linux-kernel - Re: [PATCH v5 5/8] x86/resctrl: Add user interface to enable/disable io

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 for Android: free password hash cracker in your pocket

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

Message-ID: <e768535e-444d-4fa8-bfe2-f8e8c0b1d4bc@amd.com>
Date: Thu, 22 May 2025 18:12:21 -0500
From: "Moger, Babu" <bmoger@....com>
To: "Luck, Tony" <tony.luck@...el.com>, Babu Moger <babu.moger@....com>
Cc: corbet@....net, reinette.chatre@...el.com, Dave.Martin@....com,
 james.morse@....com, tglx@...utronix.de, mingo@...hat.com, bp@...en8.de,
 dave.hansen@...ux.intel.com, x86@...nel.org, hpa@...or.com,
 akpm@...ux-foundation.org, rostedt@...dmis.org, paulmck@...nel.org,
 thuth@...hat.com, ardb@...nel.org, gregkh@...uxfoundation.org,
 thomas.lendacky@....com, seanjc@...gle.com, mario.limonciello@....com,
 perry.yuan@....com, kai.huang@...el.com, xiaoyao.li@...el.com,
 nikunj@....com, kan.liang@...ux.intel.com, xin3.li@...el.com,
 andrew.cooper3@...rix.com, ebiggers@...gle.com, xin@...or.com,
 sohil.mehta@...el.com, Xiaojian.Du@....com, gautham.shenoy@....com,
 linux-doc@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v5 5/8] x86/resctrl: Add user interface to enable/disable
 io_alloc feature

Hi Tony,

On 5/22/2025 3:28 PM, Luck, Tony wrote:
> On Thu, May 22, 2025 at 02:51:36PM -0500, Babu Moger wrote:
>>   
>> +/*
>> + * resctrl_io_alloc_closid_get - io_alloc feature uses max CLOSID to route
>> + * the IO traffic. Get the max CLOSID and verify if the CLOSID is available.
>> + *
>> + * The total number of CLOSIDs is determined in closid_init(),  based on the
>> + * minimum supported across all resources. If CDP (Code Data Prioritization)
>> + * is enabled, the number of CLOSIDs is halved. The final value is returned
>> + * by closids_supported(). Make sure this value aligns with the maximum
>> + * CLOSID supported by the respective resource.
>> + */
>> +static int resctrl_io_alloc_closid_get(struct rdt_resource *r)
>> +{
>> +	int num_closids = closids_supported();
>> +
>> +	if (resctrl_arch_get_cdp_enabled(r->rid))
>> +		num_closids *= 2;
>> +
>> +	if (num_closids != resctrl_arch_get_num_closid(r))
>> +		return -ENOSPC;
>> +
>> +	return closids_supported() - 1;
>> +}
> 
> Is using closids_supported() the right thing here? That's
> the minimum value across all resources. So suppose you had
> 16 CLOS for the L3 resource, but only 8 CLOS in one of L2/MB/SMBA.
> 
> I'd assume the your h/w doesn't care that Linux chose to
> ignore half of the available L3 CLOSIDs, and is still going
> to use CLOSID==15 for SDCIA.
> 
> I think you'll take the -ENOSPC error return. But do you
> really need to do that? Maybe you can still have SDCIA
> enabled and using CLOSID 15?

We cannot support more than closids_supported(), as the bitmaps are not 
initialized beyond that point, and other routines are also not designed 
to handle more than closid_free_map_len.

While it's technically possible to implement a workaround for this 
special case, it would be a hacky solution. It's likely unnecessary to 
go down that path.

Thanks
Babu