On 2023/11/2 1:29, Nathan Chancellor wrote:
Hi Kefeng,
On Mon, Oct 30, 2023 at 05:34:32PM +0800, Kefeng Wang wrote:
On 2023/10/26 0:37, Nathan Chancellor wrote:
On Thu, Oct 19, 2023 at 09:32:53AM +0800, Kefeng Wang wrote:
Use early_pfn_to_nid() to get correct node id from base instead of
the default NUMA_NO_NODE in cma_declare_contiguous_nid().
Signed-off-by: Kefeng Wang <wangkefeng.wang@xxxxxxxxxx>
---
mm/cma.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/mm/cma.c b/mm/cma.c
index 2b2494fd6b59..97c27e5fe1a2 100644
--- a/mm/cma.c
+++ b/mm/cma.c
@@ -375,6 +375,9 @@ int __init cma_declare_contiguous_nid(phys_addr_t base,
if (ret)
goto free_mem;
+ if (nid == NUMA_NO_NODE)
+ nid = early_pfn_to_nid(PHYS_PFN(base));
+
pr_info("Reserved %ld MiB at %pa on node %d\n", (unsigned long)size / SZ_1M,
&base, nid);
return 0;
--
2.27.0
I bisected a RISC-V boot failure in QEMU to this change in -next. It
happens with OpenSUSE's RISC-V configuration [1], which I was able to
narrow down to the follow configurations on top of defconfig:
I think the root cause is the bad node info of memory address, meanwhile,
the riscv's cma reserve is before numa init, see the following log,
[ 0.000000] cma: Reserved 16 MiB at 0x000000009f000000 on node 4
[ 0.000000] NUMA: Faking a node at [mem
0x0000000080000000-0x000000009fffffff]
[ 0.000000] NUMA: NODE_DATA [mem 0x9eff2780-0x9eff3fff]
[ 0.000000] NUMA: NODE_DATA(0) on node 4 // should be node 0
[ 0.000000] [ff1c000002000000-ff1c000002000fff] potential offnode
page_structs
additional, early_pfn_to_nid will cache the recent lookups of pfn-to-nid,
which
led to the next early_pfn_to_nid get the cache nid, not the new nid(changed
by numa init),
setup_arch
paging_init
dma_contiguous_reserve
cma_declare_contiguous_nid // 9f000000 node 4
early_pfn_to_nid // 1. lookup memblk, pfn=9f000, nid=4 cached
misc_mem_init
arch_numa_init
numa_init
dummy_numa_init
numa_add_memblk // 2. setup new nid of memblk
numa_register_nodes
setup_node_data
early_pfn_to_nid // 3. *but still use cached pfn,nid*
mm_core_init
mem_init
memblock_free_all
__free_pages_core // 4. check page and find bad page
Firstly, 9f000000 on nid=4 should be fixed in firmware(I don't know where
store this infomation), secondly, if we want to fix it or avoid
I believe the firmware for QEMU is just OpenSBI but that is about all I
know, I am not a RISC-V developer. >
I've explicitly added some RISC-V folks, the start of the thread is
available at
https://lore.kernel.org/20231025163703.GA2440148@dev-arch.thelio-3990X/.
Thanks，phy=9f000000 on nid=4 is strange, it could be fixed in firmware,
but in any case，I think the following change should be needed, as once
one use early_pfn_to_nid(), it will cache pfn-to-nid, if the pfn-to-nid
is updated, eg, dummy_numa_init() or other numa_add_memblk() callers,
the cached pfn-to-nid should be reset, or new early_pfn_to_nid() will
get wrong old numa nid.
Cheers,
Nathan
similar issue happened in other scene,a reset function to cleanup the
cached pfn-nid should be added, I try following diff, it should work.
