Re: [PATCH v2] of: Check for overlap in reserved memory regions

Re: [PATCH v2] of: Check for overlap in reserved memory regions

[Michael Ellerman]

On Tue, 2015-09-15 at 18:30 -0700, Mitchel Humpherys wrote:

> Any overlap in the reserved memory regions (those specified in the
> reserved-memory DT node) is a bug.  These bugs might go undetected as
> long as the contested region isn't used simultaneously by multiple
> software agents, which makes such bugs hard to debug.  Fix this by
> printing a scary warning during boot if overlap is detected.
> 
> diff --git a/drivers/of/of_reserved_mem.c b/drivers/of/of_reserved_mem.c
> index 726ebe792813..62f467b8ccae 100644
> --- a/drivers/of/of_reserved_mem.c
> +++ b/drivers/of/of_reserved_mem.c
> @@ -197,12 +198,52 @@ static int __init __reserved_mem_init_node(struct reserved_mem *rmem)
>  	return -ENOENT;
>  }
>  
> +static int __init __rmem_cmp(const void *a, const void *b)
> +{
> +	const struct reserved_mem *ra = a, *rb = b;
> +
> +	return ra->base - rb->base;
> +}
> +
> +static void __init __rmem_check_for_overlap(void)
> +{
> +	int i;
> +
> +	if (reserved_mem_count < 2)
> +		return;
> +
> +	sort(reserved_mem, reserved_mem_count, sizeof(reserved_mem[0]),
> +	     __rmem_cmp, NULL);
> +	for (i = 0; i < reserved_mem_count - 1; i++) {
> +		struct reserved_mem *this, *next;
> +
> +		this = &reserved_mem[i];
> +		next = &reserved_mem[i + 1];
> +		if (!(this->base && next->base))
> +			continue;
> +		if (this->base + this->size > next->base) {
> +			phys_addr_t this_end, next_end;
> +
> +			this_end = this->base + this->size;
> +			next_end = next->base + next->size;
> +			WARN(1,
> +			     "Reserved memory: OVERLAP DETECTED!\n%s (%pa--%pa) overlaps with %s (%pa--%pa)\n",
> +			     this->name, &this->base, &this_end,
> +			     next->name, &next->base, &next_end);

This is blowing up on some powerpc machines.

It's too early in boot to call WARN() on these systems.

Can we turn it into a pr_err() for now?

I'll send a patch?

cheers

Re: [PATCH v2] of: Check for overlap in reserved memory regions

[Rob Herring]

On Mon, Nov 9, 2015 at 10:29 PM, Michael Ellerman <mpe@ellerman.id.au> wrote:
> On Tue, 2015-09-15 at 18:30 -0700, Mitchel Humpherys wrote:
>
>> Any overlap in the reserved memory regions (those specified in the
>> reserved-memory DT node) is a bug.  These bugs might go undetected as
>> long as the contested region isn't used simultaneously by multiple
>> software agents, which makes such bugs hard to debug.  Fix this by
>> printing a scary warning during boot if overlap is detected.
>>
>> diff --git a/drivers/of/of_reserved_mem.c b/drivers/of/of_reserved_mem.c
>> index 726ebe792813..62f467b8ccae 100644
>> --- a/drivers/of/of_reserved_mem.c
>> +++ b/drivers/of/of_reserved_mem.c
>> @@ -197,12 +198,52 @@ static int __init __reserved_mem_init_node(struct reserved_mem *rmem)
>>       return -ENOENT;
>>  }
>>
>> +static int __init __rmem_cmp(const void *a, const void *b)
>> +{
>> +     const struct reserved_mem *ra = a, *rb = b;
>> +
>> +     return ra->base - rb->base;
>> +}
>> +
>> +static void __init __rmem_check_for_overlap(void)
>> +{
>> +     int i;
>> +
>> +     if (reserved_mem_count < 2)
>> +             return;
>> +
>> +     sort(reserved_mem, reserved_mem_count, sizeof(reserved_mem[0]),
>> +          __rmem_cmp, NULL);
>> +     for (i = 0; i < reserved_mem_count - 1; i++) {
>> +             struct reserved_mem *this, *next;
>> +
>> +             this = &reserved_mem[i];
>> +             next = &reserved_mem[i + 1];
>> +             if (!(this->base && next->base))
>> +                     continue;
>> +             if (this->base + this->size > next->base) {
>> +                     phys_addr_t this_end, next_end;
>> +
>> +                     this_end = this->base + this->size;
>> +                     next_end = next->base + next->size;
>> +                     WARN(1,
>> +                          "Reserved memory: OVERLAP DETECTED!\n%s (%pa--%pa) overlaps with %s (%pa--%pa)\n",
>> +                          this->name, &this->base, &this_end,
>> +                          next->name, &next->base, &next_end);
>
> This is blowing up on some powerpc machines.
>
> It's too early in boot to call WARN() on these systems.

