Re: [PATCH] parisc/mm: preallocate fixmap page tables at init

From: Vlastimil Babka
Date: Thu Aug 03 2023 - 04:48:48 EST

On 8/3/23 08:24, Mike Rapoport wrote:
> From: "Mike Rapoport (IBM)" <rppt@xxxxxxxxxx>
>
> Christoph Biedl reported early OOM on recent kernels:
>
> swapper: page allocation failure: order:0, mode:0x100(__GFP_ZERO),
> nodemask=(null)
> CPU: 0 PID: 0 Comm: swapper Not tainted 6.3.0-rc4+ #16
> Hardware name: 9000/785/C3600
> Backtrace:
> [<10408594>] show_stack+0x48/0x5c
> [<10e152d8>] dump_stack_lvl+0x48/0x64
> [<10e15318>] dump_stack+0x24/0x34
> [<105cf7f8>] warn_alloc+0x10c/0x1c8
> [<105d068c>] __alloc_pages+0xbbc/0xcf8
> [<105d0e4c>] __get_free_pages+0x28/0x78
> [<105ad10c>] __pte_alloc_kernel+0x30/0x98
> [<10406934>] set_fixmap+0xec/0xf4
> [<10411ad4>] patch_map.constprop.0+0xa8/0xdc
> [<10411bb0>] __patch_text_multiple+0xa8/0x208
> [<10411d78>] patch_text+0x30/0x48
> [<1041246c>] arch_jump_label_transform+0x90/0xcc
> [<1056f734>] jump_label_update+0xd4/0x184
> [<1056fc9c>] static_key_enable_cpuslocked+0xc0/0x110
> [<1056fd08>] static_key_enable+0x1c/0x2c
> [<1011362c>] init_mem_debugging_and_hardening+0xdc/0xf8
> [<1010141c>] start_kernel+0x5f0/0xa98
> [<10105da8>] start_parisc+0xb8/0xe4
>
> Mem-Info:
> active_anon:0 inactive_anon:0 isolated_anon:0
> active_file:0 inactive_file:0 isolated_file:0
> unevictable:0 dirty:0 writeback:0
> slab_reclaimable:0 slab_unreclaimable:0
> mapped:0 shmem:0 pagetables:0
> sec_pagetables:0 bounce:0
> kernel_misc_reclaimable:0
> free:0 free_pcp:0 free_cma:0
> Node 0 active_anon:0kB inactive_anon:0kB active_file:0kB
> inactive_file:0kB unevictable:0kB isolated(anon):0kB isolated(file):0kB
> mapped:0kB dirty:0kB writeback:0kB shmem:0kB
> +writeback_tmp:0kB kernel_stack:0kB pagetables:0kB sec_pagetables:0kB
> all_unreclaimable? no
> Normal free:0kB boost:0kB min:0kB low:0kB high:0kB
> reserved_highatomic:0KB active_anon:0kB inactive_anon:0kB active_file:0kB
> inactive_file:0kB unevictable:0kB writepending:0kB
> +present:1048576kB managed:1039360kB mlocked:0kB bounce:0kB free_pcp:0kB
> local_pcp:0kB free_cma:0kB
> lowmem_reserve[]: 0 0
> Normal: 0*4kB 0*8kB 0*16kB 0*32kB 0*64kB 0*128kB 0*256kB 0*512kB
> 0*1024kB 0*2048kB 0*4096kB = 0kB
> 0 total pagecache pages
> 0 pages in swap cache
> Free swap = 0kB
> Total swap = 0kB
> 262144 pages RAM
> 0 pages HighMem/MovableOnly
> 2304 pages reserved
> Backtrace:
> [<10411d78>] patch_text+0x30/0x48
> [<1041246c>] arch_jump_label_transform+0x90/0xcc
> [<1056f734>] jump_label_update+0xd4/0x184
> [<1056fc9c>] static_key_enable_cpuslocked+0xc0/0x110
> [<1056fd08>] static_key_enable+0x1c/0x2c
> [<1011362c>] init_mem_debugging_and_hardening+0xdc/0xf8
> [<1010141c>] start_kernel+0x5f0/0xa98
> [<10105da8>] start_parisc+0xb8/0xe4
>
> Kernel Fault: Code=15 (Data TLB miss fault) at addr 0f7fe3c0
> CPU: 0 PID: 0 Comm: swapper Not tainted 6.3.0-rc4+ #16
> Hardware name: 9000/785/C3600
>
> This happens because patching static key code temporarily maps it via
> fixmap and if it happens before page allocator is initialized set_fixmap()
> cannot allocate memory using pte_alloc_kernel().
>
> Make sure that fixmap page tables are preallocated early so that
> pte_offset_kernel() in set_fixmap() never resorts to pte allocation.
>
> Signed-off-by: Mike Rapoport (IBM) <rppt@xxxxxxxxxx>

Acked-by: Vlastimil Babka <vbabka@xxxxxxx>

Thanks, Mike!

