Re: [PATCH 04/21] mm: free_area_init: use maximal zone PFNs rather than zone sizes
From: Mike Rapoport <rppt@kernel.org>
Date: 2020-06-15 06:22:56
Also in:
linux-alpha, linux-arch, linux-doc, linux-m68k, linux-mips, linux-mm, linux-riscv, linux-s390, linux-sh, linux-um, lkml, sparclinux
Subsystem:
m68k architecture, the rest, uclinux (m68knommu and coldfire) · Maintainers:
Geert Uytterhoeven, Linus Torvalds, Greg Ungerer
Hi Greg, On Mon, Jun 15, 2020 at 01:53:42PM +1000, Greg Ungerer wrote:
Hi Mike, From: Mike Rapoport <redacted>quoted
Currently, architectures that use free_area_init() to initialize memory map and node and zone structures need to calculate zone and hole sizes. We can use free_area_init_nodes() instead and let it detect the zone boundaries while the architectures will only have to supply the possible limits for the zones. Signed-off-by: Mike Rapoport <redacted>This is causing some new warnings for me on boot on at least one non-MMU m68k target:
There were a couple of changes that cause this. The free_area_init() now relies on memblock data and architectural limits for zone sizes rather than on explisit pfns calculated by the arch code. I've update motorola variant and missed coldfire. Angelo sent a fix for mcfmmu.c [1] and I've updated it to include nommu as well [1] https://lore.kernel.org/linux-m68k/20200614225119.777702-1-angelo.dureghello@timesys.com (local) From 55b8523df2a5c4565b132c0691990f0821040fec Mon Sep 17 00:00:00 2001 From: Angelo Dureghello <redacted> Date: Mon, 15 Jun 2020 00:51:19 +0200 Subject: [PATCH] m68k: fix registration of memory regions with memblock Commit 3f08a302f533 ("mm: remove CONFIG_HAVE_MEMBLOCK_NODE_MAP option") introduced assumption that UMA systems have their memory at node 0 and updated most of them, but it forgot nommu and coldfire variants of m68k. The later change in free area initialization in commit fa3354e4ea39 ("mm: free_area_init: use maximal zone PFNs rather than zone sizes") exposed that and caused a lot of "BUG: Bad page state in process swapper" reports. Using memblock_add_node() with nid = 0 to register memory banks solves the problem. Fixes: 3f08a302f533 ("mm: remove CONFIG_HAVE_MEMBLOCK_NODE_MAP option") Fixes: fa3354e4ea39 ("mm: free_area_init: use maximal zone PFNs rather than zone sizes") Signed-off-by: Angelo Dureghello <redacted> Co-developed-by: Mike Rapoport <redacted> Signed-off-by: Mike Rapoport <redacted> --- arch/m68k/kernel/setup_no.c | 2 +- arch/m68k/mm/mcfmmu.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/arch/m68k/kernel/setup_no.c b/arch/m68k/kernel/setup_no.c
index e779b19e0193..0c4589a39ba9 100644
--- a/arch/m68k/kernel/setup_no.c
+++ b/arch/m68k/kernel/setup_no.c@@ -138,7 +138,7 @@ void __init setup_arch(char **cmdline_p) pr_debug("MEMORY -> ROMFS=0x%p-0x%06lx MEM=0x%06lx-0x%06lx\n ", __bss_stop, memory_start, memory_start, memory_end); - memblock_add(memory_start, memory_end - memory_start); + memblock_add_node(memory_start, memory_end - memory_start, 0); /* Keep a copy of command line */ *cmdline_p = &command_line[0];
diff --git a/arch/m68k/mm/mcfmmu.c b/arch/m68k/mm/mcfmmu.c
index 29f47923aa46..7d04210d34f0 100644
--- a/arch/m68k/mm/mcfmmu.c
+++ b/arch/m68k/mm/mcfmmu.c@@ -174,7 +174,7 @@ void __init cf_bootmem_alloc(void) m68k_memory[0].addr = _rambase; m68k_memory[0].size = _ramend - _rambase; - memblock_add(m68k_memory[0].addr, m68k_memory[0].size); + memblock_add_node(m68k_memory[0].addr, m68k_memory[0].size, 0); /* compute total pages in system */ num_pages = PFN_DOWN(_ramend - _rambase);
--
2.26.2
> ...
> NET: Registered protocol family 17
> BUG: Bad page state in process swapper pfn:20165
> page:41fe0ca0 refcount:0 mapcount:1 mapping:00000000 index:0x0
> flags: 0x0()
> raw: 00000000 00000100 00000122 00000000 00000000 00000000 00000000 00000000
> page dumped because: nonzero mapcount
> CPU: 0 PID: 1 Comm: swapper Not tainted 5.8.0-rc1-00001-g3a38f8a60c65-dirty #1
> Stack from 404c9ebc:
> 404c9ebc 4029ab28 4029ab28 40088470 41fe0ca0 40299e21 40299df1 404ba2a4
> 00020165 00000000 41fd2c10 402c7ba0 41fd2c04 40088504 41fe0ca0 40299e21
> 00000000 40088a12 41fe0ca0 41fe0ca4 0000020a 00000000 00000001 402ca000
> 00000000 41fe0ca0 41fd2c10 41fd2c10 00000000 00000000 402b2388 00000001
...
>
> System boots pretty much as normal through user space after this.
> Seems to be fully operational despite all those BUGONs.
>
> Specifically this is a M5208EVB target (arch/m68k/configs/m5208evb).
>
>
> [snip]
> > diff --git a/arch/m68k/mm/init.c b/arch/m68k/mm/init.c
> > index b88d510d4fe3..6d3147662ff2 100644
> > --- a/arch/m68k/mm/init.c
> > +++ b/arch/m68k/mm/init.c
> > @@ -84,7 +84,7 @@ void __init paging_init(void)
> > * page_alloc get different views of the world.
> > */
> > unsigned long end_mem = memory_end & PAGE_MASK;
> > - unsigned long zones_size[MAX_NR_ZONES] = { 0, };
> > + unsigned long max_zone_pfn[MAX_NR_ZONES] = { 0, };
> > high_memory = (void *) end_mem;
> > @@ -98,8 +98,8 @@ void __init paging_init(void)
> > */
> > set_fs (USER_DS);
> > - zones_size[ZONE_DMA] = (end_mem - PAGE_OFFSET) >> PAGE_SHIFT;
> > - free_area_init(zones_size);
> > + max_zone_pfn[ZONE_DMA] = end_mem >> PAGE_SHIFT;
> > + free_area_init(max_zone_pfn);
>
> This worries me a little. On this target PAGE_OFFSET will be non-0.
> Thoughts?
The initialization in free_area_init() takes into account the actual
physical memory sizing from memblock and max_zone_pfn as the
architectural limit for possible zone extents. This (and the patch
above) is enough to properly setup node and zones.
> Regards
> Greg
>
>
>
--
Sincerely yours,
Mike.