Re: [Qemu-devel] [PATCH 2/5] block/vpc: simplify vpc_read

[Peter Lieven <pl@kamp.de>]

Am 23.02.2015 um 19:29 schrieb Max Reitz:
> On 2015-02-23 at 09:27, Peter Lieven wrote:
>> Signed-off-by: Peter Lieven <pl@kamp.de>
>> ---
>>   block/vpc.c |  116 +++++++++++++++++++++++++++--------------------------------
>>   1 file changed, 52 insertions(+), 64 deletions(-)
>>
>> diff --git a/block/vpc.c b/block/vpc.c
>> index 326c2bb..4e5ba85 100644
>> --- a/block/vpc.c
>> +++ b/block/vpc.c
>> @@ -497,40 +497,70 @@ static int vpc_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
>>       return 0;
>>   }
>>   -static int vpc_read(BlockDriverState *bs, int64_t sector_num,
>> -                    uint8_t *buf, int nb_sectors)
>> +static int64_t coroutine_fn vpc_co_get_block_status(BlockDriverState *bs,
>> +        int64_t sector_num, int nb_sectors, int *pnum)
>
> How about just putting the function header here? If you really have to move vpc_co_get_block_status() up, I'd rather like it to be in a separate patch.
>
> Second, while apparently vpc_read() is actually called in a coroutine, that is pretty hard to know. Most importantly, it's not marked as a coroutine_fn. Therefore I don't think it's a good idea to call vpc_co_get_block_status() directly; I'd vote for 
> either using bdrv_get_block_status(), or moving the content of vpc_co_get_block_status() to a non-coroutine_fn (it doesn't contain an coroutine-related function calls, so this is fine) and then making vpc_co_get_block_status() a wrapper around that, or 
> just dropping this patch.
>
> The latter I'm proposing because I don't really see what this patch improves. The previous vpc_read() function was pretty straightforward, too, and I don't think it was unbearably longer.
>
> One could argue that the coroutine_fn stuff doesn't really matter in this situation because it doesn't actually do anything right now and vpc_co_get_block_status() does not call any other coroutine_fn functions in turn; however, it is a semantic 
> contract established by include/block/coroutine.h and as far as I remember, Stefan did eventually want to have something to error out on compile-time if a non-coroutine_fn function calls a coroutine_fn. I don't like breaking this contract even if it's 
> not bad in this specific case.
>
> Considering you probably think bdrv_get_block_status() to be too much overhead (it will fall down to the protocol layer on VHD_FIXED) and you probably find making vpc_read() shorter justified (again, which I don't necessarily), I think moving the 
> contents of vpc_co_get_block_status() to a non-coroutine_fn might be the best way to go.
>
>>   {
>>       BDRVVPCState *s = bs->opaque;
>> -    int ret;
>> -    int64_t offset;
>> -    int64_t sectors, sectors_per_block;
>> -    VHDFooter *footer = (VHDFooter *) s->footer_buf;
>> +    VHDFooter *footer = (VHDFooter*) s->footer_buf;
>> +    int64_t start, offset;
>> +    bool allocated;
>> +    int n;
>>         if (be32_to_cpu(footer->type) == VHD_FIXED) {
>> -        return bdrv_read(bs->file, sector_num, buf, nb_sectors);
>> +        *pnum = nb_sectors;
>> +        return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID | BDRV_BLOCK_DATA |
>> +               (sector_num << BDRV_SECTOR_BITS);
>>       }
>> -    while (nb_sectors > 0) {
>> -        offset = get_sector_offset(bs, sector_num, 0);
>>   -        sectors_per_block = s->block_size >> BDRV_SECTOR_BITS;
>> -        sectors = sectors_per_block - (sector_num % sectors_per_block);
>> -        if (sectors > nb_sectors) {
>> -            sectors = nb_sectors;
>> +    offset = get_sector_offset(bs, sector_num, 0);
>> +    start = offset;
>> +    allocated = (offset != -1);
>> +    *pnum = 0;
>> +
>> +    do {
>> +        /* All sectors in a block are contiguous (without using the bitmap) */
>> +        n = ROUND_UP(sector_num + 1, s->block_size / BDRV_SECTOR_SIZE)
>> +          - sector_num;
>> +        n = MIN(n, nb_sectors);
>> +
>> +        *pnum += n;
>> +        sector_num += n;
>> +        nb_sectors -= n;
>> +
>> +        if (allocated) {
>> +            return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start;
>>           }
>> +        if (nb_sectors == 0) {
>> +            break;
>> +        }
>> +        offset = get_sector_offset(bs, sector_num, 0);
>> +    } while (offset == -1);
>>   -        if (offset == -1) {
>> -            memset(buf, 0, sectors * BDRV_SECTOR_SIZE);
>> -        } else {
>> -            ret = bdrv_pread(bs->file, offset, buf,
>> -                sectors * BDRV_SECTOR_SIZE);
>> -            if (ret != sectors * BDRV_SECTOR_SIZE) {
>> +    return 0;
>> +}
>> +
>> +static int vpc_read(BlockDriverState *bs, int64_t sector_num,
>> +                    uint8_t *buf, int nb_sectors)
>> +{
>> +    int ret, n;
>> +    int64_t ret2;
>> +
>> +    while (nb_sectors > 0) {
>> +        ret2 = vpc_co_get_block_status(bs, sector_num, nb_sectors, &n);
>> +
>
> Superfluous whitespace here.
>
>> +        if (ret2 & BDRV_BLOCK_OFFSET_VALID) {
>> +            ret = bdrv_pread(bs->file, ret2 & BDRV_SECTOR_MASK, buf,
>> +                             n * BDRV_SECTOR_SIZE);
>> +            if (ret != n * BDRV_SECTOR_SIZE) {
>>                   return -1;
>
> Please make that "return ret" (and possibly "if (ret < 0)", if you want to).

I took this from the orignal vpc_read function. Maybe the author also wanted
to catch short reads.

I tend to drop the whole patch anyway. I was tempted to use that new vpc_co_get_block_status
function somehow because I saw that part of its logic is in vpc_read as well.

If it should stay maybe it would be an option to inline vpc_read in vpc_co_read (and the same for write)?