Reply-To:
Hi Mike!
Since commit fa3354e4ea39 (mm: free_area_init: use maximal zone PFNs rather
than zone sizes), I get the following BUG on Alpha (an AlphaServer ES47 Marvel)
and loading userspace leads to a segfault:
(I didn't notice this for a long time because of other unrelated regressions,
the pandemic, changing jobs, ...)
BUG: Bad page state in process swapper pfn:2ffc53
page:fffffc000ecf14c0 refcount:0 mapcount:1 mapping:0000000000000000 index:0x0
flags: 0x0()
raw: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
raw: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
page dumped because: nonzero mapcount
Modules linked in:
CPU: 0 PID: 0 Comm: swapper Not tainted 5.7.0-03841-gfa3354e4ea39-dirty #26
fffffc0001b5bd68 fffffc0001b5be80 fffffc00011cd148 fffffc000ecf14c0
fffffc00019803df fffffc0001b5be80 fffffc00011ce340 fffffc000ecf14c0
0000000000000000 fffffc0001b5be80 fffffc0001b482c0 fffffc00027d6618
fffffc00027da7d0 00000000002ff97a 0000000000000000 fffffc0001b5be80
fffffc00011d1abc fffffc000ecf14c0 fffffc0002d00000 fffffc0001b5be80
fffffc0001b2350c 0000000000300000 fffffc0001b48298 fffffc0001b482c0
Trace:
[<fffffc00011cd148>] bad_page+0x168/0x1b0
[<fffffc00011ce340>] free_pcp_prepare+0x1e0/0x290
[<fffffc00011d1abc>] free_unref_page+0x2c/0xa0
[<fffffc00014ee5f0>] cmp_ex_sort+0x0/0x30
[<fffffc00014ee5f0>] cmp_ex_sort+0x0/0x30
[<fffffc000101001c>] _stext+0x1c/0x20
I haven't tried reproducing this on other machines or QEMU, but I'd be glad to
if that helps.
Any ideas?
Thanks,
Matt
Hi Matt,
On Mon, Jul 26, 2021 at 12:27:50PM -0700, Matt Turner wrote:
> Reply-To:
>
> Hi Mike!
>
> Since commit fa3354e4ea39 (mm: free_area_init: use maximal zone PFNs rather
> than zone sizes), I get the following BUG on Alpha (an AlphaServer ES47 Marvel)
> and loading userspace leads to a segfault:
>
> (I didn't notice this for a long time because of other unrelated regressions,
> the pandemic, changing jobs, ...)
I suspect there will be more surprises down the road :)
> BUG: Bad page state in process swapper pfn:2ffc53
> page:fffffc000ecf14c0 refcount:0 mapcount:1 mapping:0000000000000000 index:0x0
> flags: 0x0()
> raw: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> raw: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> page dumped because: nonzero mapcount Modules linked in:
> CPU: 0 PID: 0 Comm: swapper Not tainted 5.7.0-03841-gfa3354e4ea39-dirty #26
> fffffc0001b5bd68 fffffc0001b5be80 fffffc00011cd148 fffffc000ecf14c0
> fffffc00019803df fffffc0001b5be80 fffffc00011ce340 fffffc000ecf14c0
> 0000000000000000 fffffc0001b5be80 fffffc0001b482c0 fffffc00027d6618
> fffffc00027da7d0 00000000002ff97a 0000000000000000 fffffc0001b5be80
> fffffc00011d1abc fffffc000ecf14c0 fffffc0002d00000 fffffc0001b5be80
> fffffc0001b2350c 0000000000300000 fffffc0001b48298 fffffc0001b482c0
> Trace:
> [<fffffc00011cd148>] bad_page+0x168/0x1b0
> [<fffffc00011ce340>] free_pcp_prepare+0x1e0/0x290
> [<fffffc00011d1abc>] free_unref_page+0x2c/0xa0
> [<fffffc00014ee5f0>] cmp_ex_sort+0x0/0x30
> [<fffffc00014ee5f0>] cmp_ex_sort+0x0/0x30
> [<fffffc000101001c>] _stext+0x1c/0x20
>
> I haven't tried reproducing this on other machines or QEMU, but I'd be glad to
> if that helps.
If it's reproducible on QEMU I can debug it locally.
> Any ideas?
It seems like memory map is not properly initialized. Can you enable
CONFIG_DEBUG_MEMORY_INIT and add mminit_debug=4 to the command line. The
interesting part of the log would be before "Memory: xK/yK available ..."
line.
Hopefully it'll give some clues.
--
Sincerely yours,
Mike.
On Mon, Jul 26, 2021 at 1:06 PM Mike Rapoport <[email protected]> wrote:
>
> Hi Matt,
>
> On Mon, Jul 26, 2021 at 12:27:50PM -0700, Matt Turner wrote:
> > Reply-To:
> >
> > Hi Mike!
