Thread (23 messages) flat view 23 messages, 5 authors, 2025-11-06

Re: [PATCH v4] powerpc: Use shared font data

From: Christophe Leroy <hidden>
Date: 2025-11-05 22:20:04
Also in: lkml
Subsystem: framebuffer console, library code, linux for powerpc (32-bit and 64-bit), the rest · Maintainers: Helge Deller, Thomas Zimmermann, Andrew Morton, Madhavan Srinivasan, Linus Torvalds


Le 03/11/2025 à 02:48, Dr. David Alan Gilbert a écrit :
* Finn Thain (fthain@linux-m68k.org) wrote:
quoted
On Sun, 2 Nov 2025, Dr. David Alan Gilbert wrote:
quoted
So I'm not a PPC person specifically; so lets see if the PPC people have
any suggestions, but:

    a) Do you know if there's any way to recreate the same hang/works
combination in qemu; I know it has a g3beige model but I don't know how
to get something similar to your failing combo.
I guess we could probably reproduce this in QEMU if the BootX bootloader
could be made to work there. In theory, 'qemu-system-ppc -M g3beige' might
work.
quoted
    b) Can you get any diagnostics out of the prom on the mac?  Like a PC
or anything to have some idea where it hung?
Well, that's the problem: if you enable the CONFIG_BOOTX_TEXT diagnostics,
the system hangs instead of printing stuff. If you disable the
CONFIG_BOOTX_TEXT diagnostics (in favour of serial diagnostics) the hang
goes away.
Ah, a bug that doesn't like to be seen :-)
quoted
Anyway, I imagine that the problem with your patch was that it relies on
font data from a different (read only) section, which is unavailable for
some reason (MMU not fully configured yet?)

So I've asked Stan to test a patch that simply removes the relevant
'const' keywords. It's not a solution, but might narrow-down the search.
I wonder if this is a compiler-flag-ism; I see arch/powerpc/kernel/Makefile
has a pile of special flags, and for btext.o it has a -fPIC
(as well as turning off some other flags).
I wonder if bodging those in lib/fonts/Makefile for lib/fonts/font_sun8x16.c
fixes it?
But... this is data - there's no code is there - are any of those flags
relevant for data only?
I think -fPIC is relevant for data-only here because font_sun_8x16 
contains a pointer to fontdata_sun8x16 in font_sun_8x16.data

I see two things to try:

1/ Either build font_sun8x16.o with -fPIC
diff --git a/lib/fonts/Makefile b/lib/fonts/Makefile
index e16f68492174a..844306d7b15e9 100644
--- a/lib/fonts/Makefile
+++ b/lib/fonts/Makefile
@@ -20,3 +20,5 @@ font-objs-$(CONFIG_FONT_6x8)       += font_6x8.o
  font-objs += $(font-objs-y)

  obj-$(CONFIG_FONT_SUPPORT)         += font.o
+
+CFLAGS_font_sun8x16.o             += -fPIC
2/ Or add a PTRRELOC:
diff --git a/arch/powerpc/kernel/btext.c b/arch/powerpc/kernel/btext.c
index 7f63f1cdc6c39..fc461cfaf4a34 100644
--- a/arch/powerpc/kernel/btext.c
+++ b/arch/powerpc/kernel/btext.c
@@ -20,6 +20,7 @@
  #include <asm/io.h>
  #include <asm/processor.h>
  #include <asm/udbg.h>
+#include <asm/setup.h>

  #define NO_SCROLL
@@ -463,7 +464,7 @@ static noinline void draw_byte(unsigned char c, long 
locX, long locY)
  {
  	unsigned char *base	= calc_base(locX << 3, locY << 4);
  	unsigned int font_index = c * 16;
-	const unsigned char *font	= font_sun_8x16.data + font_index;
+	const unsigned char *font	= PTRRELOC(font_sun_8x16.data) + font_index;
  	int rb			= dispDeviceRowBytes;

  	rmci_maybe_on();


Christophe


Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help