Watch Bonds traded on NSE India - Part1

On the website of the National Stock Exchange (NSE) India, the details of various Bonds traded on the exchange are published. Anyone can view this publicly available data...

...on this page which is updated in real-time (or near real-time).
Sufficient to keep an eye on numerous bonds traded on the exchange.

 

However, unlike the stock of a company that trades under single symbol on the exchange, bonds of a single company often have multiple series, each with their different yields and maturity-dates. Thus, even if one is interested in a single company, often one would need to monitor dozens of bonds of the same company to find one that has the desired yield and is also currently being traded.

To simplify this workflow, we can scrape the above NSE webpage and filter the bonds of interest to us. Here's one way we can do that using simple Python code...   

Running the above script nse-bond-watch.py
on a system that has access to the internet,
we receive the data in CSV and JSON formats.

In subsequent posts in this series,
we will explore how we can maintain this data locally, as well as filter it.

Bottom Halves on Linux

Bottom halves on Linux

git gc, loose and packed git objects

Intro:
Recently during lunch, a bunch of us started discussing git internals. One popular point of contention that came-up was how git stores the incremental changes internally?

The Question:

Does the .git subdirectory contain:
- compressed diffs that apply on the original version of each file?
    OR
- compressed copy of each version of each modified file?

The Answer:
Different blogs appeared to claim differently [1] [2], and we were too lazy to go look-up the definitive source - the source-code of git.

Warning: The following video is processed in a facility that also processes nuts.

 Contains traces of the 60's Batman TV show.

The Conclusion:

Another mystery solved. git gc to the rescue...

References:
[1] https://codewords.recurse.com/issues/two/git-from-the-inside-out
[2] https://schacon.github.io/gitbook/1_the_git_object_model.html
[3] https://schacon.github.io/gitbook/7_how_git_stores_objects.html
[4] https://schacon.github.io/gitbook/7_the_packfile.html

Meanwhile... having solved the riddle,
the caped-crusader and the wonder-boy went to Gotham city's garbage collection facility
in the Bat-mobile to foil whatever plan the Riddler's hatching...

Git prevision

Need to quickly check a previous version of a specific file?...
Do not want to switch the entire repository to an older commit and back?...

A combination of git aliases/awk/shell-functions to the rescue

Here is a quick and intuitive way to checkout a older version of a single file in git.

Basically the command does a git log on the specified file and picks the appropriate commit-id in the history of the file and then runs git checkout to the commit-id for the specified file.

Essentially, all that one would manually do in this situation, wrapped-up in one beautiful, efficient git-alias - git-prevision

Liked git-prevision? Help others discover git-prevision.
Upvote git-prevision on StackOverflow.

Use-Case is Everything

The importance of a proper requirements gathering process, including detailed use-cases at the beginning of the software development process is often underestimated.

Common problems like feature-bloat, schedule-overruns, customer dissatisfaction are easily avoided by mandating during the requirements gathering stage, the preparation of an artifact containing a comprehensive list of detailed use-cases of the final system.

  

Download the complete slide-deck - Use-Case is Everything.

MAX_JIFFY_OFFSET: msleep() beyond infinity

Ever wondered what would happen if a driver in the Linux kernel were to invoke Buzz Lightyear's oft-quoted catchphrase?

If not then today is your lucky day, you can ponder over this million-dollar(price of Lightyear's left wing-nut :P) question safe in the comfort of the fact that the definitive answer is only as far as a couple of clicks of your mouse-wheel.

NOTE: This article discusses the intricacies of the implementation of msleep(), specifically the conversion between milliseconds and jiffies within the Linux kernel. For details on implementing blocking calls and other such constructs that wait indefinitely, refer to Section 6.2 Blocking I/O of LDD3 here or here.

First a few numerical constants relevant to this discussion
Largest unsigned 32bit number = ULONG_MAX         = 0xFFFFFFFF
Largest signed 32bit number   = LONG_MAX          = 0x7FFFFFFF
Largest jiffies defined       = MAX_JIFFY_OFFSET  = 0x3FFFFFFE

A "jiffy" is defined as 1/HZ seconds. On a typical system with HZ=100, 1Jiffy works out to be 10ms in terms of real world units of time.

MAX_JIFFY_OFFSET = 0x3FFFFFFE jiffies

= 1073741822jiffies

= ~10737418seconds

= ~2982 hrs

= ~124days

= ~17weeks

msleep() uses msecs_to_jiffies(), which relies on MAX_JIFFY_OFFSET as the definition of infinite sleep timeout.

Now consider the following input/output map of the msec_to_jiffies() for the entire range of 32bit unsigned int.

As is apparent, msecs-jiffies mapping is mostly linearly. Starting from 0, for increasing values of input, msec_to_jiffies() returns correspondingly larger values. However once the input msecs exceeds LONG_MAX, the output jiffies is clamped to MAX_JIFFY_OFFSET.

Its obvious that, for certain values of input to msecs_to_jiffies() the output result exceeds its definition of "infinity" i.e. exceeds MAX_JIFFY_OFFSET. In fact, this happens for a quarter of the range of 32bit unsigned int! Pretty brain-dead, you say, eh?

