[PATCH 0/6] Mac OS 9 boot fixes

[Cormac O'Brien]

These patches fix a number of issues that hindered the Mac OS 9 boot process.

Signed-off-by: Cormac O'Brien

Cormac O'Brien (6):
  bootinfo_load: recognize ASCII EOT as end of bootinfo
  ppc: add Adler-32 checksum capability
  ppc: add ROM node to device tree
  ppc: fix stack usage in mmu_claim, mem_claim
  ppc: add CPU reservation-granule-size property
  ppc: add Apple copyright injection to adler32

arch/ppc/qemu/init.c        | 62 +++++++++++++++++++++++++++++++++++++++++++++
 arch/ppc/qemu/methods.c     | 34 ++++++++++++-------------
 arch/ppc/qemu/qemu.fs       | 28 ++++++++++++++++++++
 arch/ppc/qemu/tree.fs       |  8 ++++++
 libopenbios/bootinfo_load.c | 10 ++++++++
 5 files changed, 125 insertions(+), 17 deletions(-)

[PATCH 1/6] bootinfo_load: recognize ASCII EOT as end of bootinfo

[Cormac O'Brien]

Mac OS 9's CHRP boot info consists of the standard SGML document terminated by
an ASCII EOT, followed by a large binary blob which is relocated by the Forth
script. This patch allows the bootinfo loader to stop loading the script if it
detects an EOT, avoiding the out-of-memory error previously encountered when
booting OS 9.

Signed-off-by: Cormac O'Brien 
---
 libopenbios/bootinfo_load.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/libopenbios/bootinfo_load.c b/libopenbios/bootinfo_load.c
index fa9e36b..4b6a52c 100644
--- a/libopenbios/bootinfo_load.c
+++ b/libopenbios/bootinfo_load.c
@@ -161,6 +161,16 @@ bootinfo_init_program(void)
    feval("load-size");
    size = POP();

+	/* Check for a delimiting ASCII EOT, as in Mac OS 9 */
+	current = 0;
+	while (current < size) {
+		if (base[current] == 0x04) {
+			size = current;
+			break;
+		}
+		current++;
+	}
+
    bootscript = malloc(size);
    if (bootscript == NULL) {
    	DPRINTF("Can't malloc %d bytes\n", size);

[PATCH 2/6] ppc: add Adler-32 checksum capability

[Cormac O'Brien]

This patch provides an implementation of the adler32 Forth word as required by
Mac OS 9 and BootX.

Signed-off-by: Cormac O'Brien 
---
 arch/ppc/qemu/init.c | 57 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 57 insertions(+)

diff --git a/arch/ppc/qemu/init.c b/arch/ppc/qemu/init.c
index 4fe8b72..5cc45ae 100644
--- a/arch/ppc/qemu/init.c
+++ b/arch/ppc/qemu/init.c
@@ -680,6 +680,60 @@ static void ffilll(void)
     }   
 }

+/*
+ * adler32        ( adler buf len -- checksum )
+ *
+ * Adapted from Mark Adler's original implementation (zlib license)
+ *
+ * Both OS 9 and BootX require this word for payload validation.
+ */
+
+#define DO1(buf,i)  {s1 += buf[i]; s2 += s1;}
+#define DO2(buf,i)  DO1(buf,i); DO1(buf,i+1);
+#define DO4(buf,i)  DO2(buf,i); DO2(buf,i+2);
+#define DO8(buf,i)  DO4(buf,i); DO4(buf,i+4);
+#define DO16(buf)   DO8(buf,0); DO8(buf,8);
+
+static void adler32(void)
+{
+    uint32_t len = (uint32_t)POP();
+    char *buf = (char *)POP();
+    uint32_t adler = (uint32_t)POP();
+
+    if (buf == NULL) {
+        RET(-1);
+    }
+
+    uint32_t base = 65521;
+    uint32_t nmax = 5552;
+
+    uint32_t s1 = adler & 0xffff;
+    uint32_t s2 = (adler >> 16) & 0xffff;
+
+    uint32_t k;
+    while (len > 0) {
+        k = (len < nmax ? len : nmax);
+        len -= k;
+
+        while (k >= 16) {
+            DO16(buf);
+            buf += 16;
+            k -= 16;
+        }
+        if (k != 0) {
+            do {
+                s1 += *buf++;
+                s2 += s1;
+            } while (--k);
+        }
+
+        s1 %= base;
+        s2 %= base;
+    }
+
+    RET(s2 << 16 | s1);
+}
+
 void
 arch_of_init(void)
 {
@@ -945,6 +999,9 @@ arch_of_init(void)

