[ECOS] bug in read in a device file

[ECOS] bug in read in a device file

[wangwei]

I think I found a bug in read() when a device file is set to NON_BLOCK
When I use read() to get more than one data from device ,such as :
      read(devfd,buf,3);
it will read nothing for ever .

The reason is :
    in file io/fileio/src/devfs.cxx, linenumber 324

//-----------------------------------------------------------------------

static int dev_fo_read      (struct CYG_FILE_TAG *fp, struct CYG_UIO_TAG
*uio)
{
    Cyg_ErrNo err = 0;
    int i;

    // Now loop over the iovecs until they are all done, or
    // we get an error.
    for( i = 0; i < uio->uio_iovcnt; i++ )
    {
        cyg_iovec *iov = &uio->uio_iov[i];
        cyg_uint32 len = iov->iov_len;

        err = cyg_io_read( (cyg_io_handle_t)fp->f_data,
                           iov->iov_base,
                           &len);

        if( err < 0 ) break;

        uio->uio_resid -= len;
    }

    return -err;
}
//-------------------------------------------------------------------

now len=3, if just one byte in the read buffer of drivers . after called
cyg_io_read ,
,it will return one byte in iov->iov_base,and len =1(because set
NON_BLOCK),but
err < 0,
so the byte will lost .
I use function "select" and "read" to get device data , and the device speed
isn't quick
enough , when function select decide there are some data in device driver,
function "read"
will be call, but only one byte in the read buffer now .so the read function
will return
failure always .

Re: [ECOS] bug in read in a device file

[Jonathan Larmour]

wangwei wrote:
> 
> I think I found a bug in read() when a device file is set to NON_BLOCK
> When I use read() to get more than one data from device ,such as :
>       read(devfd,buf,3);
> it will read nothing for ever .
> 
> The reason is :
>     in file io/fileio/src/devfs.cxx, linenumber 324
> 
> //-----------------------------------------------------------------------
> 
> static int dev_fo_read      (struct CYG_FILE_TAG *fp, struct CYG_UIO_TAG
> *uio)
> {
[smo[
>         err = cyg_io_read( (cyg_io_handle_t)fp->f_data,
>                            iov->iov_base,
>                            &len);
> 
>         if( err < 0 ) break;
> 
>         uio->uio_resid -= len;
>     }
> 
>     return -err;
> }
> //-------------------------------------------------------------------
> 
> now len=3, if just one byte in the read buffer of drivers . after called
> cyg_io_read ,
> ,it will return one byte in iov->iov_base,and len =1(because set
> NON_BLOCK),but
> err < 0,
> so the byte will lost .

I believe if you just swap round the "if (err < 0) break;" and
"uio->uio_resid -= len;" lines it should work. Try it and let us know.

Nick, is this the correct permanent fix? I think so. Obviously the same
would apply to dev_fo_write().

Jifl
-- 
Red Hat, Rustat House, Clifton Road, Cambridge, UK. Tel: +44 (1223) 271062
Maybe this world is another planet's Hell -Aldous Huxley || Opinions==mine

[ECOS] Bug in read/write in the io.cxx file

[Boris V. Guzhov]

The readwrite() function in the io/fileio/current/src/io.cxx file never
returns any error code from op() function.

I think that it's necessary to add some lines in readwrite() function::
....
    if ( ret != ENOERR )              // two new lines
        FILEIO_RETURN(ret);

    CYG_REPORT_RETVAL(cnt);
    return cnt;
}

That's right?

--
Boris Guzhov,
St.Petersburg, Russia