To be fair, the patch that introduced this definition of MAX_JIFFY_OFFSET is perfectly fine as it solves a genuine practical problem.
commit 9f907c0144496e464bd5ed5a99a51227d63a9c0b
Author: Ingo Molnar
[PATCH] Fix timeout overflow with jiffies
Prevent timeout overflow if timer ticks are behind jiffies
(due to high softirq load or due to dyntick),
by limiting the valid timeout range to MAX_LONG/2.

So what can be done for msecs_to_jiffies()?

Instead of clamping negative values to MAX_JIFFY_OFFSET
502         /*
503          * Negative value, means infinite timeout:
504          */
505         if ((int)m < 0)
506                 return MAX_JIFFY_OFFSET;

Should values larger than (MAX_LONG/2) be clamped to MAX_JIFFY_OFFSET?
502         /*
503          * Prevent timeout overflow if timer ticks are behind jiffies
504          * by limiting the valid timeout range to MAX_LONG/2.
505          */
506         if (m > MAX_JIFFY_OFFSET)
507                 return MAX_JIFFY_OFFSET;

This update follows the spirit of the patch that introduced the current definition of MAX_JIFFY_OFFSET. Heck, even the comment is borrowed from it!

To answer the original question, any invocation of  msleep(N)
where MAX_JIFFY_OFFSET < N < LONG_MAX
technically sleeps for longer than "infinity" as defined in the context of msleep().

For the technically inclined, a parting question -

Q. With the current implementation in the Linux kernel,
under what circumstances will msleep(10) return sooner than 10ms?

Internet Radio : Part1 - Shoutcast Protocol

NOTE: If you spent the last decade 100miles below the surface of the earth studying the growth of algae under an antarctic glacier, then you will be surprised to learn that we can now listen to radio over the internet. Just like tuning-in to particular frequencies for particular stations in the olden days; now we can "tune-in" to specific radio stations over the internet. Without further ado, lets jump-in into the world of "Internet Radio".

Most apps claiming to support Internet Radio, in fact support a industry standard - Shoutcast. It was a protocol devised by nullsoft in the 90's and first implemented in their popular player Winamp(stop reading if you haven't heard of Winamp!). Inspite of being a proprietary protocol with not much documentation to go with, Shoutcast has become the de-facto industry standard for streaming audio. This is mainly due to its simplicity and similarity with the existing hyper-text transfer protocol (dear old http! wink wink). Icecast is a similar open-source implementation compatible with Shoutcast.

Initial handshake between Shoutcast client-server

High-level overview of Internet-radio over Shoutcast.

[STEP1] Station Listing

The client app connects to a station listing/aggregator on the internet and obtains a list of stations alongwith their details like genres, language, now-playing, bitrate among other things.

[STEP2] Station Lookup

The user can then select one of the stations as desired. Then the client obtains the ip-address(& port) of the server running that particular station from the station-listing/aggregator. Networking enthusiasts will notice that this step is exactly like a DNS lookup i.e. the client obtains the network address for a particular station name; the station-listing/aggregator acting like a DNS-server for Radio stations. Also note that sometimes the station-listing will provide only a domain-name and then an additional actual DNS lookup is needed to obtain the ip-address of the streaming server. Popular station-listing/aggregator sites like Xiph, Shoutcast.com and vTuner provide huge web-friendly lists of live radio stations.

[STEP3|4] Station Connection

1. The client attempts to connect to the server using the ip-address(and port) obtained during station lookup.

Connection request from shoutcast client (click to enlarge)

2. The server responds with "ICY 200 OK" (a custom 200 OK success code)...

ICY 200 OK reply from shoutcast server (click to enlarge)

3. ...and the stream header...

Shoutcast stream header (click to enlarge)

4. ..and finally the server starts sending encoded audio in a continuous stream of packets(which the client app can decode and playback) until the client disconnects(stops ACK-ing and signals a disconnect).

Encoded audio data stream (click to enlarge)

Download the entire WireShark capture of packets exchanged by the shoutcast client and server during initial station connection.

 The above steps are similar to what a browser does when it connects to a website (and hence in-browser streaming audio playback of shoutcast streams IS possible).

Shoutcast has subtle differences over http during the station connection step above. Shoutcast supports "Icy-MetaData" - an additonal field in the request header. When set, its a request to the shoutcast server to embed metadata about the stream at periodic intervals(once every "icy-metaint" bytes) in the encoded audio stream itself. The value of "icy-metaint" is decided by the shoutcast server configuration and is sent to the client as part of the initial reply.

Shoutcast stream format when ICY:MetaData is set to 1

This poses a slight complication during playback. If the received audio stream is directly queued for playback, then the embedded metadata appears as periodic glitches. Following is one such sample recording. This audio clip was retrieved from a radio stream whose icy:metaint = 32768; i.e. the metadata is embedded in the audio stream once every 32KBytes. Stream bit-rate is 4KBps. So during playback a glitch is present once every 32KB/4KB = 8seconds (0:08s, 0:16s, 0:24s, 0:32s,...).



To view/analyse the stream data in a hex editor, download the actual clip and check out the following offsets 

Update: Unfortunately the service i was using to host the audio clip has lost it and i was foolish enough to trust them and have no local backups. :(

R.I.P
Shoutcast-Metadata.mp3

2013-2014
Here lies a song, cut short in its prime...

