[parisc-linux] [grundler@puffin.external.hp.com: commit FP coproc enable code?]

[parisc-linux] [grundler@puffin.external.hp.com: commit FP coproc enable code?]

[Phil Schwan]

[-- Attachment #1: Type: text/plain, Size: 99 bytes --]

[ I think I've fixed the mailing list; here's the mails from Grant
  that it was dropping --phil ]

[-- Attachment #2: Forwarded message from grundler@puffin.external.hp.com --]
[-- Type: message/rfc822, Size: 7101 bytes --]

From: grundler@puffin.external.hp.com
To: parisc-linux@thepuffingroup.com
Subject: commit FP coproc enable code?
Date: Fri, 1 Oct 1999 11:25:50 -0600
Message-ID: <199910011724.LAA09983@puffin.external.hp.com>

[ 4th try...I'm not getting a copy back from the reflector...]

Hi all,
I'd like to commit changes to enable FP coproc for PA 1.1 arch.
Any objections?

o FP enable via writing cr10 (CCR) register. Not doing this the "right"
  way - which is to use PDC_COPROC but it will work for all PA1.1
  platforms.

o remove extraneous console output. I want to post console output
  for my next problem: kernel panic when trying to kill idle_task.

thanks,
grant

Index: linux/arch/parisc/kernel/head.S
===================================================================
RCS file: /home/cvs/parisc/linux/arch/parisc/kernel/head.S,v
retrieving revision 1.32
diff -u -r1.32 head.S
--- head.S      1999/08/31 19:25:23     1.32
+++ head.S      1999/09/30 22:43:03
@@ -295,9 +295,20 @@
        ldil            L%KERNEL_PSW,%r10
        ldo             R%KERNEL_PSW(%r10),%r10
        mtctl           %r10,%ipsw
-       
-       mtctl           %r0,%cr17               ; Clear two-level IIA Space Queue
-       mtctl           %r0,%cr17               ;    effectively setting kernel space.
+
+       ; Just assume we have an FP coprocessor and assume firmware
+       ; did not enable it.
+       ; Pass what firmware set CCR (cr10) to start_parisc() and print it.
+       mfctl           %cr10, %arg2
+
+#define FP_COPROC_ENABLE 0xc0
+       ldil            L%FP_COPROC_ENABLE,%r10
+       ldo             R%FP_COPROC_ENABLE(%r10),%r10
+       mtctl           %r10,%cr10      ; CCR
+
+
+       mtctl           %r0,%cr17       ; Clear two-level IIA Space Queue
+       mtctl           %r0,%cr17       ;    effectively setting kernel space.
        ldil            L%start_parisc,%r10
        ldo             R%start_parisc(%r10),%r10
 
Index: linux/arch/parisc/kernel/setup.c
===================================================================
RCS file: /home/cvs/parisc/linux/arch/parisc/kernel/setup.c,v
retrieving revision 1.59
diff -u -r1.59 setup.c
--- setup.c     1999/09/28 22:34:57     1.59
+++ setup.c     1999/09/30 22:43:15
@@ -587,13 +587,16 @@
 #define PA(x) (void *)((unsigned long)(x)&~0xc0000000)
 extern int pdc_call_x(void *,...);
 
+#if 0
 static char iodc_tbuf[4096] __attribute__((aligned(4096))) = "Hallo\n";
 static char iodc_rbuf[4096] __attribute__((aligned(4096))) = "Hallo\n";
+#endif
 
 __initfunc(void start_parisc(unsigned long new_memory_start, 
-                            char *bootloader_command_line))
+                            char *bootloader_command_line,
+                            unsigned long ccr))
 {
-#if 1
+#if 0
        pdc_call_x(
                (void *)PAGE0->mem_cons.hpa, ENTRY_IO_COUT,
                PAGE0->mem_cons.spa, PAGE0->mem_cons.dp.layers,
@@ -613,28 +616,24 @@
        pdc_console_init();
 
        printk("The Kernel has started...\n");
+#if 0
        printk("Hmm\n");
        printk("%d\n", 7);
        printk("%d\n", *(unsigned *)0xc0000000);
-
+#endif
        printk("%p\n", bootloader_command_line);
        printk("You are running a %dMHz PA-RISC-Machine with %dMB of RAM.\n",
                ((int)PAGE0->mem_10msec/10000), PAGE0->imm_max_mem>>20);
+       printk("FP CCR was set to: 0x%x\n", (int) ccr);
        parisc_memstart = new_memory_start;
        
        pdc = PAGE0->mem_pdc;
 
-       printk("You are running a %dMHz PA-RISC-Machine with %dMB of RAM.\n",
-               ((int)PAGE0->mem_10msec/10000), PAGE0->imm_max_mem>>20);
-
        if (bootloader_command_line)
            strcpy(saved_command_line,__va(bootloader_command_line));
        else
            saved_command_line[0] = 0;
 
