[parisc-linux] FB cleanups

[parisc-linux] FB cleanups

[Matthew Wilcox]

I'd like someone to test these out for me -- they compile but I'm monitorless
here.

 * Turn gsc_read/write[bwl] calls into read/write[bwl]
 * Remove unused functions memcpy_tohp, memcopy_fromhp from fbcon-sti.c
 * Delete gsc_memset_io()
 * Teach <video/fbcon.h> to use memset_io instead.
 * Make <video/fbcon.h> use read/write[bwl] instead of direct accesses
   unless the architecture specifies otherwise.

Index: drivers/video/fbcon-sti.c
===================================================================
RCS file: /home/cvs/parisc/linux/drivers/video/fbcon-sti.c,v
retrieving revision 1.5
diff -u -p -r1.5 fbcon-sti.c
--- fbcon-sti.c	2001/03/02 00:54:55	1.5
+++ fbcon-sti.c	2001/04/11 19:59:13
@@ -70,40 +70,14 @@ memcpy_fromhp_tohp(void *dest, void *src
 	count += 3;
 	count &= ~3; /* XXX */
 
-	while(count) {
-		count --;
-		gsc_writel(~gsc_readl(s), d);
+	while (count--) {
+		writel(~readl(s), d);
 		d += 32*4;
 		s += 32*4;
 	}
 }
 
 static void
-memcpy_tohp(void *dest, void *src, int count)
-{
-	unsigned long d = (unsigned long) dest;
-	u32 *s = (u32 *)src;
-
-	count += 3;
-	count &= ~3; /* XXX */
-
-	d = ram2log(dest);
-
-	while(count) {
-		count--;
-		gsc_writel(*s++, d);
-		d += 32*4;
-	}
-}
-
-static void
-memcopy_fromhp(void *dest, void *src, int count)
-{
-	/* FIXME */
-	printk("uhm ...\n");
-}
-
-static void
 memset_tohp(void *dest, u32 word, int count)
 {
 	unsigned long d = ram2log(dest);
@@ -111,9 +85,8 @@ memset_tohp(void *dest, u32 word, int co
 	count += 3;
 	count &= ~3;
 
-	while(count) {
-		count--;
-		gsc_writel(word, d);
+	while (count--) {
+		writel(word, d);
 		d += 32;
 	}
 }
@@ -123,7 +96,7 @@ readb_hp(void *src)
 {
 	unsigned long s = ram2log(src);
 
-	return ~gsc_readb(s);
+	return ~readb(s);
 }
 
 static void
@@ -137,7 +110,7 @@ writeb_hp(u8 b, void *dst)
 		return;
 	}
 
-	gsc_writeb(b, d);
+	writeb(b, d);
 }
 
 static void
Index: include/asm-parisc/gsc.h
===================================================================
RCS file: /home/cvs/parisc/linux/include/asm-parisc/gsc.h,v
retrieving revision 1.9
diff -u -p -r1.9 gsc.h
--- gsc.h	2001/03/02 00:09:47	1.9
+++ gsc.h	2001/04/11 19:59:15
@@ -32,14 +32,6 @@ extern void _gsc_writeq(u64,void *);
 #define gsc_writel(v,a)	_gsc_writel((v),(void *)(a))
 #define gsc_writeq(v,a)	_gsc_writeq((v),(void *)(a))
 
-static __inline__ void *gsc_memset_io(void *s, int c, size_t n)
-{
-	while (n--) {
-		gsc_writeb(c,s++);
-	}
-	return NULL;
-}
-
 struct gsc_dev {
 	struct gsc_bus	*bus;		/* bus this device is on */
 	struct gsc_dev	*next;		/* chain of all devices */
Index: include/video/fbcon.h
===================================================================
RCS file: /home/cvs/parisc/linux/include/video/fbcon.h,v
retrieving revision 1.6
diff -u -p -r1.6 fbcon.h
--- fbcon.h	2001/03/02 00:16:36	1.6
+++ fbcon.h	2001/04/11 19:59:52
@@ -206,16 +206,6 @@ extern int set_all_vcs(int fbidx, struct
 #define fb_writel sbus_writel
 #define fb_memset sbus_memset_io
 
