Minor tweak to silo regarding GPT labeled drives

Minor tweak to silo regarding GPT labeled drives

[Bryce]

[-- Attachment #1: Type: text/plain, Size: 426 bytes --]

I've been prompted by a couple of folk to pass along this minor patch for silo
Newer sparc firmware, mostly in the T4 and above range, along with sun(vtoc) now
support gpt labelled drives on the system boot drive (and theoretically some can
also support msdos? not tried it)

The problem with sillo is that it actively looks to see if a sun label magic is
written but GPT has no such 'magic' and silo would moan.

Phil
=--=



[-- Attachment #2: 0015-GPT-tweak.patch --]
[-- Type: text/x-patch, Size: 1295 bytes --]

From 5d5994b88513f2e45701460d275a2d29ab89c40b Mon Sep 17 00:00:00 2001
From: Dazed and confused <clueless@somewhere.com>
Date: Thu, 19 Feb 2015 05:58:30 -0800
Subject: [PATCH] GPT tweak
 quietens silo about lack of a MAGIC for the GPT label

---
 include/silo.h  |    1 +
 second/fs/iom.c |    4 ++--
 2 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/include/silo.h b/include/silo.h
index 94d6e31..135b79e 100644
--- a/include/silo.h
+++ b/include/silo.h
@@ -46,6 +46,7 @@ typedef struct {
 } sun_partition;
 
 #define SUN_LABEL_MAGIC          0xDABE
+#define GPT_LABEL_MAGIC          0x0000
 
 struct silo_inode {
     unsigned int inolen;
diff --git a/second/fs/iom.c b/second/fs/iom.c
index 28dd4b2..8629d36 100644
--- a/second/fs/iom.c
+++ b/second/fs/iom.c
@@ -37,8 +37,8 @@ static int read_sun_partition (int partno)
         silo_fatal("Cannot read partition");
         return 0;
     }
-    if (sdl.magic != SUN_LABEL_MAGIC)
-        silo_fatal("Wrong disklabel magic");
+    if ((sdl.magic != SUN_LABEL_MAGIC) && (sdl.magic != GPT_LABEL_MAGIC))
+        printf("Wrong disklabel magic [0x%X]\n", (unsigned int) sdl.magic);
     for (csum = 0, ush = ((unsigned short *) ((&sdl) + 1)) - 1; ush >= (unsigned short *) &sdl;)
         csum ^= *ush--;
     if (csum)
-- 
1.7.1

Re: Minor tweak to silo regarding GPT labeled drives

[Anatoly Pugachev]

On Mon, Jan 18, 2016 at 9:17 PM, Bryce <philip.copeland@oracle.com> wrote:
> I've been prompted by a couple of folk to pass along this minor patch for silo
> Newer sparc firmware, mostly in the T4 and above range, along with sun(vtoc) now
> support gpt labelled drives on the system boot drive (and theoretically some can
> also support msdos? not tried it)
>
> The problem with sillo is that it actively looks to see if a sun label magic is
> written but GPT has no such 'magic' and silo would moan.

having non-patched silo run on disk with gpt label (hardware is T5-2
server), breaks autoboot ( auto-boot?=true )

Boot device: vdisk0  File and args:
SILO Version 1.4.14

Fatal error: Wrong disklabel magic
boot:

I'm going to test git silo with this patch and report back tomorrow.

Re: Minor tweak to silo regarding GPT labeled drives

[Sam Ravnborg]

On Mon, Jan 18, 2016 at 06:17:34PM +0000, Bryce wrote:
> I've been prompted by a couple of folk to pass along this minor patch for silo
> Newer sparc firmware, mostly in the T4 and above range, along with sun(vtoc) now
> support gpt labelled drives on the system boot drive (and theoretically some can
> also support msdos? not tried it)
> 
> The problem with sillo is that it actively looks to see if a sun label magic is
> written but GPT has no such 'magic' and silo would moan.
> 
> Phil
> =--> 
> 

> >From 5d5994b88513f2e45701460d275a2d29ab89c40b Mon Sep 17 00:00:00 2001
> From: Dazed and confused <clueless@somewhere.com>
Can the original author be found and attributed?

I recall davem would like silo patches to be signed off, like kernel patches.

	Sam

Re: Minor tweak to silo regarding GPT labeled drives

[Bryce]

On 20/01/16 17:29, Sam Ravnborg wrote:
> On Mon, Jan 18, 2016 at 06:17:34PM +0000, Bryce wrote:
>> I've been prompted by a couple of folk to pass along this minor patch for silo
>> Newer sparc firmware, mostly in the T4 and above range, along with sun(vtoc) now
>> support gpt labelled drives on the system boot drive (and theoretically some can
>> also support msdos? not tried it)
>>
>> The problem with sillo is that it actively looks to see if a sun label magic is
>> written but GPT has no such 'magic' and silo would moan.
>>
>> Phil
>> =-->>
>>
>> >From 5d5994b88513f2e45701460d275a2d29ab89c40b Mon Sep 17 00:00:00 2001
>> From: Dazed and confused <clueless@somewhere.com>
> Can the original author be found and attributed?
>
> I recall davem would like silo patches to be signed off, like kernel patches.
>
> 	Sam
The original author is myself, that is the raw patch from the internal svn repo.

