Guys, can you help with http://bugme.osdl.org/show_bug.cgi?id=2711 ?
The recent change broke it on x86 becuase we're now doing ioremap()
of a kernel-virtual address, but the oriiginal version doesn't work on ARM.
Should it just be:
vga16fb.screen_base = VGA_MAP_MEM(VGA_FB_PHYS);
I went looking at the various platforms to see what exactly was going on.
We have:
Useage
./drivers/video/console/mdacon.c: mda_vram_base = VGA_MAP_MEM(0xb0000);
./drivers/video/console/vgacon.c: vga_vram_base = VGA_MAP_MEM(vga_vram_base);
./drivers/video/console/vgacon.c: vga_vram_end = VGA_MAP_MEM(vga_vram_end);
./drivers/video/console/vgacon.c: charmap = (char *) VGA_MAP_MEM(colourmap);
./drivers/video/console/vgacon.c: charmap = (char *) VGA_MAP_MEM(blackwmap);
./drivers/video/hgafb.c: hga_fix.smem_start = VGA_MAP_MEM(hga_vram_base);
./drivers/video/vga16fb.c: vga16fb.screen_base = ioremap(VGA_MAP_MEM(VGA_FB_PHYS), VGA_FB_PHYS_LEN);
./drivers/video/vga16fb.c: vga16fb.fix.smem_start = VGA_MAP_MEM(vga16fb.fix.smem_start);
ioremap happy
./include/asm-alpha/vga.h:#define VGA_MAP_MEM(x) ((unsigned long) ioremap((x), 0))
./include/asm-ia64/vga.h:#define VGA_MAP_MEM(x) ((unsigned long) ioremap((x), 0))
./include/asm-ppc64/vga.h:#define VGA_MAP_MEM(x) ((unsigned long) ioremap((x), 0))
can be ioremap happy because ioremap in io.h matches below definetions.
./include/asm-i386/vga.h:#define VGA_MAP_MEM(x) (unsigned long)phys_to_virt(x)
./include/asm-sparc64/vga.h:#define VGA_MAP_MEM(x) (x)
./include/asm-x86_64/vga.h:#define VGA_MAP_MEM(x) (unsigned long)phys_to_virt(x)
Not happy.
./include/asm-arm/vga.h:#define VGA_MAP_MEM(x) (PCIMEM_BASE + (x))
./include/asm-mips/vga.h:#define VGA_MAP_MEM(x) (0xb0000000L + (unsigned long)(x))
./include/asm-ppc/vga.h:#define VGA_MAP_MEM(x) (x + vgacon_remap_base)
So you can see that VGA_MAP_MEM is already ioremapping which is causing
the problem. Personally I like to see the lose ends fixed on ARM, MIPS,
and PPC so we can use just ioremap.
-------------------------------------------------------
This SF.Net email is sponsored by: SourceForge.net Broadband
Sign-up now for SourceForge Broadband and get the fastest
6.0/768 connection for only $19.95/mo for the first 3 months!
http://ads.osdn.com/?ad_id=2562&alloc_id=6184&op=click
James Simmons [off-list ref] wrote:
quoted
Guys, can you help with http://bugme.osdl.org/show_bug.cgi?id=2711 ?
The recent change broke it on x86 becuase we're now doing ioremap()
of a kernel-virtual address, but the oriiginal version doesn't work on ARM.
Should it just be:
vga16fb.screen_base = VGA_MAP_MEM(VGA_FB_PHYS);
I went looking at the various platforms to see what exactly was going on.
We have:
Useage
./drivers/video/console/mdacon.c: mda_vram_base = VGA_MAP_MEM(0xb0000);
./drivers/video/console/vgacon.c: vga_vram_base = VGA_MAP_MEM(vga_vram_base);
./drivers/video/console/vgacon.c: vga_vram_end = VGA_MAP_MEM(vga_vram_end);
./drivers/video/console/vgacon.c: charmap = (char *) VGA_MAP_MEM(colourmap);
./drivers/video/console/vgacon.c: charmap = (char *) VGA_MAP_MEM(blackwmap);
./drivers/video/hgafb.c: hga_fix.smem_start = VGA_MAP_MEM(hga_vram_base);
./drivers/video/vga16fb.c: vga16fb.screen_base = ioremap(VGA_MAP_MEM(VGA_FB_PHYS), VGA_FB_PHYS_LEN);
./drivers/video/vga16fb.c: vga16fb.fix.smem_start = VGA_MAP_MEM(vga16fb.fix.smem_start);
ioremap happy
./include/asm-alpha/vga.h:#define VGA_MAP_MEM(x) ((unsigned long) ioremap((x), 0))
./include/asm-ia64/vga.h:#define VGA_MAP_MEM(x) ((unsigned long) ioremap((x), 0))
./include/asm-ppc64/vga.h:#define VGA_MAP_MEM(x) ((unsigned long) ioremap((x), 0))
can be ioremap happy because ioremap in io.h matches below definetions.