> ---
>
> I didn't add Fixes tag with the commit Christoph bisected because that
> commit didn't change anything in the initialization of the static keys
> and I really doubt it is the actual cause of the issue.

Yeah I also don't see how the commit could cause a static key initialization
to happen sooner than it already has.
But if it's manifesting for whatever reason since 6.4 then a "Cc: stable #
6.4+" would be in order.

>
> arch/parisc/mm/fixmap.c | 3 ---
> arch/parisc/mm/init.c | 34 ++++++++++++++++++++++++++++++++++
> 2 files changed, 34 insertions(+), 3 deletions(-)
>
> diff --git a/arch/parisc/mm/fixmap.c b/arch/parisc/mm/fixmap.c
> index cc15d737fda6..ae3493dae9dc 100644
> --- a/arch/parisc/mm/fixmap.c
> +++ b/arch/parisc/mm/fixmap.c
> @@ -19,9 +19,6 @@ void notrace set_fixmap(enum fixed_addresses idx, phys_addr_t phys)
> pmd_t *pmd = pmd_offset(pud, vaddr);
> pte_t *pte;
>
> - if (pmd_none(*pmd))
> - pte = pte_alloc_kernel(pmd, vaddr);
> -
> pte = pte_offset_kernel(pmd, vaddr);
> set_pte_at(&init_mm, vaddr, pte, __mk_pte(phys, PAGE_KERNEL_RWX));
> flush_tlb_kernel_range(vaddr, vaddr + PAGE_SIZE);
> diff --git a/arch/parisc/mm/init.c b/arch/parisc/mm/init.c
> index 406c52fe23d5..389941c7f209 100644
> --- a/arch/parisc/mm/init.c
> +++ b/arch/parisc/mm/init.c
> @@ -669,6 +669,39 @@ static void __init gateway_init(void)
> PAGE_SIZE, PAGE_GATEWAY, 1);
> }
>
> +static void __init fixmap_init(void)
> +{
> + unsigned long addr = FIXMAP_START;
> + unsigned long end = FIXMAP_START + FIXMAP_SIZE;
> + pgd_t *pgd = pgd_offset_k(addr);
> + p4d_t *p4d = p4d_offset(pgd, addr);
> + pud_t *pud = pud_offset(p4d, addr);
> + pmd_t *pmd;
> +
> + BUILD_BUG_ON(FIXMAP_SIZE > PMD_SIZE);
> +
> +#if CONFIG_PGTABLE_LEVELS == 3
> + if (pud_none(*pud)) {
> + pmd = memblock_alloc(PAGE_SIZE << PMD_TABLE_ORDER,
> + PAGE_SIZE << PMD_TABLE_ORDER);
> + if (!pmd)
> + panic("fixmap: pmd allocation failed.\n");
> + pud_populate(NULL, pud, pmd);
> + }
> +#endif
> +
> + pmd = pmd_offset(pud, addr);
> + do {
> + pte_t *pte = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
> + if (!pte)
> + panic("fixmap: pte allocation failed.\n");
> +
> + pmd_populate_kernel(&init_mm, pmd, pte);
> +
> + addr += PAGE_SIZE;
> + } while (addr < end);
> +}
> +
> static void __init parisc_bootmem_free(void)
> {
> unsigned long max_zone_pfn[MAX_NR_ZONES] = { 0, };
> @@ -683,6 +716,7 @@ void __init paging_init(void)
> setup_bootmem();
> pagetable_init();
> gateway_init();
> + fixmap_init();
> flush_cache_all_local(); /* start with known state */
> flush_tlb_all_local(NULL);
>
>
> base-commit: 5d0c230f1de8c7515b6567d9afba1f196fb4e2f4