RE: [Yaffs] Optimizing the number of reads and writes to fla…

Hi Daniel,
uncomment the define of this two labels in the makefile:
CONFIG_YAFFS_DISABLE_CHUNK_ERASED_CHECK
CONFIG_YAFFS_DISABLE_WRITE_VERIFY

In this way you avoid to re-read every chunk before write (to control if it's blank) and to re-read after a succesfull write.
I've also made a little modification of the yaffs_WriteNewChunkToNAND, to completely avoid chunks read after write (the original code avoids yaffs_VerifyCompare only).
Check the following code against cvs one.

static int yaffs_WriteNewChunkToNAND(struct yaffs_DeviceStruct *dev, const __u8 *data, yaffs_Spare *spare,int useReserve)
{
    int chunk;

    int writeOk = 1;
    int attempts = 0;

    do{
        chunk = yaffs_AllocateChunk(dev,useReserve);

        if(chunk >= 0)
        {

            // First check this chunk is erased...
#ifndef CONFIG_YAFFS_DISABLE_CHUNK_ERASED_CHECK
            writeOk = yaffs_CheckChunkErased(dev,chunk);
#endif
            if(!writeOk)
            {
                T(YAFFS_TRACE_ERROR,(TSTR("**>> yaffs chunk %d was not erased" TENDSTR),chunk));
            }
            else
            {
                writeOk =  yaffs_WriteChunkToNAND(dev,chunk,data,spare);
            }
            attempts++;
#ifndef CONFIG_YAFFS_DISABLE_WRITE_VERIFY
            if(writeOk)
            {
                unsigned char rbData[YAFFS_BYTES_PER_CHUNK];
                yaffs_Spare rbSpare;

                // Readback & verify
                // If verify fails, then delete this chunk and try again
                // To verify we compare everything except the block and
                // page status bytes.
                // NB We check a raw read without ECC correction applied

                yaffs_ReadChunkFromNAND(dev,chunk,rbData,&rbSpare,0);

                if(!yaffs_VerifyCompare(data,rbData,spare,&rbSpare))
                {
                    // Didn't verify
                    T(YAFFS_TRACE_ERROR,(TSTR("**>> yaffs write verify failed on chunk %d" TENDSTR), chunk));

                    writeOk = 0;
                }

            }
#endif
            if(writeOk)
            {
                // Copy the data into the write buffer.
                // NB We do this at the end to prevent duplicates in the case of a write error.
                //Todo
                yaffs_HandleWriteChunkOk(dev,chunk,data,spare);
            }
            else
            {
                yaffs_HandleWriteChunkError(dev,chunk);
            }
        }

    } while(chunk >= 0 && ! writeOk);

    if(attempts > 1)
    {
        T(YAFFS_TRACE_ERROR,(TSTR("**>> yaffs write required %d attempts" TENDSTR),attempts));
        dev->nRetriedWrites+= (attempts - 1);
    }

    return chunk;
}

To reduce writes I think you must wait (or aid ;) for Yaffs2, which will use a zero page rewrites to spare area and will have generally better performance over Yaffs.
Look at www.aleph1.co.uk/yaffs/yaffs2.html


Regards
Adamo



> -----Original Message-----
> From: yaffs-admin@stoneboat.aleph1.co.uk
> [mailto:yaffs-admin@stoneboat.aleph1.co.uk]On Behalf Of Daniel
> Gustafsson
> Sent: lunedì 11 ottobre 2004 14.47
> To: yaffs@stoneboat.aleph1.co.uk
> Subject: [Yaffs] Optimizing the number of reads and writes to flash
>
>
>
>
> I am using yaffs/direct and I want to reduce the reads and
> writes towards
> the flash to an absolute minimum. I do not want any
> verifications of flash
> etc. I don not want to use ECC.
>
> I would greatly appreatiate any tips regarding the above
> issue but also how
> to reduce the footprint in RAM and size.
>
> Further more I am interested in some documetation what
> happens (reads/writes
> to flash) when performing some simple fs operations like
> read/write/open
> etc.
>
>
> I hope someone have the answers I am looking for,
>
> Best Regards
> Daniel Gustavsson
>
>
> _______________________________________________
> yaffs mailing list
> yaffs@stoneboat.aleph1.co.uk
> http://stoneboat.aleph1.co.uk/cgi-bin/mailman/listinfo/yaffs
>