| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <20240116122304.qwzy7san2vgspt2x@Board-3A3000>
Date: Tue, 16 Jan 2024 20:23:04 +0800
From: Huang Pei <huangpei@...ngson.cn>
To: Mike Rapoport <rppt@...nel.org>
Cc: Jiaxun Yang <jiaxun.yang@...goat.com>, linux-mm@...ck.org,
Bibo Mao <maobibo@...ngson.cn>, linux-mips@...r.kernel.org,
Paul Burton <paulburton@...nel.org>,
Li Xuefeng <lixuefeng@...ngson.cn>,
Yang Tiezhu <yangtiezhu@...ngson.cn>,
Gao Juxin <gaojuxin@...ngson.cn>,
Huacai Chen <chenhuacai@...ngson.cn>,
Thomas Bogendoerfer <tsbogend@...ha.franken.de>,
linux-kernel@...r.kernel.org, Yajun Deng <yajun.deng@...ux.dev>
Subject: Re: memblock_reserve for unadded region (was: [PATCH] MIPS:
loongson64: fix boot failure)
On Tue, Jan 16, 2024 at 10:39:04AM +0200, Mike Rapoport wrote:
> On Mon, Jan 15, 2024 at 02:08:21PM +0000, Jiaxun Yang wrote:
> > Hi mm folks,
> >
> > Just a quick question, what is the expected behavior of memblock_reserve
> > a region that is not added to memblock with memblock_add?
> >
> > I'm unable to find any documentation about memblock_reserve in comments and
> > boot-time-mm, but as per my understanding to the code, this should be a
> > legit usage?
>
> Yes, memblock allows reserving memory that was not added to memblock with
> memblock_add().
I think arch/platform specific code should fix this bug, like,
--------------------------------------------------------------------------
//for loongson64
memblock_set_node(0, 1ULL << 44, &memblock.reserved, 0);
--------------------------------------------------------------------------
or maybe memblock provide something like memblock_reserve_node
>
> > In practical we run into uninitialized nid of reserved block problem, should
> > we fix it
> > in our usage, or on memblock side?
>
> Apparently it's a bug in memblock :(
>
> If you revert 61167ad5fecd ("mm: pass nid to reserve_bootmem_region()")
> does the issue disappear?
Yes, I git bisect this commit.
But I don't think it is a bug in memblock. IMO, memblock_reserve under
NUMA set nid of reserved region to MAX_NUMNODES, which is the point
that cause the "memblock_get_region_node from memmap_init_reserved_pages "
passing a invalid node id(aka MAX_NUMNODES) to "reserver_bootmem_region
-> init_reserved_page -> early_pfn_to_nid". If arch-specific code DOES NOT
initialize the nid of reserved region(only it know that), or the reserved
region NOT within a memblock added by memblock_add, memblock can not
give a valid node id to the reserved region. Commit 61167ad5fecd ("mm: pass nid to
reserve_bootmem_region()") just reveals the embarrassment case by an
out of bound memory access.
>
> > Thanks
> >
> > 在 2023/12/25 09:30, Huang Pei 写道:
> > > Since commit 61167ad5fecd("mm: pass nid to reserve_bootmem_region()),
> > > loongson64 booting failed with CONFIG_DEFERRED_STRUCT_PAGE_INIT like
> > > this:
> > > ----------------------------------------------------------------------
> > > Call Trace:
> > > [<ffffffff8235d088>] reserve_bootmem_region+0xa8/0x184
> > > [<ffffffff82333940>] memblock_free_all+0x104/0x2a8
> > > [<ffffffff8231d8e4>] mem_init+0x84/0x94
> > > [<ffffffff82330958>] mm_core_init+0xf8/0x308
> > > [<ffffffff82318c38>] start_kernel+0x43c/0x86c
> > >
> > > Code: 10400028 2402fff0 de420000 <dc432880> 0203182b 14600022
> > > 64420070 00003025 24040003
> > >
> > > ---[ end trace 0000000000000000 ]---
> > > Kernel panic - not syncing: Attempted to kill the idle task!
