if a kernel device returns an I/O error, it is better not to chew
the CPU resources retrying I/O operation on a not ready device.
Fixes periodical 100%-core load in audioserver after a voice call
completion.
Change-Id: I2b66ffd18c85823a11e50598a6b4e3c0245a0083
On some devices, like SM-T805 (Galaxy Tab S 10.5 LTE), the RIL client
code produces errors "SendOemRequestHookRaw: No token" due to large
amount of tokens allocated by OEM messages requiring no response from
RIL daemon (like set call audio path, set call volume, set mite etc).
The indicator of an OEM message with no response expected is:
RegisterRequestCompleteHandler(client, REQ_XXX, NULL);
inside the RIL client message method. To prevent the token pool from
overflows, the proposed patch de-registers the token and frees its call
history if the handler for this message req_id is set to NULL.
Change-Id: Id414263bf471115797cae1a9ed628249734b9347
Exynos4 devices send two ints as a response to this request,
causing responseFailCause to think the result is a LastCallFailCause struct,
which it then tries to treat as a Java UTF-8 encoded string, causing a
crash.
Work around this by forcing responseInts (the original behaviour)
on Exynos4 devices.
Change-Id: Ied14948db3049066ffc760b6d3df09b4cefa2df0
* Instead of messing around with indices, look up
the requestNumber in the array.
* This has a cost of O(N) instead of O(1) with the
previous implementation, but we don't receive unsol
response codes frequently enough to be worried about
this.
* This was needed because a few vendor reponses, aka
RIL_UNSOL_SNDMGR_WB_AMR_REPORT at index 33 and
RIL_UNSOL_SNDMGR_CLOCK_CTRL at index 34
could not be addressed by their array indices anymore
because we cannot calculate their index by the unsol
response code we receive from the modem.
Change-Id: I27319e621c777fe19ae8908d7e0c4a46d6dd6d3b
* Adjusted whitespace to minimize diffs
* Plugged memory leaks
* Fix use-after-free
* Use calloc instead of alloca
* Reworked the Samsung request/response code to be
more readable
Change-Id: Ifd07b525fc84e8a3fca584d4d412785974ca3482
The difference between a v5 and v6 CardStatus response is the addition
of ims_subscription_app_index. In previous releases libril added a -1
to the Parcel in its place for transparent handling of GET_SIM_STATUS
everywhere else regardless of v5 or v6 responses, but it was erroneously
removed in change I4e53c6125995064b8ac0cdfc9376ef2cefe69ec8
Change-Id: I078b61d2f6dc28f738e8eb87f51193fa0bd94c7d
Get the address,sub address,Bearer with dynamic length in the data parcel,
to avoid the OutOfBoundsException with the MAX macro.
BUG=26856156
Change-Id: I1d7437143c1369d77791136b68e06088acd72937
This gerrit adds error codes for messages related
to SMS and set_mute, get_mute requests
Bug: 27038563
Change-Id: I4991e4a38c81501c39e91cfd5df90d8288f542ad
wakelock.
+ New type is added as ack from RIL java should be sent only for unsolicited mesgs
which hold wakelock in ril.cpp.
Bug: 22127517
Change-Id: Ib7e920d168ab92665283f80476bad3f5ede033e4
Modify condition to use SimRefreshResponse struct in
all RIL Versions greater than 7.
BUG=21075857
Change-Id: Ifde00434bfcaecf3b4c32392fc4964fe3b43f6b5
+ Added documentation for new RIL versioning format.
+ Changes associated to new RIl versioning format.
Bug: 25779475
Change-Id: I4e53c6125995064b8ac0cdfc9376ef2cefe69ec8
* These workarounds were introduced before http://review.cyanogenmod.org/#/c/159520/
and are no longer needed because the framework now has dynamic QAN element support.
Setting the ro.ril.telephony.mqanelements to a proper value should fix the operator
search on all the devices and these workarounds are no longer required.
* http://review.cyanogenmod.org/#/c/100398/ discards every 5th QAN element which breaks
devices with libsec-ril that returns more than 5 QAN elements. zeroflte and noblelte
return 6 QAN elements and the 5th element was being discarded instead of 6th.
* The code assumes 5 QAN elements per operator which doesn't hold true for all devices.
* In case of devices with 6 QAN elements, the size of the string array was a multiple
of 6, but since we are discarding the 5th element, the contents of the string pertaining
to each operator is a multiple of 5 following by trailing null strings in the array,
which caused issues in the framework while trying to read the strings, since we are
expecting 6 strings per operator and with a discarded string, we only receive 5.
Thanks to Javi Ferrer for diagnosing the issue on zeroflte
Change-Id: I2e99291f3438998a253755a19a063d15a19d63e0
Copying headers makes fixing compile errors hard, and breaks dependency
restrictions. Move ril_ex.h to an exported include directory instead of
copying it.
Bug: 18903857
Change-Id: I4b59a01e442db0d3198def69a5a5ff17b2a41c66
There is a buffer overflow which causing rild crash. For
appendPrintBuf, "sprintf" is replaced with "snprintf" for
the safe copy operation.
Bug:23289936
Change-Id: I25430a5c6f2a8b19a3271235323499b6ee802746
If someone needs this, it has to be fixed to read
TARGET_OUT_HEADERS (notice the extra 's').
We are only getting away with it right now because M ignores empty
include paths, but N will barf on this.
Change-Id: I7f010398a40ebaf6811a58a121590944445fbcba
xmm7260 on SLTE doesn't have this flag anymore. Devices which need it
should specify the flag.
Change-Id: I166f579274667919756d3bb90e0b182165525975
Signed-off-by: Andreas Schneider <asn@cryptomilk.org>
This is not part of the standard API, so remove it.
Samsung devices almost always use the RIL_REQUEST_DIAL_EMERGENCY_CALL
hook, so devices should implement that in their specific RIL subclass.
Change-Id: I2cf9eee00a4b5b02e998d8efcbefa87a9ca7b7b4
The modems used in recent Samsung devices use different paths
than older devices apparently.
These values were extracted from hexdumps at the correct data
positions from the device specific libsecril-client.so blobs.
Change-Id: I0547cb1de40130843fee3f1310e1e6aed95195c7
Martin Bouchet