[PATCH] MIPS: vpe-mt: provide a default 'physical_memsize'

[PATCH] MIPS: vpe-mt: provide a default 'physical_memsize'

[Randy Dunlap]

When neither LANTIQ nor MIPS_MALTA is set, 'physical_memsize' is not
declared. This causes the build to fail with:

mips-linux-ld: arch/mips/kernel/vpe-mt.o: in function `vpe_run':
arch/mips/kernel/vpe-mt.c:(.text.vpe_run+0x280): undefined reference to `physical_memsize'

Fix this by declaring a 0-value physical_memsize with neither LANTIQ
nor MIPS_MALTA is set, like LANTIQ does.

Fixes: 1a2a6d7e8816 ("MIPS: APRP: Split VPE loader into separate files.")
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/all/202302030625.2g3E98sY-lkp@intel.com/
Cc: Dengcheng Zhu <dzhu@wavecomp.com>
Cc: John Crispin <john@phrozen.org>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: linux-mips@vger.kernel.org
---
How has this build error not been detected for 10 years?

 arch/mips/kernel/vpe-mt.c |    9 +++++++++
 1 file changed, 9 insertions(+)

diff -- a/arch/mips/kernel/vpe-mt.c b/arch/mips/kernel/vpe-mt.c
--- a/arch/mips/kernel/vpe-mt.c
+++ b/arch/mips/kernel/vpe-mt.c
@@ -22,6 +22,15 @@ static int major;
 /* The number of TCs and VPEs physically available on the core */
 static int hw_tcs, hw_vpes;
 
+#if !defined(CONFIG_MIPS_MALTA) && !defined(CONFIG_LANTIQ)
+/* The 2 above provide their own 'physical_memsize' variable. */
+/*
+ * This is needed by the vpe-mt loader code, just set it to 0 and assume
+ * that the firmware hardcodes this value to something useful.
+ */
+unsigned long physical_memsize = 0L;
+#endif
+
 /* We are prepared so configure and start the VPE... */
 int vpe_run(struct vpe *v)
 {

Re: [PATCH] MIPS: vpe-mt: provide a default 'physical_memsize'

[Philippe Mathieu-Daudé]

Hi Randy,

On 14/2/23 02:09, Randy Dunlap wrote:
> When neither LANTIQ nor MIPS_MALTA is set, 'physical_memsize' is not
> declared. This causes the build to fail with:
> 
> mips-linux-ld: arch/mips/kernel/vpe-mt.o: in function `vpe_run':
> arch/mips/kernel/vpe-mt.c:(.text.vpe_run+0x280): undefined reference to `physical_memsize'
> 
> Fix this by declaring a 0-value physical_memsize with neither LANTIQ
> nor MIPS_MALTA is set, like LANTIQ does.
> 
> Fixes: 1a2a6d7e8816 ("MIPS: APRP: Split VPE loader into separate files.")
> Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
> Reported-by: kernel test robot <lkp@intel.com>
> Link: https://lore.kernel.org/all/202302030625.2g3E98sY-lkp@intel.com/
> Cc: Dengcheng Zhu <dzhu@wavecomp.com>
> Cc: John Crispin <john@phrozen.org>
> Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
> Cc: linux-mips@vger.kernel.org
> ---
> How has this build error not been detected for 10 years?
> 
>   arch/mips/kernel/vpe-mt.c |    9 +++++++++
>   1 file changed, 9 insertions(+)
> 
> diff -- a/arch/mips/kernel/vpe-mt.c b/arch/mips/kernel/vpe-mt.c
> --- a/arch/mips/kernel/vpe-mt.c
> +++ b/arch/mips/kernel/vpe-mt.c
> @@ -22,6 +22,15 @@ static int major;
>   /* The number of TCs and VPEs physically available on the core */
>   static int hw_tcs, hw_vpes;
>   
> +#if !defined(CONFIG_MIPS_MALTA) && !defined(CONFIG_LANTIQ)
> +/* The 2 above provide their own 'physical_memsize' variable. */

Which seems dubious. The variable should be defined once, likely in
arch/mips/kernel/vpe-mt.c, since arch/mips/include/asm/vpe.h declares
it.

I'm surprised CONFIG_MIPS_MALTA always links malta-dtshim.o, but
malta-dtshim.o depends on MIPS_VPE_LOADER_MT, and I can't find a
CONFIG_MIPS_MALTA -> MIPS_VPE_LOADER_MT Kconfig dep.

> +/*
> + * This is needed by the vpe-mt loader code, just set it to 0 and assume
> + * that the firmware hardcodes this value to something useful.
> + */
> +unsigned long physical_memsize = 0L;

I agree this is where this variable has be be declared / initialized,
but having this dependent on CONFIG_MIPS_MALTA/CONFIG_LANTIQ machines
doesn't seem right.

