Enable forceclear only supported for producer pipe so
avoid sending the force clear command to consumer pipes.
Signed-off-by: Ashok Vuyyuru <avuyyuru@codeaurora.org>
Change-Id: I850f360f18c6598af6914c9cd2c89cacad1ac4c9
if there are packets coming from rndis client to device
IPA driver tries to gives those packets to network driver.
If at the same time USB disconnection is in progress, there
can be a possibility of race where callback is called with NULL
private data and it can result in a undefined behavior at client
driver. Make changes to add proper checks to handle the scenario
gracefully.
Change-Id: I7b3a5731ae21362dc67c53a2de464b17a223d4ba
Signed-off-by: Praveen Kurapati <pkurapat@codeaurora.org>
Signed-off-by: Chaitanya Pratapa <cpratapa@codeaurora.org>
When USB is disconnected USB driver deregisters callback
with IPA driver. If there are exception packets coming to USB
driver at the same time there is a possiblity of race condition
where callback is de-registered and IPA driver tries to access it.
Make changes to restrict the callback access when disconnect is
in progress.
Change-Id: I85d2876847bfd0b23f19b2b81e1aab68729b2e31
Acked-by: Chaitanya Pratapa <cpratapa@qti.qualcomm.com>
Signed-off-by: Praveen Kurapati <pkurapat@codeaurora.org>
Signed-off-by: Bojun Pan <bojunp@codeaurora.org>
Sg table tables are not contiguous page due to this while copying
total sg table it failing. To avoid this copying the each sg
table individually.
Acked-by: Akshay Pandit <pandit@codeaurora.org>
Signed-off-by: Ashok Vuyyuru <avuyyuru@codeaurora.org>
scatterlist pages not unmapped properly, it was leading to
issue when mapping the same pages again. Adding changes
to store scatterlist information properly and unmap the
page with correct length.
Change-Id: If11926608f633ce1d66150b1e4b6c515adc48049
Signed-off-by: Ashok Vuyyuru <avuyyuru@codeaurora.org>
Make changes to add additional trace points for rx
and tx path. Also enhance existing traces to capture more
information.
Change-Id: I40d5bfc4080a47d986e16ac157524c341949e51e
Signed-off-by: Chaitanya Pratapa <cpratapa@codeaurora.org>
Make changes to memcopy before preload end.
Change-Id: Icc056a3bcd5b739b8165813202c87dd84e72c78a
Signed-off-by: Michael Adisumarta <madisuma@codeaurora.org>
Make changes to delete NAPI TX object when pipe setup fails.
Change-Id: Ie7124fc9f996795b1a955166a9cae198f6a57f03
Signed-off-by: Chaitanya Pratapa <cpratapa@codeaurora.org>
Ethenet driver npn phy address was same for UL/DL
pipes, currently this address was unmapped during
DL pipe disconnect where UL pipe still avtive,
move the logic to do the unmamp to the last pipe
disconnect.
Signed-off-by: Skylar Chang <chiaweic@codeaurora.org>
Ethenet driver npn phy address was same for UL/DL
pipes, currently this address was mapped twice.
Added changes to map the map npn phy address only
once.
Change-Id: Icd713943326aad38ecf8e080a73d1d6bb8f641d2
Signed-off-by: Ashok Vuyyuru <avuyyuru@codeaurora.org>
observing the shadow stack overflow due to skb is
chained using the frag_list more than 64. To reduce the
skb chained reducing the NAPI weight to 32.
Change-Id: I1164c7d8e886a1079d30c213c91827b0aa191db2
Signed-off-by: Ashok Vuyyuru <avuyyuru@codeaurora.org>
Changes done to support wdi gsi db polling based on
the polling flag whether enabled in dtsi or not.
Change-Id: Ib5559da136866dafc0b3ab3908c6073111dbab21
Signed-off-by: Armaan Siddiqui <asiddiqu@codeaurora.org>
When ssr happen in middle of egress ioctl, only tx pipe is setup
and the current state in teardown cannot handle it.
The fix is to add state for tx/rx setup separately.
Change-Id: I0eb4011ce91139acce3a619a8179d84066873953
The stats counter which saved in idr was some stack memory which
could be freed and cause memory corruption. The fix is to use
the memory allocated in heap instead.