-#elif defined(__hppa__)
-
-#define fb_readb gsc_readb
-#define fb_readw gsc_readw
-#define fb_readl gsc_readl
-#define fb_writeb gsc_writeb
-#define fb_writew gsc_writew
-#define fb_writel gsc_writel
-#define fb_memset gsc_memset_io
-
 #elif defined(__i386__) || defined(__alpha__)
 
 #define fb_readb __raw_readb
@@ -228,13 +218,13 @@ extern int set_all_vcs(int fbidx, struct
 
 #else
 
-#define fb_readb(addr) (*(volatile u8 *) (addr))
-#define fb_readw(addr) (*(volatile u16 *) (addr))
-#define fb_readl(addr) (*(volatile u32 *) (addr))
-#define fb_writeb(b,addr) (*(volatile u8 *) (addr) = (b))
-#define fb_writew(b,addr) (*(volatile u16 *) (addr) = (b))
-#define fb_writel(b,addr) (*(volatile u32 *) (addr) = (b))
-#define fb_memset memset
+#define fb_readb(addr) readb(addr)
+#define fb_readw(addr) readw(addr) 
+#define fb_readl(addr) readl(addr) 
+#define fb_writeb(b,addr) writeb(b,addr) 
+#define fb_writew(b,addr) writew(b,addr) 
+#define fb_writel(b,addr) writel(b,addr) 
+#define fb_memset(addr, c, len) memset_io((unsigned long)addr, c, len)
 
 #endif
 

-- 
Revolutions do not require corporate support.

Re: [parisc-linux] FB cleanups

[Jes Sorensen]

>>>>> "Matthew" == Matthew Wilcox <matthew@wil.cx> writes:

Matthew> I'd like someone to test these out for me -- they compile but
Matthew> I'm monitorless here.

Matthew>  * Turn gsc_read/write[bwl] calls into read/write[bwl] *

Question, does this STI device appear on PCI/ISA cards?

Jes

Re: [parisc-linux] FB cleanups

[Matthew Wilcox]

On Wed, Apr 11, 2001 at 10:30:09PM +0200, Jes Sorensen wrote:
> >>>>> "Matthew" == Matthew Wilcox <matthew@wil.cx> writes:
> 
> Matthew> I'd like someone to test these out for me -- they compile but
> Matthew> I'm monitorless here.
> 
> Matthew>  * Turn gsc_read/write[bwl] calls into read/write[bwl] *
> 
> Question, does this STI device appear on PCI/ISA cards?

STI isn't a device, it's an interface.  STI applies to GSC, EISA & PCI
cards.  The particular chip (Artist) the driver deals with currently is
only on the GSC bus.  But we don't need to have gsc_read/write functions;
they will all go away eventually, I suspect.

-- 
Revolutions do not require corporate support.

Re: [parisc-linux] FB cleanups

[Jes Sorensen]

>>>>> "Matthew" == Matthew Wilcox <matthew@wil.cx> writes:

Matthew> STI isn't a device, it's an interface.  STI applies to GSC,
Matthew> EISA & PCI cards.  The particular chip (Artist) the driver
Matthew> deals with currently is only on the GSC bus.  But we don't
Matthew> need to have gsc_read/write functions; they will all go away
Matthew> eventually, I suspect.

Point is that readb/writeb are *only* defined for PCI and PCI
look-alike devices, even for ISA you're supposed to use
isa_readb/isa_writeb

Jes

Re: [parisc-linux] FB cleanups

[Matthew Wilcox]

On Wed, Apr 11, 2001 at 10:43:52PM +0200, Jes Sorensen wrote:
> Point is that readb/writeb are *only* defined for PCI and PCI
> look-alike devices, even for ISA you're supposed to use
> isa_readb/isa_writeb

No, you're wrong.  isa_* are due to go away in 2.5.  After you've
called ioremap(), you can call readb/writeb on the returned address.
You might want to refer to Documentation/IO-mapping.txt and/or
Documentation/DocBook/deviceiobook.* (only in Alan's tree right now).

