bug in asm-ppc/div64.h

bug in asm-ppc/div64.h

[Bastien Nocera]

[-- Attachment #1: Type: text/plain, Size: 885 bytes --]

Hello,

While chasing a date bug in smbfs with Urban Widmark, it so happened
that we stumbled across a bug in do_div on PPC 32bit, ie. it doesn't
work.

I attached a test case, provided by Urban. do_div is used in quite a few
places like vsprintf, the matrox fb code, etc.

When running on x86:
now: 1047238073
adjusted: 24edd64059e100
/ 10000000: 3df4e80a
1039460362

Check the run on PPC. It's wrong by quite a scale.

Any ideas on how to solve this properly ?

Cheers

--
/Bastien Nocera
http://www.redhat.com

Perfection is reached, not when there is no longer anything to add, but
when there is no longer anything to take away.
                                                Antoine de Saint-Exupery
--
/Bastien Nocera
http://hadess.net

#2  0x4205a2cc in printf ("Oh my %s\n", preferred_deity) from
/lib/i686/libc.so.6 printf ("Oh my %s\n", preferred_deity);
Segmentation fault

[-- Attachment #2: smbfs-time.c --]
[-- Type: text/x-c, Size: 550 bytes --]

#include <stdio.h>
#include <time.h>
#include <asm/types.h>
#include <asm/div64.h>

typedef  __u64 u64;

#define NTFS_TIME_OFFSET ((u64)(369*365 + 89) * 24 * 3600 * 10000000)

static time_t
smb_ntutc2unixutc(u64 ntutc)
{
	u64 t = ntutc - NTFS_TIME_OFFSET;
	printf("adjusted: %Lx\n", t);

	do_div(t, 10000000);
	printf("/ 10000000: %Lx\n", t);

	return (time_t)t;
}

int main()
{
    time_t t;

    t = time(NULL);
    printf("now: %ld\n", t);

    t = smb_ntutc2unixutc(0x01c29fb515986100LL);
    printf("%ld\n", t);
}

Debugging

[Giuliano Pochini]

Does xmon work ?  When something oopses I see the xmon prompt, but
I can't type anything - machine completely freezed. SysReq commands
also are disabled and I must hit the hard reset button.
I disabled xmon so oops messages show up on the console and in the
logs, but I have problems with them too. ksymoops spits a lot of
warnings about non-matching addresses, etc. and the function where
the oops occurred is always "may not be reliable". To locate the
bug I put a lot of printk in my driver and - murphy's surprise -
oopses do not have a backtrace anymore, just one unresolvable
address. #:-6
I'm using 2.4.21p4 on a blue G3. I recompiled Kernel, modules, ALSA
to be sure System.map is the right one.

Bye.


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: Debugging

[Olaf Hering]

 On Tue, Mar 11, Giuliano Pochini wrote:

> I'm using 2.4.21p4 on a blue G3. I recompiled Kernel, modules, ALSA
> to be sure System.map is the right one.

Get an ADB keyboard and use that for xmon.


--
A: No.
Q: Should I include quotations after my reply?

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: Debugging

[Giuliano Pochini]

On 11-Mar-2003 Olaf Hering wrote:
>
>  On Tue, Mar 11, Giuliano Pochini wrote:
>
>> I'm using 2.4.21p4 on a blue G3. I recompiled Kernel, modules, ALSA
>> to be sure System.map is the right one.
>
> Get an ADB keyboard and use that for xmon.

:( I sold it a few months ago. Can macs without ADB run xmon ?


Bye.


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: Debugging

[Olaf Hering]

 On Thu, Mar 13, Giuliano Pochini wrote:

>
> On 11-Mar-2003 Olaf Hering wrote:
> >
> >  On Tue, Mar 11, Giuliano Pochini wrote:
> >
> >> I'm using 2.4.21p4 on a blue G3. I recompiled Kernel, modules, ALSA
> >> to be sure System.map is the right one.
> >
> > Get an ADB keyboard and use that for xmon.
>
> :( I sold it a few months ago. Can macs without ADB run xmon ?

Via Firewire maybe, but your onboard pcilynx controller ist not supported.

--
A: No.
Q: Should I include quotations after my reply?

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

bug in asm-ppc/div64.h

[Bastien Nocera]

[-- Attachment #1: Type: text/plain, Size: 706 bytes --]

Hello,

While chasing a date bug in smbfs with Urban Widmark, it so happened
that we stumbled across a bug in do_div on PPC 32bit, ie. it doesn't
work.

I attached a test case, provided by Urban. do_div is used in quite a few
places like vsprintf, the matrox fb code, etc.

When running on x86:
now: 1047238073
adjusted: 24edd64059e100
/ 10000000: 3df4e80a
1039460362

Check the run on PPC. It's wrong by quite a scale.

Any ideas on how to solve this properly ?

Cheers

--
/Bastien Nocera
http://www.redhat.com

Perfection is reached, not when there is no longer anything to add, but
when there is no longer anything to take away.
                                                Antoine de Saint-Exupery

[-- Attachment #2: smbfs-time.c --]
[-- Type: text/x-c, Size: 550 bytes --]

#include <stdio.h>
#include <time.h>
#include <asm/types.h>
#include <asm/div64.h>

typedef  __u64 u64;

#define NTFS_TIME_OFFSET ((u64)(369*365 + 89) * 24 * 3600 * 10000000)

static time_t
smb_ntutc2unixutc(u64 ntutc)
{
	u64 t = ntutc - NTFS_TIME_OFFSET;
	printf("adjusted: %Lx\n", t);

	do_div(t, 10000000);
	printf("/ 10000000: %Lx\n", t);

	return (time_t)t;
}

int main()
{
    time_t t;

    t = time(NULL);
    printf("now: %ld\n", t);

    t = smb_ntutc2unixutc(0x01c29fb515986100LL);
    printf("%ld\n", t);
}

Re: bug in asm-ppc/div64.h

[Paul Mackerras]

Bastien Nocera writes:

> While chasing a date bug in smbfs with Urban Widmark, it so happened
> that we stumbled across a bug in do_div on PPC 32bit, ie. it doesn't
> work.

Grab a copy of include/asm-ppc/div64.h and arch/ppc/lib/div64.S from
a 2.5 tree.

Paul.

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: bug in asm-ppc/div64.h

[Bastien Nocera]

Hi Paul,

On Tue, 2003-03-11 at 11:04, Paul Mackerras wrote:
>
> Bastien Nocera writes:
>
> > While chasing a date bug in smbfs with Urban Widmark, it so happened
> > that we stumbled across a bug in do_div on PPC 32bit, ie. it doesn't
> > work.
>
> Grab a copy of include/asm-ppc/div64.h and arch/ppc/lib/div64.S from
> a 2.5 tree.

Ben just pointed me to these files. I'll give it a try. Are there any
plans to backport these ? All my files on an SMB share appear with the
wrong dates because of that bug.

Cheers

PS: And sorry about the spam, and badly signed mails...
--
/Bastien Nocera
http://hadess.net

#2  0x4205a2cc in printf ("Oh my %s\n", preferred_deity) from
/lib/i686/libc.so.6 printf ("Oh my %s\n", preferred_deity);
Segmentation fault

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/