Change-Id: Ie398b0271571fcff41cdb85de4d77d202b6552b5
Signed-off-by: Bojun Pan <bojunp@codeaurora.org>
In case if modem crashes even before uC is loaded, there is
a possibility that we hold the proxy vote twice and the clock
will not be disabled. Make changes to avoid duplicate proxy vote.
Change-Id: Id62994c79eff6868e3a491aa53963ec127ea1213
Signed-off-by: Chaitanya Pratapa <cpratapa@codeaurora.org>
Due scheduling the NAPI in suspend path observing race conditions
in updating channel polling context and pipe polling context. To
avoid these scenario scheduling NAPI only from interrupt context.
Change-Id: I8eb0438618aedb9873e6eafc269435cc67c03dd5
Signed-off-by: Ashok Vuyyuru <avuyyuru@codeaurora.org>
In low latency ping scenario clock unvoting fastly due to this
ping RTT was increased. To avoid these scenario increased the
inactivity timer to 100msec.
Change-Id: I9b11e2adbe087aa67fa2e7ed751b190abb09d0ed
Signed-off-by: Ashok Vuyyuru <avuyyuru@codeaurora.org>
Observing the race condition if channel mode configured after
channel start. To avoid these race conditions configuring
the channel mode and starting the channel.
Change-Id: I8f8ec2d9faed320539e2b18de90ab24059a5616c
IPA drvier free the qmi server hdl without notify the qmi
framework which is causing the use-after-free on QMI framework.
The fix is to notify qmi framework before freeing the qmi handle.
Change-Id: I1ec9d3efd29283fddd958561a538b2995222a53c
Signed-off-by: Bojun Pan <bojunp@codeaurora.org>
Adding checks if the rt_rule is not freed
before committing.
Change-Id: I4a2c86c24587a6a3ea7318bac438a9f27784f2f7
Signed-off-by: Michael Adisumarta <madisuma@codeaurora.org>
packet_rcv_spkt is expecting to strip out the skb_mac_header.
Part of the alloc_skb, the mac header is setup ~0 which is causing
the skb data point to wrong location.
Change-Id: Ib8c14c029145da6a779ead21fb851b26677f4d0f
Signed-off-by: Bojun Pan <bojunp@codeaurora.org>
As part of HOLB monitoring, uc will clear the HOLB discard if
it was set as part of HOLB monitoring or it can clear if
peripheral recovers just before receiving HOLB_DEL command.
Make change to set HOLB discard back if it was intended to be
configured from AP as part of disconnect scenarios.
Change-Id: I6659a62499150844c3b540fadb4ba8b9535f5e45
Signed-off-by: Chaitanya Pratapa <cpratapa@codeaurora.org>
When we have no buffer available in HW and switch to irq, a dead-loop is
observed between sending force close comamnd and replenish workqueue.
The fix is to keep in polling mode, wait for napi-poll and replenish again.
Change-Id: Ied0504564f86f41b8b5f9aa25170768027bf3391
Rate limitting the warning if client doesn't have
callback.
Change-Id: I09b02181e4c34ecf342e150de76eace9341a0c0a
Signed-off-by: Michael Adisumarta <madisuma@codeaurora.org>
ADPLv3 targets do not require status to be enabled on ODL_DPL pipe.
Make changes to disable the status for targets with IPAv4.5 and above.
Change-Id: Ica07a25a01e742928d3ecb19a88b7ee2305235ae
Signedo-off-by: Chaitanya Pratapa <cpratapa@codeaurora.org>
Correcting the sequence to dynamically change the HOLB timer.
Change-Id: Iefcac36b93f3aee037628e3b0ec81aa5f5636caa
Acked-by: Suhas mallesh <smallesh@codeaurora.org>
Signed-off-by: Michael Adisumarta <madisuma@codeaurora.org>
In holi target due uC memory constraints uC monitoring stats
not support. So disabling the sending monitor command to uC.
Change-Id: I34e319118287fcd82876344bd85ab53c50cf5651
Signed-off-by: Ashok Vuyyuru <avuyyuru@codeaurora.org>
Adding changes to configure the destination resource group for DRB-IP.
Change-Id: I41248e7e7e92a0e4635b8661d1f324be165bd2fa
Signed-off-by: Ashok Vuyyuru <avuyyuru@codeaurora.org>
The while loop in gsi_handle_irq will cause an extra register
read after handle last interrupt. This read can be the unclocked
read. The fix is to break on ch/evt control interrupt where
could potientially cause IPA clock off.
