When sg_table is imported, avoid allocating memory. This will
help in reducing alloc/free overhead there by improving
the composition cycle time.

Change-Id: I1514e88f3ef7f7320719445884142f9ded43c49d
Signed-off-by: Kalyan Thota <kalyant@codeaurora.org>

Increase ahb bus vote on first frame after idle
pc because that will program the complete mdss
register space.

Change-Id: I360d43656f753c70acb51395a978998412354b64
Signed-off-by: Dhaval Patel <pdhaval@codeaurora.org>
Signed-off-by: Kalyan Thota <kalyant@codeaurora.org>

No need to process response messages from video hardware
after device went into invalid state. Processing responses
may result in use-after-free memory fault because client
might free all the resources after error.
Change-Id: Ia2c22a2740466e6368e61437aa4246927150858b
Signed-off-by: Maheshwar Ajja <majja@codeaurora.org>
Signed-off-by: Sanjay Singh <sisanj@codeaurora.org>

Enable secure camera buffer client driver on SDM670 VM.

Change-Id: I98ea9b5f94baf67b33d9e9eb68cc05f5de1ec901
Signed-off-by: Neeraj Upadhyay <neeraju@codeaurora.org>

Some use cases for compute VMs on Qualcomm devices require them
to have access to secure camera buffers (i.e. buffers that are
not writable by the HLOS VM). These buffers are mapped into the
compute VM's physical address space by the hypervisor's resource
manager, and their addresses can be queried from the resource
manager using handles that are passed directly from the HLOS VM's
camera driver.

This patch adds a driver that performs the address queries and
exposes the resulting buffers as fds that implement mmap().

Change-Id: I464dd03231de5fb6c0ce60eeb6a4cde94dd86134
Signed-off-by: Philip Derrin <philip@cog.systems>
Git-commit: 1c92a87e2750ff6f4f1c294d866474a5406e56ea
Git-repo: https://github.com/CogSystems/linux-msm


[neeraju@codeaurora: Resolve trivial merge conflicts]
Signed-off-by: Neeraj Upadhyay <neeraju@codeaurora.org>

Fix potential race between tty release and echo_delayed_work,
which could result in use after free scenario for tty struct.

Change-Id: I1ea0f2a17d1131c686b3fe387a3576f3d037a3a5
Signed-off-by: Neeraj Upadhyay <neeraju@codeaurora.org>

Merge "sched/rt: Up the root domain ref count when passing it around via IPIs" into kernel.lnx.4.9.r18-rel

* changes:
  drm/msm/sde: queue display failure work to event thread
  drm/msm/sde: enable backlight in ESD recovery case

* changes:
  qcom: hyp_core_ctl: Add a debugfs interface to set reserve CPUs
  qcom: hyp_core_ctl: Add sysfs interface to read the driver status
  qcom: hyp_core_ctl: Use hyp system call for vcpu to pcpu assignment

* changes:
  qcom: hyp_core_ctl: Add support for thermal management
  drivers: thermal: cpu_cooling: Add a notifier for max level transitions
  qcom: hyp_core_ctl: Get reserve CPUs from Hypervisor
  microvisor: update system headers to version 5.3.6m15

On sdm670, VDD_PX10 regulator is updated to be default OFF
in sleep for UFS configuration. This change updates UFS
system pm-level to 5 which ensures higher power savings by
turning link off and power down device.

Change-Id: I055c948bf00960c93c295e9191b69d383e69bd19
Signed-off-by: Sayali Lokhande <sayalil@codeaurora.org>

UFS_RESET control register offset has got updated
in the design document.

Update the offset value as per the design recomendation.

Change-Id: I55cdf0f83b3f6cd8db3fc1e0b7a9e17e5045d343
Signed-off-by: Chintan Pandya <cpandya@codeaurora.org>

This change adds pin control settings to support UFS
device reset operation for sdm670.

Change-Id: I649610b7ad8274641ad0b36fafd7dbe7b3e9dda6
Signed-off-by: Sayali Lokhande <sayalil@codeaurora.org>

commit 364f5665 upstream.

When issuing an IPI RT push, where an IPI is sent to each CPU that has more
than one RT task scheduled on it, it references the root domain's rto_mask,
that contains all the CPUs within the root domain that has more than one RT
task in the runable state. The problem is, after the IPIs are initiated, the
rq->lock is released. This means that the root domain that is associated to
the run queue could be freed while the IPIs are going around.

Add a sched_get_rd() and a sched_put_rd() that will increment and decrement
the root domain's ref count respectively. This way when initiating the IPIs,
the scheduler will up the root domain's ref count before releasing the
rq->lock, ensuring that the root domain does not go away until the IPI round
is complete.

