Skip to contentSkip to conversation

parport: add parallel port support for SGI O2

PATCH In review Arnaud Giersch 2 participants 4 messages Jan 20, 2006 – Jan 24, 2006 Parallel port

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 parport: add parallel port support for SGI O2

From: Arnaud Giersch <[email protected]>

Add support for the built-in parallel port on SGI O2 (a.k.a. IP32). Define a new configuration option: PARPORT_IP32. The module is named parport_ip32.

Hardware support for SPP, EPP and ECP modes along with DMA support when available are currently implemented.

---

Changes since previous version sent on January 9, 2006
(http://marc.theaimsgroup.com/?l=linux-kernel&m=113684000618245):

* Address comments from Randy Dunlap <rdunlap_at_xenotime.net>, that is:
- Remove useless memory barriers.
- Review function inlining.
- Review kernel docs.
- Fix style.
- Fix typos.
* Remove useless parport_ip32_{in,out} functions.
* Review locking in DMA operations.
* Simplify parport_ip32_ecp_supported().
* Fix includes.

 drivers/parport/Kconfig        |    9 
 drivers/parport/Makefile       |    1 
 drivers/parport/parport_ip32.c | 2217 +++++++++++++++++++++++++++++++
 include/linux/parport.h        |    6 
 4 files changed, 2233 insertions(+)
drivers/parport/Kconfig +9 -0
diff -Naurp linux-2.6.16-rc1.orig/drivers/parport/Kconfig linux-2.6.16-rc1/drivers/parport/Kconfig
@@ -90,6 +90,15 @@ config PARPORT_ARC
 	depends on ARM && PARPORT
 	select PARPORT_NOT_PC
 
+config PARPORT_IP32
+	tristate "SGI IP32 builtin port (EXPERIMENTAL)"
+	depends on SGI_IP32 && PARPORT && EXPERIMENTAL
+	select PARPORT_NOT_PC
+	help
+	  Say Y here if you need support for the parallel port on
+	  SGI O2 machines. This code is also available as a module (say M),
+	  called parport_ip32.  If in doubt, saying N is the safe plan.
+
 config PARPORT_AMIGA
 	tristate "Amiga builtin port"
 	depends on AMIGA && PARPORT
drivers/parport/Makefile +1 -0
diff -Naurp linux-2.6.16-rc1.orig/drivers/parport/Makefile linux-2.6.16-rc1/drivers/parport/Makefile
@@ -17,3 +17,4 @@ obj-$(CONFIG_PARPORT_MFC3)	+= parport_mf
 obj-$(CONFIG_PARPORT_ATARI)	+= parport_atari.o
 obj-$(CONFIG_PARPORT_SUNBPP)	+= parport_sunbpp.o
 obj-$(CONFIG_PARPORT_GSC)	+= parport_gsc.o
+obj-$(CONFIG_PARPORT_IP32)	+= parport_ip32.o
drivers/parport/parport_ip32.c +2217 -0

2,217 lines not on this page · raw message

include/linux/parport.h +6 -0
diff -Naurp linux-2.6.16-rc1.orig/include/linux/parport.h linux-2.6.16-rc1/include/linux/parport.h
@@ -128,6 +128,11 @@ struct amiga_parport_state {
        unsigned char statusdir;/* ciab.ddrb & 7 */
 };
 
+struct ip32_parport_state {
+	unsigned int dcr;
+	unsigned int ecr;
+};
+
 struct parport_state {
 	union {
 		struct pc_parport_state pc;
@@ -135,6 +140,7 @@ struct parport_state {
 		struct ax_parport_state ax;
 		struct amiga_parport_state amiga;
 		/* Atari has not state. */
+		struct ip32_parport_state ip32;
 		void *misc; 
 	} u;
 };
#1 lore raw
Quoted 9 lines · Arnaud Giersch

From: Arnaud Giersch <[email protected]>

Add support for the built-in parallel port on SGI O2 (a.k.a. IP32). Define a new configuration option: PARPORT_IP32. The module is named parport_ip32.

Hardware support for SPP, EPP and ECP modes along with DMA support when available are currently implemented.

Nice looking driver. Big.

It does rather a lot of

	if (foo)	do_something();

whereas we prefer

	if (foo)
		do_something();

but if that was a blocker, we wouldn't have any drivers.

drivers/parport/parport_ip32.c line 474

Quoted 9 lines
drivers/parport/parport_ip32.c +9 -0
+static void parport_ip32_dma_setup_context(unsigned int limit)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&parport_ip32_dma.lock, flags);
+	if (parport_ip32_dma.left > 0) {
+		volatile u64 __iomem *ctxreg = (parport_ip32_dma.ctx == 0) ?
+			&mace->perif.ctrl.parport.context_a :
+			&mace->perif.ctrl.parport.context_b;

Does this need to be volatile? writeq() should do the right thing.

drivers/parport/parport_ip32.c line 1088

Quoted 11 lines
drivers/parport/parport_ip32.c +11 -0
+static size_t parport_ip32_epp_read(void __iomem *eppreg,
+				    struct parport *p, void *buf,
+				    size_t len, int flags)
+{
+	struct parport_ip32_private * const priv = p->physport->private_data;
+	size_t got;
+	parport_ip32_set_mode(p, ECR_MODE_EPP);
+	parport_ip32_data_reverse(p);
+	parport_ip32_write_control(p, DCR_nINIT);
+	if ((flags & PARPORT_EPP_FAST) && (len > 1)) {
+		readsb(eppreg, buf, len);

readsb() is a mips thing, and doesn't seem to be documented. What does it do, and why does the driver use it (only) here?

