2018-02-13 16:45:07

by Gustavo A. R. Silva

[permalink] [raw]
Subject: [PATCH] x86/microcode/intel: Use 64-bit arithmetic instead of 32-bit

Add suffix ULL to constant 1024 in order to give the compiler complete
information about the proper arithmetic to use. Notice that this
constant is used in a context that expects an expression of type
u64 (64 bits, unsigned).

The expression c->x86_cache_size * 1024 is currently being evaluated
using 32-bit arithmetic.

Addresses-Coverity-ID: 1464429
Signed-off-by: Gustavo A. R. Silva <[email protected]>
---
arch/x86/kernel/cpu/microcode/intel.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/x86/kernel/cpu/microcode/intel.c b/arch/x86/kernel/cpu/microcode/intel.c
index f7c55b0..e5edb92 100644
--- a/arch/x86/kernel/cpu/microcode/intel.c
+++ b/arch/x86/kernel/cpu/microcode/intel.c
@@ -982,7 +982,7 @@ static struct microcode_ops microcode_intel_ops = {

static int __init calc_llc_size_per_core(struct cpuinfo_x86 *c)
{
- u64 llc_size = c->x86_cache_size * 1024;
+ u64 llc_size = c->x86_cache_size * 1024ULL;

do_div(llc_size, c->x86_max_cores);

--
2.7.4



2018-02-13 17:04:40

by Thomas Gleixner

[permalink] [raw]
Subject: Re: [PATCH] x86/microcode/intel: Use 64-bit arithmetic instead of 32-bit

On Tue, 13 Feb 2018, Gustavo A. R. Silva wrote:

> Add suffix ULL to constant 1024 in order to give the compiler complete
> information about the proper arithmetic to use. Notice that this
> constant is used in a context that expects an expression of type
> u64 (64 bits, unsigned).
>
> The expression c->x86_cache_size * 1024 is currently being evaluated
> using 32-bit arithmetic.
>
> Addresses-Coverity-ID: 1464429
> Signed-off-by: Gustavo A. R. Silva <[email protected]>
> ---
> arch/x86/kernel/cpu/microcode/intel.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/arch/x86/kernel/cpu/microcode/intel.c b/arch/x86/kernel/cpu/microcode/intel.c
> index f7c55b0..e5edb92 100644
> --- a/arch/x86/kernel/cpu/microcode/intel.c
> +++ b/arch/x86/kernel/cpu/microcode/intel.c
> @@ -982,7 +982,7 @@ static struct microcode_ops microcode_intel_ops = {
>
> static int __init calc_llc_size_per_core(struct cpuinfo_x86 *c)
> {
> - u64 llc_size = c->x86_cache_size * 1024;
> + u64 llc_size = c->x86_cache_size * 1024ULL;

x86_cache_size is 'int', so you really want to cast c->x86_cache_size to
(u64) for correctness sake.

Aside of that the patch is really purely cosmetic at the moment because the
largest LLC sizes are still below the 3 digit MB range which fits into
32bit quite well. You'd need to have a CPU with >= 2G LLC to create a
problem.

But looking at c->x86_cache_size again. It's int because it's set to -1
initially which is then changed if CPUid or general CPU info gives real
information about the cache size. The only place where that matters is the
/proc/cpuinfo output:

if (c->x86_cache_size >= 0)
seq_printf(m, "cache size\t: %d KB\n", c->x86_cache_size);

which is silly, because that really can be done with:

if (c->x86_cache_size)

as there is no point in printing 'cache size 0KB', which means
x86_cache_size can be made unsigned int, which makes sense because cache
size < 0 does not at all.

So instead of doing this purely mechanical cosmetic change to make a static
checker shut up, I'd like to see a proper cleanup of that thing.