> >
> > Since commit fa3354e4ea39 (mm: free_area_init: use maximal zone PFNs rather
> > than zone sizes), I get the following BUG on Alpha (an AlphaServer ES47 Marvel)
> > and loading userspace leads to a segfault:
> >
> > (I didn't notice this for a long time because of other unrelated regressions,
> > the pandemic, changing jobs, ...)
>
> I suspect there will be more surprises down the road :)
>
> > BUG: Bad page state in process swapper pfn:2ffc53
> > page:fffffc000ecf14c0 refcount:0 mapcount:1 mapping:0000000000000000 index:0x0
> > flags: 0x0()
> > raw: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> > raw: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> > page dumped because: nonzero mapcount Modules linked in:
> > CPU: 0 PID: 0 Comm: swapper Not tainted 5.7.0-03841-gfa3354e4ea39-dirty #26
> > fffffc0001b5bd68 fffffc0001b5be80 fffffc00011cd148 fffffc000ecf14c0
> > fffffc00019803df fffffc0001b5be80 fffffc00011ce340 fffffc000ecf14c0
> > 0000000000000000 fffffc0001b5be80 fffffc0001b482c0 fffffc00027d6618
> > fffffc00027da7d0 00000000002ff97a 0000000000000000 fffffc0001b5be80
> > fffffc00011d1abc fffffc000ecf14c0 fffffc0002d00000 fffffc0001b5be80
> > fffffc0001b2350c 0000000000300000 fffffc0001b48298 fffffc0001b482c0
> > Trace:
> > [<fffffc00011cd148>] bad_page+0x168/0x1b0
> > [<fffffc00011ce340>] free_pcp_prepare+0x1e0/0x290
> > [<fffffc00011d1abc>] free_unref_page+0x2c/0xa0
> > [<fffffc00014ee5f0>] cmp_ex_sort+0x0/0x30
> > [<fffffc00014ee5f0>] cmp_ex_sort+0x0/0x30
> > [<fffffc000101001c>] _stext+0x1c/0x20
> >
> > I haven't tried reproducing this on other machines or QEMU, but I'd be glad to
> > if that helps.
>
> If it's reproducible on QEMU I can debug it locally.
>
> > Any ideas?
>
> It seems like memory map is not properly initialized. Can you enable
> CONFIG_DEBUG_MEMORY_INIT and add mminit_debug=4 to the command line. The
> interesting part of the log would be before "Memory: xK/yK available ..."
> line.
>
> Hopefully it'll give some clues.
Sure thing. Please find attached.
On Mon, Jul 26, 2021 at 02:23:20PM -0700, Matt Turner wrote:
> On Mon, Jul 26, 2021 at 1:06 PM Mike Rapoport <[email protected]> wrote:
> >
> > Hi Matt,
> >
> > On Mon, Jul 26, 2021 at 12:27:50PM -0700, Matt Turner wrote:
> > > Reply-To:
> > >
> > > Hi Mike!
> > >
> > > Since commit fa3354e4ea39 (mm: free_area_init: use maximal zone PFNs rather
> > > than zone sizes), I get the following BUG on Alpha (an AlphaServer ES47 Marvel)
> > > and loading userspace leads to a segfault:
> > >
> > > (I didn't notice this for a long time because of other unrelated regressions,
> > > the pandemic, changing jobs, ...)
> >
> > I suspect there will be more surprises down the road :)
> >
> > > BUG: Bad page state in process swapper pfn:2ffc53
> > > page:fffffc000ecf14c0 refcount:0 mapcount:1 mapping:0000000000000000 index:0x0
> > > flags: 0x0()
> > > raw: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> > > raw: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> > > page dumped because: nonzero mapcount Modules linked in:
> > > CPU: 0 PID: 0 Comm: swapper Not tainted 5.7.0-03841-gfa3354e4ea39-dirty #26
> > > fffffc0001b5bd68 fffffc0001b5be80 fffffc00011cd148 fffffc000ecf14c0
> > > fffffc00019803df fffffc0001b5be80 fffffc00011ce340 fffffc000ecf14c0
> > > 0000000000000000 fffffc0001b5be80 fffffc0001b482c0 fffffc00027d6618
> > > fffffc00027da7d0 00000000002ff97a 0000000000000000 fffffc0001b5be80
> > > fffffc00011d1abc fffffc000ecf14c0 fffffc0002d00000 fffffc0001b5be80
> > > fffffc0001b2350c 0000000000300000 fffffc0001b48298 fffffc0001b482c0
> > > Trace:
> > > [<fffffc00011cd148>] bad_page+0x168/0x1b0
> > > [<fffffc00011ce340>] free_pcp_prepare+0x1e0/0x290
> > > [<fffffc00011d1abc>] free_unref_page+0x2c/0xa0
> > > [<fffffc00014ee5f0>] cmp_ex_sort+0x0/0x30
> > > [<fffffc00014ee5f0>] cmp_ex_sort+0x0/0x30
> > > [<fffffc000101001c>] _stext+0x1c/0x20
> > >
> > > I haven't tried reproducing this on other machines or QEMU, but I'd be glad to
> > > if that helps.