[0:08s] 0x0815A - 0x0817A count N = 2, meta = 1+ (16 x 2) = 33(0x21h)bytes
[0:16s] 0x1017B - 0x1017B count N = 0, meta = 1byte
[0:24s] 0x1817C - 0x1817C count N = 0, meta = 1byte
[0:32s] 0x2017D - 0x2017D count N = 0, meta = 1byte

Embedded metadata from 0x0815A to 0x0817a.
Note the first byte is 02 i.e. metadata is 2x16=32(0x20h)bytes following it.

Also note that the first 345(0x159h)bytes of the clip are the reply header of the stream(plain-text in ASCII) sent by the shoutcast server. Technically these are NOT part of the audio stream as well.

NOTE: If you simply want to obtain the audio stream (no embedded metadata) then set the "Icy-MetaData" field in the request header to 0 or simply do NOT pass it as part of the initial request header.

Finally here is a small bit of code that implements all that we have learnt so far - a simple shoutcast client in a few lines of C, that connects to any shoutcast server and logs the audio stream data to stdout. It uses the curl library to initiate connection requests to the shoutcast server.

https://gist.github.com/TheCodeArtist/2f1b9fa68197e39ca9bc
Stripping off the comments and the clean-up code following line:50, it comes down to 13 lines of C code. Pretttty neat eh?...

Usage:
$> sudo apt-get install libcurl4-gnutls-dev
$> gcc simple.c -o simple -l libcurl
$> ./simple <shoutcast-server-ip-addr:port> > <test-file>

After running the above commands, the <test-file> will contain the audio stream of that particular internet radio station. The can be played back in any player that supports decoding the stream format(AAC, MP3, OGG etc. depending on the radio station) Make sure to comment out line 38 in simple.c to have a glitch-free(no embedded metadata) audio stream.

This concludes part 1 of the series on how internet radio works. In part2 we will analyse the challenges and issues faced during de-packetising, parsing and queuing the audio stream buffers for local playback. Stay tuned for updates.

5 ways to improve HDD speed on Linux

LINUX LINUX LINUX LINUX

LINUX LINUX LINUX LINUX

(If you still think this post is about making windows load faster, then press ALT+F4 to continue)

Our dear Mr. Client

Its a fine sunny sunday morning. Due tomorrow, is your presentation to a client on improving disk-I/O. You pull yourself up by your boots and manage to climb out of bed and onto your favorite chair...

I think you forgot to wear your glasses...

Aahh thats better... You jump into the couch and turn on your laptop and launch (your favorite presentation app here). As you sip your morning coffee and wait for the app to load, you look out of the window and wonder what it could be doing.

Looks simple, right? Then why is it taking so long?

If you would be so kind enough to wipe your rose-coloured glasses clean, you would see that this is what is ACTUALLY happening

0. The app (running in RAM) decides that it wants to play spin-the-wheel with your hard-disk.

1. It initiates a disk-I/O request to read some data from the HDD to RAM(userspace).

2. The kernel does a quick check in its page-cache(again in RAM) to see if it has this data from any earlier request. Since you just switched-on your computer,...

3. ...the kernel did NOT find the requested data in the page-cache. "Sigh!" it says and starts its bike and begins it journey all the way to HDD-land. On its way, the kernel decides to call-up its old friend "HDD-cache" and tells him that he will be arriving shortly to collect a package of data from HDD-land. HDD-cache, the good-friend as always, tells the kernel not to worry and that everything will be ready by the time he arrives in HDD-land.

4. HDD-cache starts spinning the HDD-disk...

5. ...and locates and collects the data.

6. The kernel reaches HDD-land and picks-up the data package and starts back.

7. Once back home, it saves a copy of the package in its cache in case the app asks for it again. (Poor kernel has NO way of knowing that the app has no such plans).

8. The kernel gives the package of data to the app...

9. ...which promptly stores it in RAM(userspace).

Do keep in mind that this is how it works in case of extremely disciplined, well-behaved apps. At this point misbehaving apps tend to go -

"Yo kernel, ACTUALLY, i didn't allocate any RAM, i wanted to just see if the file existed. Now that i know it does, can you please send me this other file from some other corner of HDD-land."

...and the story continues...

Time to go refill your coffee-cup. Go go go...

Hmmm... you are back with some donuts too. Nice caching!

So as you sit there having coffee and donuts, you wonder how does one really improve the disk-I/O performance. Improving performance can mean different things to different people:

1. Apps should NOT slow down waiting for data from disk.
2. One disk-I/O-heavy app should NOT slow down another app's disk-I/O.
3. Heavy disk-I/O should NOT cause increased cpu-usage.
4. (Enter your client's requirement here)

So when the disk-I/O throughput is PATHETIC, what does one do?...