/* Implementation of filll word (required by BootX) */
     bind_func("filll", ffilll);
+
+    /* Implementation of adler32 word (required by OS 9, BootX) */
+    bind_func("adler32", adler32);

bind_func("platform-boot", boot);
     bind_func("(go)", go);

[PATCH 3/6] ppc: add ROM node to device tree

[Cormac O'Brien]

This patch adds a ROM node to the PPC device tree as is required by Mac OS 9.

Based on a patch by John Arbuckle

Signed-off-by: Cormac O'Brien 
---
 arch/ppc/qemu/tree.fs | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/arch/ppc/qemu/tree.fs b/arch/ppc/qemu/tree.fs
index 1ed8383..5b6bbc6 100644
--- a/arch/ppc/qemu/tree.fs
+++ b/arch/ppc/qemu/tree.fs
@@ -42,6 +42,14 @@ new-device
    : close ;
 finish-device

+new-device
+	" rom" device-name
+	h# ff800000 encode-int 0 encode-int encode+ " reg" property
+	1 encode-int " #address-cells" property
+	h# ff800000 encode-int h# 800000 encode-int encode+
+	h# ff800000 encode-int encode+ " ranges" property
+finish-device
+
 \ -------------------------------------------------------------
 \ /packages
 \ -------------------------------------------------------------

[PATCH 4/6] ppc: fix stack usage in mmu_claim, mem_claim

[Cormac O'Brien]

The 'phys' argument to mem_claim() and 'virt' argument to mmu_claim() are now
only popped from the stack if the 'align' argument is provided. Exception
throws have been removed to simplify crossing C <-> Forth boundaries and to
maintain consistency with other architectures.

This patch also fixes two stack diagrams with 'phys' arguments instead of
'virt' ones.

Signed-off-by: Cormac O'Brien 
---
 arch/ppc/qemu/methods.c | 34 +++++++++++++++++-----------------
 1 file changed, 17 insertions(+), 17 deletions(-)

diff --git a/arch/ppc/qemu/methods.c b/arch/ppc/qemu/methods.c
index fd993da..cb72bc1 100644
--- a/arch/ppc/qemu/methods.c
+++ b/arch/ppc/qemu/methods.c
@@ -164,21 +164,21 @@ DECLARE_UNNAMED_NODE( mmu, INSTALL_OPEN, 0 );
 DECLARE_NODE( mmu_ciface, 0, 0, "+/openprom/client-services" );

-/* ( phys size align --- base ) */
+/* ( [phys] size align --- base ) */
 static void
 mem_claim( void )
 {
    ucell align = POP();
    ucell size = POP();
-	ucell phys = POP();
-	ucell ret = ofmem_claim_phys( phys, size, align );
+	phys_addr_t phys = -1;

-	if( ret == -1 ) {
-		printk("MEM: claim failure\n");
-		throw( -13 );
-		return;
+	if (!align) {
+		phys = POP();
    }
-	PUSH( ret );
+
+	phys = ofmem_claim_phys(phys, size, align);
+
+	PUSH(phys);
 }

/* ( phys size --- ) */
@@ -188,24 +188,24 @@ mem_release( void )
    POP(); POP();
 }

-/* ( phys size align --- base ) */
+/* ( [virt] size align --- base ) */
 static void
 mmu_claim( void )
 {
    ucell align = POP();
    ucell size = POP();
-	ucell phys = POP();
-	ucell ret = ofmem_claim_virt( phys, size, align );
+	ucell virt = -1;

-	if( ret == -1 ) {
-		printk("MMU: CLAIM failure\n");
-		throw( -13 );
-		return;
+	if (!align) {
+		virt = POP();
    }
-	PUSH( ret );
+
+	virt = ofmem_claim_virt(virt, size, align);
+
+	PUSH(virt);
 }

-/* ( phys size --- ) */
+/* ( virt size --- ) */
 static void
 mmu_release( void )
 {

[PATCH 5/6] ppc: add CPU reservation-granule-size property

[Cormac O'Brien]

This patch adds the 'reservation-granule-size' property to the CPU node, as is
required by Mac OS 9.