> +#endif
> +
>   /* We are prepared so configure and start the VPE... */
>   int vpe_run(struct vpe *v)
>   {

Regards,

Phil.

Re: [PATCH] MIPS: vpe-mt: provide a default 'physical_memsize'

[Randy Dunlap]

Hi,

On 2/13/23 23:40, Philippe Mathieu-Daudé wrote:
> Hi Randy,
> 
> On 14/2/23 02:09, Randy Dunlap wrote:
>> When neither LANTIQ nor MIPS_MALTA is set, 'physical_memsize' is not
>> declared. This causes the build to fail with:
>>
>> mips-linux-ld: arch/mips/kernel/vpe-mt.o: in function `vpe_run':
>> arch/mips/kernel/vpe-mt.c:(.text.vpe_run+0x280): undefined reference to `physical_memsize'
>>
>> Fix this by declaring a 0-value physical_memsize with neither LANTIQ
>> nor MIPS_MALTA is set, like LANTIQ does.
>>
>> Fixes: 1a2a6d7e8816 ("MIPS: APRP: Split VPE loader into separate files.")
>> Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
>> Reported-by: kernel test robot <lkp@intel.com>
>> Link: https://lore.kernel.org/all/202302030625.2g3E98sY-lkp@intel.com/
>> Cc: Dengcheng Zhu <dzhu@wavecomp.com>
>> Cc: John Crispin <john@phrozen.org>
>> Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
>> Cc: linux-mips@vger.kernel.org
>> ---
>> How has this build error not been detected for 10 years?
>>
>>   arch/mips/kernel/vpe-mt.c |    9 +++++++++
>>   1 file changed, 9 insertions(+)
>>
>> diff -- a/arch/mips/kernel/vpe-mt.c b/arch/mips/kernel/vpe-mt.c
>> --- a/arch/mips/kernel/vpe-mt.c
>> +++ b/arch/mips/kernel/vpe-mt.c
>> @@ -22,6 +22,15 @@ static int major;
>>   /* The number of TCs and VPEs physically available on the core */
>>   static int hw_tcs, hw_vpes;
>>   +#if !defined(CONFIG_MIPS_MALTA) && !defined(CONFIG_LANTIQ)
>> +/* The 2 above provide their own 'physical_memsize' variable. */
> 
> Which seems dubious. The variable should be defined once, likely in
> arch/mips/kernel/vpe-mt.c, since arch/mips/include/asm/vpe.h declares
> it.
> 
> I'm surprised CONFIG_MIPS_MALTA always links malta-dtshim.o, but
> malta-dtshim.o depends on MIPS_VPE_LOADER_MT, and I can't find a
> CONFIG_MIPS_MALTA -> MIPS_VPE_LOADER_MT Kconfig dep.
> 
>> +/*
>> + * This is needed by the vpe-mt loader code, just set it to 0 and assume
>> + * that the firmware hardcodes this value to something useful.
>> + */
>> +unsigned long physical_memsize = 0L;
> 
> I agree this is where this variable has be be declared / initialized,
> but having this dependent on CONFIG_MIPS_MALTA/CONFIG_LANTIQ machines
> doesn't seem right.
> 
>> +#endif
>> +
>>   /* We are prepared so configure and start the VPE... */
>>   int vpe_run(struct vpe *v)
>>   {

OK, I'll try to consolidate all of these into one location.

Thanks.
-- 
~Randy

Re: [PATCH] MIPS: vpe-mt: provide a default 'physical_memsize'

[Randy Dunlap]

Hi,

On 2/13/23 23:40, Philippe Mathieu-Daudé wrote:
> Hi Randy,
> 
> On 14/2/23 02:09, Randy Dunlap wrote:
>> When neither LANTIQ nor MIPS_MALTA is set, 'physical_memsize' is not
>> declared. This causes the build to fail with:
>>
>> mips-linux-ld: arch/mips/kernel/vpe-mt.o: in function `vpe_run':
>> arch/mips/kernel/vpe-mt.c:(.text.vpe_run+0x280): undefined reference to `physical_memsize'
>>
>> Fix this by declaring a 0-value physical_memsize with neither LANTIQ
>> nor MIPS_MALTA is set, like LANTIQ does.
>>
>> Fixes: 1a2a6d7e8816 ("MIPS: APRP: Split VPE loader into separate files.")
>> Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
>> Reported-by: kernel test robot <lkp@intel.com>
>> Link: https://lore.kernel.org/all/202302030625.2g3E98sY-lkp@intel.com/
>> Cc: Dengcheng Zhu <dzhu@wavecomp.com>
>> Cc: John Crispin <john@phrozen.org>
>> Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
>> Cc: linux-mips@vger.kernel.org
>> ---
>> How has this build error not been detected for 10 years?
>>
>>   arch/mips/kernel/vpe-mt.c |    9 +++++++++
>>   1 file changed, 9 insertions(+)
>>
>> diff -- a/arch/mips/kernel/vpe-mt.c b/arch/mips/kernel/vpe-mt.c
>> --- a/arch/mips/kernel/vpe-mt.c
>> +++ b/arch/mips/kernel/vpe-mt.c
>> @@ -22,6 +22,15 @@ static int major;
>>   /* The number of TCs and VPEs physically available on the core */
>>   static int hw_tcs, hw_vpes;
>>   +#if !defined(CONFIG_MIPS_MALTA) && !defined(CONFIG_LANTIQ)
>> +/* The 2 above provide their own 'physical_memsize' variable. */
> 
> Which seems dubious. The variable should be defined once, likely in
> arch/mips/kernel/vpe-mt.c, since arch/mips/include/asm/vpe.h declares
> it.

