Message-ID: <543a1cdb-30a1-4aea-aa8a-e6dd7be0aba6@amd.com>
Date: Wed, 19 Jun 2024 09:43:41 -0500
From: "Moger, Babu" <babu.moger@....com>
To: Reinette Chatre <reinette.chatre@...el.com>,
 Tony Luck <tony.luck@...el.com>, Fenghua Yu <fenghua.yu@...el.com>,
 Maciej Wieczor-Retman <maciej.wieczor-retman@...el.com>,
 Peter Newman <peternewman@...gle.com>, James Morse <james.morse@....com>,
 Drew Fustini <dfustini@...libre.com>, Dave Martin <Dave.Martin@....com>
Cc: x86@...nel.org, linux-kernel@...r.kernel.org, patches@...ts.linux.dev
Subject: Re: [PATCH v20 06/18] x86/resctrl: Introduce snc_nodes_per_l3_cache



On 6/18/24 17:58, Reinette Chatre wrote:
> Hi Babu and Tony,
> 
> On 6/17/24 3:36 PM, Moger, Babu wrote:
>> On 6/10/2024 1:35 PM, Tony Luck wrote:
> 
>>> diff --git a/arch/x86/kernel/cpu/resctrl/monitor.c
>>> b/arch/x86/kernel/cpu/resctrl/monitor.c
>>> index 89d7e6fcbaa1..f2fd35d294f2 100644
>>> --- a/arch/x86/kernel/cpu/resctrl/monitor.c
>>> +++ b/arch/x86/kernel/cpu/resctrl/monitor.c
>>> @@ -97,6 +97,8 @@ unsigned int resctrl_rmid_realloc_limit;
>>>   #define CF(cf)    ((unsigned long)(1048576 * (cf) + 0.5))
>>> +static int snc_nodes_per_l3_cache = 1;
>>> +
>>>   /*
>>>    * The correction factor table is documented in
>>> Documentation/arch/x86/resctrl.rst.
>>>    * If rmid > rmid threshold, MBM total and local values should be
>>> multiplied
>>> @@ -185,7 +187,43 @@ static inline struct rmid_entry *__rmid_entry(u32
>>> idx)
>>>       return entry;
>>>   }
>>> -static int __rmid_read(u32 rmid, enum resctrl_event_id eventid, u64 *val)
>>> +/*
>>> + * When Sub-NUMA Cluster (SNC) mode is not enabled (as indicated by
>>> + * "snc_nodes_per_l3_cache  == 1") no translation of the RMID value is
>>> + * needed. The physical RMID is the same as the logical RMID.
>>> + *
>>> + * On a platform with SNC mode enabled, Linux enables RMID sharing mode
>>> + * via MSR 0xCA0 (see the "RMID Sharing Mode" section in the "Intel
>>> + * Resource Director Technology Architecture Specification" for a full
>>> + * description of RMID sharing mode).
>>> + *
>>> + * In RMID sharing mode there are fewer "logical RMID" values available
>>> + * to accumulate data ("physical RMIDs" are divided evenly between SNC
>>> + * nodes that share an L3 cache). Linux creates an rdt_mon_domain for
>>> + * each SNC node.
>>> + *
>>> + * The value loaded into IA32_PQR_ASSOC is the "logical RMID".
>>> + *
>>> + * Data is collected independently on each SNC node and can be retrieved
>>> + * using the "physical RMID" value computed by this function and loaded
>>> + * into IA32_QM_EVTSEL. @cpu can be any CPU in the SNC node.
>>> + *
>>> + * The scope of the IA32_QM_EVTSEL and IA32_QM_CTR MSRs is at the L3
>>> + * cache.  So a "physical RMID" may be read from any CPU that shares
>>> + * the L3 cache with the desired SNC node, not just from a CPU in
>>> + * the specific SNC node.
>>> + */
>>> +static int logical_rmid_to_physical_rmid(int cpu, int lrmid)
>>
>> How about ? (or something similar)
>>
>> static int get_snc_node_rmid(int cpu, int rmid)
>>
>>> +{
>>> +    struct rdt_resource *r =
>>> &rdt_resources_all[RDT_RESOURCE_L3].r_resctrl;
>>> +
>>> +    if (snc_nodes_per_l3_cache  == 1)
> 
> (nit: unnecessary space)