Signed-off-by: Cormac O'Brien 
---
 arch/ppc/qemu/init.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/arch/ppc/qemu/init.c b/arch/ppc/qemu/init.c
index 5cc45ae..180679c 100644
--- a/arch/ppc/qemu/init.c
+++ b/arch/ppc/qemu/init.c
@@ -302,6 +302,11 @@ cpu_generic_init(const struct cpudef *cpu)
     fword("encode-string");
     push_str("state");
     fword("property");
+
+    PUSH(0x20);
+    fword("encode-int");
+    push_str("reservation-granule-size");
+    fword("property");
 }

static void

[PATCH 6/6] ppc: add Apple copyright injection to adler32

[Cormac O'Brien]

Since Mac OS 9 and BootX are the only known users of the adler32 word, we use
the word to add an Apple copyright message to the device tree in order to
allow OS 9 to boot.

Signed-off-by: Cormac O'Brien 
---
 arch/ppc/qemu/init.c  |  2 +-
 arch/ppc/qemu/qemu.fs | 28 ++++++++++++++++++++++++++++
 2 files changed, 29 insertions(+), 1 deletion(-)

diff --git a/arch/ppc/qemu/init.c b/arch/ppc/qemu/init.c
index 180679c..2b5b8e1 100644
--- a/arch/ppc/qemu/init.c
+++ b/arch/ppc/qemu/init.c
@@ -1006,7 +1006,7 @@ arch_of_init(void)
     bind_func("filll", ffilll);

/* Implementation of adler32 word (required by OS 9, BootX) */
-    bind_func("adler32", adler32);
+    bind_func("(adler32)", adler32);

bind_func("platform-boot", boot);
     bind_func("(go)", go);
diff --git a/arch/ppc/qemu/qemu.fs b/arch/ppc/qemu/qemu.fs
index 458af1b..11e344a 100644
--- a/arch/ppc/qemu/qemu.fs
+++ b/arch/ppc/qemu/qemu.fs
@@ -93,3 +93,31 @@ variable keyboard-phandle 0 keyboard-phandle !
 :noname
   set-defaults
 ; PREPOST-initializer
+
+\ -------------------------------------------------------------------------
+\ Adler-32 wrapper
+\ -------------------------------------------------------------------------
+
+: adler32 ( adler buf len -- checksum )
+  \ Since Mac OS 9 is the only system using this word, we take this
+  \ opportunity to inject a copyright message that is necessary for the
+  \ system to boot.
+  " Copyright 1983-2001 Apple Computer, Inc. THIS MESSAGE FOR COMPATIBILITY ONLY"
+  encode-string " copyright"
+  " /" find-package if
+    " set-property" $find if
+      execute
+    else
+      3drop drop
+    then
+  then
+
+  ( adler buf len )
+
+  " (adler32)" $find if
+    execute
+  else
+    ." Can't find " ( adler32-name ) type cr
+    3drop 0
+  then
+;

[Programmingkid]

On Jun 20, 2015, at 1:48 PM, Cormac O'Brien wrote:
> These patches fix a number of issues that hindered the Mac OS 9 boot process.
>
> Signed-off-by: Cormac O'Brien
>
> Cormac O'Brien (6):
>  bootinfo_load: recognize ASCII EOT as end of bootinfo
>  ppc: add Adler-32 checksum capability
>  ppc: add ROM node to device tree
>  ppc: fix stack usage in mmu_claim, mem_claim
>  ppc: add CPU reservation-granule-size property
>  ppc: add Apple copyright injection to adler32

Have you know about github? You could keep a working repo there. It would allow you
to have all the changes for QEMU and OpenBIOS in one place. Instead of having to download
and apply all the patches you made, someone wanting to help can simply download your
repository. This would be a *lot* easier than having to keep track of multiple versions of 
patches.