Quoted 6 lines
drivers/parport/parport_ip32.c +5 -0
+		writesb(eppreg, buf, len);
 
+static unsigned int parport_ip32_fifo_wait_break(struct parport *p,
+						 unsigned long expire)
+{
+	cond_resched();

drivers/parport/parport_ip32.c line 1230

Quoted 10 lines
drivers/parport/parport_ip32.c +10 -0
+	if (time_after(jiffies, expire)) {
+		printk(KERN_DEBUG PPIP32
+		       "%s: FIFO write timed out\n", p->name);
+		return 1;
+	}
+	if (signal_pending(current)) {
+		printk(KERN_DEBUG PPIP32
+		       "%s: Signal pending\n", p->name);
+		return 1;
+	}

This printk could be a bit noisy, if someone hoses a signal stream at a printing program.

#2 lore raw
Quoted 7 lines · Andrew Morton
Quoted 5 lines

From: Arnaud Giersch <[email protected]>

Add support for the built-in parallel port on SGI O2 (a.k.a. IP32). Define a new configuration option: PARPORT_IP32. The module is named parport_ip32.

Nice looking driver. Big.

Thanks.

Quoted 8 lines

It does rather a lot of

	if (foo)	do_something();

whereas we prefer

	if (foo)
		do_something();

Those are limited to the parport_ip32_dump_state() function. The rationale is that this function is only here for debugging purpose, and I did not want to make it longer than it already is.

I will correct the style.

Quoted 1 line
Quoted 1 line

+static void parport_ip32_dma_setup_context(unsigned int limit)

[...]

Quoted 5 lines

drivers/parport/parport_ip32.c line 474

Quoted 3 lines
drivers/parport/parport_ip32.c +3 -0
+		volatile u64 __iomem *ctxreg = (parport_ip32_dma.ctx == 0) ?
+			&mace->perif.ctrl.parport.context_a :
+			&mace->perif.ctrl.parport.context_b;

Does this need to be volatile? writeq() should do the right thing.

The "volatile" is here for type consistency. Both variables "mace->perif.ctrl.parport.context_a" and "context_b" (defined in include/asm-mips/ip32/mace.h) are from type "volatile u64". Omitting the "volatile" qualifier for "ctxreg" would make gcc and sparse complain.

I will add a comment to explain it in the code.

Quoted 1 line
Quoted 1 line

+static size_t parport_ip32_epp_read(void __iomem *eppreg,

[...]

Quoted 7 lines

drivers/parport/parport_ip32.c line 1088

Quoted 2 lines
drivers/parport/parport_ip32.c +2 -0
+	if ((flags & PARPORT_EPP_FAST) && (len > 1)) {
+		readsb(eppreg, buf, len);

readsb() is a mips thing, and doesn't seem to be documented. What does it do, and why does the driver use it (only) here?

drivers/parport/parport_ip32.c line 1127

Quoted 1 line
drivers/parport/parport_ip32.c +1 -0
+		writesb(eppreg, buf, len);

readsb() and writesb() are like ioread8_rep() and iowrite8_rep(). The io{read,write} family functions are however not available in the linux-mips.org tree.

The usage of readsb() here is similar to that of insb() in the corresponding code of the parport_pc driver.

Quoted 1 line
Quoted 1 line

+static unsigned int parport_ip32_fifo_wait_break(struct parport *p,

[...]

Quoted 8 lines

drivers/parport/parport_ip32.c line 1230

Quoted 5 lines
drivers/parport/parport_ip32.c +5 -0
+	if (signal_pending(current)) {
+		printk(KERN_DEBUG PPIP32
+		       "%s: Signal pending\n", p->name);
+		return 1;
+	}

This printk could be a bit noisy, if someone hoses a signal stream at a printing program.

I did not realized that. I will try to correct the problem.

Thank you for your attention.

Arnaud

#3 lore raw

Mardi 24 janvier 2006, vers 00:38:57 (+0100), j'ai écrit:

Quoted 1 line

I will correct the style.

[...]

Quoted 1 line

I will add a comment to explain it in the code.

[...]

Quoted 1 line

I did not realized that. I will try to correct the problem.

An updated version is available at

     http://arnaud.giersch.free.fr/parport_ip32/parport_ip32-0.6.patch.gz

Since all changes are non-functional ones, I did not want to flood the lists again with the full patch.

Changes are:

* Address comments from Andrew Morton <akpm_at_osdl.org>:
- Style: avoid single line "if" and "case" statements.
- Turn some printk() into pr_debug1(), so that they only exist in debugging mode.
- Add a comment about "volatile".
- Add a comment about readsb() and writesb().
* Only define DEBUG_PARPORT_IP32 if it is not already defined.
* Remove the CVS $Id$ line.

Arnaud