updated based
on ALE type. The header of the file is also updated to indicate it
supports N port switch ALE, not just 3 port. The version mask is
3 bits in NU Switch ALE vs 8 bits on other ALE types.
While at it, change the debug print to info print so that ALE
version gets displayed in boot log.
Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: Sekhar Nori <nsekhar@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some of the newer Ethernet switch hw (such as that on k2e/l/g) can
strip the Etherenet FCS from packet at the port 0 egress of the switch.
So use this capability instead of doing it in software.
Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: Sekhar Nori <nsekhar@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently to parse phy-handle, driver doesn't check if the interface is
MAC to PHY. This patch add this check for all MAC to PHY interface types
supported by the driver.
Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: Sekhar Nori <nsekhar@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Previously the network statistics were stored in 32 bit variable
which can cause some stats to roll over after several minutes of
high traffic. This implements 64 bit storage so larger numbers
can be stored.
Signed-off-by: Michael Scherban <m-scherban@ti.com>
Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: Sekhar Nori <nsekhar@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The psdata is populated with command data by netcp modules
to the tail of the buffer and set_words() copy the same
to the front of the psdata. So remove the redundant memmov
function call.
Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Extract the eflag bits from the received desc and pass it down
the rx_hook chain to be available for netcp modules. Also the
psdata and epib data has to be inspected by the netcp modules.
So the desc can be freed only after returning from the rx_hook.
So move knav_pool_desc_put() after the rx_hook processing.
Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Grygorii Strashko says:
====================
net: ethernet: ti: cpsw: support placing CPDMA descriptors into DDR
This series intended to add support for placing CPDMA descriptors into
DDR by introducing new module parameter "descs_pool_size" to specify
size of descriptor's pool. The "descs_pool_size" defines total number
of CPDMA CPPI descriptors to be used for both ingress/egress packets
processing. If not specified - the default value 256 will be used
which will allow to place descriptor's pool into the internal CPPI
RAM.
In addition, added ability to re-split CPDMA pool of descriptors
between RX and TX path via ethtool '-G' command wich will allow to
configure and fix number of descriptors used by RX and TX path, which,
then, will be split between RX/TX channels proportionally depending on
number of RX/TX channels and its weight.
This allows significantly to reduce UDP packets drop rate for
bandwidth >301 Mbits/sec (am57x).
Before enabling this feature, the am437x SoC has to be fixed as it's
proved that it's not working when CPDMA descriptors placed in DDR.
So, the patch 1 fixes this issue.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Even if no_bd_ram property is described in TI CPSW bindings the support for
it has never been introduced in CPSW driver, so there are no real users of
it. Hence, remove no_bd_ram property from documentation and DT files.
Cc: 'Rob Herring <robh+dt@kernel.org>'
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The CPDMA uses one pool of descriptors for both RX and TX which by default
split between all channels proportionally depending on total number of
CPDMA channels and number of TX and RX channels. As result, more
descriptors will be consumed by TX path if there are more TX channels and
there is no way now to dedicate more descriptors for RX path.
So, add the ability to re-split CPDMA pool of descriptors between RX and TX
path via ethtool '-G' command wich will allow to configure and fix number
of descriptors used by RX and TX path, which, then, will be split between
RX/TX channels proportionally depending on RX/TX channels number and
weight. ethtool '-G' command will accept only number of RX entries and rest
of descriptors will be arranged for TX automatically.