Change-Id: If4b3bb127ee66648d24b8ca7c16bd6fec42d9fcc
Signed-off-by: Bojun Pan <bojunp@codeaurora.org>
Becauase of using wrong pair of PM operation device
was not resuming if suspend fails. Adding changes to
use correct set of PM operations.
Change-Id: Ib9fdea8f93f0a6dafbd5908dc8fc2bbf4f1e0904
Signed-off-by: Ashok Vuyyuru <avuyyuru@codeaurora.org>
Make changes to use the max budget of 64 for the LAN polling.
Otherwise we will switch between Polling and Interrupt mode
even when there is continuous data.
Change-Id: I6bbf49b77d258e88044c03a33fe20bb015a55019
Signed-off-by: Chaitanya Pratapa <cpratapa@codeaurora.org>
In IPA4.9, the wrong end_init_deaggr ipa hal implementation had
been chosen. The fix is to correc the setting.
Change-Id: I63e2c50bc42b13ef9213d4b75d5c22346170a0d3
With HOLB monitoring, it is beneficial to enable drop
stats for USB and WLAN pipes. Also enable drop stats for
DPL pipe.
Change-Id: I24df7041a4056d2875f6b14d50ab52beb1459eeb
Signed-off-by: Chaitanya Pratapa <cpratapa@codeaurora.org>
Avoiding NULL pointer de-referencing for an internal variable.
Change-Id: I9fa3bc24c747c7c04d1d64b29fc63677206252e0
Signed-off-by: Michael Adisumarta <madisuma@codeaurora.org>
Removing the files copied from the kernel:
ipa.h
msm_gsi.h
ipa_fmwk.h
ipa_mhi.h
ipa_odu_bridge.h
ipa_uc_offload.h
ipa_usb.h
ipa_wdi3.h
ipa_wigig.h
and including the original files instead.
Depends-on: 3177533
Change-Id: I485f5545505bfbc4e775a75ae3d0809f9a977a97
Signed-off-by: Ilia Lin <ilial@codeaurora.org>
Currently the ipa_debugfs_print_tethering_stats()
prints values from zeroed buffer.
The ipa_get_teth_stats() has to be called only once,
and ipa_query_teth_stats() to fill the buffer.
Change-Id: I72e6db3c9f2c342e8b635341eee06795b2be9d67
Signed-off-by: Ilia Lin <ilialin@codeaurora.org>
When USB is suspended there are DL packets on consumer
channel leading to stall. to avoid the stall enabling
the holb as part of suspend and removing it in resume.
Change-Id: I08e2fe248ee675698da16b6df6aec8f109c73608
Signed-off-by: sivakanth vaka <svaka@codeaurora.org>
If in the __ipa3_stop_gsi_channel all iterafions
of the gsi_stop_channel returned GSI_STATUS_AGAIN,
channel->stop_in_proc flag has to be set,
otherwise the teardown of the ipa_mhi_suspend_internal
won't resume this channel.
For that the return value of the gsi_stop_channel
has to be forwarded (in case no other errors exist).
Additionally ipa_mhi_resume_channels should check both
IPA_HW_MHI_CHANNEL_STATE_SUSPEND state
and stop_in_proc flag.
Change-Id: I369211456fa949f8c1869b577028ff44ed4e2c95
Signed-off-by: Ilia Lin <ilialin@codeaurora.org>
PM resources are released even when there are outstanding
packets leading to data stall hence resources are released
only when there are no outstanding packets.
Change-Id: I630737edfc9cadea1d8a99bdb6bb6fcbe76615d8
Signed-off-by: sivakanth vaka <svaka@codeaurora.org>
Fixed the if condition for enabling the NAPI netdev to init the netdev
if at least one of both features is enabled.
Change-Id: I14d454bf6eaa070a9ba5bab9e0b2d2f3bd4a8cf7
Acked-by: Tal Gelbard <tgelbard@qti.qualcomm.com>
Signed-off-by: Amir Levy <alevy@codeaurora.org>
Duplicate code which is causing we call probe_cb twice to rmnet_ctl
module.
Remove the un-necessary return after pipe setup but rmnet_ctl not
registered yet.
Change-Id: I935ca6c09801ab2882a196d739f0de55a4ab3806
ipa_api provided a wrapping layer for various functions in order to
mitigate between different ipa version. This mitigation is no longer
needed, hence ipa_api is obsolete.