./include/asm-i386/vga.h:#define VGA_MAP_MEM(x) (unsigned long)phys_to_virt(x)
./include/asm-sparc64/vga.h:#define VGA_MAP_MEM(x) (x)
./include/asm-x86_64/vga.h:#define VGA_MAP_MEM(x) (unsigned long)phys_to_virt(x)
Not happy.
./include/asm-arm/vga.h:#define VGA_MAP_MEM(x) (PCIMEM_BASE + (x))
./include/asm-mips/vga.h:#define VGA_MAP_MEM(x) (0xb0000000L + (unsigned long)(x))
./include/asm-ppc/vga.h:#define VGA_MAP_MEM(x) (x + vgacon_remap_base)
So you can see that VGA_MAP_MEM is already ioremapping which is causing
the problem. Personally I like to see the lose ends fixed on ARM, MIPS,
and PPC so we can use just ioremap.
I have pondered your email at length and have failed to understand it.
I _think_ you're saying that we need to do this, which will fix x86:
--- 25/drivers/video/vga16fb.c~vga16fb-fix Tue May 18 17:10:14 2004
+++ 25-akpm/drivers/video/vga16fb.c Tue May 18 17:10:39 2004
@@ -1347,7 +1347,7 @@ int __init vga16fb_init(void)
/* XXX share VGA_FB_PHYS and I/O region with vgacon and others */
- vga16fb.screen_base = ioremap(VGA_MAP_MEM(VGA_FB_PHYS), VGA_FB_PHYS_LEN);
+ vga16fb.screen_base = VGA_MAP_MEM(VGA_FB_PHYS);
if (!vga16fb.screen_base) {
printk(KERN_ERR "vga16fb: unable to map device\n");
ret = -ENOMEM;
_
and that ARM and others need to teach their VGA_MAP_MEM() to do an internal
ioremap().
Or do you mean something else? Please be more clear?
quoted hunk
I have pondered your email at length and have failed to understand it.
I _think_ you're saying that we need to do this, which will fix x86:
--- 25/drivers/video/vga16fb.c~vga16fb-fix Tue May 18 17:10:14 2004
+++ 25-akpm/drivers/video/vga16fb.c Tue May 18 17:10:39 2004
@@ -1347,7 +1347,7 @@ int __init vga16fb_init(void)
/* XXX share VGA_FB_PHYS and I/O region with vgacon and others */
- vga16fb.screen_base = ioremap(VGA_MAP_MEM(VGA_FB_PHYS), VGA_FB_PHYS_LEN);
+ vga16fb.screen_base = VGA_MAP_MEM(VGA_FB_PHYS);
if (!vga16fb.screen_base) {
printk(KERN_ERR "vga16fb: unable to map device\n");
ret = -ENOMEM;
This will make the driver on all platforms.
_
and that ARM and others need to teach their VGA_MAP_MEM() to do an internal
ioremap().
Or do you mean something else? Please be more clear?
I like to see the VGA_MAP_MEM hack go away and be replaced with ioremap.
James Simmons [off-list ref] wrote:
quoted
I have pondered your email at length and have failed to understand it.
I _think_ you're saying that we need to do this, which will fix x86:
--- 25/drivers/video/vga16fb.c~vga16fb-fix Tue May 18 17:10:14 2004
+++ 25-akpm/drivers/video/vga16fb.c Tue May 18 17:10:39 2004
@@ -1347,7 +1347,7 @@ int __init vga16fb_init(void)
/* XXX share VGA_FB_PHYS and I/O region with vgacon and others */
- vga16fb.screen_base = ioremap(VGA_MAP_MEM(VGA_FB_PHYS), VGA_FB_PHYS_LEN);
+ vga16fb.screen_base = VGA_MAP_MEM(VGA_FB_PHYS);
if (!vga16fb.screen_base) {
printk(KERN_ERR "vga16fb: unable to map device\n");
ret = -ENOMEM;
This will make the driver on all platforms.
There's a missing word in that sentence. Was it "work" or "crash"? This
matters ;)
_
quoted
and that ARM and others need to teach their VGA_MAP_MEM() to do an internal
ioremap().
Or do you mean something else? Please be more clear?
I like to see the VGA_MAP_MEM hack go away and be replaced with ioremap.
If you can cut a patch we can ask arch maintainers to review and test it.
quoted
quoted
/* XXX share VGA_FB_PHYS and I/O region with vgacon and others */
- vga16fb.screen_base = ioremap(VGA_MAP_MEM(VGA_FB_PHYS), VGA_FB_PHYS_LEN);
+ vga16fb.screen_base = VGA_MAP_MEM(VGA_FB_PHYS);
if (!vga16fb.screen_base) {
printk(KERN_ERR "vga16fb: unable to map device\n");
ret = -ENOMEM;
This will make the driver on all platforms.
There's a missing word in that sentence. Was it "work" or "crash"? This
matters ;)
works :-)
quoted
quoted
and that ARM and others need to teach their VGA_MAP_MEM() to do an internal
ioremap().
Or do you mean something else? Please be more clear?
I like to see the VGA_MAP_MEM hack go away and be replaced with ioremap.
If you can cut a patch we can ask arch maintainers to review and test it.
I could but that would take some time. For now apply your fix.
Actually Ihave a newer vga16 driver for you I can push you.