[gPXE-devel] [PATCH 4/4] [myri10ge] Add NonVolatile Option (nvo) support

[Stefan Hajnoczi]

On Tue, Jun 22, 2010 at 6:45 PM, Glenn Brown <glenn at myri.com> wrote:
> +/*
> + * Initialize NonVolatile storage support for a device.
> + *
> + * @v priv     Device private data for the device.
> + * @ret rc     0 on success, else an error code.
> + */
> +
> +static int myri10ge_nv_init ( struct myri10ge_private *priv )
> +{
> +       int rc;
> +       struct myri10ge_eeprom_header
> +       {
> +               uint8 __jump[8];
> +               uint32 eeprom_len;
> +               uint32 eeprom_segment_len;
> +               uint32 mcp1_offset;
> +               uint32 mcp2_offset;
> +               uint32 version;
> +       } hdr;
> +       uint32 mcp2_len;
> +       unsigned int nvo_fragment_pos;
> +
> +       DBGP ( "myri10ge_nv_init\n" );
> +
> +       /* Read the EEPROM header, and byteswap the fields we will use.
> +          This is safe even though priv->nvs is not yet initialized. */
> +
> +       rc = myri10ge_nvs_read ( &priv->nvs, 0, &hdr, sizeof ( hdr ) );
> +       if ( rc ) {
> +               DBG ( "EEPROM header unreadable\n" );
> +               return rc;
> +       }
> +       hdr.eeprom_len         = bswap_32 ( hdr.eeprom_len );
> +       hdr.eeprom_segment_len = bswap_32 ( hdr.eeprom_segment_len );
> +       hdr.mcp2_offset        = bswap_32 ( hdr.mcp2_offset );
> +       hdr.version            = bswap_32 ( hdr.version );

Does be32_to_cpu() make sense instead of bswap_32()?  It would be nice not to
assume a little-endian CPU.  There is another bswap_32() a few lines down.

> +       DBG2 ( "eelen:%xh seglen:%xh mcp2@%xh ver%d\n", hdr.eeprom_len,
> +              hdr.eeprom_segment_len, hdr.mcp2_offset, hdr.version );
> +
> +       /* If the firmware does not support EEPROM writes, simply return. */
> +
> +       if ( hdr.version < 1 ) {
> +               DBG ( "No EEPROM write support\n" );
> +               return 0;
> +       }
> +
> +       /* Read the length of MCP2. */
> +
> +       rc = myri10ge_nvs_read ( &priv->nvs, hdr.mcp2_offset, &mcp2_len, 4 );
> +       mcp2_len = bswap_32 ( mcp2_len );

Stefan