5 WAYS to optimise your HDD throughput!

1. Bypass page-cache for "read-once" data.

What exactly does page-cache do? It caches recently accessed pages from the HDD. Thus reducing seek-times for subsequent accesses to the same data. The key here being subsequent. The page-cache does NOT improve the performance the first time a page is accessed from the HDD. So if an app is going to read a file once and just once, then bypassing the page-cache is the better way to go. This is possible by using the O_DIRECT flag. This means that the kernel does NOT considered this particular data for the page-cache. Reducing cache-contention means that other pages (which wouold be accessed repeatedly) have a better chance of being retained in the page-cache. This improves the cache-hit ratio i.e better performance.


void ioReadOnceFile()
{
/*  Using direct_fd and direct_f bypasses kernel page-cache.
 *  - direct_fd is a low-level file descriptor
 *  - direct_f is a filestream similar to one returned by fopen()
 *  NOTE: Use getpagesize() for determining optimal sized buffers.
 */

int direct_fd = open("filename", O_DIRECT | O_RDWR);
FILE *direct_f = fdopen(direct_fd, "w+");

/* direct disk-I/O done HERE*/

fclose(f);
close(fd);
}

2. Bypass page-cache for large files.

Consider the case of a reading in a large file (ex: a database) made of a huge number of pages. Every subsequent page accessed get into the page-cache only to be dropped out later as more and more pages are read. This severely reduces the cache-hit ratio. In this case the page-cache does NOT provide any performance gains. Hence one would be better off bypassing the page-cache when accessing large files.

void ioLargeFile()
{
/*  Using direct_fd and direct_f bypasses kernel page-cache.
 *  - direct_fd is a low-level file descriptor
 *  - direct_f is a filestream similar to one returned by fopen()
 *  NOTE: Use getpagesize() for determining optimal sized buffers.
 */

int direct_fd = open("largefile.bin", O_DIRECT | O_RDWR | O_LARGEFILE);
FILE *direct_f = fdopen(direct_fd, "w+");

/* direct disk-I/O done HERE*/

fclose(f);
close(fd);
}

3. If (cpu-bound) then scheduler == no-op;

The io-scheduler optimises the order of I/O operations to be queued on to the HDD. As seek-time is the heaviest penalty on a HDD, most I/O schedulers attempt to minimise the seek-time. This is implemented as a variant of the elevator algorithm i.e. re-ordering the randomly ordered requests from numerous processes to the order in which the data is present on the HDD. require a significant amount of CPU-time.

Certain tasks that involve complex operations tend to be limited by how fast the cpu can process vast amounts of data. A complex I/O-scheduler running in the background can be consuming precious CPU cycles, thereby reducing the system performance. In this case, switching to a simpler algorithm like no-op reduces the CPU load and can improve system performance.

echo noop > /sys/block/<block-dev>/queue/scheduler


4. Block-size: Bigger is Better

Q. How will you move Mount Fuji to bangalore?
Ans. Bit by bit.

While this will eventually get the job done, its definitely NOT the most optimal way. From the kernel's perspective, the most optimal size for I/O requests is the the filesystem blocksize (i.e the page-size). As all I/O in the filesystem (and the kernel page-cache) is in terms of pages, it makes sense for the app to do transfers in multiples of pages-size too. Also with multi-segmented caches making their way into HDDs now, one would hugely benefit by doing I/O in multiples of block-size.

Barracuda 1TB HDD : Optimal I/O block size 2M (=4blocks)

The following command can be used to determine the optimal block-size
stat --printf="bs=%s optimal-bs=%S\n" --file-system /dev/<block-dev> 
 

5. SYNC vs. ASYNC (& read vs. write)

ASYNC I/O i.e. non-blocking mode is effectively faster with cache

When an app initiates a SYNC I/O read, the kernel queues a read operation for the data and returns only after the entire block of requested data is read back. During this period, the Kernel will mark the app's process as blocked for I/O. Other processes can utilise the CPU, resulting in a overall better performance for the system.

When an app initiates a SYNC I/O write, the kernel queues a write operation for the data puts the app's process in a blocked I/O. Unfortunately what this means is that the current app's process is blocked and cannot do any other processing (or I/O for that matter) until this write operation completes.

When an app initiates an ASYNC I/O read, the read() function usually returns after reading a subset of the large block of data. The app needs to repeatedly call read() with the size of data remaining to be read, until the entire required data is read-in. Each additional call to read introduces some overhead as it introduces a context-switch between the userspace and the kernel. Implementing a tight loop to repeatedly call read() wastes CPU cycles that other processes could have used. Hence one usually implements blocking using select() until the next read() returns non-zero bytes read-in. i.e the ASYNC is made to block just like the SYNC read does.

When an app initiates an ASYNC I/O write, the kernel updates the corresponding pages in the page-cache and marks them dirty. Then the control quickly returns to the app which can continue to run. The data is flushed to HDD later at a more optimal time(low cpu-load) in a more optimal way(sequentially bunched writes).

Hence, SYNC-reads and ASYNC-writes are generally a good way to go as they allow the kernel to optimise the order and timing of the underlying I/O requests.

There you go. I bet you now have quite a lot of things to say in your presentation about improving disk-IO. ;-)

PS: If your client fails to comprehend all this (just like when he saw inception for the first time), then do not despair. Ask him to go buy a freaking-fast SSD and he will never bother you again.

More on How data-barriers affect HDD Throughput and Data-integrity.