Re: Minor tweak to silo regarding GPT labeled drives

[Sam Ravnborg]

Hi Bryce.

On Wed, Jan 20, 2016 at 05:40:14PM +0000, Bryce wrote:
> On 20/01/16 17:29, Sam Ravnborg wrote:
> > On Mon, Jan 18, 2016 at 06:17:34PM +0000, Bryce wrote:
> >> I've been prompted by a couple of folk to pass along this minor patch for silo
> >> Newer sparc firmware, mostly in the T4 and above range, along with sun(vtoc) now
> >> support gpt labelled drives on the system boot drive (and theoretically some can
> >> also support msdos? not tried it)
> >>
> >> The problem with sillo is that it actively looks to see if a sun label magic is
> >> written but GPT has no such 'magic' and silo would moan.
> >>
> >> Phil
> >> =--> >>
> >>
> >> >From 5d5994b88513f2e45701460d275a2d29ab89c40b Mon Sep 17 00:00:00 2001
> >> From: Dazed and confused <clueless@somewhere.com>
> > Can the original author be found and attributed?
> >
> > I recall davem would like silo patches to be signed off, like kernel patches.
> >
> > 	Sam
> The original author is myself, that is the raw patch from the internal svn repo.
Great.

Then I suggest to await feedback from Anatoly and then please re-submit 
like a proper patch with a proper changelog including signed-off-by,
and Tested-by.
And with a real name in the From part- so git picks up the real name.

	Sam

Re: Minor tweak to silo regarding GPT labeled drives

[Anatoly Pugachev]

On Wed, Jan 20, 2016 at 07:29:55PM +0100, Sam Ravnborg wrote:
> On Wed, Jan 20, 2016 at 05:40:14PM +0000, Bryce wrote:
> > On 20/01/16 17:29, Sam Ravnborg wrote:
> > > On Mon, Jan 18, 2016 at 06:17:34PM +0000, Bryce wrote:
> > >> I've been prompted by a couple of folk to pass along this minor patch for silo
> > >> Newer sparc firmware, mostly in the T4 and above range, along with sun(vtoc) now
> > >> support gpt labelled drives on the system boot drive (and theoretically some can
> > >> also support msdos? not tried it)
> > >>
> > >> The problem with sillo is that it actively looks to see if a sun label magic is
> > >> written but GPT has no such 'magic' and silo would moan.
> > >>
> > >> Phil
> > >> =--> > >>
> > >>
> > >> >From 5d5994b88513f2e45701460d275a2d29ab89c40b Mon Sep 17 00:00:00 2001
> > >> From: Dazed and confused <clueless@somewhere.com>
> > > Can the original author be found and attributed?
> > >
> > > I recall davem would like silo patches to be signed off, like kernel patches.
> > >
> > > 	Sam
> > The original author is myself, that is the raw patch from the internal svn repo.
> Great.
> 
> Then I suggest to await feedback from Anatoly and then please re-submit 
> like a proper patch with a proper changelog including signed-off-by,
> and Tested-by.
> And with a real name in the From part- so git picks up the real name.


My quick report.

Used silo git from https://kernel.googlesource.com/pub/scm/linux/kernel/git/davem/silo

I wasn't able to compile git silo on debian sparc64 (full 64-bit,
kernel+userland), and used debian 7.9 (32-bit userland) instead.

I have 2 test machines: bare-metal v215 (no gpt disks, 2 disks with sun
partition table) with debian 7.9 installed and T5-2 with solaris 11 control
domain and few LDOMs (Oracle Linux on Sparc, debian sid 64-bit, solaris)

So, I used v215 to compile silo git, installed and tried to boot - silo
timeout doesn't work, other than that Linux boots - seems like git silo has
broken timeout.

Next, compiled with gpt-patch, and applied another patch from oracle linux on
sparc silo.src.rpm package
(http://yum.oracle.com/repo/linux_sparc64/latest//silo-1.4.14-4.0.18.el6.src.rpm),
called silo-1.4.14-timer-timeout.patch

Compiled, installed - timeout settings in silo.conf works, linux boots.

Copied compiled silo directory to another debian 64-bit LDOM (where I have
debian sid sparc 64-bit installed to gpt disk partition), made "make install"
and run "silo -f", it complained, that there's no such file, cause don't have
32-bit C library, so I've installed 32-bit libc6-sparc, run "silo -f" again
succesfully.

