Basic set of tree - upbringing #1
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Change-Id: If818650729fb0c01dff69c37bddaee81c9c67ed7
Signed-off-by: Park Ju Hyung <qkrwngud825@gmail.com>
SELinux doesn't actually need audit support. In fact, auditing is not particularly useful on production kernels, where minimal debugging is needed. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
/dev/random on Android cannot even properly generate a UUID; it's practically unusable. There are no programs/applications on Android that should use /dev/random. Signed-off-by: Park Ju Hyung <qkrwngud825@gmail.com> Signed-off-by: Alex Naidis <alex.naidis@linux.com>
We won't be using this with an Adreno 3xx/4xx/5xx GPU. Signed-off-by: kdrag0n <dragon@khronodragon.com> Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Qualcomm devices actually use LPM-based governors (QCOM), therefore this is a useless driver to have. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
This driver does not actually depend on CONFIG_PROFILING. Remove this false dependency. Test: compiles and boots with profiling disabled Signed-off-by: Danny Lin <danny@kdrag0n.dev>
Some kernel code currently depends on the profiling subsystem solely for its task exit notifier. The rest of the profiling subsystem is left unused. Add a simple task exit notifier implemented in kernel/exit.c to make such code work properly even when profiling is disabled. This will allow unnecessary profiling code to be disabled without breakage. Signed-off-by: Danny Lin <danny@kdrag0n.dev> Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Now that we have a simple task exit notifier system to notify uid_sys_stats when tasks exit independently of the profiling subsystem, remove this unnecessary dependency. Test: /proc/uid_cputime shows valid stats with profiling disabled Signed-off-by: Danny Lin <danny@kdrag0n.dev>
Updated for 4.14 kernel support. The purpose of Anxiety is to improve throughput on latency-sensitive tasks on embedded flash storage devices. On modern devices, embedded systems contain relatively fast storage, so spending too much time with complicated queue-shifting scheduling algorithms may end up bottlenecking the device. Similarly to the No-op I/O scheduler, Anxiety works with a FIFO queue; however, Anxiety includes basic request logic based on READ v. WRITE requests. WRITE requests are starved at a fixed ratio, defined as the `max_writes_starved` tunable. READ requests are given priority over WRITE requests due to their essentiality in determining load-time latency. ------------------TEST-------------------- SCHEDULERS Anxiety CFQ S. READ 286.22 280.74 S. WRITE 221.01 174.0 R. READ 41.35 47.89 R. WRITE 10.65 10.7 SQL INS. 1276.57 1121.25 SQL UPD. 2920.97 2978.74 SQL DEL. 2503.16 2642.3 ------------------------------------------ Tests were done using AndroBench on a Xiaomi Mi A2 on the latest 4.4.178 patch. This application has been known to be misrepresentative on I/O scores in the past. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
By scheduling the last woken task first, we can increase cache locality since that task is likely to touch the same data as before. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Some requests would like to be treated as a synchronous request. In this case, we can allow these special requests to be part of the READ FIFO list. They are treated as READ requests, inheriting their request biases. Also, READ requests are always treated as sync. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Red Hat linux states that vmstat updates can be mildly expensive. Increase from 1 second to 10 seconds. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Since we have an excess of available ram, we can bump the dirty ratios for some additional performance improvements. Keeping the dirty ratio high and the dirty background ratio low allows for less blocking writeback while still allowing most dirty pages to stay in ram when they can benefit by doing so. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
There's no sense in creating scheduler-specific methods for handling former / latter requests. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
This is mainly for readability. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
This adds a great deal of latency to block requests due to disabling preemption temporarily during statistics collection. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
This takes RCU and tasks locks to simply print debugging information, which with certain log levels will not even display. DIsable this by default. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Allow maintainers to select ASLR_FULL, ASLR_PARTIAL, or ASLR_NONE. Be warned, this commit allows users to disable ASLR entirely in the kernel, which can be a critical security issue. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
When using video rendering with vidc, CMA fills up quickly and LMKd begins killing off other tasks to free up more contiguous memory. Since we have 8 GB of RAM, lets bump our CMA to 256 MiB. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Sacrafice some bandwidth for lower latency on lossy networks. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