Change-Id: I84c3fafef0f578b86080ac8457493ced4ec442e3
Acked-by: Eliad Ben Yishay <ebenyish@qti.qualcomm.com>
Signed-off-by: Amir Levy <alevy@codeaurora.org>
Leaving this field 0 may lead to a crash during
the first gsi_poll_n_channel() call.
Change-Id: I6c8e9e1e5bacdddd41023618881d761d730a1931
Signed-off-by: Ilia Lin <ilial@codeaurora.org>
IPA HW will treat aggeragtion byte limit value 1 consider as 1K.
Adding changes to configuring the aggeragtion byte value w.r.t
IPA HW requirement.
Change-Id: Ibe9353a39971296f7e4b04f30039457c07c63033
Signed-off-by: Ashok Vuyyuru <avuyyuru@codeaurora.org>
When we get a bad peripheral event from uC when there is a pending
HOLB ADD/DEL command response from uC we will run into a deadlock
if we try to acquire mutex as part of event log. Make a change
to remove mutex lock when logging the events.
Change-Id: I60ca3ecc4fe4d090278f745bdf2bb9e1bd09ca60
Signed-off-by: Chaitanya Pratapa <cpratapa@codeaurora.org>
TX napi context is a mutable variable and so must be deleted
whenever the pipe is disconnected. Otherwise we will see a
memory leak.
Change-Id: I9309130c786e1092fd1a717359774fca9a37fcd1
Signed-off-by: Chaitanya Pratapa <cpratapa@codeaurora.org>
Make changes to delete the route entry even when the hdr/proc_ctx
handles are already freed. Otherwise the entry will be dangling.
Change-Id: Icbab6b96fa137c5214b37ea33c6203c5a54273ac
Signed-off-by: Chaitanya Pratapa <cpratapa@codeaurora.org>
Moving the defines used only internally to the local files:
ipa_i.h
ipa_common_i.h
gsi.h
Moving completely the following files from the kernel tree:
ecm_ipa.h
rndis_ipa.h
Adjusting the #include statements accordingly.
Temporary moving the following files to the techpack
as part of 3 stage internal definitions moving:
ipa.h
msm_gsi.h
ipa_uc_offload.h
ipa_mhi.h
ipa_wigig.h
ipa_wdi3.h
ipa_usb.h
ipa_odu_bridge.h
ipa_fmwk.h
On the second stage (after this change merge)
the kernel change will be merged independently.
On the third stage the local copies will be removed
from the techpack and the original files from the kernel
will be included instead.
Change-Id: If5eecc5f96a6a98075db5819f18acaec581b456e
Signed-off-by: Ilia Lin <ilial@codeaurora.org>
GSI interrupt storm can happen due to un-clocked gsi isr.
If we enable clock and assert, there is a chance the gsi isr will
be already handled.
To debug it further, the change is to cache last 20 gsi isr with
timestamp.
Change-Id: I30be03a20df9f11be7f5d8c06b049591305940bb
Signed-off-by: Bojun Pan <bojunp@codeaurora.org>
NAPI object is deleted as part of free_netdev. No need
to free it seperately
Change-Id: I20a2b6a8f73f9341030188d00dadc6acc7671c30
Signed-off-by: Michael Adisumarta <madisuma@codeaurora.org>
Fix using the right client enum when trying to add
NAPI TX function.
Change-Id: Ie45b5b72efdd76c27da854f1a22ed7e8f5853592
Signed-off-by: Chaitanya Pratapa <cpratapa@codeaurora.org>
Addressed unintialized pointers and P1 errors and
fixing them accordingly.
Change-Id: Iba70c5fdf91fffd823c3473383e687fe8b549b64
Signed-off-by: Michael Adisumarta <madisuma@codeauror.org>
Receiving the global interrupt INT1 without sending any
generic commands. So disabling interrupt by default.
Enabling the interrupt only before sending the generic command.
Change-Id: I55011933d92d29e19cc478ed1dbea9d08b95d239
Signed-off-by: Ashok Vuyyuru <avuyyuru@codeaurora.org>
To make sure detach and attach netif instead of stop
and wakeup and also to update transfer timer update.