So far works, linux boots, silo timeout works, no fatal error with "Wrong
disklabel magic".

PS: oracle linux on sparc silo.src.rpm has a few more patches, one of them is
to make silo to compile as 64-bit.

Tested-by: Anatoly Pugachev <matorola@gmail.com>
Signed-off-by: Bryce <philip.copeland@oracle.com>

--

 include/silo.h  |    1 +
 second/fs/iom.c |    4 ++--
 second/timer.c  |    4 ++--
 3 files changed, 5 insertions(+), 4 deletions(-)

diff --git a/include/silo.h b/include/silo.h
index 94d6e31..135b79e 100644
--- a/include/silo.h
+++ b/include/silo.h
@@ -46,6 +46,7 @@ typedef struct {
 } sun_partition;
 
 #define SUN_LABEL_MAGIC          0xDABE
+#define GPT_LABEL_MAGIC          0x0000
 
 struct silo_inode {
     unsigned int inolen;
diff --git a/second/fs/iom.c b/second/fs/iom.c
index 28dd4b2..c25e41c 100644
--- a/second/fs/iom.c
+++ b/second/fs/iom.c
@@ -37,8 +37,8 @@ static int read_sun_partition (int partno)
         silo_fatal("Cannot read partition");
         return 0;
     }
-    if (sdl.magic != SUN_LABEL_MAGIC)
-        silo_fatal("Wrong disklabel magic");
+    if ((sdl.magic != SUN_LABEL_MAGIC) && (sdl.magic != GPT_LABEL_MAGIC))
+        printf("Wrong disklabel magic [0x%X]\n", (unsigned int) sdl.magic);
     for (csum = 0, ush = ((unsigned short *) ((&sdl) + 1)) - 1; ush >= (unsigned short *) &sdl;)
         csum ^= *ush--;
     if (csum)
diff --git a/second/timer.c b/second/timer.c
index 7f03996..b559f93 100644
--- a/second/timer.c
+++ b/second/timer.c
@@ -145,7 +145,7 @@ static inline int sun4u_init_timer ()
            prom_getstring(node, "device_type", node_str, sizeof(node_str));
             if (!strcmp(node_str, "cpu")) {
                 foundcpu = 1;
-                clock_frequency = prom_getintdefault(node, "clock-frequency", 0) / 100;
+                clock_frequency = (unsigned int) prom_getintdefault(node, "clock-frequency", 0) / 100;
             }
         }
         if (notimer) {
@@ -155,7 +155,7 @@ static inline int sun4u_init_timer ()
         }
     }
     if (!foundcpu || !clock_frequency)
-        clock_frequency = prom_getint(prom_root_node, "clock-frequency") / 100;
+        clock_frequency = (unsigned int) prom_getint(prom_root_node, "clock-frequency") / 100;
     if (notimer && !sun4v_cpu) {
         sun4u_notimer = 1;
         __asm__ __volatile__ ("\t"

Re: Minor tweak to silo regarding GPT labeled drives

[Anatoly Pugachev]

On Thu, Jan 21, 2016 at 3:11 AM, Anatoly Pugachev <matorola@gmail.com> wrote:
> Next, compiled with gpt-patch, and applied another patch from oracle linux on
> sparc silo.src.rpm package
> (http://yum.oracle.com/repo/linux_sparc64/latest//silo-1.4.14-4.0.18.el6.src.rpm),
> called silo-1.4.14-timer-timeout.patch
>
> diff --git a/second/timer.c b/second/timer.c
> index 7f03996..b559f93 100644
> --- a/second/timer.c
> +++ b/second/timer.c
> @@ -145,7 +145,7 @@ static inline int sun4u_init_timer ()
>             prom_getstring(node, "device_type", node_str, sizeof(node_str));
>              if (!strcmp(node_str, "cpu")) {
>                  foundcpu = 1;
> -                clock_frequency = prom_getintdefault(node, "clock-frequency", 0) / 100;
> +                clock_frequency = (unsigned int) prom_getintdefault(node, "clock-frequency", 0) / 100;
>              }
>          }
>          if (notimer) {
> @@ -155,7 +155,7 @@ static inline int sun4u_init_timer ()
>          }
>      }
>      if (!foundcpu || !clock_frequency)
> -        clock_frequency = prom_getint(prom_root_node, "clock-frequency") / 100;
> +        clock_frequency = (unsigned int) prom_getint(prom_root_node, "clock-frequency") / 100;
>      if (notimer && !sun4v_cpu) {
>          sun4u_notimer = 1;
>          __asm__ __volatile__ ("\t"

found an owner of this patch in silo.spec changelog, quote:

* Fri Aug 30 2013 Nikita Gerasimov <nikita.gerasimov@oracle.com> 1.4.14-4.0.14
- silo.conf timeout property does not work due to overflow. Basically
  the CPU is so fast now that it overslows the signed 32bit value.
  I guess at some point it'll be even faster and we'll have to promte
  prom_getintdefault() to returning an unsigned double int.

