Re: [PATCH v3 2/3] drivers/block/mtip32xx: Adding source for hardware related operations

[Asai Thambi S P <asamymuthupa@micron.com>]

On 8/12/2011 2:42 AM, Jens Axboe wrote:
>> +/*
>> + * Obtain an empty command slot.
>> + *
>> + * This function needs to be reentrant since it could be called
>> + * at the same time on multiple CPUs. The allocation of the
>> + * command slot must be atomic.
>> + *
>> + * @port Pointer to the port data structure.
>> + *
>> + * return value
>> + *     >= 0    Index of command slot obtained.
>> + *     -1      No command slots available.
>> + */
>> +static int get_slot(struct mtip_port *port)
>> +{
>> +       int slot, i;
>> +       unsigned int num_command_slots = port->dd->slot_groups * 32;
>> +
>> +       /*
>> +        * Try 10 times, because there is a small race here.
>> +        *  that's ok, because it's still cheaper than a lock.
>> +        */
> 
> Might want to expand on what that race is.

find_next_zero_bit and test_and_set, in order, could be executed in
different process contexts in different processor. So both(2 or more)
contexts could land in choosing the same bit, but test_and_set will
succeed only for one. The failing contexts will try to get the next bit
in subsequent retries. This operation is cheaper than a lock.

This procedure is protected under the counting semaphore, so there is a
slot request for every context that calls this routine.

Is there an equivalent of doing find_next_zero_bit and test_and_set_bit
atomically?


>> +static irqreturn_t mtip_irq_handler(int irq, void *instance)
>> +{
>> +       struct driver_data *dd = instance;
>> +#ifdef MTIP_USE_TASKLET
>> +       tasklet_schedule(&dd->tasklet);
>> +       return IRQ_HANDLED;
>> +#else
>> +       return mtip_handle_irq(dd);
>> +#endif
>> +}
> 
> Have you experimented with blk-iopoll?

Working on this. Believe this experiment could go in parallel to
submission. Let us know.

>> +/*
>> + * Byte-swap ATA ID strings.
>> + *
>> + * ATA identify data contains strings in byte-swapped 16-bit words.
>> + * They must be swapped (on all architectures) to be usable as C
>> strings.
>> + * This function swaps bytes in-place.
>> + *
>> + * @buf The buffer location of the string
>> + * @len The number of bytes to swap
>> + *
>> + * return value
>> + *     None
>> + */
>> +static inline void ata_swap_string(u16 *buf, unsigned int len)
>> +{
>> +       int i;
>> +       for (i = 0; i < (len/2); i++)
>> +               be16_to_cpus(&buf[i]);
>> +}
> 
> I'm pretty sure we have the exact same thing in IDE and libata, so lets
> put this somewhere we they can all use it.

Is ata.h an appropriate location? Would it be better to make this change
after inclusion of this driver?

> 
> Overall looks very clean. If you take care of the little things
> mentioned by me and Christoph, I don't see anything stopping this from
> being merged for 3.2.

Appreciate your valuable feedback.

-- 
Regards,
Asai Thambi