-- 
Revolutions do not require corporate support.

Re: [parisc-linux] FB cleanups

[Jes Sorensen]

>>>>> "Matthew" == Matthew Wilcox <matthew@wil.cx> writes:

Matthew> On Wed, Apr 11, 2001 at 10:43:52PM +0200, Jes Sorensen wrote:
>> Point is that readb/writeb are *only* defined for PCI and PCI
>> look-alike devices, even for ISA you're supposed to use
>> isa_readb/isa_writeb

Matthew> No, you're wrong.  isa_* are due to go away in 2.5.  After
Matthew> you've called ioremap(), you can call readb/writeb on the
Matthew> returned address.  You might want to refer to
Matthew> Documentation/IO-mapping.txt and/or
Matthew> Documentation/DocBook/deviceiobook.* (only in Alan's tree
Matthew> right now).

Then Linus changed his mind. You are still *only* supposed to use
readb/writeb on PCI (and possible ISA) on PCI like devices.

Jes

Re: [parisc-linux] FB cleanups

[Alan Cox]

> Then Linus changed his mind. You are still *only* supposed to use
> readb/writeb on PCI (and possible ISA) on PCI like devices.

ISA is a PCI like device, its a wacky slow motion PCI bridge with no 
configuration space ;).

Re: [parisc-linux] FB cleanups

[Matthew Wilcox]

On Thu, Apr 12, 2001 at 12:56:01AM +0200, Jes Sorensen wrote:
> Then Linus changed his mind. You are still *only* supposed to use
> readb/writeb on PCI (and possible ISA) on PCI like devices.

I've heard this before.  But no-one ever came up with an example of what
_was_ `PCI-like'.  Since right now our writeb is #defined to gsc_writeb,
that leads me to believe GSC is sufficiently PCI-like.

-- 
Revolutions do not require corporate support.

Re: [parisc-linux] FB cleanups

[Jes Sorensen]

>>>>> "Alan" == Alan Cox <alan@lxorguk.ukuu.org.uk> writes:

>> Then Linus changed his mind. You are still *only* supposed to use
>> readb/writeb on PCI (and possible ISA) on PCI like devices.

Alan> ISA is a PCI like device, its a wacky slow motion PCI bridge
Alan> with no configuration space ;).

True, but is GSC sufficiently similar? (note, I don't have intimate
knowledge about GSC).

Jes

Re: [parisc-linux] FB cleanups

[Grant Grundler]

Jes Sorensen wrote:
> Then Linus changed his mind. You are still *only* supposed to use
> readb/writeb on PCI (and possible ISA) on PCI like devices.

Jes,
AFAIK, STI devices will all be memory mapped devices.
readb/writeb will do the right thing since GSC is "PCI-like"
in this regard.

BTW, I'm not aware of any HP EISA graphics devices - only GSC devices 
with an EISA form-factor. So the answer is GSC and PCI device.

grant

Grant Grundler
parisc-linux {PCI|IOMMU|SMP} hacker
+1.408.447.7253

Re: [parisc-linux] FB cleanups

[Grant Grundler]

Alan Cox wrote:
> > Then Linus changed his mind. You are still *only* supposed to use
> > readb/writeb on PCI (and possible ISA) on PCI like devices.
> 
> ISA is a PCI like device, its a wacky slow motion PCI bridge with no 
> configuration space ;).

And no memory mapped I/O. Just I/O port space afaik.

grant

Re: [parisc-linux] FB cleanups

[Alan Cox]

> > ISA is a PCI like device, its a wacky slow motion PCI bridge with no 
> > configuration space ;).
> 
> And no memory mapped I/O. Just I/O port space afaik.

ISA bus has MMIO, including busmastering. 

Re: [parisc-linux] FB cleanups

[Jes Sorensen]

>>>>> "Matthew" == Matthew Wilcox <matthew@wil.cx> writes:

Matthew> On Thu, Apr 12, 2001 at 12:56:01AM +0200, Jes Sorensen wrote:
>> Then Linus changed his mind. You are still *only* supposed to use
>> readb/writeb on PCI (and possible ISA) on PCI like devices.

