Support flow control with IPA on the low latency channel.
Change-Id: I1044abf3b8f59f928f9df3c29c50497b64b934a6
Signed-off-by: Weiyi Chen <quic_weiyic@quicinc.com>
If a grant indication for a bearer has "removed" flag set, caching
the bearer could lead to data stall for the next data call because
modem will not send grant for this bearer during powersave.
Do not cache grant for the removed or disabled bearers. Also make sure
a bearer is flow enabled when it is associated with a tx queue.
Change-Id: I7eca597e3cc7d5a0bfe523201c454fb45e66a3a0
Signed-off-by: Weiyi Chen <quic_weiyic@quicinc.com>
When a bearer is removed, calling qdisc reset on a tx queue
could have a race condition with qdisc dequeue for lockless
qdisc such as pfifo_fast.
This change uses a different mechanism not relying on qdisc
implementation to achieve packet purge on bearer remove.
Change-Id: I8f9201809853b07293896d6cb8e010e9e0904e46
Acked-by: Weiyi Chen <weiyic@qti.qualcomm.com>
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Descriptors that came from coalesced frames will now carry usage info.
In order to prevent overcoutning, only the first segment from each frame
will report this information; the rest will report 0.
This allows a simple calculation of summing the coal_bytes and coal_bufsize
values reported and dividing to determine the total ratio over a given
period.
These values will be reported in the SKB via a new Control Block struct,
accessable via RMNET_SKB_CB(skb);
Change-Id: I8ea31921afa3afad4d2c570ec5728118bd635b86
Signed-off-by: Sean Tranchetti <stranche@codeaurora.org>
In the DFC powersave work, the separate spin lock and unlock
of multiple qos structures could be out of sync during SSR
and results in spinlock leak after exiting the work.
This change consolidated the spinlock operations to avoid
multiple locking and unlocking, and fixed below issue:
BUG: workqueue leaked lock or atomic: kworker/0:9/0x00000201/1361
last function: qmi_rmnet_check_stats_2.cfi_jt [rmnet_core]
1 lock held by kworker/0:9/1361:
(&qos->qos_lock){....}-{2:2}, at: rmnet_lock_unlock_all_flows+0xa4/0xdc
Change-Id: I10c1687a4f9993363dc631dee0b347faaa1067ab
Acked-by: Weiyi Chen <weiyic@qti.qualcomm.com>
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Add traces to track TCP and UDP non fragmented packets in uplink and downlink.
CRs-Fixed: 3002908
Change-Id: I35643c38dff564964dcd9e04518cbe4655dd050b
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
The current condition only looks for packets which have only the ACK flag set.
This can cause SYN ACKs to get stuck behind other queued data.
This potentially manifests as delayed TCP connection setup.
Update the condition to allow all control packets through.
CRs-Fixed: 3001492
Change-Id: I0baab53bf4226eb891919040ee1c37c7b17d2910
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Keep count of single fragments and other fragment counts in increments
of 6 upto the maximum possible 16.
CRs-Fixed: 2992022
Change-Id: I1b1c474e51cf1fe8856d53511ae7c374caa0b4a2
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Powersave alarm timer could fail and delay suspend for 2 seconds
if the timer expires in 2 seconds. Change to hold a wakelock for
the actual duration of the timer so the suspend can be triggered
after timer expiration without delay.
Change-Id: Icc3cb835d4e174955e128c5a0c80198513213b2d
Acked-by: Weiyi Chen <weiyic@qti.qualcomm.com>
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Keep a count in increments of 10 upto 60 and a count for greater
than 60.
CRs-Fixed: 2984565
Change-Id: I20d81e03b1a592e3e69c6e83d1a827ae0c72f6b9
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Add support for UDP_GSO and also print the address, port, ip family
and transport protocol in each trace.
CRs-Fixed: 2983027
Change-Id: If98c9423f77c0f566b004b167c2b45eca560b0d7
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
UDP GSO packets from network stack currently do not honor the
gso_max_size limits set in a driver. As a result, there needs
to be some enforcement done at the driver itself prior to the
transmit to hardware.
Instead of setting the gso_max_size on the rmnet devices, the
gso_max_size is instead set on the physical device. This
ensures that the network stack processing happens with the
maximum gso size possible for the TCP case.
CRs-Fixed: 2981039
Change-Id: I5280ea79f868e2b933f2604f8a33fbf33687f76c
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Add strings to display the cumulative result of the segmentation of
TSO packets on LL channel per rmnet device
CRs-Fixed: 2982506
Change-Id: I2454080d89c7a39a67004e0143463cd35772d45c
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
The csum_valid bit being set to zero in the v5 csum offload header has two
possible meanings. Either 1) the checksum is bad, or 2) the checksum was
not calculated. When a packet is received with such a header, we need to
manually checksum the packet to avoid reporting potentially valid packets
as having bad checksums.