That doesn't work for CONFIG_MIPS_MALTA=y and MIPS_VPE_LOADER is not set.

In the current (before a consolidation patch) code, mti-malta/malta-memory.c declares
'physical_memsize' and malta-dtshim.c uses it (thru an 'extern'), so
MIPS_VPE_LOADER and MIPS_VPE_LOADER_MT are not required to be enabled.

> I'm surprised CONFIG_MIPS_MALTA always links malta-dtshim.o, but
> malta-dtshim.o depends on MIPS_VPE_LOADER_MT, and I can't find a
> CONFIG_MIPS_MALTA -> MIPS_VPE_LOADER_MT Kconfig dep.

Why does malta-dtshim.o depend on MIPS_VPE_LOADER_MT?

MIPS_MALTA selects SUPPORTS_VPE_LOADER and SYS_SUPPORTS_MULTITHREADING
but does not require MIPS_VPE_LOADER or MIPS_VPE_LOADER_MT.
It builds fine with those symbols being enabled (before any patch).

>> +/*
>> + * This is needed by the vpe-mt loader code, just set it to 0 and assume
>> + * that the firmware hardcodes this value to something useful.
>> + */
>> +unsigned long physical_memsize = 0L;
> 
> I agree this is where this variable has be be declared / initialized,
> but having this dependent on CONFIG_MIPS_MALTA/CONFIG_LANTIQ machines
> doesn't seem right.

So far I have been able to consolidate the LANTIQ code into a general
patch, but not MALTA.

>> +#endif
>> +
>>   /* We are prepared so configure and start the VPE... */
>>   int vpe_run(struct vpe *v)
>>   {
> 
> Regards,
> 
> Phil.

Thanks.
-- 
~Randy

Re: [PATCH] MIPS: vpe-mt: provide a default 'physical_memsize'

[Thomas Bogendoerfer]

On Wed, Feb 15, 2023 at 10:59:35PM -0800, Randy Dunlap wrote:
> > I agree this is where this variable has be be declared / initialized,
> > but having this dependent on CONFIG_MIPS_MALTA/CONFIG_LANTIQ machines
> > doesn't seem right.
> 
> So far I have been able to consolidate the LANTIQ code into a general
> patch, but not MALTA.

if I didn't miss something physical_memory is always 0 for LANTIQ 
and something for MALTA depending on command line/DT. Now

arch/mips/kernel/vpe-mt.c contains

        /*
         * The sde-kit passes 'memsize' to __start in $a3, so set something
         * here...  Or set $a3 to zero and define DFLT_STACK_SIZE and
         * DFLT_HEAP_SIZE when you compile your program
         */
        mttgpr(6, v->ntcs);
        mttgpr(7, physical_memsize);

so the 0 for LANTIQ is fine with the correct VPE payload. But for
MALTA could cause major problems, if the VPE payload uses the top
of memory for it's stack. So I would guess nobody uses this "mode".
Therefore let's get rid of physical_memory in vpe.c completly.

Thomas.

-- 
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea.                                                [ RFC1925, 2.3 ]

Re: [PATCH] MIPS: vpe-mt: provide a default 'physical_memsize'

[Randy Dunlap]

On 2/17/23 03:57, Thomas Bogendoerfer wrote:
> On Wed, Feb 15, 2023 at 10:59:35PM -0800, Randy Dunlap wrote:
>>> I agree this is where this variable has be be declared / initialized,
>>> but having this dependent on CONFIG_MIPS_MALTA/CONFIG_LANTIQ machines
>>> doesn't seem right.
>>
>> So far I have been able to consolidate the LANTIQ code into a general
>> patch, but not MALTA.
> 
> if I didn't miss something physical_memory is always 0 for LANTIQ 
> and something for MALTA depending on command line/DT. Now
> 
> arch/mips/kernel/vpe-mt.c contains
> 
>         /*
>          * The sde-kit passes 'memsize' to __start in $a3, so set something
>          * here...  Or set $a3 to zero and define DFLT_STACK_SIZE and
>          * DFLT_HEAP_SIZE when you compile your program
>          */
>         mttgpr(6, v->ntcs);
>         mttgpr(7, physical_memsize);
> 
> so the 0 for LANTIQ is fine with the correct VPE payload. But for
> MALTA could cause major problems, if the VPE payload uses the top
> of memory for it's stack. So I would guess nobody uses this "mode".
> Therefore let's get rid of physical_memory in vpe.c completly.

