Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 14 Sep 2022 16:55:04 +0100
From:   Mike Rapoport <rppt@kernel.org>
To:     Christophe Leroy <christophe.leroy@csgroup.eu>
CC:     =?ISO-8859-1?Q?Pali_Roh=E1r?= <pali@kernel.org>,
        Ash Logan <ash@heyquark.com>,
        "paulus@samba.org" <paulus@samba.org>,
        "mpe@ellerman.id.au" <mpe@ellerman.id.au>,
        "robh+dt@kernel.org" <robh+dt@kernel.org>,
        "benh@kernel.crashing.org" <benh@kernel.crashing.org>,
        "linuxppc-dev@lists.ozlabs.org" <linuxppc-dev@lists.ozlabs.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        "j.ne@posteo.net" <j.ne@posteo.net>,
        "linux-mm@kvack.org" <linux-mm@kvack.org>
Subject: Re: Fragmented physical memory on powerpc/32
User-Agent: K-9 Mail for Android
In-Reply-To: <ed8ff681-4182-3f9f-a65f-21cf5012fff0@csgroup.eu>
References: <20220609222420.ponpoodiqmaqtwht@pali> <20220808184034.lskqrk6z3gb5q76r@pali> <219cda7b-da4b-7a5a-9809-0878e0fc02ba@csgroup.eu> <20220908153511.57ceunyusziqfcav@pali> <20220908201701.sd3zqn5hfixmjvhh@pali> <9fbc5338-5e10-032a-8f55-e080bd93f74b@csgroup.eu> <Yx9GpV1XT8r2a++R@kernel.org> <20220912211623.djb7fckgknyfmof7@pali> <1c95875c-29f8-68b7-e480-fed8614f3037@csgroup.eu> <4f540391-37dc-8e22-be0a-74543082504d@csgroup.eu> <YyGfkDKgeW7/nNlr@kernel.org> <ed8ff681-4182-3f9f-a65f-21cf5012fff0@csgroup.eu>
Message-ID: <9779CA34-E40D-4035-A319-A92D2F6E4DDF@kernel.org>
MIME-Version: 1.0
Content-Type: text/plain;
 charset=utf-8
Content-Transfer-Encoding: quoted-printable
Precedence: bulk


On September 14, 2022 10:43:52 AM GMT+01:00, Christophe Leroy <christophe=
=2Eleroy@csgroup=2Eeu> wrote:
>
>
>Le 14/09/2022 =C3=A0 11:32, Mike Rapoport a =C3=A9crit=C2=A0:
>> On Tue, Sep 13, 2022 at 02:36:13PM +0200, Christophe Leroy wrote:
>>>
>>>
>>> Le 13/09/2022 =C3=A0 08:11, Christophe Leroy a =C3=A9crit=C2=A0:
>>>>
>>>>
>>>> Le 12/09/2022 =C3=A0 23:16, Pali Roh=C3=A1r a =C3=A9crit=C2=A0:
>>>>>>
>>>>>> My guess would be that something went wrong in the linear map
>>>>>> setup, but it
>>>>>> won't hurt running with "memblock=3Ddebug" added to the kernel
>>>>>> command line
>>>>>> to see if there is anything suspicious there=2E
>>>>>
>>>>> Here is boot log on serial console with memblock=3Ddebug command lin=
e:
>>>>>
>>>> =2E=2E=2E
>>>>>
>>>>> Do you need something more for debug?
>>>>
>>>> Can you send me the 'vmlinux' used to generate the above Oops so that=
 I
>>>> can see exactly where we are in function mem_init()=2E
>>>>
>>>> And could you also try without CONFIG_HIGHMEM just in case=2E
>>>>
>>>
>>> I looked at the vmlinux you sent me, the problem is in the loop for hi=
ghmem
>>> in mem_init()=2E It crashes in the call to free_highmem_page()
>>>
>>> #ifdef CONFIG_HIGHMEM
>>> 	{
>>> 		unsigned long pfn, highmem_mapnr;
>>>
>>> 		highmem_mapnr =3D lowmem_end_addr >> PAGE_SHIFT;
>>> 		for (pfn =3D highmem_mapnr; pfn < max_mapnr; ++pfn) {
>>> 			phys_addr_t paddr =3D (phys_addr_t)pfn << PAGE_SHIFT;
>>> 			struct page *page =3D pfn_to_page(pfn);
>>> 			if (!memblock_is_reserved(paddr))
>>> 				free_highmem_page(page);
>>> 		}
>>> 	}
>>> #endif /* CONFIG_HIGHMEM */
>>>
>>>
>>> As far as I can see in the memblock debug lines, the holes don't seem =
to be
>>> marked as reserved by memblock=2E So it is above valid ? Other archite=
ctures
>>> seem to do differently=2E
>>>
>>> Can you try by replacing !memblock_is_reserved(paddr) by
>>> memblock_is_memory(paddr) ?
>>=20
>> The holes should not be marked as reserved, we just need to loop over t=
he
>> memory ranges rather than over pfns=2E Then the holes will be taken int=
o
>> account=2E
>>=20
>> I believe arm and xtensa got this right:
>>=20
>> (from arch/arm/mm/init=2Ec)
>>=20
>> static void __init free_highpages(void)
>> {
>> #ifdef CONFIG_HIGHMEM
>> 	unsigned long max_low =3D max_low_pfn;
>> 	phys_addr_t range_start, range_end;
>> 	u64 i;
>>=20
>> 	/* set highmem page free */
>> 	for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
>> 				&range_start, &range_end, NULL) {
>> 		unsigned long start =3D PFN_UP(range_start);
>> 		unsigned long end =3D PFN_DOWN(range_end);
>>=20
>> 		/* Ignore complete lowmem entries */
>> 		if (end <=3D max_low)
>> 			continue;
>>=20
>> 		/* Truncate partial highmem entries */
>> 		if (start < max_low)
>> 			start =3D max_low;
>>=20
>> 		for (; start < end; start++)
>> 			free_highmem_page(pfn_to_page(start));
>> 	}
>> #endif
>> }
>>=20
>
>
>And what about the way MIPS does it ?
>
>static inline void __init mem_init_free_highmem(void)
>{
>#ifdef CONFIG_HIGHMEM
>	unsigned long tmp;
>
>	if (cpu_has_dc_aliases)
>		return;
>
>	for (tmp =3D highstart_pfn; tmp < highend_pfn; tmp++) {
>		struct page *page =3D pfn_to_page(tmp);
>
>		if (!memblock_is_memory(PFN_PHYS(tmp)))
>			SetPageReserved(page);
>		else
>			free_highmem_page(page);
>	}
>#endif
>}

This iterates over all PFNs in the highmem range and skips those in holes=
=2E
for_each_free_mem_range() skips the holes altogether=2E

Largely, I think we need to move, say, arm's version to mm/ and use it eve=
rywhere, except, perhaps, x86=2E

>Christophe
--=20
Sincerely yours,
Mike