Omnivison ov3640 i2c sccb

TASK : Write a device-driver for Omnivision ov3640 camera

Timeline : A.S.A.P (is there any other way? :P)

For the aptitude champs out there, here is a quick one:

Q. If you are sitting facing west and running I2C at 0.00000000055kbps and a bear appears in front of you, what color is the bear?

Not sure? Read on...

DAY 1 : Initial study

A bit about ov3640: The ov3640 (color) image sensor is a 1/4-inch 3.2-megapixel CMOS image sensor that is capable of QXGA(2048x1536)@15FPS using OmniPixel3™ technology in a small footprint package. It provides full-frame, sub-sampled, windowed or arbitrarily scaled 8-bit/10-bit images in various formats via the control of the Serial Camera Control Bus (SCCB) interface or MIPI interface. It supports both a digital video parallel port and a serial MIPI port.

Searching the "internets", an old "v4l2-int" styled driver for ov3640 is available for the linux-kernel. This will have to do for now. Can scavenge the camera configuration register-settings from it.

The Omnivision ov3640 product-brief contains the following functional block-diagram of ov3640:

The camera is controlled using the SCCB bus. Again back to the "internets". SCCB is an i2c-clone i.e a two-wire serial protocol that has significant differences from I2C to merits its own specification.

1. According to spec, SCCB supports only upto 100Khz (not more).
2. I2C spec requires pullups with open-collector(drain) drivers everywhere. SCCB requires CMOS-like drivers which are always either +VDD or GND i.e no pullups.
3. In I2C, after every 8bits transferred, the 9th bit is designated ACK. The slave pulls SDA low to ack. SCCB designates the 9th bit "dont-care". SCCB spec states that the master continues regardless of ACK/NACK in the 9th bit.

DAY 2 : First attempt

Following the omnivision product-brief and the datasheet, the ov3640 camera-module is connected with the CPU as follows:

So far SCCB looked to be a simpler less restrictive version of I2C. Having worked extensively on I2C previously, was under the impression that setting up the basic comunication between the CPU and ov3640 would be a walk in the park. Wrote a simple skeleton i2c-driver and registered it with the kernel. Scavenged the I2C read/write routines from the old ov3640 driver and booted-up the device...

...and the driver failed to load as I2C-read failed. The ID register of ov3640 did NOT match the expected ID. Inserting logs in the code showed that the I2C-read routine was failing. The CPU was NOT getting an ACK from the ov3640 sensor. A true WTF moment as the I2C routines in the driver were tested to be working properly in earlier devices.

Oh well, maybe should really not expect an ACK i suppose. What with ov3640 being an SCCB device and not I2C. Started digging into the I2C core driver. found a provision to ignore this NACK(absense of ACK from slave). Updated the ov3640 driver to set the IGNORE_NACK flag and tried again. Now the I2C-read routine completed successfully despite there being no ACK from the slave. But still the driver failed to load. Turns out the contents of the ID register, read over I2C, did NOT match the expected value. The I2C-read routine was returning the contents of the ID register as "0". Further debugging showed that attempting to read any register of ov3640 over I2C gave the same result- a nice big ZERO. It was evident now that something was terribly wrong.

DAY 3 : Challenge Accepted

Time to bring out the big guns. Switched to multimeter and oscilloscope. Tested the lines from CPU to the ov3640 connecters for proper continuity. Booted-up the device and probed the I2C lines. The master was sending in the right values alright. But ov3640 was simply not responding. Suspect no.1 ? the I2C slave-id.

Ov3640 spec mentions that it responds to 2 I2C slave-IDs. 0x78 & 0x79. Had tried 0x78 so far. 0x79 also makes no difference - still no data from ov3640. Further digging through the docs i find one interesting line which mentions that the addresses are special in the sense that 0x78 is used to write and 0x79 to read the device. Hmmm... interesting. Lokks like these are 8bit addresses including the read/write bit of I2C. Which means the actual device slave-id is just the 7MSBs (common to 0x78 & 0x79) i.e. 0x3C. Face-palm!

Changed the slave-id of the ov3640-driver and booted-up the device, but still no dice. It would be easier to light a fire with 2 stones and twig.

DAY 4 : ...and let there be light

Lost all hopes of getting this to work. Swapped other camera modules. Tried a couple of other boards. But the ov3640 just does not seem to respond to anything. It is as if the module is not even powered-on.

Maybe, mayyyybe thats wat it IS!

Back to the schematics.

I2C-CLK? check.

I2C-DATA? check.

CAM_XCLK? check.

CAM-IO? check.

CAM-digital? check.

CAM-Analog? Do we really need to power the sensor array at this stage?

Well what the heck nothing else seems to work anyway. Might as well try this. So quickly pulled down a line from an existing 3.3V power-rail on the board. Placed a diode along it to drop it down a bit and powered-on the board.

And VOILA! It worked. The driver was able to read the ov3640 module properly.

The ov3640 even responded to the default settings (QXGA@15FPS). Pretty neat eh?