Change-Id: I9d589b7f9f6fe98f778df509d3c16f339dfdeea1
Signed-off-by: Michael Adisumarta <madisuma@codeaurora.org>
If gsi_start_channel() or gsi_stop_channel() are called
on a channel with no event ring allocated, there will be
null pointer dereference error.
Adding null pointer check.
Change-Id: I200aa09f49c2394041f66c33c458f7daa9aeaf8c
Signed-off-by: Ilia Lin <ilial@codeaurora.org>
The ipa3_get_hw_type_index() uses the ipa3_ctx->ipa_config_is_mhi
for MHI flavor detection. Therefore the ipa3_ctx->ipa_config_is_mhi
has to be initialized before the ipa3_get_hw_type_index() call.
Change-Id: I8446c68f8b1fb0e2a56838e92f59198bd6890462
Signed-off-by: Ilia Lin <ilial@codeaurora.org>
Rate limit the warning which could cause latency on low_lat data path.
Add stats count for buffer replenish issues.
Change-Id: I8ce8c7584be211e215a05b6c5c9a1ff1b2512240
In suspend just before stopping the channel possible to receive
the IEOB interrupt and xfer pointer will not be processed in this
mode but gsi stats are updated. In resume after starting the channel
will receive the IEOB interrupt and xfer pointer will be overwritten
because of this observing the gsi/sys len stats are not matching and
buffers won't be replinished properly leading to HW stall.
To avoid this process all data in polling context, gsi/sys len
stats are updated properly.
Change-Id: Id665448165b6aa51b251cddd72573c6df64ee541
Signed-off-by: Ashok Vuyyuru <avuyyuru@codeaurora.org>
Caching gsi_ver in static global variable ipa3_res.
Added gsi_ver to struct ipa3_plat_drv_res.
Caching gsi_ver in function ipa3_post_init in gsi_props of type
struct gsi_per_props.
Change-Id: I63e286d907be8d6151ed3a19bb1ef3c0601b9dc7
Acked-by: Eliad Ben Yishay <ebenyish@qti.qualcomm.com>
Signed-off-by: Amir Levy <alevy@codeaurora.org>
Addressed NULL check and P1 errors and fixing them accordingly.
Change-Id: I1a7e03fad2bd54d4c035a783f16709be12cd5be8
Signed-off-by: Michael Adisumarta <madisuma@codeauror.org>
Defined proper include search path for the header files
in the project, and changed the "#include" lines accordingly.
Change-Id: I759177e2e2c32b2e062774d2a68d40b63a0f6906
Signed-off-by: Ilia Lin <ilial@codeaurora.org>
Processing status on every packet increases MIPS consumption
on Q6. Enable status supression to disable sending status
for every packet on UL path.
Change-Id: Ie818bcac683ee8b46e2e8a7a64e62e17ffda67ad
Signed-off-by: Chaitanya Pratapa <cpratapa@codeaurora.org>
Add default cases to avoid compilation error for unsupported
or new hw version
Change-Id: I37d8c753225a71ef6f1f28261e5938e40435fe50
Signed-off-by: Armaan Siddiqui <asiddiqu@codeaurora.org>
Default return value of netif_receive_skb is drop. Make
changes to ignore the return value to avoid printing error
for every packet.
Change-Id: I002e53c85d6bfa781240e65ee38175dae68a80bb
Signed-off-by: Chaitanya Pratapa <cpratapa@codeaurora.org>
After adding the tx_pkt_wrapper cache the allocation
was done in GFP_KERNEL while SoftIrqs are disabled
since spin_lock_bh was used.
This change releases the spinlock before allocation is done.
Change-Id: I3c3cee3e82d581c79a0258d3fcec378ed3d9a07c
Acked-by: Tal Gelbard <tgelbard@qti.qualcomm.com>
Signed-off-by: Amir Levy <alevy@codeaurora.org>
1. Fix the issue where IPA adds the fragments into
the skb in reverse.
2. Fix the size update to stack for EOB packet.
Change-Id: I24e33391859dfe735cd3258a797e2c86d90bdfd6
Netmgr requires upstream iface info whenever tethering is
enabled to toggle RSB advanced feature on/off for the
respective upstream iface. Make changes to give provision
to get upstream iface whenever tethering is enabled and
post a uevent.
Change-Id: I841e334fa9000dd5a0f0cc4fce52c5432f67cba8
Signed-off-by: Chaitanya Pratapa <cpratapa@codeaurora.org>
register the following ipa clients to ipa_framework:
ipa_wdi3
ipa_gsb
ipa_uc_offload
ipa_mhi
ipa_wigig
this comes as an addition to ipa_usb support.
