>>> +static void test_erase_sector(void)
>>> +{
>>> + uint32_t some_page_addr = 0x600 * PAGE_SIZE;
>>> + uint32_t page[PAGE_SIZE / 4];
>>> + int i;
>>> +
>>> + spi_conf(CONF_ENABLE_W0);
>>> +
>>> + spi_ctrl_start_user();
>>> + writeb(AST2400_FLASH_BASE, WREN);
>>> + writeb(AST2400_FLASH_BASE, EN_4BYTE_ADDR);
>>> + writeb(AST2400_FLASH_BASE, ERASE_SECTOR);
>>> + writel(AST2400_FLASH_BASE, cpu_to_be32(some_page_addr));
>> Hi,
>>
>> I think you should not make any byte swapping because
>> qtest for all write* instructions (see qtest.c:377).
>
> yes. on a BE, we should not byte swap.
>
> When on a LE host, the cpu_to_be32() call in the test byte-swaps
> the address, tswap32() in qtest.c does nothing as the host endian
> and guest endian are in sync and so the address reaches the
> controller region in the expected order.
>
> When on a BE host, the cpu_to_be32() does nothing but tswap32() in
> qtest.c reverses the order and so the address reaches the controller
> region in the wrong order (LE).
>
> I see two possible fixes :
>
> 1 - replace the read* routines by out* (no tswap there)
> 2 - add an extra byteswap on BE hosts using qtest_big_endian() as
> in virtio-blk-test.c. see virtio_blk_fix_request()
or :
3 - use memwrite/read
which is I think the appropriate fix.
Thanks,
C.
