Re: [PATCH v3] memblock: make memblock_find_in_range method private

[Guenter Roeck <linux@roeck-us.net>]

On 8/10/21 11:55 AM, Mike Rapoport wrote:
> On Mon, Aug 09, 2021 at 12:06:41PM -0700, Guenter Roeck wrote:
>> On Tue, Aug 03, 2021 at 09:42:18AM +0300, Mike Rapoport wrote:
>>> From: Mike Rapoport <rppt@linux.ibm.com>
>>>
>>> There are a lot of uses of memblock_find_in_range() along with
>>> memblock_reserve() from the times memblock allocation APIs did not exist.
>>>
>>> memblock_find_in_range() is the very core of memblock allocations, so any
>>> future changes to its internal behaviour would mandate updates of all the
>>> users outside memblock.
>>>
>>> Replace the calls to memblock_find_in_range() with an equivalent calls to
>>> memblock_phys_alloc() and memblock_phys_alloc_range() and make
>>> memblock_find_in_range() private method of memblock.
>>>
>>> This simplifies the callers, ensures that (unlikely) errors in
>>> memblock_reserve() are handled and improves maintainability of
>>> memblock_find_in_range().
>>>
>>> Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
>>
>> I see a number of crashes in next-20210806 when booting x86 images from efi.
>>
>> [    0.000000] efi: EFI v2.70 by EDK II
>> [    0.000000] efi: SMBIOS=0x1fbcc000 ACPI=0x1fbfa000 ACPI 2.0=0x1fbfa014 MEMATTR=0x1f25f018
>> [    0.000000] SMBIOS 2.8 present.
>> [    0.000000] DMI: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 0.0.0 02/06/2015
>> [    0.000000] last_pfn = 0x1ff50 max_arch_pfn = 0x400000000
>> [    0.000000] x86/PAT: Configuration [0-7]: WB  WC  UC- UC  WB  WP  UC- WT
>> [    0.000000] Kernel panic - not syncing: alloc_low_pages: can not alloc memory
>> [    0.000000] CPU: 0 PID: 0 Comm: swapper Not tainted 5.14.0-rc4-next-20210806 #1
>> [    0.000000] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 0.0.0 02/06/2015
>> [    0.000000] Call Trace:
>> [    0.000000]  ? dump_stack_lvl+0x57/0x7d
>> [    0.000000]  ? panic+0xfc/0x2c6
>> [    0.000000]  ? alloc_low_pages+0x117/0x156
>> [    0.000000]  ? phys_pmd_init+0x234/0x342
>> [    0.000000]  ? phys_pud_init+0x171/0x337
>> [    0.000000]  ? __kernel_physical_mapping_init+0xec/0x276
>> [    0.000000]  ? init_memory_mapping+0x1ea/0x2aa
>> [    0.000000]  ? init_range_memory_mapping+0xdf/0x12e
>> [    0.000000]  ? init_mem_mapping+0x1e9/0x26f
>> [    0.000000]  ? setup_arch+0x5ff/0xb6d
>> [    0.000000]  ? start_kernel+0x71/0x6b4
>> [    0.000000]  ? secondary_startup_64_no_verify+0xc2/0xcb
>>
>> Bisect points to this patch. Reverting it fixes the problem. Key seems to
>> be the amount of memory configured in qemu; the problem is not seen if
>> there is 1G or more of memory, but it is seen with all test boots with
>> 512M or 256M of memory. It is also seen with almost all 32-bit efi boots.
>>
>> The problem is not seen when booting without efi.
> 
> It looks like this change uncovered a problem in
> x86::memory_map_top_down().
> 
> The allocation in alloc_low_pages() is limited by min_pfn_mapped and
> max_pfn_mapped. The min_pfn_mapped is updated at every iteration of the
> loop in memory_map_top_down, but there is another loop in
> init_range_memory_mapping() that maps several regions below the current
> min_pfn_mapped without updating this variable.
> 
> The memory layout in qemu with 256M of RAM and EFI enabled, causes
> exhaustion of the memory limited by min_pfn_mapped and max_pfn_mapped
> before min_pfn_mapped is updated.
> 
> Before this commit there was unconditional "reservation" of 2M in the end
> of the memory that moved the initial min_pfn_mapped below the memory
> reserved by EFI. The addition of check for xen_domain() removed this
> reservation for !XEN and made alloc_low_pages() use the range already busy
> with EFI data.
> 
> The patch below moves the update of min_pfn_mapped near the update of
> max_pfn_mapped so that every time a new range is mapped both limits will be
> updated accordingly.
> 
> diff --git a/arch/x86/mm/init.c b/arch/x86/mm/init.c
> index 1152a29ce109..be279f6e5a0a 100644
> --- a/arch/x86/mm/init.c
> +++ b/arch/x86/mm/init.c
> @@ -1,3 +1,4 @@
> +#define DEBUG
>   #include <linux/gfp.h>
>   #include <linux/initrd.h>
>   #include <linux/ioport.h>
> @@ -485,6 +486,7 @@ static void add_pfn_range_mapped(unsigned long start_pfn, unsigned long end_pfn)
>   	nr_pfn_mapped = clean_sort_range(pfn_mapped, E820_MAX_ENTRIES);
>   
>   	max_pfn_mapped = max(max_pfn_mapped, end_pfn);
> +	min_pfn_mapped = min(min_pfn_mapped, start_pfn);
>   
>   	if (start_pfn < (1UL<<(32-PAGE_SHIFT)))
>   		max_low_pfn_mapped = max(max_low_pfn_mapped,
> @@ -643,7 +645,6 @@ static void __init memory_map_top_down(unsigned long map_start,
>   		mapped_ram_size += init_range_memory_mapping(start,
>   							last_start);
>   		last_start = start;
> -		min_pfn_mapped = last_start >> PAGE_SHIFT;
>   		if (mapped_ram_size >= step_size)
>   			step_size = get_new_step_size(step_size);
>   	}
>   

The offending patch was removed from next-20210810, but I applied the above change
to next-20210809 and it does indeed fix the problem. If it is added as separate patch,
please feel free to add

Tested-by: Guenter Roeck <linux@roeck-us.net>

Thanks,
Guenter