> > > ---[ end Kernel panic - not syncing: Attempted to kill the idle task! ]---
> > > ----------------------------------------------------------------------
> > >
> > > The root cause is no memory region "0x0-0x1fffff" paired with
> > > memory-reserved region "0x0-0x1fffff" and "0x0-0xfff", with "memblock
> > > =debug":
> > >
> > > ----------------------------------------------------------------------
> > > memory[0x0] [0x0000000000200000-0x000000000effffff],
> > > 0x000000000ee00000 bytes on node 0 flags: 0x0 !!!!here
> > > memory[0x1] [0x0000000090000000-0x00000000fdffffff],
> > > 0x000000006e000000 bytes on node 0 flags: 0x0
> > > memory[0x2] [0x0000000100000000-0x000000027fffffff],
> > > 0x0000000180000000 bytes on node 0 flags: 0x0
> > > memory[0x3] [0x0000100000000000-0x000010000fffffff],
> > > 0x0000000010000000 bytes on node 1 flags: 0x0
> > > memory[0x4] [0x0000100090000000-0x000010027fffffff],
> > > 0x00000001f0000000 bytes on node 1 flags: 0x0
> > > reserved.cnt = 0x1f
> > > reserved[0x0] [0x0000000000000000-0x000000000190c80a],
> > > 0x000000000190c80b bytes flags: 0x0 !!!!oops 0x0-0x1fffff not in memory[0]
> > > reserved[0x1] [0x000000000190c810-0x000000000190eea3],
> > > 0x0000000000002694 bytes flags: 0x0
> > > ----------------------------------------------------------------------
> > >
> > > It caused memory-reserved region "0x0-0x1fffff" without valid node id
> > > in "memblock_get_region_node" from "memmap_init_reserved_pages", lead to
> > > "reserve_bootmem_region-> init_reserved_page -> early_pfn_to_nid()"
> > > accessing "node_data" out of bound.
> > >
> > > To fix this bug, we should remove unnecessary memory block reservation.
> > >
> > > +. no need to reserve 0x0-0x1fffff below kernel loading address, since
> > > it is not registered by "memblock_add_node"
> > >
> > > +. no need to reserve 0x0-0xfff for exception handling if it is not
> > > registered by "memblock_add" either.
> > >
> > > Fixes: commit 61167ad5fecd("mm: pass nid to reserve_bootmem_region())
> > > Signed-off-by: Huang Pei <huangpei@...ngson.cn>
> > > ---
> > > arch/mips/kernel/traps.c | 3 ++-
> > > arch/mips/loongson64/numa.c | 2 --
> > > 2 files changed, 2 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/arch/mips/kernel/traps.c b/arch/mips/kernel/traps.c
> > > index 246c6a6b0261..9b632b4c10c3 100644
> > > --- a/arch/mips/kernel/traps.c
> > > +++ b/arch/mips/kernel/traps.c
> > > @@ -2007,7 +2007,8 @@ unsigned long vi_handlers[64];
> > > void reserve_exception_space(phys_addr_t addr, unsigned long size)
> > > {
> > > - memblock_reserve(addr, size);
> > > + if(memblock_is_region_memory(addr, size))
> > > + memblock_reserve(addr, size);
> > > }
> > > void __init *set_except_vector(int n, void *addr)
> > > diff --git a/arch/mips/loongson64/numa.c b/arch/mips/loongson64/numa.c
> > > index 8f61e93c0c5b..0f516dde81da 100644
> > > --- a/arch/mips/loongson64/numa.c
> > > +++ b/arch/mips/loongson64/numa.c
> > > @@ -130,8 +130,6 @@ static void __init node_mem_init(unsigned int node)
> > > memblock_reserve((node_addrspace_offset | 0xfe000000),
> > > 32 << 20);
> > > - /* Reserve pfn range 0~node[0]->node_start_pfn */
> > > - memblock_reserve(0, PAGE_SIZE * start_pfn);
> > > }
> > > }
> >
> > --
> > ---
> > Jiaxun Yang
> >
>
> --
> Sincerely yours,
> Mike.
Powered by blists - more mailing lists