Re: Sun4/110 "Panic: Crazy Interrupts"

[Julian Coleman <jdc%coris.org.uk@localhost>]

Hi,

> I commented out that block and compiled again - I had to add
> -Wno-unused to the generated Makefile to get it to build since it
> didn't like 'bt' and 'i' hanging around unused.
> 
> It looks like the stray interrupts still happen, and we still panic in
> that further along spot from before. So I assume we'd just disable
> these interrupts?

I wondered if that might happen.  Looking at the P4 register definitions:

  https://nxr.netbsd.org/xref/src/sys/dev/sun/pfourreg.h

there is PFOUR_REG_INTEN, so I wonder if this is set and we can clear it on
attach.  I've attached a patch that tries to adjust the FBC DAC registers to
be a mix of char and int, plus unset the interrupt enable bit in the P4 reg.
I assume that because we pass a shift of 24 to BT_INIT in the old code that
we want to access bt_addr[3] and bt_ctrl[3].  I hope that this will work,
otherwise we'll need to think of something else.

There is also an interrupt enable register in the FBC control register:

  https://nxr.netbsd.org/xref/src/sys/dev/sun/btreg.h#118

so we might need to also disable that, but I didn't do that in the patch
(we should be able to access it via &sc->sc_fbc->fbc_ctrl).

Regards,

Julian

-- 

Index: sys/arch/sparc/dev/cgfour.c
===================================================================
RCS file: /cvsroot/src/sys/arch/sparc/dev/cgfour.c,v
retrieving revision 1.50
diff -u -r1.50 cgfour.c
--- sys/arch/sparc/dev/cgfour.c	18 Oct 2014 08:33:26 -0000	1.50
+++ sys/arch/sparc/dev/cgfour.c	19 Apr 2024 14:01:49 -0000
@@ -124,12 +124,35 @@
 #include <dev/sun/pfourreg.h>
 
 /* per-display variables */
+struct p4bt_regs {
+	u_char	bt_addr[4];
+	u_int	bt_cmap;
+	u_char	bt_ctrl[4];
+	u_int	bt_omap;
+};
+#define P4BT_INIT(bt) do { \
+ 	(bt)->bt_addr[3] = 0x06;	/* command reg */ \
+ 	(bt)->bt_ctrl[3] = 0x73;	/* overlay plane */ \
+ 	(bt)->bt_addr[3] = 0x04;	/* read mask */ \
+ 	(bt)->bt_ctrl[3] = 0xff;	/* color planes */ \
+} while(0)
+
+
+struct p4fbcontrol {
+	struct	p4bt_regs fbc_dac;
+	u_char	fbc_ctrl;
+ 	u_char	fbc_status;
+ 	u_char	fbc_cursor_start;
+ 	u_char	fbc_cursor_end;
+ 	u_char	fbc_vcontrol[12];	/* 12 bytes of video timing goo */
+};
+
 struct cgfour_softc {
 	struct fbdevice	sc_fb;		/* frame buffer device */
 	bus_space_tag_t	sc_bustag;
 	bus_addr_t	sc_paddr;	/* phys address for device mmap() */
 
-	volatile struct fbcontrol *sc_fbc;	/* Brooktree registers */
+	volatile struct p4fbcontrol *sc_fbc;	/* Brooktree registers */
 	union bt_cmap	sc_cmap;	/* Brooktree color map */
 };
 
@@ -305,7 +328,7 @@
 	for (i = 0; i < 256 * 3 / 4; i++)
 		((char *)&sc->sc_cmap)[i] = bt->bt_cmap >> 24;
 
-	BT_INIT(bt, 24);
+	P4BT_INIT(bt);
 
 #if 0	/* See above. */
 	if (isconsole) {
@@ -317,6 +340,10 @@
 #endif /* 0 */
 		printf("\n");
 
+	/* Disable interrupts */
+printf("P4 register = 0x%04x\n", fb->fb_pfour);
+	fb->fb_pfour &= ~(PFOUR_REG_INTEN);
+
 	/*
 	 * Even though we're not using rconsole, we'd still like
 	 * to notice if we're the console framebuffer.