Change-Id: I6a85d7a01c844be625c11c80ba381ac4dbd0366d
Signed-off-by: Sean Tranchetti <stranche@codeaurora.org>
maxtype should be set to the value of the highest attribute type that
is valid, not the number of valid attributes accepted. This ensures that
attributes with types higher than this number are ignored properly.
Change-Id: Ife53a4a9a7327a8d89e709320eb1c6cb50992922
Signed-off-by: Sean Tranchetti <stranche@codeaurora.org>
Some drivers may not set the transport header when queueing packets
to the network stack. Network stack will end up setting the
transport offset to 0 in this case.
When these packets arrive in rmnet and have to be checked for
TCP ACKs, usage of standard helpers for tcp header is not possible
as it relies on the transport header offset to be set.
CRs-Fixed: 2965067
Change-Id: I50d53b7762fd75d1b17ccc6765abe65568aaa4e0
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Break dependency between rmnet_core and rmnet_ctl
so that we can compile rmnet_core without
rmnet_ctl on yocto based targets.
Change-Id: I3bfa3dbcd24f9343d073107f40b6d98c77aba881
Signed-off-by: Conner Huff <chuff@codeaurora.org>
Add check to ensure that rmnet_port value returned
by rmnet_get_port is not null before we go ahead
and try to queue up an skb to be transmitted via it.
Change-Id: I7890f4f2bcbb8dd300957c4fb12ec77f0412f4a6
Signed-off-by: Conner Huff <chuff@codeaurora.org>
The tcp ancillary bit has been updated to be sent in almost all cases
unlike earlier where this was sent in case of TCP DL traffic.
This bit will now not be set only in case of adverse scenarios, so
this change ensures that the ACK queue state is updated in case the
ancillary bit is unset.
CRs-Fixed: 2957344
Change-Id: I4e1f26c9d3fabc64401284e36b48d82a4f3a5161
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
This provides the flexibility to use TARGET_BOARD_PLATFORM to maintain
a common codebase while differentiating between products with
TARGET_PRODUCT.
CRs-Fixed: 2955427
Change-Id: I824730142b6a68968c6bb44fec841ed49642ef7a
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Steer packets from LLC to different CPU cores for rmnet processing
by assigning a different RX queue. Also bypass rmnet_offload and
rmnet_shs for LL packets.
Change-Id: I459dabe8dd02132614f0e2cf461c89274f18223c
Acked-by: Weiyi Chen <weiyic@qti.qualcomm.com>
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
LLC acks are now sent in a workqueue context.
Change-Id: Ic162e6ad7575f9a6e73e5c96d7bc14c26c8ffe61
Acked-by: Weiyi Chen <weiyic@qti.qualcomm.com>
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
ULSO is not supported on the LL endpoint. If such skbs are received by the
rmnet driver, they must be segmented in software before transmitting them.
Change-Id: I0103d06c6bfa8eb96cfbde85f68b1b45034a93e5
Signed-off-by: Sean Tranchetti <stranche@codeaurora.org>
Allows the use of different uplink aggregation parameters for the
default and low-latency uplink aggregation states. To faciliate this,
both contexts now have their own page recycling elements instead of a
single port-wide list, as well as their own instance of the
rmnet_egress_agg_params struct.
To configure these paramters, a new netlink attribute has been created
for specifying which aggregation state the given IFLA_RMNET_UL_AGG_PARAMS
attribute should apply to. For compatibility with user space, the default
state will be chosen if this element is not provided.
Change-Id: Ia340bbb479d9427658a153a2019f4891da0b741c
Signed-off-by: Sean Tranchetti <stranche@codeaurora.org>
DL markers received over this channel need to be silently dropped, as
processing them will interfere with the standard DL marker processing on
the default channel.
Change-Id: Id6b36c3f877bf15768e3ac0a5ea8803656375a2b
Signed-off-by: Sean Tranchetti <stranche@codeaurora.org>
Allows the use of the LL channel on IPA based targetds. MHI specific
functionality is split into the new rmnet_ll_mhi.c file, and IPA is
placed in rmnet_ll_ipa.c. rmnet_ll.c works as a generic interface to the
core rmnet module, and handles calling specific functions in the active
HW module to provide the low latency channel functionality.
Change-Id: Id3e77b8433134872eba09818fc662fc109687d80
Signed-off-by: Sean Tranchetti <stranche@codeaurora.org>
Add support to switch bearers between default and lower latency
channels via QMAP commands.
Change-Id: I6662f59c713e8e3ab7409f50871bec11d9908c67
Acked-by: Weiyi Chen <weiyic@qti.qualcomm.com>
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Packets are now sent over a dedicated MHI channel when indicated by the
DFC driver.