Oh well... sure makes me look foolish, now that it works. :-)
Ah well there's always the first time for everything. ;-) ;-)

And now that it works, was able to summarise the following

Hardware connections:

• CAM-Analog (2.8V for powering-on the module)
• CAM-IO 1.8V (1.8V for i2c-communication)
• CAM-Digital (1.5V generated by module)
• I2C_CLK (1.8V 400Khz-MAX for i2c-communication)
• I2C_DATA (1.8V bi-directional)
• CAM_XCLK (24Mhz reqd. for internal PLL)
• CAM_PCLK (generated by module)

Software configurations:

• I2C slave-id 0x3c
• ov3640 DOES provide an ACK (its I2C, NOT sccb)
• Works on I2C@400KHz

With the above specs, surely this begs the question that why does someone go all the way to define their own bus-specs when the hardware obviously works on I2C!!!! WHY Omnivision? WHY????

Designing you own serial-bus? == $1,000,000

Not using it in your own products? == $0

The smile on my face when i finally figure it out? PRICELESS

Somethings, money can't buy. For everything else , there is... Ah wait, i'm forgetting something now, right? Well here goes...

Q. If you are sitting facing west and running I2C at 0.00000000055kbps and a bear appears in front of you, what color is the bear?

Ans. If it takes 4 days to transfer a byte, do you REALLY think i care!!

NOTE:

[i]. No bears were harmed in the development of this camera-module.

[ii]. The image captured above did NOT appear out of the blue with only the driver in place. Several days of tweaking the exposure/white-balance settings and an earthquake later, managed to get the Kernel-driver, Android camera-HAL and app to work together.

Android Double-buffering, Page-Flip and HDMI

(a.k.a The case of the disappearing charging-icon)

The following is an account of the final development stages of an android phone, which for obvious reasons will not be named. The bug in itself was a very simple matter and the consequent fix too. But the entire process of discovering what exactly was happening was quite fun. (Ya right! tell that to my manager :P)

The android phone, i was working on, supported connecting an external-display/TV via MHL(Mobile-HD link i.e. HDMI-over-USB). Once connected, the entire UI would be displayed on the TV. The phone battery would also charge while connected.

The issue in question was initially raised as an application issue. It so happened that the charging status was not being displayed properly on the lockscreen. 

I. The issue...

With the device locked, when a external MHL-cable was connected to the android phone, it used to update the charging-icon. But if removed immediately, the charging-icon would continue to be displayed. This would not happen always. But sometimes even upto a minute after the cable was removed, the status would continue to be displayed as "charging".

The application developers banged their heads over the weekend and finally pushed the issue onto the underlying kernel drivers, stating that they were updating the charging-status as-&-when they get an update from the framework, which in turn depends on the fuelguage/battery-driver to obtain the battery-status.

 

It was now the kernel developers turn to use the "stress-reduction kit". After hours of logging almost every single instruction in every single interrupt routine, it was quite evident that the battery driver was not at fault. It was promptly reporting the connect/disconnect events when(and only when) an MHL cable was inserted/removed. The android framework was getting the events and eventually the lockscreen application too.

So now the question was that if EVERYTHING was working as it was supposed to, why the charging-status was not being displayed correctly?

II. The peculiar observation...

The peculiar thing about the disappearing charging-icon was that it was almost never for the same amount of time. Every time we tested it by plugging-in the cable, if it would disappear, it would do so for varying periods of time and then appear again onscreen.

III. What it meant...

We finally got onto the right track after we saw that the icon ALWAYS re-appeared onscreen just as the clock on the lock-screen updated itself. As it turned out the culprit was the display driver. When plugging in the MHL cable, there was some amount of tinkering going on in the background to handle the multiple displays and/or switch to the secondary(external HDTV over MHL)  from the primary(mobile-LCD). As is the norm, the display was double-buffered to improve performance and prevent onscreen flickering and tearing. Plugging-in the MHL-cable just as the display driver was initiating a swapbuffer() (i.e. a page-flip operation to pick the back-buffer to display onscreen) the device would then initiate another swapbuffer() which meant the stale buffer was displayed onscreen. to add to the misery the "smart" display driver was programmed to skip redundant swapbuffer() calls. i.e. unless the display contents had changed from the time the previous call to swapbuffer() it would not refresh the display unnecessarily. This meant that after plugging-in the MHL-cable, once the wrong screen (one without the chargin-icon) was displayed, it would not be refreshed unless something else changed onscreen.

Usually the onscreen clock forced a refresh of the buffers when the time was updated. As it showed time only down to the minute, it would mean that sometimes the display could be "stale" for as long as (but no longer than) a minute. An additional forced-refresh in the MHL-cable detection routine fixed the issue properly.

A simple example of double-buffering is shown below:

IV. Could Triple-buffering have prevented this issue?

Triple-buffering involves 2 back-buffers. at any given moment, the display-driver can immediately pick one that is not being updated by the graphics h/w to display into the front-buffer.

Triple buffering itself has 2 variants:
(A) Triple-buffering with no-sync. In this method the back-buffers are alternately updated by the graphics-h/w as fast as it can. At each Vsync, the display driver picks one of the buffers which is currently not being written to and swaps it with the front-buffer.