-       printk("You are running a %dMHz PA-RISC-Machine with %dMB of RAM.\n",
-               ((int)PAGE0->mem_10msec/10000), PAGE0->imm_max_mem>>20);
-
        printk("Free memory was at 0x%08lX (=stackbase, stacksize=0x%08lX).\n",
                new_memory_start,(unsigned long)KERNEL_STACKSIZE );
 
@@ -659,7 +658,7 @@
 #endif
        initrd_below_start_ok = 1;
        
-       printk("initrd: %08x-%08x\n", initrd_start, initrd_end);
+       printk("initrd: %08x-%08x\n", (int) initrd_start, (int) initrd_end);
        *memory_start_p = parisc_memstart;
        *memory_end_p   = 0xc2000000;
 
Index: linux/drivers/block/rd.c
===================================================================
RCS file: /home/cvs/parisc/linux/drivers/block/rd.c,v
retrieving revision 1.3
diff -u -r1.3 rd.c
--- rd.c        1999/09/22 18:10:17     1.3
+++ rd.c        1999/09/30 22:43:26
@@ -219,7 +219,6 @@
 {
        int left;
 
-       printk("initrd_read\n");
        left = initrd_end - initrd_start - *ppos;
        if (count > left) count = left;
        if (count == 0) return 0;

Index: linux/fs/buffer.c
===================================================================
RCS file: /home/cvs/parisc/linux/fs/buffer.c,v
retrieving revision 1.3
diff -u -r1.3 buffer.c
--- buffer.c    1999/09/22 18:10:35     1.3
+++ buffer.c    1999/09/30 22:44:21
@@ -840,8 +840,6 @@
 
        struct buffer_head * bh;
 
-       printk("bread bh=%08x\n", bh);
-//     printk("%08x\n", &buffer_page(bh)->flags);
        bh = getblk(dev, block, size);
        if (buffer_uptodate(bh))
                return bh;

Re: [parisc-linux] commit FP coproc enable code?

[Philipp Rumpf]

> o FP enable via writing cr10 (CCR) register. Not doing this the "right"
>   way - which is to use PDC_COPROC but it will work for all PA1.1
>   platforms.

Why is not using PDC_COPROC not the right way ?  It should do exactly the same
thing, modulo machines without FPU (for which we do not provide emulation
anyway) and firmware bugs ...

> ===================================================================
> RCS file: /home/cvs/parisc/linux/arch/parisc/kernel/head.S,v
> retrieving revision 1.32
> diff -u -r1.32 head.S
> --- head.S      1999/08/31 19:25:23     1.32
> +++ head.S      1999/09/30 22:43:03

I would prefer to do this somewhere after the kernel is loaded, in C, surrounded
by a pair of printks, so we know whether it is this that hangs the machine.

Re: [parisc-linux] commit FP coproc enable code?

[Grant Grundler]

Philipp Rumpf wrote:
> > o FP enable via writing cr10 (CCR) register. Not doing this the "right"
> >   way - which is to use PDC_COPROC but it will work for all PA1.1
> >   platforms.
> 
> Why is not using PDC_COPROC not the right way ?
> It should do exactly the same
> thing, modulo machines without FPU (for which we do not provide emulation
> anyway) and firmware bugs ...

You are right.
But I'm lazy and I don't care right now about machines older than PA 1.1.
Someone who cares about PA 1.0 can implement that.

> 
> > ===================================================================
> > RCS file: /home/cvs/parisc/linux/arch/parisc/kernel/head.S,v
> > retrieving revision 1.32
> > diff -u -r1.32 head.S
> > --- head.S      1999/08/31 19:25:23     1.32
> > +++ head.S      1999/09/30 22:43:03
> 
> I would prefer to do this somewhere after the kernel is loaded, in C,
> surrounded
> by a pair of printks, so we know whether it is this that hangs the machine.

Agreed...I'll do that....thanks!

grant


Grant Grundler
Communications Infrastructure Computer Operations
+1.408.447.7253

Re: [parisc-linux] commit FP coproc enable code?

[Jeffrey A Law]

  In message <199910012052.NAA13364@milano.cup.hp.com>you write:

  > > Why is not using PDC_COPROC not the right way ?
  > > It should do exactly the same
  > > thing, modulo machines without FPU (for which we do not provide emulation
  > > anyway) and firmware bugs ...
  > 
  > You are right.
  > But I'm lazy and I don't care right now about machines older than PA 1.1.
  > Someone who cares about PA 1.0 can implement that.
I would recommend ignoring PA1.0 -- the only machines with PA1.0 processors
are radically different than the machines y'all are targeting.  The effort
would be spent on PA2.0 :-)


jeff