Add all remaining IPA core driver API functions.
remove internal functions.
remove deprecated code from mhi_client and ipa_rm.
Change-Id: Ia56520863446c42ca7b4602f1633e0887abb0b40
Signed-off-by: Amir Levy <alevy@codeaurora.org>
Because of less NAPI budget it's need to reschedule multiple
times in high tput scenario.To avoid multiple reschedule
configuring maximum NAPI budget value.
Change-Id: Id8745b54c3909d01df69376a823396ea4c94eebb
Signed-off-by: Ashok Vuyyuru <avuyyuru@codeaurora.org>
Signed-off-by: Chaitanya Pratapa <cpratapa@codeaurora.org>
ECONNRESET is returned if Q6 QMI service is down. Make changes to
check for ECONNRESET as well to detect SSR.
Signed-off-by: Michael Adisumarta <madisuma@codeaurora.org>
Observing race condition when we are starting the channel and
modem trying to offload flow_control on same channel. Add changes
to retrun success if channel state is flow control when trying to
start the channel.
Change-Id: Id254cd0e3e17c74e931d27cb96577224559c091e
Signed-off-by: Praveen Kurapati <pkurapat@codeaurora.org>
When the RMNET pipe channel in the started state receiving the
junk packets, it leading IPA HW stall. Add changes to support
the new flow control channel state for rmnet pipe, it avoids
entering the junk packet to IPA HW.
Change-Id: If2b3992e8cdc93343e70f9d2c1783bbbde329bc2
Signed-off-by: Ashok Vuyyuru <avuyyuru@codeaurora.org>
This is a revert of a temporary change:
I178bf6fe4e17bf107f98086a9fd52f4cad484950
Was done since issues with the dts was resolved.
Change-Id: I7fec695989bee54f96c8cbe6c8f30dcce4a1e005
Acked-by: Tal Gelbard <tgelbard@qti.qualcomm.com>
Signed-off-by: Amir Levy <alevy@codeaurora.org>
When switching from USB DPL to ODL or vice versa, HOLB configuration
should not be resetted as there can be a race where HOLB set by Modem
is removed by AP.
Change-Id: I8d7feb43f2c28b3a8f150e64b5089f9d8b991545
Signed-off-by: Chaitanya Pratapa <cpratapa@codeaurora.org>
Fetch channel stats from local variable instead of
reading it from MMIO, which would have resulted in
Unclocked access.
Change-Id: I9a4eedeaa28cd52fade6ab9bea7151b702ca46d0
Signed-off-by: Michael Adisumarta <madisuma@codeaurora.org>
skb->len is still used to update the tx stats, after the skb
could be freed in tx completion context.
Change-Id: I2c490dfdbc9401613816058c1c8207ea6b6f3450
Canceling the nop descriptor timer in case we eventually
do send an EOT packet since we cross the threshold
Change-Id: Ifcf8615c19cf8ce36a2198d214c7c7ff68501e43
Acked-by: Tal Gelbard <tgelbard@qti.qualcomm.com>
Signed-off-by: Amir Levy <alevy@codeaurora.org>
ECONNRESET is returned if Q6 QMI service is down. Make changes to
check for ECONNRESET as well to detect SSR.
Change-Id: If466106ed6726aa7d60af86bdcf5e55ef03b6e2c
Signed-off-by: Chaitanya Pratapa <cpratapa@codeaurora.org>
enabled the debug support for drop packets on the basis of
pipe number.
Change-Id: Ia1811dd7a5b28bfc787ed7e28e661dfac3e41a17
Signed-off-by: Rishabh Garg <rishgarg@codeaurora.org>
SMMU mapping resources are not freed as part of wdi disconnect.
Change-Id: I702ac3bc0b5643424d051c45ec607d925f9c369d
Signed-off-by: sivakanth vaka <svaka@codeaurora.org>
Part of allocations made by the ipa_pre_init() are not freed correctly.
Fixing this by calling the relevant cleanup's.
Since the ipa_odl_init() is not the last allocation now,
the appropriate ipa_odl_cleanup added.
Change-Id: Iedbe1dff829f5bedbf788ffd89928982a498d065
Signed-off-by: Ilia Lin <ilial@codeaurora.org>