(B) Triple-buffering with Vsync. In this method, the back-buffers are updated by the graphics h/w as fast as it can. But the update stops if both the back-buffers are updated but have not been displayed in the front-buffer yet. The display-driver as usual swaps one of the back-buffers witht he fornt-buffer at each Vsync. at this point the previous front-buffer which is now a back buffer is considered "stale" and the graphics h/w fills it up with the updated frame.

Triple-bufffering used could potentially correct the issue as one of the back-buffers would hold the properly updated screen data and it even if it was not picked-up right away, it would be picked immediately in the following next swapbuffer() call. Also in double-buffering, the graphics h/w doesn't have to wait for access to the backbuffer till the swapbuffer() completes the flip operation between the front and back buffers. This is not the case in triple-buffering, thus allowing the graphics h/w to run at full throttle thereby reducing the time that either of the backbuffers contains stale display data.


Further reading: A detailed description of double/triple buffering.

[patch] [resend] Preparing a modified patch

In case of any collaborated project (eg. linux-kernel), often after submitting a patch for review,  we often receive several comments and need to make appropriate changes and generate a new "version2" of the patch containing the changes. If we are using Git for revision control, then the entire process becomes a snap.

How to prepare a modified patch for resend in 5 easy steps:

STEP1.
git rebase -i <commit-id-just-before-our-changes>

STEP2.
As discussed in the review, make the new changes to the source-files.

STEP3.
git add <modified-filenames>

STEP4.
git commit --amend
(shows editor with original commit-msg)
Edit the commit-msg (or leave as-is) and quit.
New commit is generated in the place of old commit.

STEP5.
git format-patch HEAD~1
DONE!! New patch version2 is ready for review now. :-)

If we do a diff between the PREV and NEW patch, we can see :

+ Changes made after review.

+ Time-Stamp change.

+ Hash change.

Booting Android completely over ethernet

When developing embedded-systems, initial development stages often involve huge number of "Modify-Build-Flash-Test" cycles. Test-Driven-Development methodology further promotes this style of development. This leads to a break in the "flow" at the Flash stage. Flashing the device with a newly built set of binaries interrupts the otherwise smooth "Modify-Build-Test" flow. Also errors tend to creep-in in the form of an older binary being copied/flashed, often causing confusion during debugging and  endless grief to the developer.

A simple way to avoid this is to have the binaries on the host-machine (a PC) and boot the embedded device directly using those binaries. In case of Android embedded system development, these binaries are the Linux-Kernel and the Android filesystem image.

Pre-requisites:

• The embedded device
• A linux PC
• Ethernet connectivity between the two

NOTE: Below listed parts 1, 2 & 3 involve setting-up the "host" Linux PC. Part 4 describes configuring the device to boot directly using the binaries present on the "host". It is assumed that a functional bootloader (u-boot) is present on the device (internal-flash/mmc-card) and that ethernet-support(either direct or over usb) is enabled.

Part1: Linux kernel over tftp

1. Install tftpd and related packages

host-PC$ sudo apt-get install xinetd tftpd tftp

2. Create /etc/xinetd.d/tftp

host-PC$ cat <<EOF | sudo tee /etc/xinetd.d/tftp
service tftp
{
    protocol        = udp
    port            = 69
    socket_type     = dgram
    wait            = yes
    user            = nobody
    server          = /usr/sbin/in.tftpd
    server_args     = /srv/tftp
    disable         = no
}
EOF

3. Make tftp-server directory

host-PC$ mkdir <tftp-server-path>

host-PC$ chmod -R 777 <tftp-server-path>

host-PC$ chown -R nobody <tftp-server-path>

4. Start tftpd through xinetd

host-PC$ sudo /etc/init.d/xinetd restart 
This concludes the tftp part of the setup process on the host.

Part2: Android fs over NFS

1. Install nfs packages

host-PC$ sudo apt-get install nfs-kernel-server nfs-common 

2. Add this line to /etc/exports

<rootfs-path> *(rw,sync,no_subtree_check,no_root_squash)

3. Restart service

host-PC$ sudo service nfs-kernel-server restart

4. Update exports for the NFS server

host-PC$ exportfs -a

5. Check NFS server

host-PC$ showmount -e

If everything went right, the <rootfs-path> will be listed in the output of showmount.

Part3: Where to put the files

1. Linux Kernel uImage

On the "host" PC,
Copy the Linux-Kernel uImage into <tftp-server-path>

2. Android rootfs

On the "host" PC,
Copy the contents of the Android rootfs into <rootfs-path>

Part4: Configuring the bootloader

1. Update bootargs

Connect the embedded device to the host-PC over ethernet (either directly or via a switch/router) and power it on. As shown below, configure the bootloader to pick-up the kernel from the host-PC over tftp and to mount the filesystem from the host-PC over NFS. As both support configuring a static-ip for the embedded-device or obtaining one dynamically using dhcp, 4 combinations are possible (2 shown below).

nfs(static-ip) and tftp(dhcp)
U-Boot# setenv bootargs 'console=ttyO0,115200n8 androidboot.console=ttyO0 mem=256M root=/dev/nfs ip=<client-device-ip> nfsroot=<nfs-server-ip>:<rootfs-path> rootdelay=2'

U-Boot# setenv serverip 'host-pc-ip'

