Re: [PATCH v2 10/13] arch, mm: set high_memory in free_area_init()

[Mike Rapoport <rppt@kernel.org>]

On Wed, Apr 02, 2025 at 06:31:02PM +0200, Thomas Weißschuh wrote:
> On Wed, Apr 02, 2025 at 03:07:51PM +0200, Thomas Weißschuh wrote:
> > On Wed, Apr 02, 2025 at 03:46:37PM +0300, Mike Rapoport wrote:
> > > On Wed, Apr 02, 2025 at 02:19:01PM +0200, Thomas Weißschuh wrote:
> > > > (drop all the non-x86 and non-mm recipients)
> > > > 
> > > > On Thu, Mar 13, 2025 at 03:50:00PM +0200, Mike Rapoport wrote:
> > > > > From: "Mike Rapoport (Microsoft)" <rppt@kernel.org>
> > > > > 
> > > > > high_memory defines upper bound on the directly mapped memory.
> > > > > This bound is defined by the beginning of ZONE_HIGHMEM when a system has
> > > > > high memory and by the end of memory otherwise.
> > > > > 
> > > > > All this is known to generic memory management initialization code that
> > > > > can set high_memory while initializing core mm structures.
> > > > > 
> > > > > Add a generic calculation of high_memory to free_area_init() and remove
> > > > > per-architecture calculation except for the architectures that set and
> > > > > use high_memory earlier than that.
> > > > 
> > > > This change (in mainline as commit e120d1bc12da ("arch, mm: set high_memory in free_area_init()")
> > > > breaks booting i386 on QEMU for me (and others [0]).
> > > > The boot just hangs without output.
> > > > 
> > > > It's easily reproducible with kunit:
> > > > ./tools/testing/kunit/kunit.py run --arch i386
> > > > 
> > > > See below for the specific problematic hunk.
> > > > 
> > > > [0] https://lore.kernel.org/lkml/CA+G9fYtdXHVuirs3v6at3UoKNH5keuq0tpcvpz0tJFT4toLG4g@mail.gmail.com/
> > > > 
> > > > 
> > > > > diff --git a/arch/x86/mm/init_32.c b/arch/x86/mm/init_32.c
> > > > > index 6d2f8cb9451e..801b659ead0c 100644
> > > > > --- a/arch/x86/mm/init_32.c
> > > > > +++ b/arch/x86/mm/init_32.c
> > > > > @@ -643,9 +643,6 @@ void __init initmem_init(void)
> > > > >  		highstart_pfn = max_low_pfn;
> > > > >  	printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
> > > > >  		pages_to_mb(highend_pfn - highstart_pfn));
> > > > > -	high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
> > > > > -#else
> > > > > -	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
> > > > >  #endif
> > > > 
> > > > Reverting this hunk fixes the issue for me.
> > >  
> > > This is already done by d893aca973c3 ("x86/mm: restore early initialization
> > > of high_memory for 32-bits").
> > 
> > Thanks. Of course I only noticed this shortly after sending my mail.
> > But this usecase is indeed broken on mainline.
> > Some further bisecting lead to the mm merge commit being broken, while both its
> > parents work. That lead the bisection astray.
> > eb0ece16027f ("Merge tag 'mm-stable-2025-03-30-16-52' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm")
> > 
> > As unlikely as it sounds, it's reproducible. I'll investigate a bit.
> 
> The issue is fixed with the following diff:
> 
> diff --git a/mm/memblock.c b/mm/memblock.c
> index 284154445409..8cd95f60015d 100644
> --- a/mm/memblock.c
> +++ b/mm/memblock.c
> @@ -2165,7 +2165,8 @@ static unsigned long __init __free_memory_core(phys_addr_t start,
>                                  phys_addr_t end)
>  {
>         unsigned long start_pfn = PFN_UP(start);
> -       unsigned long end_pfn = PFN_DOWN(end);
> +       unsigned long end_pfn = min_t(unsigned long,
> +                                     PFN_DOWN(end), max_low_pfn);

This will leave HIGHMEM completely unusable. The proper fix is

diff --git a/mm/memblock.c b/mm/memblock.c
index 64ae678cd1d1..d7ff8dfe5f88 100644
--- a/mm/memblock.c
+++ b/mm/memblock.c
@@ -2166,6 +2166,9 @@ static unsigned long __init __free_memory_core(phys_addr_t start,
 	unsigned long start_pfn = PFN_UP(start);
 	unsigned long end_pfn = PFN_DOWN(end);
 
+	if (!IS_ENABLED(CONFIG_HIGHMEM) && end_pfn > max_low_pfn)
+		end_pfn = max_low_pfn;
+
 	if (start_pfn >= end_pfn)
 		return 0;

I've sent it along with the fix for x86 [1] (commit 7790c9c9265e
("memblock: don't release high memory to page allocator when HIGHMEM is
off") in mm-unstable), but for some reason it didn't make it to the Linus
tree :/

@Andrew, are you going to send it to Linus or you prefer if I take it via
memblock tree? 

[1] https://lore.kernel.org/all/20250325114928.1791109-3-rppt@kernel.org/

>         if (start_pfn >= end_pfn)
>                 return 0;
> 
> 
> Background:
> 
> This reverts part of commit 6faea3422e3b ("arch, mm: streamline HIGHMEM freeing")
> which is the direct child of the partially reverted 
> commit e120d1bc12da ("arch, mm: set high_memory in free_area_init()").
> The assumptions the former commit became invalid with the partial revert the latter.
> 
> This bug only triggers when CONFIG_HIGHMEM=n. When mm was branched from mainline
> the i386 configuration generated by kunit ended up with CONFIG_HIGHMEM=y.
> With some recent changes in mainline the kunit configuration switched to
> CONFIG_HIGHMEM=n, triggering this specific reproducer only when mm got merged
> into mainline again.
> 
> New kunit reproducer:
> ./tools/testing/kunit/kunit.py run --arch i386 example --timeout 10 --kconfig_add CONFIG_HIGHMEM=n
> 
> Does this sound reasonable?  If so I'll send a patch tomorrow.
> 
> @Naresh, could you test this, too?
> 
> 
> Thomas

-- 
Sincerely yours,
Mike.