New dedicated channel is controlled by rmnet driver. Buffers are allocated
and supplied to it as needed from a recyclable pool for RX on the channel,
and packets will be sent to it and freed manually once the channel
indicates that they have been sent.
Low latency packets can be aggregated like standard QMAP packets, but have
their own aggregation state to prevent mixing default and low latency
flows, and to allow each type of flow to use their own send functions
(i.e. dev_queue_xmit() versus rmnet_ll_send_skb()).
Low latency packets also have their own load-balancing scheme, and do not
need to use the SHS module for balancing. To facilitate this, we mark the
low latency packets with a non-zero priority value upon receipt from the
MHI chainnel and avoid sending any such marked packets to the SHS ingress
hook.
DFC has been updated with a new netlink message type to handle swapping a
list of bearers from one channel to another. The actual swap is performed
asynchronously, and separate netlink ACKs will be sent to the userspace
socket when the switch has been completed.
Change-Id: I93861d4b004f399ba203d76a71b2f01fa5c0d5d2
Signed-off-by: Sean Tranchetti <stranche@codeaurora.org>
Define the hooks to be used be perf tether ingress and egress.
CRs-Fixed: 2813607
Change-Id: I68c4cc1e73c60e784fd4117679b3a373d29f539c
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
This adds support for the new transmit offload header as well as
the handling needed for it in the core driver.
CRs-Fixed: 2810638
Change-Id: I8ce2e0772209faf3d585e7d9d8d56eceb695d586
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
RmNet would previously update the page offset and page length values
contained within each skb_frag_t in the SKB received from the physical
driver during deaggregation. This ensured that the next data to be
deaggregated was always at the "start" of the SKB.
This approach is problematic as it creates a race between the RmNet
deaggregation logic and any usespace application listening on a standard
packet socket (i.e. PACKET_RX/TX_RING socket options were not set, so
packet_rcv() is used by the kernel for handling the socket). Since
packet_rcv() creates a clone of the incoming SKB and queues it until the
point where userspace can read it, the cloned SKB in the queue and the
original SKB being processed by RmNet will refer to the same
skb_shared_info struct lying at the end of the buffer pointed to by
skb->head. This means that when RmNet updates these values inside of the
skb_frag_t struct in the SKB as it processes them, the same changes will
be reflected in the cloned SKB waiting in the queue for the packet socket.
When userspace calls recv() to listen for data, this SKB will then be
copied into the user provided buffer via skb_copy_datagram_iter(). This
copy will result in -EFAULT being returned to the user since each page in
the SKB will have length 0.
This updates the deaggregation logic to maintain the current position in
the SKB being deaggregated via different means to avoid this race with
userspace. Instead of receiving the current fragment to start checking,
rmnet_frag_deaggregate_one() recieves the offset into the SKB to start at
and now returns the number of bytes that it has placed into a descriptor
struct to the calling function.
Change-Id: I9d0f5d8be6d47a69d6b0260fd20907ad69e377ff
Signed-off-by: Sean Tranchetti <stranche@codeaurora.org>
Kbuild flag for LE based target was not being picked up
by cflags as expected, but as an unintended consequence
when rmnet_shs would try to include rmnet_trace.h,
the LE flag which wasn't used to build the rmnet_core.ko
was seen as unheard of causing compilation issue. To
get around this constraint, we are instead changing
the featurization flags in rmnet_trace.h to filter
by kernel version first for LE targets, and then
from there filtering by target name specifically.
Change-Id: I0f9a40f2ed7bfacc492cf2bb99b816c86ec710ed
Signed-off-by: Conner Huff <chuff@codeaurora.org>
These changes will help to have TRACE_INCLUDE_PATH according to
target directory structure. Target specific flag was added to detect
correct source path.
Change-Id: I04463c7c30a700f6d697a5de5df69cf9de7805ce
Signed-off-by: Mayank Vishwari <mayankvi@codeaurora.org>
Port and adjust code as needed to enable compilation
of rmnet_core.ko and rmnet_ctl.ko for taro.
Change-Id: I1ef4ea71115827f49dc3bd49aaf516eff91c2138
Signed-off-by: Conner Huff <chuff@codeaurora.org>
Ensure that the generic netlink response has a memset payload
before sending to userspace
Change-Id: Ie2fa92ce80bb3c0716e779cebeaedb2d31d759c1
Acked-by: Ryan Chapman <rchapman@qti.qualcomm.com>
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: Kaustubh Pandey <kapandey@codeaurora.org>
Include in .bb recipe file mention of kernel header files
so that we no longer need to put a relative path to include
qmi header.
Change-Id: I266e167eb5970da3cd9206d289135248ddc4f791
Signed-off-by: Conner Huff <chuff@codeaurora.org>