By disabling CONFIG_SCHED_DEBUG, the scheduler sysctl parameters can be configured kernel-side without Android init scripts overwriting our values. According to the Linux Kernel CFS documentation, the scheduling period will be sysctl_sched_latency as long as the number of running threads is less than sysctl_sched_latency / sysctl_sched_min_granularity (10 with this commit). Otherwise, the scheduling period becomes sysctl_sched_min_granularity * number_of_running_threads. On average, /proc/stat's procs_running remains at under 10 processes, excluding the occasional outliers. At this state, the scheduler will be able to provide equal timeslices in under sysctl_sched_latency ms (10ms with this commit) as long as there 10 or less processes running at any one moment. Testing has yet to be done. Do not pick unless you are willing to test. This commit may not be stable or beneficial. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
In an ideal world, Android would request only the amount of memory it needs, and nothing more or less. Unfortunately, some apps request too little memory, requiring the kernel to guess how much (if any) memory to overcommit. We could skip this process by setting overcommit_memory to 1, so applications can request up to 50% of the system memory. We will rely on lmkd to garbage collect unfreed memory. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
There is a high probability that the specific CPU that makes an I/O request may be repeated. For example, the same CPU may read the same data multiple times when reading a database. This flag forces requests to be completed on the same CPU that requested them. Additionally, this prevents a scenario where one CPU is pinned with handling multiple block requests from other CPUs. Since we have an 8 core system, we are in no shortage of available CPUs to handle block requests. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
This value is set to the number of pages in which to calculate whether or not to force writeback (likely blocking). On Android systems as of now, the page size is 4 KiB. The default value for ratelimit_pages is 32, meaning that after 128 KiB of dirtied pages, the function will be called. Let's increase this to 256, meaning that after 1024 KiB (or 1 MiB) of dirtied pages, the function will be called. Regardless, the dirty background ratio is set to 10, which is (for our size of RAM), equivelent to around 1638 KiB. In theory, we should reduce overhead by calling the writeback checking function less. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
We can more utilize our available RAM better like this. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
We don't want sched_nr_migrate to be too high, as that would impact real-time latencies as the load balancing process for SCHED_OTHER disallows IRQ interrupts. Instead of making this a constant value, let's use an amount that relates to the CPU performance of a device. Consider that a device with a very weak 4 core processor needs to load balance 32 tasks from the most busy-but-idle CPU. That balancing would take significantly longer than a device with 8 cores. On the contrary view, that same 8 core device balancing 32 tasks would have no issues with SCHED_OTHER performance, however, real-time tasks would suffer more jitter than is necessary. Let's only balance as many tasks as there are CPUs in a device for optimal SCHED_OTHER performance and SCHED_FIFO / SCHED_RR latency. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Mobile devices do not benefit from hispeed frequency boosting. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
There is little benefit to increasing CPU frequencies when we are waiting on our flash chip to finish up. I have seen commits that make iowait boosting optional, but I would like to erase all traces of the boosting as to avoid any unnecessary timestamping or overhead in the governor. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
This is not a great decision for real-time workloads. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Don't take priority over other workqueues. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
An article from 2014 (old but relevant) discussed the impact of readahead on flash storage and its effects on SQLine and DBMS performance. What was discovered was that turning off readahead could improve performance and power consumption by 70% (sequential; 60% for random). https://www.semanticscholar.org/paper/Revisiting-read-ahead-efficiency-for-raw-NAND-flash-Olivier-Boukhobza/297f3d189dcd9b099f8f6b5fbc7e304db395a683 Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
We can prevent overflowing our scheduling period by allowing double the number of round-robin tasks to get a timeslice per scheduling period. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
If we are not busy with a userspace task, quickly check for wakelocks that could be cleaned up. Otherwise, fallback to the previous counter method until the max count is reached or the current CPU becomes idle. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Also, abolish the anxiety_can_dispatch method. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Otherwise, if no requests are dispatched, the counter could be initialized to anything, causing a possible panic. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Both options in the dispath method will already check the sync and async lists to make sure they are not empty. It might not be worth checking twice. Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Signed-off-by: Tyler Nijmeh <tylernij@gmail.com>
Inkypen79
force-pushed
the
lineage-16.0
branch
from
9f7f28d
to
0fe50c4
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.