[Under Powercut Testing] os/fs/driver/mtd: Read optimization when CRC and Journaling are enabled

[gSahitya-samsung]

• When CRC and Journaling are both enabled, CRC verification of written data is done by reading it back immediately after writing.
• Hence, we do not need to repeat this verification at the time of reading again.

[amogh-samsung]

Test pseudocode: -

for i iterations
{
    for n files
    {
        create(file)
        for t times
        {
            open(file)
            seek(END)
            write(buffer[bufsize])
            close(file)
        }
    }
}

	Byte Reads	Block Reads	Total Read Bytes
Before this PR	29,367	8,620	22,36,087
After this PR	7,17,100	4,613	1897925.55
		% Improvement =	15.123

The improvements may vary in different configuration settings and testcases.

[Taejun-Kwon]

In my opinion, we should compare 1 time of bread & 2 times of read, in otherwords, read count & read size should be checked.
please verify this with tp1x, tp2x, 8730.
I think read count is more important factor because every time it should lock flash & wait for dma interrupt, for most of case.

[Taejun-Kwon]

useless errout

[amogh-samsung]

In my opinion, we should compare 1 time of bread & 2 times of read, in otherwords, read count & read size should be checked. please verify this with tp1x, tp2x, 8730. I think read count is more important factor because every time it should lock flash & wait for dma interrupt, for most of case.

We want to check that total read calls to flash are increased/decreased/constant? We should check this at a lower hardware level.
Is that right @Taejun-Kwon ? Total number of read bytes is definitely decreased.
Please keep a check on this @gSahitya-samsung

We should do this for our most commercialized boards - 8721csm, 8730e and mt87686_hdk

[amogh-samsung]

@gSahitya-samsung
Please also repeat the test with all intermediate logs disabled and with timing enabled.
Lets see if more number of byte reads leads to more time consumption as TJ suggested.

We have to evaluate
Lesser total bytes read VS More total read calls to hardware and time/energy consumed

[Taejun-Kwon]

In my opinion, we should compare 1 time of bread & 2 times of read, in otherwords, read count & read size should be checked. please verify this with tp1x, tp2x, 8730. I think read count is more important factor because every time it should lock flash & wait for dma interrupt, for most of case.

We want to check that total read calls to flash are increased/decreased/constant? We should check this at a lower hardware level. Is that right @Taejun-Kwon ? Total number of read bytes is definitely decreased. Please keep a check on this @gSahitya-samsung

We should do this for our most commercialized boards - 8721csm, 8730e and mt87686_hdk

I mean count of read will affect to performance more, because flow of read is...

External flash (ext flash of tp2x need to be tested)

1. Lock flash
2. SPI Lock
3. send data through the spi
4. read data through the spi
5. SPI unlick
6. Flash unlock

internal flash

1. Lock Flash
2. Copy data from flash
3. Unlock Flash

[kishore-sn]

There is no cleanup code at errout. So i think we can directly return from here.

[kishore-sn]

Pls add a tab at the beginning for this comment and split it into two lines.