> 
>>> +        return lrmid;
>>> +
>>> +    return lrmid + (cpu_to_node(cpu) % snc_nodes_per_l3_cache) *
>>> r->num_rmid;
>>> +}
>>> +
>>> +static int __rmid_read_phys(u32 prmid, enum resctrl_event_id eventid,
>>> u64 *val)
>>
>> You don't need to write new function.  Just update the rmid.
>>
>>
>>>   {
>>>       u64 msr_val;
>>> @@ -197,7 +235,7 @@ static int __rmid_read(u32 rmid, enum
>>> resctrl_event_id eventid, u64 *val)
>>>        * IA32_QM_CTR.Error (bit 63) and IA32_QM_CTR.Unavailable (bit 62)
>>>        * are error bits.
>>>        */
>>> -    wrmsr(MSR_IA32_QM_EVTSEL, eventid, rmid);
>>> +    wrmsr(MSR_IA32_QM_EVTSEL, eventid, prmid);
>>>       rdmsrl(MSR_IA32_QM_CTR, msr_val);
>>>       if (msr_val & RMID_VAL_ERROR)
>>> @@ -233,14 +271,17 @@ void resctrl_arch_reset_rmid(struct rdt_resource
>>> *r, struct rdt_mon_domain *d,
>>>                    enum resctrl_event_id eventid)
>>>   {
>>>       struct rdt_hw_mon_domain *hw_dom = resctrl_to_arch_mon_dom(d);
>>> +    int cpu = cpumask_any(&d->hdr.cpu_mask);
>>>       struct arch_mbm_state *am;
>>> +    u32 prmid;
>>
>> snc_rmid?
>>
>>>       am = get_arch_mbm_state(hw_dom, rmid, eventid);
>>>       if (am) {
>>>           memset(am, 0, sizeof(*am));
>>> +        prmid = logical_rmid_to_physical_rmid(cpu, rmid);
>>>           /* Record any initial, non-zero count value. */
>>> -        __rmid_read(rmid, eventid, &am->prev_msr);
>>> +        __rmid_read_phys(prmid, eventid, &am->prev_msr);
>>
>> How about ? Feel free to simplify.
>>
>>            if (snc_nodes_per_l3_cache > 1) {
>>                   snc_rmid = get_snc_node_rmid(cpu, rmid);
>>                  __rmid_read(snc_rmid, eventid, &am->prev_msr);
>>            } else {
>>                __rmid_read(rmid, eventid, &am->prev_msr);
>>            }
>>
> 
> When considering something like this I think it would be better to contain
> the
> SNC checking in a single place so that all places needing to read RMID
> need not
> remember to have the same copied "if (snc_nodes_per_l3_cache > 1)" check.
> This then essentially becomes logical_rmid_to_physical_rmid() in this
> patch so
> now it just becomes a question about what name to pick for variables and
> functions.
> 
> I do prefer a name like __rmid_read_phys()  with a unique "prmid"
> parameter since that
> should prompt developer to give a second thought to what rmid parameter is
> provided
> instead of just blindly calling __rmid_read() that implies that it is
> reading the
> data for the RMID used by resctrl without considering that a conversion
> may be needed.

Ok. That sounds reasonable.

> 
> I do understand and agree that "logical" vs "physical" is not intuitive
> here but
> to that end I find that the comments explain the distinction well. If
> there are
> better suggestions then they are surely welcome.
> 
> In summary, I do think that the "__rmid_read()" function needs a name
> change to make
> clear that it may not be reading the RMID used internally by resctrl and
> this function
> should be accompanied by a function with similar term in its name that
> does the
> conversion and includes the SNC check.
> 
> Reinette
> 

-- 
Thanks
Babu Moger

Powered by blists - more mailing lists