Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: Re: [RESEND PATCH v2] mm/sparse: clarify pgdat_to_phys
To:     Miles Chen <miles.chen@mediatek.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Mike Rapoport <rppt@kernel.org>,
        Mark Rutland <mark.rutland@arm.com>
Cc:     linux-mm@kvack.org, linux-kernel@vger.kernel.org,
        linux-mediatek@lists.infradead.org, wsd_upstream@mediatek.com
References: <20210723070137.23321-1-miles.chen@mediatek.com>
From:   David Hildenbrand <david@redhat.com>
Organization: Red Hat
Message-ID: <a042906d-7d07-d1af-57d2-ab4cbdc53f36@redhat.com>
Date:   Fri, 23 Jul 2021 09:10:53 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101
 Thunderbird/78.11.0
MIME-Version: 1.0
In-Reply-To: <20210723070137.23321-1-miles.chen@mediatek.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 23.07.21 09:01, Miles Chen wrote:
> Clarify pgdat_to_phys() by testing if
> pgdat == &contig_page_data when CONFIG_NUMA=n.
> 
> contig_page_data is only available when CONFIG_NUMA=n
> so we have to use #ifndef here.
> 
> No functional change intended.
> 
> Comment from Mark [1]:
> "
> ... and I reckon it'd be clearer and more robust to define
> pgdat_to_phys() in the same ifdefs as contig_page_data so
> that these, stay in-sync. e.g. have:
> 
> | #ifdef CONFIG_NUMA
> | #define pgdat_to_phys(x)	virt_to_phys(x)
> | #else /* CONFIG_NUMA */
> |
> | extern struct pglist_data contig_page_data;
> | ...
> | #define pgdat_to_phys(x)	__pa_symbol(&contig_page_data)
> |
> | #endif /* CONIFIG_NUMA */
> "
> 
> Comment from Mike [2]:
> "
> I'm not sure a macro is better than a static inline.
> 
> Maybe we'd want to warn if pgdat passed to pgtat_to_phys() is not
> &contig_page_data, e.g something like
> 
> static inline phys_addr_t pgdat_to_phys(struct pglist_data *pgdat)
> {
> 	if (!IS_ENABLED(CONFIG_NUMA)) {
> 		if (pgdat == &contig_page_data)
> 			return __pa_symbol(&contig_page_data);
> 		else
> 			pr_warn("Unexpected pglist_data pointer!\n");
> 	}
> 
> 	return __pa(pgdat);
> }
> "
> 
> [1] https://lore.kernel.org/linux-arm-kernel/20210615131902.GB47121@C02TD0UTHF1T.local/
> [2] https://lore.kernel.org/patchwork/patch/1452903/#1650759
> 
> Cc: Mike Rapoport <rppt@kernel.org>
> Cc: Mark Rutland <mark.rutland@arm.com>
> Signed-off-by: Miles Chen <miles.chen@mediatek.com>
> 
> ---
> 
> Change since v1:
> Thanks for Mike's comment, check if pgdat == &contig_page_data,
> so it is clearer that we only expect contig_page_data when
> CONFIG_NUMA=n.
> ---
>   mm/sparse.c | 7 ++++---
>   1 file changed, 4 insertions(+), 3 deletions(-)
> 
> diff --git a/mm/sparse.c b/mm/sparse.c
> index 6326cdf36c4f..f73ff3c124c5 100644
> --- a/mm/sparse.c
> +++ b/mm/sparse.c
> @@ -348,10 +348,11 @@ size_t mem_section_usage_size(void)
>   static inline phys_addr_t pgdat_to_phys(struct pglist_data *pgdat)
>   {
>   #ifndef CONFIG_NUMA
> -	return __pa_symbol(pgdat);
> -#else
> +	if (pgdat == &contig_page_data)
> +		return __pa_symbol(&contig_page_data);
> +	pr_warn("Unexpected pglist_data pointer!\n");

Shouldn't this rather be a VM_BUG_ON()?

Because it looks like something that should barely happen and we might 
not want to perform runtime checks on each and every system?


-- 
Thanks,

David / dhildenb