Re: [PATCH 2/2 v2] sata highbank: add bit-banged SGPIO driver support

From: Mark Langsdorf
Date: Tue Jun 04 2013 - 11:09:51 EST

On 06/03/2013 03:37 PM, Tejun Heo wrote:
> Hello, Mark.
> In general, please try to reply to reviews addressing each point. It
> gives much better sense of what's going on to the reviewer and also
> helps the reviewee to avoid misunderstandings or missing points.
>> +static DEFINE_SPINLOCK(sgpio_lock);
>> +#define SCLOCK 0
>> +#define SLOAD 1
>> +#define SDATA 2
>> +#define SGPIO_PINS 3
>> +#define SGPIO_PORTS 8
>> +
>> +/* can be cast as an ahci_host_priv for compatibility with most functions */
> This sounds awfully scary. What's going on here? Are you actually
> overriding ahci_host_priv? If so, please don't ever do things like
> that. Do it properly. Add host_priv->platform_priv or whatever and
> chain the pointer there. If you're worried about the extra deref,
> update ahci core such that it allows specifying extra size and you can
> embed ahci_host_priv in your own priv. ie.
> struct ecx_host_priv {
> struct ahci_host_priv ahci_priv; /* must be the first field */
> /* your own stuff */
> };
> And tell ahci core sizeof(ecx_host_priv) some way, but really, just
> having a plain pointer should be enough, I think.

I think I want to do the opposite. For 90% of the AHCI EM functions,
I want ecx_host_priv to be an ahci_host_priv so that I can use those
functions without having to keep a local copy of them.

Would something like this:
struct ahci_host_priv {
/* standard AHCI existing stuff */
void *private_data;

I shied away from that because a private data structure having a private
data structure doesn't seem right.

>> +static inline int sgpio_bit_shift(struct ecx_host_priv *hpriv, u32 port,
>> + u32 shift)
>> +{
>> + return 1 << (3 * hpriv->port_to_sgpio[port] + shift);
>> +}
>> +
>> +static void ecx_parse_sgpio(struct ecx_host_priv *hpriv, u32 port, u32 state)
> Would be kinda nice to have comment explaining what the function does
> as @state == 0 turns off everything while anything else would just
> turn things on.


>> +static ssize_t ecx_transmit_led_message(struct ata_port *ap, u32 state,
>> + ssize_t size)
>> +{
> ...
>> + if (!hpriv->em_msg_type & EM_MSG_TYPE_LED)
>> + return size;
> Is this really correct? You first negate and convert it to bool and
> then bit-wise and it with a mask? How is supposed to work?

Am I confused about the order of operations? It's meant to be "continue
if hpriv->em_msg_type doesn't have EM_MSG_TYPE_LED set".

>> - ahci_save_initial_config(dev, hpriv, 0, 0);
>> + ahci_save_initial_config(dev, (struct ahci_host_priv *) hpriv, 0, 0);
> Ugh....... how is this supposed to work? What if ahci_host_priv grows
> larger than ecx one in the future? :(

For functions like ahci_save_initial_config, I just want to use the
already defined ahci_ functions with my extra data along for the ride.
What's the best way to do that?

--Mark Langsdorf
Calxeda, Inc.

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at