diff --git a/drivers/base/arch_numa.c b/drivers/base/arch_numa.c
index eaa31e567d1e..24100e45971c 100644
--- a/drivers/base/arch_numa.c
+++ b/drivers/base/arch_numa.c
@@ -210,6 +210,7 @@ int __init numa_add_memblk(int nid, u64 start, u64 end)
}
node_set(nid, numa_nodes_parsed);
+ early_pfn_reset_nid();
return ret;
}
diff --git a/include/linux/mm.h b/include/linux/mm.h
index 418d26608ece..f20a8da22b35 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -3173,9 +3173,11 @@ static inline int early_pfn_to_nid(unsigned long pfn)
{
return 0;
}
+static inline void early_pfn_reset_nid(void) {}
#else
/* please see mm/page_alloc.c */
extern int __meminit early_pfn_to_nid(unsigned long pfn);
+extern void __meminit early_pfn_reset_nid(void);
#endif
extern void set_dma_reserve(unsigned long new_dma_reserve);
diff --git a/mm/mm_init.c b/mm/mm_init.c
index 077bfe393b5e..fb7751b233c4 100644
--- a/mm/mm_init.c
+++ b/mm/mm_init.c
@@ -586,6 +586,7 @@ struct mminit_pfnnid_cache {
};
static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
+static DEFINE_SPINLOCK(early_pfn_lock);
/*
* Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
@@ -611,7 +612,6 @@ static int __meminit __early_pfn_to_nid(unsigned long
pfn,
int __meminit early_pfn_to_nid(unsigned long pfn)
{
- static DEFINE_SPINLOCK(early_pfn_lock);
int nid;
spin_lock(&early_pfn_lock);
@@ -623,6 +623,15 @@ int __meminit early_pfn_to_nid(unsigned long pfn)
return nid;
}
+void __meminit early_pfn_reset_nid(void)
+{
+ spin_lock(&early_pfn_lock);
+ early_pfnnid_cache.last_start = 0;
+ early_pfnnid_cache.last_end = 0;
+ early_pfnnid_cache.last_nid = 0;
+ spin_unlock(&early_pfn_lock);
+}
+
int hashdist = HASHDIST_DEFAULT;
static int __init set_hashdist(char *str)
<hangs after OpenSBI output>
Without CONFIG_ACPI_SPCR_TABLE=y, there is a visible crash.
[ 0.000000] Linux version 6.6.0-rc7-next-20231025 (nathan@dev-fedora.c3-large-arm64) (riscv64-linux-gcc (GCC) 13.2.0, GNU ld (GNU Binutils) 2.41) #1 SMP Wed Oct 25 16:14:59 UTC 2023
...
[ 0.000000] mem auto-init: stack:all(zero), heap alloc:off, heap free:off
[ 0.000000] page:ff1c000002200000 is uninitialized and poisoned
[ 0.000000] page dumped because: VM_BUG_ON_PAGE(1 && PageCompound(page))
[ 0.000000] ------------[ cut here ]------------
[ 0.000000] kernel BUG at include/linux/page-flags.h:493!
[ 0.000000] Kernel BUG [#1]
[ 0.000000] Modules linked in:
[ 0.000000] CPU: 0 PID: 0 Comm: swapper Not tainted 6.6.0-rc7-next-20231025 #1
[ 0.000000] Hardware name: riscv-virtio,qemu (DT)
[ 0.000000] epc : __free_pages_core+0x78/0x126
[ 0.000000] ra : __free_pages_core+0x78/0x126
[ 0.000000] epc : ffffffff8018dd8e ra : ffffffff8018dd8e sp : ffffffff81403d40
[ 0.000000] gp : ffffffff815013a0 tp : ffffffff8140db00 t0 : 6d75642065676170
[ 0.000000] t1 : 0000000000000070 t2 : 706d756420656761 s0 : ffffffff81403d50
[ 0.000000] s1 : 0000000000000004 a0 : 000000000000003c a1 : ffffffff814866a8
[ 0.000000] a2 : 0000000000000000 a3 : 0000000000000001 a4 : 0000000000000000
[ 0.000000] a5 : 0000000000000000 a6 : 0000000000000008 a7 : 0000000000000038
[ 0.000000] s2 : 0000000000088000 s3 : ff1c000002200000 s4 : 0000000000000009
[ 0.000000] s5 : 00000000ffffffff s6 : 0000000000081800 s7 : 0000000000088200
[ 0.000000] s8 : 00000000000001c0 s9 : 0040000000000000 s10: ffffffff81500bdd
[ 0.000000] s11: ffffffff81500bdc t3 : ffffffff81515aa7 t4 : ffffffff81515aa7
[ 0.000000] t5 : ffffffff81515aa8 t6 : ffffffff81403b58
[ 0.000000] status: 0000000200000100 badaddr: 0000000000000000 cause: 0000000000000003
[ 0.000000] [<ffffffff8018dd8e>] __free_pages_core+0x78/0x126
[ 0.000000] [<ffffffff80a12fee>] memblock_free_pages+0x52/0x62
[ 0.000000] [<ffffffff80a15f02>] memblock_free_all+0x1fc/0x27e
[ 0.000000] [<ffffffff80a061fa>] mem_init+0x34/0x22c
[ 0.000000] [<ffffffff80a13114>] mm_core_init+0x116/0x2d0
[ 0.000000] [<ffffffff80a00a6e>] start_kernel+0x3c6/0x742
[ 0.000000] Code: 0405 8399 8b85 d7f1 9597 00e2 8593 2ae5 90ef e5dd (9002) 6597
[ 0.000000] ---[ end trace 0000000000000000 ]---
[ 0.000000] Kernel panic - not syncing: Fatal exception in interrupt
The rootfs is available at [2] if necessary. If there is any more
information I can provide or patches I can test, I am more than happy to
do so.
[1]: https://github.com/openSUSE/kernel-source/raw/master/config/riscv64/default
[2]: https://github.com/ClangBuiltLinux/boot-utils/releases
Cheers,
Nathan