So we need to make Nikita Gerasimov <nikita.gerasimov@oracle.com> as
Signed-off-by as well.
Thanks.

Re: Minor tweak to silo regarding GPT labeled drives

[David Miller]

From: Anatoly Pugachev <matorola@gmail.com>
Date: Thu, 21 Jan 2016 15:52:29 +0300

> On Thu, Jan 21, 2016 at 3:11 AM, Anatoly Pugachev <matorola@gmail.com> wrote:
>> Next, compiled with gpt-patch, and applied another patch from oracle linux on
>> sparc silo.src.rpm package
>> (http://yum.oracle.com/repo/linux_sparc64/latest//silo-1.4.14-4.0.18.el6.src.rpm),
>> called silo-1.4.14-timer-timeout.patch
>>
>> diff --git a/second/timer.c b/second/timer.c
>> index 7f03996..b559f93 100644
>> --- a/second/timer.c
>> +++ b/second/timer.c
>> @@ -145,7 +145,7 @@ static inline int sun4u_init_timer ()
>>             prom_getstring(node, "device_type", node_str, sizeof(node_str));
>>              if (!strcmp(node_str, "cpu")) {
>>                  foundcpu = 1;
>> -                clock_frequency = prom_getintdefault(node, "clock-frequency", 0) / 100;
>> +                clock_frequency = (unsigned int) prom_getintdefault(node, "clock-frequency", 0) / 100;
>>              }
>>          }
>>          if (notimer) {
>> @@ -155,7 +155,7 @@ static inline int sun4u_init_timer ()
>>          }
>>      }
>>      if (!foundcpu || !clock_frequency)
>> -        clock_frequency = prom_getint(prom_root_node, "clock-frequency") / 100;
>> +        clock_frequency = (unsigned int) prom_getint(prom_root_node, "clock-frequency") / 100;
>>      if (notimer && !sun4v_cpu) {
>>          sun4u_notimer = 1;
>>          __asm__ __volatile__ ("\t"
> 
> found an owner of this patch in silo.spec changelog, quote:
> 
> * Fri Aug 30 2013 Nikita Gerasimov <nikita.gerasimov@oracle.com> 1.4.14-4.0.14
> - silo.conf timeout property does not work due to overflow. Basically
>   the CPU is so fast now that it overslows the signed 32bit value.
>   I guess at some point it'll be even faster and we'll have to promte
>   prom_getintdefault() to returning an unsigned double int.
> 
> So we need to make Nikita Gerasimov <nikita.gerasimov@oracle.com> as
> Signed-off-by as well.

And maybe we can ask them how to get access to this amazing "unsigned
double int" type.  I'd like to start using it too.

Re: Minor tweak to silo regarding GPT labeled drives

[David Miller]

From: Bryce <philip.copeland@oracle.com>
Date: Mon, 18 Jan 2016 18:17:34 +0000

> diff --git a/include/silo.h b/include/silo.h
> index 94d6e31..135b79e 100644
> --- a/include/silo.h
> +++ b/include/silo.h
> @@ -46,6 +46,7 @@ typedef struct {
>  } sun_partition;
>  
>  #define SUN_LABEL_MAGIC          0xDABE
> +#define GPT_LABEL_MAGIC          0x0000
>  
>  struct silo_inode {
>      unsigned int inolen;
> diff --git a/second/fs/iom.c b/second/fs/iom.c
> index 28dd4b2..8629d36 100644
> --- a/second/fs/iom.c
> +++ b/second/fs/iom.c
> @@ -37,8 +37,8 @@ static int read_sun_partition (int partno)
>          silo_fatal("Cannot read partition");
>          return 0;
>      }
> -    if (sdl.magic != SUN_LABEL_MAGIC)
> -        silo_fatal("Wrong disklabel magic");
> +    if ((sdl.magic != SUN_LABEL_MAGIC) && (sdl.magic != GPT_LABEL_MAGIC))
> +        printf("Wrong disklabel magic [0x%X]\n", (unsigned int) sdl.magic);
>      for (csum = 0, ush = ((unsigned short *) ((&sdl) + 1)) - 1; ush >= (unsigned short *) &sdl;)
>          csum ^= *ush--;
>      if (csum)

This can't be right.

If it's a GPT label then we have to add all of the necessary GPT
datastructures and code necessary to interpret and verify the
partition label properly.

You can't just fudge it by making the label magic check pass, the code
right afterwards verifies the checksum and then we calculate the disk
offset from the partition entry 'partno' to determine where to start
reading things.