linux-kernel - Re: [PATCH 4/8] x86/intel_rct/mba: Add MBA structures and initialize MBA

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <alpine.DEB.2.20.1703011541070.4005@nanos>
Date:   Wed, 1 Mar 2017 16:24:41 +0100 (CET)
From:   Thomas Gleixner <tglx@...utronix.de>
To:     Vikas Shivappa <vikas.shivappa@...ux.intel.com>
cc:     vikas.shivappa@...el.com, linux-kernel@...r.kernel.org,
        x86@...nel.org, hpa@...or.com, mingo@...nel.org,
        peterz@...radead.org, ravi.v.shankar@...el.com,
        tony.luck@...el.com, fenghua.yu@...el.com, andi.kleen@...el.com
Subject: Re: [PATCH 4/8] x86/intel_rct/mba: Add MBA structures and initialize
 MBA

On Fri, 17 Feb 2017, Vikas Shivappa wrote:
> --- a/arch/x86/include/asm/intel_rdt.h
> +++ b/arch/x86/include/asm/intel_rdt.h
> @@ -11,6 +11,9 @@
>  #define IA32_L3_QOS_CFG		0xc81
>  #define IA32_L3_CBM_BASE	0xc90
>  #define IA32_L2_CBM_BASE	0xd10
> +#define IA32_MBA_THRTL_BASE	0xd50
> +#define MAX_MBA_THRTL		100u
> +#define MBA_IS_LINEAR		0x4

I have a hard time to figure out how the latter two constants are related
to this list of registers. MBA_IS_LINEAR is used to check the CPUID bit and
MAX_MBA_THRTL is obviously a pure software constant because with a
non-linear scale the maximum value is not 100.

Just slapping defines to random places is equally bad as using hard coded
constants.

> +/*
> + * rdt_get_mb_table() - get a mapping of b/w percentage values
> + * exposed to user interface and the h/w understandable delay values.
> + *
> + * The non-linear delay values have the granularity of power of two
> + * and also the h/w does not guarantee a curve for configured delay
> + * values vs. actual b/w throttled.
> + * Hence we need a mapping that is pre caliberated for user to express
> + * the b/w in terms of any sensible number.

... calibrated so the user can express the bandwidth as a percentage value.

> +static inline int rdt_get_mb_table(struct rdt_resource *r)
> +{
> +	/*
> +	 * There are no Intel SKUs as of now to support non-linear delay.
> +	 */
> +	r->mb_map = NULL;

What's the point of setting this to NULL?

Also it would be helpful to emit log info here so people don't have to
start digging around.

	pr_info("Bandwidth map not implemented for ....", ... model);

> +
> +	return -ENODEV;

Returning -ENODEV to a function which just returns a boolean value is
pointless.

>  static void rdt_get_cache_config(int idx, struct rdt_resource *r)
>  {
>  	union cpuid_0x10_1_eax eax;
> @@ -184,9 +237,8 @@ static inline bool get_rdt_resources(void)
>  		ret = true;
>  	}
>  
> -	if (boot_cpu_has(X86_FEATURE_MBA)) {
> -		ret = true;
> -	}
> +	if (boot_cpu_has(X86_FEATURE_MBA))
> +		ret = rdt_get_mem_config(&rdt_resources_all[RDT_RESOURCE_MBA]);

Groan. When rdt_get_mem_config() returns false (because the map is not
implemented), then the whole function returns false and CAT is disabled.

> +static int domain_setup_ctrlval(struct rdt_resource *r, struct rdt_domain *d)
> +{
> +	int i;
> +
> +	d->ctrl_val = kmalloc_array(r->num_closid,
> +				     sizeof(*d->ctrl_val), GFP_KERNEL);
> +	if (!d->ctrl_val)
> +		return -ENOMEM;
> +
> +	/*
> +	 * Initialize the Control MSRs to having no control.
> +	 * For Cache Allocation: Set all bits in cbm
> +	 * For Memory Allocation: Set b/w requested to 100
> +	 */
> +	for (i = 0; i < r->num_closid; i++) {
> +		int idx = cbm_idx(r, i);
> +
> +		d->ctrl_val[i] = r->default_ctrl;
> +		wrmsrl(r->msr_base + idx, d->ctrl_val[i]);
> +	}

So if you use a local pointer for that, this whole mess becomes readable.

static int domain_setup_ctrlval(struct rdt_resource *r, struct rdt_domain *d)
{
	u32 *p;
	int i;

	p = kmalloc_array(r->num_closid, sizeof(*d->ctrl_val), GFP_KERNEL);
	if (!p)
		return -ENOMEM;

	d->ctrl_val = p;

	/* Initialize the Control MSRs to the default value */
	for (i = 0; i < r->num_closid; i++, p++) {
		int idx = cbm_idx(r, i);

		*p = r->default_ctrl;
		wrmsrl(r->msr_base + idx, *p);
	}
> +
> +	return 0;
> +}

>  static void domain_add_cpu(int cpu, struct rdt_resource *r)
>  {
> -	int i, id = get_cache_id(cpu, r->cache_level);
> +	int id = get_cache_id(cpu, r->cache_level), ret;

Bah. If you have the same type in one line, then please move the
uninitialized variables to the front.

	int ret, id = get_cache_id(cpu, r->cache_level);

But a s/i/ret/ would have been to simple and kept the code readable.

> @@ -298,19 +374,12 @@ static void domain_add_cpu(int cpu, struct rdt_resource *r)
>  
>  	d->id = id;
>  
> -	d->ctrl_val = kmalloc_array(r->num_closid, sizeof(*d->ctrl_val), GFP_KERNEL);
> -	if (!d->ctrl_val) {
> +	ret = domain_setup_ctrlval(r, d);
> +	if (ret) {
>  		kfree(d);
>  		return;
>  	}

What's the point of this 'ret' variable if the function is void?

	if (domain_setup_ctrlval(r, d)) {
		kfree(d);
		return;
	}

would have been to easy to read, right?

Thanks,

	tglx