I didn't realize WARN could not be used early. Good to know.

> Can we turn it into a pr_err() for now?

Sounds fine.

> I'll send a patch?

Great.

Rob

Re: [PATCH v2] of: Check for overlap in reserved memory regions

[Michael Ellerman]

On Mon, 2015-11-09 at 22:41 -0600, Rob Herring wrote:
> On Mon, Nov 9, 2015 at 10:29 PM, Michael Ellerman <mpe@ellerman.id.au> wrote:
> > On Tue, 2015-09-15 at 18:30 -0700, Mitchel Humpherys wrote:
> > > Any overlap in the reserved memory regions (those specified in the
> > > reserved-memory DT node) is a bug.  These bugs might go undetected as
> > > long as the contested region isn't used simultaneously by multiple
> > > software agents, which makes such bugs hard to debug.  Fix this by
> > > printing a scary warning during boot if overlap is detected.
> > > 
> > > diff --git a/drivers/of/of_reserved_mem.c b/drivers/of/of_reserved_mem.c
> > > index 726ebe792813..62f467b8ccae 100644
> > > --- a/drivers/of/of_reserved_mem.c
> > > +++ b/drivers/of/of_reserved_mem.c
> > > @@ -197,12 +198,52 @@ static int __init __reserved_mem_init_node(struct reserved_mem *rmem)
> > >       return -ENOENT;
> > >  }
> > > 
> > > +static int __init __rmem_cmp(const void *a, const void *b)
> > > +{
> > > +     const struct reserved_mem *ra = a, *rb = b;
> > > +
> > > +     return ra->base - rb->base;
> > > +}
> > > +
> > > +static void __init __rmem_check_for_overlap(void)
> > > +{
> > > +     int i;
> > > +
> > > +     if (reserved_mem_count < 2)
> > > +             return;
> > > +
> > > +     sort(reserved_mem, reserved_mem_count, sizeof(reserved_mem[0]),
> > > +          __rmem_cmp, NULL);
> > > +     for (i = 0; i < reserved_mem_count - 1; i++) {
> > > +             struct reserved_mem *this, *next;
> > > +
> > > +             this = &reserved_mem[i];
> > > +             next = &reserved_mem[i + 1];
> > > +             if (!(this->base && next->base))
> > > +                     continue;
> > > +             if (this->base + this->size > next->base) {
> > > +                     phys_addr_t this_end, next_end;
> > > +
> > > +                     this_end = this->base + this->size;
> > > +                     next_end = next->base + next->size;
> > > +                     WARN(1,
> > > +                          "Reserved memory: OVERLAP DETECTED!\n%s (%pa--%pa) overlaps with %s (%pa--%pa)\n",
> > > +                          this->name, &this->base, &this_end,
> > > +                          next->name, &next->base, &next_end);
> > 
> > This is blowing up on some powerpc machines.
> > 
> > It's too early in boot to call WARN() on these systems.
> 
> I didn't realize WARN could not be used early. Good to know.

Yeah, it's a bit horrible.

It used to be even worse, we'd take a recursive trap and you'd get nothing
useful at all. Ben fixed that, which makes BUG work but not WARN, because WARN
requires you to take the trap _and return_. We should be able to fix it in the
medium term, but not immediately.

> > Can we turn it into a pr_err() for now?
> 
> Sounds fine.

> > I'll send a patch?
> 
> Great.

Thanks. I'll just test the final version and then send.

cheers