Change-Id: I6976ed3df620a1a23d2ae7d7def98b79b44dc64b
Reported-by: Pavan Kondeti <pkondeti@codeaurora.org>
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Fixes: 4bdced5c ("sched/rt: Simplify the IPI based RT balancing logic")
Link: http://lkml.kernel.org/r/CAEU1=PkiHO35Dzna8EQqNSKW1fr1y1zRQ5y66X117MG06sQtNA@mail.gmail.com


Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Git-commit: 364f5665
Git-repo: git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git


Signed-off-by: Lingutla Chandrasekhar <clingutla@codeaurora.org>

There are possibilities of deadlock if esd trigger work is
queued to display thread. There is a corner case where
during bridge disable, display thread is waiting for esd
trigger work to complete to cancel esd thread, while esd
trigger work is blocked on display thread. To fix this
queue esd trigger work to event thread.

Below is the call stack of both threads:

ESD thread
----------
__switch_to+0xb8/0xec
kthread_flush_work+0xdc/0x150
sde_encoder_display_failure_notification+0xc0/0x1e8
_sde_connector_report_panel_dead.part.8+0x6c/0xd4
sde_connector_check_status_work+0x10c/0x1d4
process_one_work+0x168/0x468
worker_thread+0x140/0x460
kthread+0xf4/0x108
ret_from_fork+0x10/0x50
0xffffffffffffffff

Commit thread
-------------
__switch_to+0xb8/0xec
flush_work+0x44/0x68
__cancel_work_timer+0x138/0x1e0
cancel_delayed_work_sync+0x24/0x30
sde_connector_schedule_status_work.part.16+0xc0/0x114
_sde_connector_update_power_locked+0x1c4/0x224
_sde_connector_update_dirty_properties+0x80/0x100
sde_connector_helper_bridge_disable+0x24/0x100
dsi_bridge_disable+0x30/0x98
drm_bridge_disable+0x128/0x144
msm_atomic_helper_commit_modeset_disables+0x148/0x578
complete_commit+0x74/0x668
_msm_drm_commit_work_cb+0x4c/0x1c4
kthread_worker_fn+0xc0/0x1e4
kthread+0xf4/0x108
ret_from_fork+0x10/0x50
0xffffffffffffffff

Change-Id: Ifd95664113857fd5c06098a3a2f87f28c1975513
Signed-off-by: Jayant Shekhar <jshekhar@codeaurora.org>

In case of ESD recovery display disable/enable happens
without the knowledge of backlight framework. So backlight
does not change during this period. This causes issues
where panel used DCS backlight update method where backlight
is lost during ESD trigger and it is not set again after
recovery is complete. This change fixes the issue by
calling backlight update API during ESD recovery display
enable/disable case.

Change-Id: I157120688603dd0558b773bbf41996b6d7373d94
Signed-off-by: Sandeep Panda <spanda@codeaurora.org>
Signed-off-by: Krishna Manikandan <mkrishn@codeaurora.org>

The reserve CPUs are initialized via device tree. For testing
purpose provide a debugfs interface to override the default
reserve CPUs.

Change-Id: Ic2a81829212e2ed6b3129d3eac75a8618e5c7c2d
Signed-off-by: Pavankumar Kondeti <pkondeti@codeaurora.org>

/sys/devices/system/cpu/hyp_core_ctl/status file can be read to
dump the status of the driver. The example output would be
like this.

cat /sys/devices/system/cpu/hyp_core_ctl/status

enabled=1
reserve_cpus=4-5
reserved_cpus=4-5
our_isolated_cpus=4-5
online_cpus=0-7
isolated_cpus=4-5
thermal_cpus=
Vcpu to Pcpu mappings:
vcpu=0 pcpu=4 curr_pcpu=4 hyp_pcpu=4
vcpu=1 pcpu=5 curr_pcpu=5 hyp_pcpu=5

Change-Id: I1722663baa9b4108023ddff1c4079e194603b199
Signed-off-by: Pavankumar Kondeti <pkondeti@codeaurora.org>

When an original reserved CPUs is managed by thermal, a replacement
CPU is assigned as the current physical CPU. Use hypervisor system call
to notify the current physical CPU corresponding the given virtual CPU.
When the thermal condition is gone, restore the original mapping.

Change-Id: Ib219a59db51b975286185f1b43d07491ba2262f0
Signed-off-by: Pavankumar Kondeti <pkondeti@codeaurora.org>

Currently we are honoring the reservation by isolating CPUs. An offline
CPU is considered as reserved since this guest can't use it. However a
CPU can be isolated/offlined by thermal and we shouldn't reserve such
CPU for other guest. Find a replacement CPU and reserve it by isolating it.
When the original CPU is unthrottled by thermal, restore the reservation.

We should also take thermal into consideration when reservation is
not active. i.e hypervisor should be notified about the thermal
condition of a physical CPU which could be used for the other guests.
Handle this by finding a replacement CPU for the originally assigned
CPU which is throttled by thermal.

