Message-ID: <ff6c9ac0-dce2-4d3f-b5f7-15f8fff3379b@redhat.com>
Date: Tue, 20 May 2025 17:08:52 +0200
From: David Hildenbrand <david@...hat.com>
To: Mike Rapoport <rppt@...nel.org>
Cc: Oscar Salvador <osalvador@...e.de>,
 Andrew Morton <akpm@...ux-foundation.org>,
 Alexandre Ghiti <alexghiti@...osinc.com>, Pratyush Yadav
 <ptyadav@...zon.de>, linux-arch@...r.kernel.org,
 linux-kernel@...r.kernel.org, linux-mm@...ck.org
Subject: Re: [PATCH] mm/cma: make detection of highmem_start more robust

On 20.05.25 17:06, Mike Rapoport wrote:
> On Tue, May 20, 2025 at 11:14:28AM +0200, David Hildenbrand wrote:
>> On 20.05.25 10:30, Oscar Salvador wrote:
>>> On Mon, May 19, 2025 at 08:18:05PM +0300, Mike Rapoport wrote:
>>>> From: "Mike Rapoport (Microsoft)" <rppt@...nel.org>
>>>>
>>>> Pratyush Yadav reports the following crash:
>>>>
>>>>       ------------[ cut here ]------------
>>>>       kernel BUG at arch/x86/mm/physaddr.c:23!
>>>>       ception 0x06 IP 10:ffffffff812ebbf8 error 0 cr2 0xffff88903ffff000
>>>>       CPU: 0 UID: 0 PID: 0 Comm: swapper Not tainted 6.15.0-rc6+ #231 PREEMPT(undef)
>>>>       Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Arch Linux 1.16.3-1-1 04/01/2014
>>>>       RIP: 0010:__phys_addr+0x58/0x60
>>>>       Code: 01 48 89 c2 48 d3 ea 48 85 d2 75 05 e9 91 52 cf 00 0f 0b 48 3d ff ff ff 1f 77 0f 48 8b 05 20 54 55 01 48 01 d0 e9 78 52 cf 00 <0f> 0b 90 0f 1f 44 00 00 90 90 90 90 90 90 90 90 90 90 90 90 90 90
>>>>       RSP: 0000:ffffffff82803dd8 EFLAGS: 00010006 ORIG_RAX: 0000000000000000
>>>>       RAX: 000000007fffffff RBX: 00000000ffffffff RCX: 0000000000000000
>>>>       RDX: 000000007fffffff RSI: 0000000280000000 RDI: ffffffffffffffff
>>>>       RBP: ffffffff82803e68 R08: 0000000000000000 R09: 0000000000000000
>>>>       R10: ffffffff83153180 R11: ffffffff82803e48 R12: ffffffff83c9aed0
>>>>       R13: 0000000000000000 R14: 0000001040000000 R15: 0000000000000000
>>>>       FS:  0000000000000000(0000) GS:0000000000000000(0000) knlGS:0000000000000000
>>>>       CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>>>       CR2: ffff88903ffff000 CR3: 0000000002838000 CR4: 00000000000000b0
>>>>       Call Trace:
>>>>        <TASK>
>>>>        ? __cma_declare_contiguous_nid+0x6e/0x340
>>>>        ? cma_declare_contiguous_nid+0x33/0x70
>>>>        ? dma_contiguous_reserve_area+0x2f/0x70
>>>>        ? setup_arch+0x6f1/0x870
>>>>        ? start_kernel+0x52/0x4b0
>>>>        ? x86_64_start_reservations+0x29/0x30
>>>>        ? x86_64_start_kernel+0x7c/0x80
>>>>        ? common_startup_64+0x13e/0x141
>>>>
>>>>     The reason is that __cma_declare_contiguous_nid() does:
>>>>
>>>>             highmem_start = __pa(high_memory - 1) + 1;
>>>>
>>>>     If dma_contiguous_reserve_area() (or any other CMA declaration) is
>>>>     called before free_area_init(), high_memory is uninitialized. Without
>>>>     CONFIG_DEBUG_VIRTUAL, it will likely work but use the wrong value for
>>>>     highmem_start.
>>>>
>>>> The issue occurs because commit e120d1bc12da ("arch, mm: set high_memory in
>>>> free_area_init()") moved initialization of high_memory after the call to
>>>> dma_contiguous_reserve() -> __cma_declare_contiguous_nid() on several
>>>> architectures.
>>>>
>>>> In the case CONFIG_HIGHMEM is enabled, some architectures that actually
>>>> support HIGHMEM (arm, powerpc and x86) have initialization of high_memory
>>>> before a possible call to __cma_declare_contiguous_nid() and some
>>>> initialized high_memory late anyway (arc, csky, microblase, mips, sparc,
>>>> xtensa) even before the commit e120d1bc12da so they are fine with using
>>>> uninitialized value of high_memory.
>>>>
>>>> And in the case CONFIG_HIGHMEM is disabled high_memory essentially becomes
>>>> the first address after memory end, so instead of relying on high_memory to
>>>> calculate highmem_start use memblock_end_of_DRAM() and eliminate the
>>>> dependency of CMA area creation on high_memory in majority of
>>>> configurations.
>>>>
>>>> Reported-by: Pratyush Yadav <ptyadav@...zon.de>
>>>> Tested-by: Alexandre Ghiti <alexghiti@...osinc.com>
>>>> Signed-off-by: Mike Rapoport (Microsoft) <rppt@...nel.org>
>>>
>>> Reviewed-by: Oscar Salvador <osalvador@...e.de>
>>>
>>> I will note though that it is a bit akward to have highmem involved here
>>> when we might not have CONFIG_HIGHMEM enabled.
>>> I get that for !CONFIG_HIGHMEM it is a no-op situation, but still I
>>> wonder whether we could abstract that from this function.
> 
> Highmem is there for some time now (see f7426b983a6a ("mm: cma: adjust
> address limit to avoid hitting low/high memory boundary"))
> We might try abstracting it from that function but I'd prefer not doing it
> that late in the release cycle.

Agreed, assuming this will still make it into this release.

>   
>> Same thought here.
>>
>> Can't we do some IS_ENABLED(CONFIG_HIGHMEM) magic or similar to not even use
>> that variable without CONFIG_HIGHMEM?
> 
> You mean highmem_start or high_memory?

highmem_start in this function.

> 
> high_memory is one of the ways to say "end of directly/linearly addressable
> memory" and some other places in the kernel (outside arch) still use it
> regardless of CONFIG_HIGHMEM.
> 
> And I don't think we have another way to say where directly addressable
> memory ends, and this IMHO is something that should replace high_memory.

Agreed.


-- 
Cheers,

David / dhildenb

Powered by blists - more mailing lists