U-Boot# bootm <Load address>

nfs(dhcp) and tftp(static-ip)
U-Boot# setenv bootargs 'console=ttyO0,115200n8 androidboot.console=ttyO0 mem=256M root=/dev/nfs ip=dhcp nfsroot=<nfs-server-ip>:<rootfs-path> rootdelay=2'

U-Boot# setenv serverip 'host-pc-ip'

U-Boot# setenv ipaddr 'client-device-ip'

U-Boot# tftp

U-Boot# bootm <Load address>

2. Boot ;-)

Linux-Kernel loaded over tftp

Filesystem mounted over NFS

Why __read_mostly does NOT work as it should

In modern SMP(multicore) systems, any processor can write to a memory location. The other processors have to update their caches immediately. For that reason, SMP systems implement the concept of "cacheline bouncing" to move "ownership" of cached-data between cores. This is effective but expensive.

Individual cores have private L1 caches which are extremely faster than the L2 and L3 caches which are shared between multiple cores. Typically, when a memory location is going to be ONLY read repeatedly, but never written to (for example a variable tagged with the const modifier), each core on the SMP system can safely store its own copy of that variable in its private(non-shared) cache. As the variable is NEVER written, the cache-entry never gets invalidated or "dirty". Hence the cores never need to get into "cache line bouncing" for that variable.

Take the case of the x86 architecture,

An Intel core i5 die showing the various caches present

• [NON-SMP] Intel Pentium 4 processor has to communicate between threads over the front-side bus, thus requiring at least a 400-500 cycle delay.
• [SMP] Intel Core processor family allowed for communication over a shared L2 cache with a delay of only 20 cycles between pairs of cores and the front-side bus between multiple pairs on a quad-core design.
• [SMP] The use of a shared L3 cache in the Intel Core i7 processor means that going across a bus to synchronize with another core is NEVER required unless a multiple-socket system is being used.

The copies of "read-only" locations usually end-up being cached in the private caches of the individual cores, which are several orders of magnitude faster than the shared L3 cache.

How __read_mostly is supposed to work: 

When a variable is tagged with the __read_mostly annotation, it is a signal to the compiler that accesses to the variable will be mostly reads and rarely(but NOT never) a write.

All variables tagged __read_mostly are grouped together into a single section in the final executable. This is to improve performance by allowing the system to optimise access time to those variables in SMP systems by allowing each core to maintain its own copy of them variable in it local cache. Once in a while when the variable does get written to, "cacheline bouncing" takes place. But this is  acceptable as the the time spent by the cores constantly synchronising using locks and using the slower shared-cache would be far more than the time it takes for the multiple cores to operate on own copies in their independent caches.

What actually happens:

(NOTE: In the following section, "elements" refers to memory blocks which are are smaller than a single cache-line and are so sparse in main-memory that a single cache-line cannot contain them both simultaneously.)

The problem with the above approach is that once all the __read_mostly variables are grouped into one section, the remaining "non-read-mostly" variables end-up  together too. This increases the chances that two frequently used elements (in the "non-read-mostly" region) will end-up competing for the same position (or cache-line, the basic fixed-sized block for memory<-->cache transfers) in the cache. Thus frequent accesses will cause excessive cache thrashing on that particular cache-line thereby degrading the overall system performance.

This situation is slightly alleviated by the fact that modern cpu caches are mostly 8way or 16way set-associative. In a 16way associative cache, each element has a choice of 16 different cache-slots. This means that two very frequently accessed elements, though closely located in memory, can still end-up in 2 different slots in the cache, thereby preventing cache-thrashing (which would have occurred had both continued competing for the same cache-line slot). In other words a minimum of 17 elements frequently accessed and closely located in memory are required for 2 of them to begin competing for a common cache-line slot.

While this is true in the case of INTEL and its x86 architecture, ARM still sticks to 4way & 2way set-associative caches even in its Cortex A8, which means that just 3 or 5 closely located, frequently accessed elements can result in cache-thrashing on an ARM system. (Update: "Anonymous" rightly points out in the comments that 16-way set associative caches have made their way into modern SoCs, ARM Cortex A9 onwards.)

kernel/arch/x86/include/asm/cache.h contains
#define __read_mostly __attribute__((__section__(".data..read_mostly")))
kernel/arch/arm/include/asm/cache.h: does NOT, thereby defaulting to the empty definition in
kernel/include/linux/cache.h
#ifndef __read_mostly
#define __read_mostly
#endif

UPDATE: The patch daf8741675562197d4fb4c4e9d773f53494203a5 enables support for __read_mostly  in the linux kernel for ARM architecture as well.

The reason for this? It turns out that most modern ARM SoCs have started using 8/16-way set associative caches. For example, the ARM PL310 cache controller (as "Anonymous" rightly points out in the comments) available on the ARM Cortex-A9 supports 16-way set associativity. The above patch now makes sense on modern ARM SoCs as the probability of cache-thrashing is reduced by the larger "N" in the N-way associative caches.

With the number of cores increasing rapidly and the on-die cache size growing slowly, one must always aim to:

• Minimise access to the last level of shared cache to improve performance on multicore systems.
• Increase associativity of private caches (of individual cores) to eliminate cache-slot contention and reduce cache-thrashing.