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

On Tue, Mar 11, 2025 at 05:51:06PM +0000, Mark Brown wrote:
> On Thu, Mar 06, 2025 at 08:51:20PM +0200, Mike Rapoport wrote:
> > From: "Mike Rapoport (Microsoft)" <[email protected]>
> > 
> > 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.
> > 
> > Remove per-architecture calculation of high_memory and add a generic
> > version to free_area_init().
> 
> This patch appears to be causing breakage on a number of 32 bit arm
> platforms, including qemu's virt-2.11,gic-version=3.  Affected platforms
> die on boot with no output, a bisect with qemu points at this commit and
> those for physical platforms appear to be converging on the same place.

I'm not convinced that the old and the new code is doing the same
thing.

The new code:

+       phys_addr_t highmem = memblock_end_of_DRAM();
+
+#ifdef CONFIG_HIGHMEM
+       unsigned long pfn = arch_zone_lowest_possible_pfn[ZONE_HIGHMEM];
+
+       if (arch_has_descending_max_zone_pfns() || highmem > PFN_PHYS(pfn))
+               highmem = PFN_PHYS(pfn);
+#endif
+
+       high_memory = phys_to_virt(highmem - 1) + 1;

First, when CONFIG_HIGHMEM is disabled, this code assumes that the last
byte of DRAM declared to memblock is the highmem limit. This _could_
overflow phys_to_virt() and lead to an invalid value for high_memory.

Second, arch_zone_lowest_possible_pfn[ZONE_HIGHMEM] is the _start_ of
highmem. This is not what arch code sets high_memory to - because
the start of highmem may not contiguously follow on from lowmem.

In arch/arm/mm/mmu.c, lowmem_limit is computed to be the highest + 1
physical address that lowmem can possibly be, taking into account the
amount of vmalloc memory that is required. This is used to set
high_memory.

We also limit the amount of usable RAM via memblock_set_current_limit()
which memblock_end_of_DRAM() doesn't respect.

I don't think the proposed generic version is suitable for 32-bit arm.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 80Mbps down 10Mbps up. Decent connectivity at last!