Thanks,

tglx

2018-02-13 17:25:45

by Gustavo A. R. Silva

[permalink] [raw]
Subject: Re: [PATCH] x86/microcode/intel: Use 64-bit arithmetic instead of 32-bit

Hi Thomas,

Quoting Thomas Gleixner <[email protected]>:

> On Tue, 13 Feb 2018, Gustavo A. R. Silva wrote:
>
>> Add suffix ULL to constant 1024 in order to give the compiler complete
>> information about the proper arithmetic to use. Notice that this
>> constant is used in a context that expects an expression of type
>> u64 (64 bits, unsigned).
>>
>> The expression c->x86_cache_size * 1024 is currently being evaluated
>> using 32-bit arithmetic.
>>
>> Addresses-Coverity-ID: 1464429
>> Signed-off-by: Gustavo A. R. Silva <[email protected]>
>> ---
>> arch/x86/kernel/cpu/microcode/intel.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/arch/x86/kernel/cpu/microcode/intel.c
>> b/arch/x86/kernel/cpu/microcode/intel.c
>> index f7c55b0..e5edb92 100644
>> --- a/arch/x86/kernel/cpu/microcode/intel.c
>> +++ b/arch/x86/kernel/cpu/microcode/intel.c
>> @@ -982,7 +982,7 @@ static struct microcode_ops microcode_intel_ops = {
>>
>> static int __init calc_llc_size_per_core(struct cpuinfo_x86 *c)
>> {
>> - u64 llc_size = c->x86_cache_size * 1024;
>> + u64 llc_size = c->x86_cache_size * 1024ULL;
>
> x86_cache_size is 'int', so you really want to cast c->x86_cache_size to
> (u64) for correctness sake.
>
> Aside of that the patch is really purely cosmetic at the moment because the
> largest LLC sizes are still below the 3 digit MB range which fits into
> 32bit quite well. You'd need to have a CPU with >= 2G LLC to create a
> problem.
>
> But looking at c->x86_cache_size again. It's int because it's set to -1
> initially which is then changed if CPUid or general CPU info gives real
> information about the cache size. The only place where that matters is the
> /proc/cpuinfo output:
>
> if (c->x86_cache_size >= 0)
> seq_printf(m, "cache size\t: %d KB\n", c->x86_cache_size);
>
> which is silly, because that really can be done with:
>
> if (c->x86_cache_size)
>
> as there is no point in printing 'cache size 0KB', which means
> x86_cache_size can be made unsigned int, which makes sense because cache
> size < 0 does not at all.
>
> So instead of doing this purely mechanical cosmetic change to make a static
> checker shut up, I'd like to see a proper cleanup of that thing.
>

Yeah, actually I was curious about why x86_cache_size is signed
instead of unsigned. You've made it clear now.

I will change it to be of type unsigned int and make the proper
changes to the rest of code in which x86_cache_size is being used.

Also, I'm curious about the types of the rest of the related variables:

/* Cache QoS architectural values: */
int x86_cache_max_rmid; /* max index */
int x86_cache_occ_scale; /* scale to bytes */
int x86_power;


Maybe they need some cleanup too.

Thanks for the feedback.
--
Gustavo





2018-02-13 23:20:40

by Alan Cox

[permalink] [raw]
Subject: Re: [PATCH] x86/microcode/intel: Use 64-bit arithmetic instead of 32-bit

> if (c->x86_cache_size >= 0)
> seq_printf(m, "cache size\t: %d KB\n", c->x86_cache_size);
>
> which is silly, because that really can be done with:
>
> if (c->x86_cache_size)
>
> as there is no point in printing 'cache size 0KB', which means
> x86_cache_size can be made unsigned int, which makes sense because cache
> size < 0 does not at all.

Currently 0MB means "I know you have no cache" (early slot 1 celeron),
while not printing it means 'I have no clue'

Alan

2018-02-14 11:40:09

by Thomas Gleixner

[permalink] [raw]
Subject: Re: [PATCH] x86/microcode/intel: Use 64-bit arithmetic instead of 32-bit

On Tue, 13 Feb 2018, Alan Cox wrote:
> > if (c->x86_cache_size >= 0)
> > seq_printf(m, "cache size\t: %d KB\n", c->x86_cache_size);
> >
> > which is silly, because that really can be done with:
> >
> > if (c->x86_cache_size)
> >
> > as there is no point in printing 'cache size 0KB', which means
> > x86_cache_size can be made unsigned int, which makes sense because cache
> > size < 0 does not at all.
>
> Currently 0MB means "I know you have no cache" (early slot 1 celeron),
> while not printing it means 'I have no clue'

Cute. I hope the 3 slot1 celeron users will not go wild if their
/proc/cpuinfo now claims to have no clue about caches.

Thanks,

tglx