Thread (11 messages) 11 messages, 5 authors, 2015-05-23

Re: [PATCH v3] serial: 8250_uniphier: add UniPhier serial driver

From: Matthias Brugger <matthias.bgg@gmail.com>
Date: 2015-05-19 11:31:30
Also in: lkml

2015-05-19 13:18 GMT+02:00 Masahiro Yamada [off-list ref]:
Hi Alan,


2015-05-19 18:48 GMT+09:00 One Thousand Gnomes [off-list ref]:
quoted
quoted
I intentionally use deeper indentation for *_SHIFT
because I want to clearly show  UNIPHIER_UART_LCR_SHIFT
belongs to UNIPHIER_UART_LCR_MCR register.
Seems sensible to do it that way to me and a lot of other bits of the
kernel do.

The only other question I have is about the unipher_serial_out. If I am
writing a "special" register then the sequence becomes

- read 32bits
- modify
- write 32bits

That means that it's no longer atomic safe as the kernel expects.
Checking the users FCR seems safe, LCR is probably safe and MCR likewise
so I don't see a problem but I think it's worth noting in case anyone
else does.
Uh, I missed this.
I am a bit afraid what if LCR and MCR are updated at the same time.

Is it better to add mutex for writing special case registers?

    if (normal) {
            writel(value, p->membase + offset);
    } else {
            /* special case: two registers share the same address. */
            u32 tmp = readl(p->membase + offset);
            struct uniphier8250_priv *priv = p->private_data;

            mutex_lock(&priv->atomic_write_lock);
            tmp &= ~(0xff << valshift);
            tmp |= value << valshift;
            writel(tmp, p->membase + offset);
            mutex_unlock(&priv->atomic_write_lock);
    }


If it is OK, I can fix it in v4.



quoted
Finally can you add a comment in serial_in and serial_out where one
switch case drops through into the next so its obvious to anyone looking
at Coverity and other analyser output that this drop through was
intentional ?
I thought about it, too.

My previous version was as follows:


+#define UNIPHIER_UART_CHAR_FCR 3
+#define     UNIPHIER_UART_CHAR_SHIFT   8       /* Character Register */
+#define     UNIPHIER_UART_FCR_SHIFT    0       /* FIFO Control Register */
+#define UNIPHIER_UART_LCR_MCR  4
+#define     UNIPHIER_UART_LCR_SHIFT    8       /* Line Control Register */
+#define     UNIPHIER_UART_MCR_SHIFT    0       /* Modem Control Register */
+#define UNIPHIER_UART_DLR      9               /* Divisor Latch Register */

[snip]

+static void uniphier_serial_out(struct uart_port *p, int offset, int value)
+{
+       int valshift = 0;
+       bool normal = false;
+
+       switch (offset) {
+       case UART_FCR:
+               offset = UNIPHIER_UART_CHAR_FCR;
+               valshift = UNIPHIER_UART_FCR_SHIFT;
+               break;
+       case UART_LCR:
+               offset = UNIPHIER_UART_LCR_MCR;
+               valshift = UNIPHIER_UART_LCR_SHIFT;
+               /* Divisor latch access bit does not exist. */
+               value &= ~(UART_LCR_DLAB << valshift);
+               break;
+       case UART_MCR:
+               offset = UNIPHIER_UART_LCR_MCR;
+               valshift = UNIPHIER_UART_MCR_SHIFT;
+               break;
+       default:
+               normal = true;
+               break;
+       }



I thought it was clear to anyone although it was a bit redundant and
Matthias was opposed to it.

I personally prefer clear code to tricky code that requires comments.
Me too :)
What I wanted to say was, that in the case statement you don't need to
set valshift to zero as this is the default value when entering
uniphier_serial_out. This way you can get rid of
UNIPHIER_UART_MCR_SHIFT and UNIPHIER_UART_FCR_SHIFT.

Sorry, I didn't explained myself.

Cheers,
Matthias
-- 
motzblog.wordpress.com
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help