Hi Thomas,

I had already concluded that MALTA's dtshim and physical_memsize
are independent so I should ignore those in any changes/patches.

I'll check into your suggestion to see what that looks like.

Thanks for the comments.
-- 
~Randy

Re: [PATCH] MIPS: vpe-mt: provide a default 'physical_memsize'

[Randy Dunlap]

On 2/17/23 03:57, Thomas Bogendoerfer wrote:
> On Wed, Feb 15, 2023 at 10:59:35PM -0800, Randy Dunlap wrote:
>>> I agree this is where this variable has be be declared / initialized,
>>> but having this dependent on CONFIG_MIPS_MALTA/CONFIG_LANTIQ machines
>>> doesn't seem right.
>>
>> So far I have been able to consolidate the LANTIQ code into a general
>> patch, but not MALTA.
> 
> if I didn't miss something physical_memory is always 0 for LANTIQ 
> and something for MALTA depending on command line/DT. Now
> 
> arch/mips/kernel/vpe-mt.c contains
> 
>         /*
>          * The sde-kit passes 'memsize' to __start in $a3, so set something
>          * here...  Or set $a3 to zero and define DFLT_STACK_SIZE and
>          * DFLT_HEAP_SIZE when you compile your program
>          */
>         mttgpr(6, v->ntcs);
>         mttgpr(7, physical_memsize);
> 
> so the 0 for LANTIQ is fine with the correct VPE payload. But for
> MALTA could cause major problems, if the VPE payload uses the top
> of memory for it's stack. So I would guess nobody uses this "mode".
> Therefore let's get rid of physical_memory in vpe.c completly.

Hi Thomas,

What is this line doing?
          mttgpr(6, v->ntcs);

Does it need to stay?
But the comment and mttgpr(7, physical_memsize); can be deleted?

thanks.
-- 
~Randy

Re: [PATCH] MIPS: vpe-mt: provide a default 'physical_memsize'

[Thomas Bogendoerfer]

On Fri, Feb 17, 2023 at 03:24:04PM -0800, Randy Dunlap wrote:
> 
> 
> On 2/17/23 03:57, Thomas Bogendoerfer wrote:
> > On Wed, Feb 15, 2023 at 10:59:35PM -0800, Randy Dunlap wrote:
> >>> I agree this is where this variable has be be declared / initialized,
> >>> but having this dependent on CONFIG_MIPS_MALTA/CONFIG_LANTIQ machines
> >>> doesn't seem right.
> >>
> >> So far I have been able to consolidate the LANTIQ code into a general
> >> patch, but not MALTA.
> > 
> > if I didn't miss something physical_memory is always 0 for LANTIQ 
> > and something for MALTA depending on command line/DT. Now
> > 
> > arch/mips/kernel/vpe-mt.c contains
> > 
> >         /*
> >          * The sde-kit passes 'memsize' to __start in $a3, so set something
> >          * here...  Or set $a3 to zero and define DFLT_STACK_SIZE and
> >          * DFLT_HEAP_SIZE when you compile your program
> >          */
> >         mttgpr(6, v->ntcs);
> >         mttgpr(7, physical_memsize);
> > 
> > so the 0 for LANTIQ is fine with the correct VPE payload. But for
> > MALTA could cause major problems, if the VPE payload uses the top
> > of memory for it's stack. So I would guess nobody uses this "mode".
> > Therefore let's get rid of physical_memory in vpe.c completly.
> 
> Hi Thomas,
> 
> What is this line doing?
>           mttgpr(6, v->ntcs);

from a quick glance over the code, it's probably the number of thread
contexts (TCs) the VPE is allowed to use.

> Does it need to stay?

yes.

> But the comment and mttgpr(7, physical_memsize); can be deleted?

replace it with

	/*
	 * we don't pass the memsize here, so VPE programs need to be
	 * compiled with DFLT_STACK_SIZE and DFLT_HEAP_SIZE defined
	 */
	mttgpr(7, 0);

Thomas.

-- 
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea.                                                [ RFC1925, 2.3 ]