Matthew> I've heard this before.  But no-one ever came up with an
Matthew> example of what _was_ `PCI-like'.  Since right now our writeb
Matthew> is #defined to gsc_writeb, that leads me to believe GSC is
Matthew> sufficiently PCI-like.

Well time to go study the bus behaviors and compare them then. How
does gsc behave wrt cache coherency, read-around-write, write ordering
etc etc.

Using writeb on a gsc is misleading everybody else who reads the code
into believing that it's a PCI device.

Jes

Re: [parisc-linux] FB cleanups

[Jes Sorensen]

>>>>> "Grant" == Grant Grundler <grundler@puffin.external.hp.com> writes:

Grant> Jes Sorensen wrote:
>> Then Linus changed his mind. You are still *only* supposed to use
>> readb/writeb on PCI (and possible ISA) on PCI like devices.

Grant> Jes, AFAIK, STI devices will all be memory mapped devices.
Grant> readb/writeb will do the right thing since GSC is "PCI-like" in
Grant> this regard.

I still don't like it as it is misleading.

Jes

Re: [parisc-linux] FB cleanups

[Grant Grundler]

Alan Cox wrote:
> ISA bus has MMIO, including busmastering. 

really? ok.  I thought the 8237A (DMA chip) on pc's provided the
DMA capability. I didn't think it wasn't inherent to the ISA bus.
But it's been so long (late 80's) since I've seriously mucked with
ISA devices.

ta,
grant

Grant Grundler
parisc-linux {PCI|IOMMU|SMP} hacker
+1.408.447.7253

Re: [parisc-linux] FB cleanups

[Alan Cox]

> Well time to go study the bus behaviors and compare them then. How
> does gsc behave wrt cache coherency, read-around-write, write ordering
> etc etc.

And ISA is noticable for the fact it doesnt behave quite like PCI. You use
readb/readw/readl on ISA.

> Using writeb on a gsc is misleading everybody else who reads the code
> into believing that it's a PCI device.

Rubbish.

Alan

Re: [parisc-linux] FB cleanups

[Alan Cox]

> really? ok.  I thought the 8237A (DMA chip) on pc's provided the
> DMA capability. I didn't think it wasn't inherent to the ISA bus.

It provides some DMA capabilities. Cards can also drive DMA themselves, and
for anything > 2Mhz basically have to.

Re: [parisc-linux] FB cleanups

[Jes Sorensen]

>>>>> "Alan" == Alan Cox <alan@lxorguk.ukuu.org.uk> writes:

>> Well time to go study the bus behaviors and compare them then. How
>> does gsc behave wrt cache coherency, read-around-write, write
>> ordering etc etc.

Alan> And ISA is noticable for the fact it doesnt behave quite like
Alan> PCI. You use readb/readw/readl on ISA.

Yes, except that ISA has stronger semantics than PCI.

>> Using writeb on a gsc is misleading everybody else who reads the
>> code into believing that it's a PCI device.

Alan> Rubbish.

Unfortunately not, look at the fb code to see the chaos.

Jes

Re: [parisc-linux] FB cleanups

[Matthew Wilcox]

On Tue, Apr 17, 2001 at 08:04:53PM +0200, Jes Sorensen wrote:
> >>>>> "Alan" == Alan Cox <alan@lxorguk.ukuu.org.uk> writes:
> 
> Yes, except that ISA has stronger semantics than PCI.

You need to issue a read after a write to be sure the write hit the
device on ISA?

> >> Using writeb on a gsc is misleading everybody else who reads the
> >> code into believing that it's a PCI device.
> 
> Alan> Rubbish.
> 
> Unfortunately not, look at the fb code to see the chaos.

huh?  the fb code is ugly precisely because people aren't using
readb/writeb to access device memory.

-- 
Revolutions do not require corporate support.

Re: [parisc-linux] FB cleanups

[Alan Cox]

> > Yes, except that ISA has stronger semantics than PCI.
> 
> You need to issue a read after a write to be sure the write hit the
> device on ISA?

In some cases yes.