Change-Id: I4a664353faee13e1b8c7a14047b5e6d841fdf0d4
Signed-off-by: Pavankumar Kondeti <pkondeti@codeaurora.org>

Add a blocking notifier to send a notification when a CPU enters or
leaves the max cooling level state. Also add an API to retrieve
the cpumask that contains all CPUs that are in max cooling state.

Interested clients like core_ctl subscribe to these notifications
and manage CPUs accordingly.

Change-Id: I4828ed572fa965b5bfce8184ef813fddfb246cb1
Signed-off-by: Pavankumar Kondeti <pkondeti@codeaurora.org>

Hypervisor provides system call interface for getting/setting
the virtual CPU to physical CPU mappings used for the other
guests. Use of this interface to get the physical CPUs to be
reserved for the other guest when reservation is enabled.
The system call requires resource ids as arguments to refer the
virtual CPUs. Convert the driver to a platform driver to receive
these resouce ids from the device tree.

Since the reserve CPUs are obtained dynamically from the hypervisor
system calls, remove the support for obtaining the reserve CPUs from
kernel config.

This patch also limits the number of CPUs that can be reserved to
half of the possible CPUs to make sure that the guest running this
driver is not starved.

Change-Id: Icfbbc9972936926d65a457e369b192334e2725bd
Signed-off-by: Pavankumar Kondeti <pkondeti@codeaurora.org>

This patch updates the include/microvisor headers to the latest version.
Notably, the new debug and affinity system calls are defined and the
unused scheduler_suspend system call is removed.

The remainder of the changes are related type, error code and tracing
definition updates.

Change-Id: I93560e59e5e64dbf4aa7b15de01b16ce92b0a172
Git-repo:  https://github.com/CogSystems/linux-msm.git


Git-commit: b0d7b140ec7ebade7d91af32983455aa2842c1be
Signed-off-by: Carl van Schaik <carl@cog.systems>
Signed-off-by: Pavankumar Kondeti <pkondeti@codeaurora.org>

A CPU can't be left in isolated state while it is offline. The
scheduler provides sched_unisolate_cpu_unlocked() API to un-isolate
a CPU while it is going offline. Currently we are calling this API
from CPUHP_AP_ONLINE_DYN callback which gets called before the CPU
is marked offline. Because of this CPU comes out of isolation state
and soon after enters offline state. To avoid, this register for a
callback that gets invoked after CPU is dead.

Also use nocalls() version for hotplug callbacks, Since this driver
state machine is kicked from the user space and there is no need for
the hotplug notifications during driver's initialization.

Change-Id: I7eb4af2f0784f12fd314096fe23dc7600ef9094c
Signed-off-by: Pavankumar Kondeti <pkondeti@codeaurora.org>

When isolating a CPU, a check is performed to see if there is
only 1 active CPU in the system. If that is the case, the
CPU is not isolated. However this check is done before testing
if the requested CPU is already isolated or not. If the
requested CPU is already isolated, there is no need to fail
the isolation even when there is only 1 active CPU in the system.

For example, 0-6 CPUs are isolated on a 8 CPU machine. When
an isolation request comes for CPU6, which is already isolated,
the current code fail the requesting thinking we end up
with no active CPU in the system.

Change-Id: I28fea4ff67ffed82465e5cfa785414069e4a180a
Signed-off-by: Pavankumar Kondeti <pkondeti@codeaurora.org>

Enable DM verity, by enabling CONFIG_DM_VERITY, for SDM670 VM.

Change-Id: I716c766c90ea5f8b9f7aa14c9c99790781c91449
Signed-off-by: Neeraj Upadhyay <neeraju@codeaurora.org>

Add an optional arg (device_wait) to allow indefinite
wait for target dm devices. This could be required for
devices which come up asynchronously. As existing
implementation had some of the optional args dependent
on processing of required args, introduce a new function
to allow optional args pre processing, for args which
affect the processing of required args.

Change-Id: I61af9a6136cefaf5e20cc08c0c891d838ad75b7f
Signed-off-by: Neeraj Upadhyay <neeraju@codeaurora.org>

While we hotplug a CPU if any one of the callback (which is to be
called after notify_online) fails, while doing rollback notify_dead()
has not being called, as for step `notify:prepare` the flag
`skipone_err` set to true and due to which sysfs nodes don't
get released.

If we call notify_dead(), it will send CPU_DEAD notification to the
drivers and they could do their cleanup. So, there is no need to send
CPU_UP_CANCEL notification on hotplug failure. This eventually
reverts commit 5ec3ec14 ("kernel: cpu: send CPU_UP_CANCELLED
notifcation").

Change-Id: I4ba42becec25a9d92acccae28a9fc28b9d9dedc1
Signed-off-by: Mukesh Ojha <mojha@codeaurora.org>