Command:
ethtool -G <devname> rx <number of descriptors>
defaults and limitations:
- minimum number of rx descriptors is 10% of total number of descriptors in
CPDMA pool
- maximum number of rx descriptors is 90% of total number of descriptors in
CPDMA pool
- by default, descriptors will be split equally between RX/TX path
- any values passed in "tx" parameter will be ignored
Usage:
# ethtool -g eth0
Pre-set maximums:
RX: 7372
RX Mini: 0
RX Jumbo: 0
TX: 0
Current hardware settings:
RX: 4096
RX Mini: 0
RX Jumbo: 0
TX: 4096
# ethtool -G eth0 rx 7372
# ethtool -g eth0
Ring parameters for eth0:
Pre-set maximums:
RX: 7372
RX Mini: 0
RX Jumbo: 0
TX: 0
Current hardware settings:
RX: 7372
RX Mini: 0
RX Jumbo: 0
TX: 820
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The CPSW CPDMA can process buffer descriptors placed as in internal
CPPI RAM as in DDR. This patch adds support in CPSW and CPDMA for
descs_pool_size mudule parameter, which defines total number of CPDMA CPPI
descriptors to be used for both ingress/egress packets processing:
- memory size, required for CPDMA descriptor pool, is calculated basing
on number of descriptors specified by user in descs_pool_size and
CPDMA descriptor size and allocated from coherent memory (CMA area);
- CPDMA descriptor pool will be allocated in DDR if pool memory size >
internal CPPI RAM or use internal CPPI RAM otherwise;
- if descs_pool_size not specified in DT - the default value 256 will
be used which will allow to place CPDMA descriptors pool into the
internal CPPI RAM (current default behaviour);
- CPDMA will ignore descs_pool_size if descs_pool_size = 0 for
backward comaptiobility with davinci_emac.
descs_pool_size is boot time setting and can't be changed once
CPSW/CPDMA is initialized.
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use devm_ioremap() and simplify the code.
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Update cpdma_desc_pool_create/destroy() to accept only one parameter
struct cpdma_ctlr*, as this structure contains all required
information for pool creation/destruction.
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The currently processing cpdma descriptor with EOQ flag set may
contain two values in Next Descriptor Pointer field:
- valid pointer: means CPDMA missed addition of new desc in queue;
- null: no more descriptors in queue.
In the later case, it's not required to write to HDP register, but now
CPDMA does it.
Hence, add additional check for Next Descriptor Pointer != null in
cpdma_chan_process() function before writing in HDP register.
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It's observed that cpsw/cpdma is not working properly when CPPI
descriptors are placed in DDR instead of internal CPPI RAM on am437x
SoC:
- rx/tx silently stops processing packets;
- or - after boot it's working for sometime, but stuck once Network
load is increased (ping is working, but iperf is not).
(The same issue has not been reproduced on am335x and am57xx).
It seems that write to HDP register processed faster by interconnect
than writing of descriptor memory buffer in DDR, which is probably
caused by store buffer / write buffer differences as these functions
are implemented differently across devices. So, to fix this i come up
with two minimal, required changes:
1) all accesses to the channel register HDP/CP/RXFREE registers should
be done using sync IO accessors readl()/writel(), because all previous
memory writes writes have to be completed before starting channel
(write to HDP) or completing desc processing.
2) the change 1 only doesn't work on am437x and additional reading of
desc's field is required right after the new descriptor was filled
with data and before pointer on it will be stored in
prev_desc->hw_next field or HDP register.
In addition, to above changes this patch eliminates all relaxed ordering
I/O accessors in this driver as suggested by David Miller to avoid such
kind of issues in the future, but with one exception - relaxed IO accessors
will still be used to fill desc in cpdma_chan_submit(), which is safe as
there is read barrier at the end of write sequence, and because sync IO
accessors usage here will affect on net performance.
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Guillaume Nault says:
====================
l2tp: cleanup socket lookup code in l2tp_ip and l2tp_ip6
First three patches remove redundant tests and add missing "const"
qualifiers.
Fourth patch splits the conditionals found in __l2tp_ip*_bind_lookup(),
to make these functions easier to review. In the process, I found that
some corner cases were still not handled properly. So I've added the
missing tests in this patch too, because they're pretty simple and the
whole "if" statements are modified anyway.
I expect it to be easier to review this way. If not, I can split up
patch #4, post the missing tests separately to -net, and later repost
this series as pure cleanup. Just let me know.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Jason A. Donenfeld