[Cormac O'Brien]

On Sat, Jun 20, 2015 at 1:14 PM, Programmingkid
 wrote:
> Have you know about github? You could keep a working repo there. It would allow you
> to have all the changes for QEMU and OpenBIOS in one place. Instead of having to download
> and apply all the patches you made, someone wanting to help can simply download your
> repository. This would be a *lot* easier than having to keep track of multiple versions of
> patches.

https://github.com/cormac-obrien/openbios-devel/

Way ahead of ya :) New progress is on the 'hacking' branch, 'upstream'
is for finalizing changes before submitting to the mailing list. Note
that hacking and upstream are in the same place right now as no
meaningful changes have been made.

~Cormac

[PATCH 1/6] bootinfo_load: recognize ASCII EOT as end of bootinfo

[Mark Cave-Ayland]

On 20/06/15 18:48, Cormac O'Brien wrote:
> Mac OS 9's CHRP boot info consists of the standard SGML document terminated by
> an ASCII EOT, followed by a large binary blob which is relocated by the Forth
> script. This patch allows the bootinfo loader to stop loading the script if it
> detects an EOT, avoiding the out-of-memory error previously encountered when
> booting OS 9.
> Signed-off-by: Cormac O'Brien
>
> libopenbios/bootinfo_load.c | 10 ++++++++++
>  1 file changed, 10 insertions(+)
>
> diff --git a/libopenbios/bootinfo_load.c b/libopenbios/bootinfo_load.c
> index fa9e36b..4b6a52c 100644
> --- a/libopenbios/bootinfo_load.c
> +++ b/libopenbios/bootinfo_load.c
> @@ -161,6 +161,16 @@ bootinfo_init_program(void)
>    feval("load-size");
>    size = POP();
> /* Check for a delimiting ASCII EOT, as in Mac OS 9 */
> current = 0;
> while (current < size) {
> if (base[current] == 0x04) {
> 	size = current;
> 	break;
> }
> current++;
> }
> bootscript = malloc(size);
>  if (bootscript == NULL) {
>   DPRINTF("Can't malloc %d bytes\n", size);

I've checked this patch with a debugger, and I think the use of EOT to
detect the end of the bootscript isn't correct. AFAICT with gdb the
Forth string is always NULL-terminated which is eluded to in the
IEEE1275 specification in an indirect manner in section 6.3.2.6. Note
that the patch above does work due to the position of a suitable EOT
character but in my testing here it is several KB into the binary image
rather than directly at the end of the Forth boundary.

I have an even simpler patch to replace this that I'll post to the list
shortly.

ATB,

Mark.

[Mark Cave-Ayland]

On 20/06/15 18:48, Cormac O'Brien wrote:
> These patches fix a number of issues that hindered the Mac OS 9 boot process.
>
> Signed-off-by: Cormac O'Brien
>
> Cormac O'Brien (6):
>   bootinfo_load: recognize ASCII EOT as end of bootinfo
>   ppc: add Adler-32 checksum capability
>   ppc: add ROM node to device tree
>   ppc: fix stack usage in mmu_claim, mem_claim
>   ppc: add CPU reservation-granule-size property
>   ppc: add Apple copyright injection to adler32
>
> arch/ppc/qemu/init.c        | 62 +++++++++++++++++++++++++++++++++++++++++++++
>  arch/ppc/qemu/methods.c     | 34 ++++++++++++-------------
>  arch/ppc/qemu/qemu.fs       | 28 ++++++++++++++++++++
>  arch/ppc/qemu/tree.fs       |  8 ++++++
>  libopenbios/bootinfo_load.c | 10 ++++++++
>  5 files changed, 125 insertions(+), 17 deletions(-)

While I've dropped patch 1 in favour of the revised strnlen() version
posted over the weekend, the remainder of the patches all apply with no
regressions on any of my PPC images so I've committed them to trunk.

Thanks for your hard work on this!

ATB,

Mark.

[Note for others interested in testing the OS 9 patches: as a temporary
measure you also need a patch such as the one at
http://www.openfirmware.info/pipermail/openbios/2015-June/008722.html in
order to successfully execute the Forth boot script. The proper fix
requires more work.]