We check the pool->count against the ION_POOL_LOW_MARK before deciding
to wakeup the refill threads. But as we never refill the order-0 pool,
we will keep waking up these refill threads only they goto sleep again
without doing any work. The overhead of waking up of these threads
is quite high. Fix it.
Change-Id: I0d10b96a46de24b45822ebeda1f20b75e43c2a5d
Signed-off-by: Charan Teja Reddy <charante@codeaurora.org>
[cgoldswo@codeaurora.org: Fix merge conflicts, apply change elsewhere]
Signed-off-by: Chris Goldsworthy <cgoldswo@codeaurora.org>
When performing a secure system heap allocation, and allocating from a
secure pool fails, attempt to steal pages from the uncached system heap
pools, before falling back to the buddy allocator.
Additionally, remove redundant code that was once used for zeroing out
pages allocated from buddy.
Change-Id: Id0b08f49b1fb037b26cd09e6336a0b0dc5768d82
Signed-off-by: Chris Goldsworthy <cgoldswo@codeaurora.org>
When performing secure pool allocations, it's possible to
spuriously trigger pool refill attempts for the cached and
uncached pools. Fix this by adding a check to see if we're
performing a secure allocation or not, before refilling a pool.
Change-Id: I9a415a302ca92790e6a793784020b09dea6a86e1
[cgoldswo@codeaurora.org: Fix merge conflicts]
Signed-off-by: Chris Goldsworthy <cgoldswo@codeaurora.org>
Clients may need irq number to handle doorbell, e.g. to use a common irq
handler framework.
Change-Id: Iab64a0ce3330eab2b8fa052e9d8b9b9856510e6a
Signed-off-by: Elliot Berman <eberman@codeaurora.org>
When the buffer pointed to by test_virt_addr is allocated, there
is no synchronization with respect to checking if it is necessary
to allocate the buffer. This can lead to various adversarial
scenarios, where multiple concurrent allocations of the same
buffer can occur, resulting in memory leaks.
To avoid these scenarios, make the check and allocation of the
buffer atomic with respect to other accesses to test_virt_addr.
Change-Id: Ie7e05930042b0bbf4bec0a4f1f5b325970023a5a
Signed-off-by: Isaac J. Manjarres <isaacm@codeaurora.org>
Remove this interface in preparation for a more modern replacement.
Change-Id: I84e6827167be54366bb78c6389131119ca6ccd33
Signed-off-by: Patrick Daly <pdaly@codeaurora.org>
[isaacm@codeaurora.org: resolved trivial merge conflicts]
Signed-off-by: Isaac J. Manjarres <isaacm@codeaurora.org>
Disable this Kconfig in preparation for its removal.
Change-Id: Ifff0c6d2f92b56b67b31a9c4caa8dc4d42e1ee5c
Signed-off-by: Isaac J. Manjarres <isaacm@codeaurora.org>
Enable QUPv3(Common driver, I2C, SPI, UART, I3C and GSI) drivers
support on Lahaina QGKI defconfig for perf image.
In addition also enable Slimbus satellite driver called NGD.
Change-Id: I0201cc33e98b06f56ce57e2be83a9e3466039c98
Signed-off-by: Mukesh Kumar Savaliya <msavaliy@codeaurora.org>
Enable CPUFREQ-HW for enabling cpu frequency scaling.
Change-Id: I8fadd29676063361d8d4ea2237e9ec0e94ce8c85
Signed-off-by: Naveen Yadav <naveenky@codeaurora.org>
qti_flash_led_prepare() API is used by flash clients (E.g. Camera) to
query maximum current available for flash. Add support for this API
which initially returns maximum current that is supported for flash
LED module (2000 mA). Also, add a sysfs parameter under switch devices
to show max available flash current.
Change-Id: I3aed009a0d7e0f34a6d05a45af34fcd0f37a6fe7
Signed-off-by: Shyam Kumar Thella <sthella@codeaurora.org>
Signed-off-by: Guru Das Srinagesh <gurus@codeaurora.org>
This reverts commit fe2731211e.
By default the commit muxes the pin to a GPIO upon a request_irq()
call from the client driver. This is incorrect as the client driver
could have muxed the pin already for a different purpose. Hence,
revert this default behavior to not enforce this on the clients.
Change-Id: Ib2500f57cf7e49a7bffd3309801e42166b748db9
Signed-off-by: Raghavendra Rao Ananta <rananta@codeaurora.org>
Enable the mem-buf driver on Lahaina for QGKI configurations.
Change-Id: I97fd90588130d0b18efba946ece5e72926904e20
Signed-off-by: Isaac J. Manjarres <isaacm@codeaurora.org>
Add the initial support required to share memory buffers
between VMs. This driver exposes an interface to userspace
clients to request memory to be allocated from another VM,
and added to the current VM. The VMs communicate with each
other via the use of the hypervisor message queues, as well
as utilize various facilities offered by the Haven resource
manager to manage memory that is transferred from one VM
to another.
Change-Id: I3181a1b14e61db30ed6b86752750722e24fd62e0
Signed-off-by: Isaac J. Manjarres <isaacm@codeaurora.org>
Enable the flag so that kernel panics on oops.
Change-Id: I92378f74edcfa6e6ddd1c72b387c7cf059141e13
Signed-off-by: Prakruthi Deepak Heragu <pheragu@codeaurora.org>
Add the LOCALVERSION strings for qgki, and qgki-debug
variants of Lahaina platform.
Change-Id: I95af815e9de3b906f6740b10ce4bd41de015b80d
Signed-off-by: Raghavendra Rao Ananta <rananta@codeaurora.org>
When set_rate is called on unprepared clocks new vdd_level
requirement is not cached, which leads to incorrect votes
when later clk_prepare is called. Clocks may not be prepared
when clk pre change rate is called for set_rate change request,
thus cache vdd level and only apply vote during clk prepare.
Change-Id: I54d9a3c75034cd9e67e683c3f71088f2c7cd3db1
Signed-off-by: Vivek Aknurwar <viveka@codeaurora.org>
