Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
Message-ID: <02d1e060-e0e4-92e0-ef88-7f237e36fa5c@linux.ibm.com>
Date:   Mon, 4 Apr 2022 10:00:53 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.15; rv:91.0)
 Gecko/20100101 Thunderbird/91.7.0
Subject: Re: [PATCH v2] powerpc/drmem: Don't compute the NUMA node for each
 LMB
Content-Language: en-US
To:     Christophe Leroy <christophe.leroy@csgroup.eu>,
        linuxppc-dev@lists.ozlabs.org, linux-kernel@vger.kernel.org
Cc:     nathanl@linux.ibm.com, cheloha@linux.ibm.com
References: <202008051807.Vi8NDJtX%lkp@intel.com>
 <20200805133502.33723-1-ldufour@linux.ibm.com>
 <9d02405a-793e-bcf5-a424-470d9c82ec7d@csgroup.eu>
From:   Laurent Dufour <ldufour@linux.ibm.com>
In-Reply-To: <9d02405a-793e-bcf5-a424-470d9c82ec7d@csgroup.eu>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 02/04/2022, 18:35:15, Christophe Leroy wrote:
> 
> 
> Le 05/08/2020 à 15:35, Laurent Dufour a écrit :
>> All the LMB from the same set of ibm,dynamic-memory-v2 property are
>> sharing the same NUMA node. Don't compute that node for each one.
>>
>> Tested on a system with 1022 LMBs spread on 4 NUMA nodes, only 4 calls to
>> lmb_set_nid() have been made instead of 1022.
>>
>> This should prevent some soft lockups when starting large guests
>>
>> Code has meaning only if CONFIG_MEMORY_HOTPLUG is set, otherwise the nid
>> field is not present in the drmem_lmb structure.
>>
>> Signed-off-by: Laurent Dufour <ldufour@linux.ibm.com>
> 
> It looks like this patch was superseded by e5e179aa3a39 ("pseries/drmem:
> don't cache node id in drmem_lmb struct").

That patch has been superseded and can be dropped.

Thanks,
Laurent.


> If not, anyway it conflicts with that patch so it has to be rebased.
> 
> Thanks
> Christophe
> 
> 
>> ---
>>   arch/powerpc/mm/drmem.c | 25 ++++++++++++++++++++++++-
>>   1 file changed, 24 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/powerpc/mm/drmem.c b/arch/powerpc/mm/drmem.c
>> index b2eeea39684c..c11b6ec99ea3 100644
>> --- a/arch/powerpc/mm/drmem.c
>> +++ b/arch/powerpc/mm/drmem.c
>> @@ -402,6 +402,9 @@ static void __init init_drmem_v2_lmbs(const __be32
>> *prop)
>>       const __be32 *p;
>>       u32 i, j, lmb_sets;
>>       int lmb_index;
>> +#ifdef CONFIG_MEMORY_HOTPLUG
>> +    struct drmem_lmb *first = NULL;
>> +#endif
>>         lmb_sets = of_read_number(prop++, 1);
>>       if (lmb_sets == 0)
>> @@ -426,6 +429,15 @@ static void __init init_drmem_v2_lmbs(const __be32
>> *prop)
>>       for (i = 0; i < lmb_sets; i++) {
>>           read_drconf_v2_cell(&dr_cell, &p);
>>   +#ifdef CONFIG_MEMORY_HOTPLUG
>> +        /*
>> +         * Fetch the NUMA node id for the fist set or if the
>> +         * associativity index is different from the previous set.
>> +         */
>> +        if (first && dr_cell.aa_index != first->aa_index)
>> +            first = NULL;
>> +#endif
>> +
>>           for (j = 0; j < dr_cell.seq_lmbs; j++) {
>>               lmb = &drmem_info->lmbs[lmb_index++];
>>   @@ -438,7 +450,18 @@ static void __init init_drmem_v2_lmbs(const __be32
>> *prop)
>>               lmb->aa_index = dr_cell.aa_index;
>>               lmb->flags = dr_cell.flags;
>>   -            lmb_set_nid(lmb);
>> +#ifdef CONFIG_MEMORY_HOTPLUG
>> +            /*
>> +             * All the LMB in the set share the same NUMA
>> +             * associativity property. So read that node only once.
>> +             */
>> +            if (!first) {
>> +                lmb_set_nid(lmb);
>> +                first = lmb;
>> +            } else {
>> +                lmb->nid = first->nid;
>> +            }
>> +#endif
>>           }
>>       }
>>   }