> >
> > If it's reproducible on QEMU I can debug it locally.
> >
> > > Any ideas?
> >
> > It seems like memory map is not properly initialized. Can you enable
> > CONFIG_DEBUG_MEMORY_INIT and add mminit_debug=4 to the command line. The
> > interesting part of the log would be before "Memory: xK/yK available ..."
> > line.
> >
> > Hopefully it'll give some clues.
>
> Sure thing. Please find attached.
> aboot: loading uncompressed vmlinuz-5.7.0-03841-gfa3354e4ea39-dirty...
> aboot: loading compressed vmlinuz-5.7.0-03841-gfa3354e4ea39-dirty...
> aboot: PHDR 0 vaddr 0xfffffc0001010000 offset 0xc0 size 0x17c5ae0
> aboot: bss at 0xfffffc00027d5ae0, size 0xe4ea0
> aboot: zero-filling 937632 bytes at 0xfffffc00027d5ae0
> aboot: loading initrd (5965252 bytes/5825 blocks) at 0xfffffc05ff2cc000
> aboot: starting kernel vmlinuz-5.7.0-03841-gfa3354e4ea39-dirty with arguments ro panic=5 domdadm root=/dev/md1 console=srm mminit_debug=4
> Linux version 5.7.0-03841-gfa3354e4ea39-dirty (mattst88@ivybridge) (gcc version 11.1.0 (Gentoo 11.1.0-r2 p3), GNU ld (Gentoo 2.36.1 p3) 2.36.1) #26 SMP Sun Jul 25 18:20:06 PDT 2021
> printk: bootconsole [srm0] enabled
> Booting on Marvel variation Marvel/EV7 using machine vector MARVEL/EV7 from SRM
> Major Options: SMP EV67 VERBOSE_MCHECK DEBUG_SPINLOCK MAGIC_SYSRQ
> Command line: ro panic=5 domdadm root=/dev/md1 console=srm mminit_debug=4
> memcluster 0, usage 1, start 0, end 1984
> memcluster 1, usage 0, start 1984, end 1048576
> memcluster 2, usage 1, start 2097152, end 2097224
> memcluster 3, usage 0, start 2097224, end 3145728
> Initial ramdisk at: 0x(____ptrval____) (5965252 bytes)
> Found an IO7 at PID 0
> Initializing IO7 at PID 0
> FIXME: disabling master aborts
> FIXME: disabling master aborts
> FIXME: disabling master aborts
> FIXME: disabling master aborts
> SMP: 2 CPUs probed -- cpu_present_mask = 3
> Zone ranges:
> DMA [mem 0x0000000000f80000-0x00000fffffffdfff]
> Normal empty
> Movable zone start for each node
> Early memory node ranges
> node 0: [mem 0x0000000000f80000-0x00000001ffffffff]
> node 0: [mem 0x0000000400090000-0x00000005ffffffff]
I think that the issue is that memory marked as used in memcluster is never
added to memblock and it skews node/zone sizing calculations.
Can you try this patch:
diff --git a/arch/alpha/kernel/setup.c b/arch/alpha/kernel/setup.c
index 7d56c217b235..b4fbbba30aa2 100644
--- a/arch/alpha/kernel/setup.c
+++ b/arch/alpha/kernel/setup.c
@@ -319,18 +319,19 @@ setup_memory(void *kernel_end)
i, cluster->usage, cluster->start_pfn,
cluster->start_pfn + cluster->numpages);
- /* Bit 0 is console/PALcode reserved. Bit 1 is
- non-volatile memory -- we might want to mark
- this for later. */
- if (cluster->usage & 3)
- continue;
-
end = cluster->start_pfn + cluster->numpages;
if (end > max_low_pfn)
max_low_pfn = end;
memblock_add(PFN_PHYS(cluster->start_pfn),
cluster->numpages << PAGE_SHIFT);
+
+ /* Bit 0 is console/PALcode reserved. Bit 1 is
+ non-volatile memory -- we might want to mark
+ this for later. */
+ if (cluster->usage & 3)
+ memblock_reserve(PFN_PHYS(cluster->start_pfn),
+ cluster->numpages << PAGE_SHIFT);
}
/*
> Initmem setup node 0 [mem 0x0000000000f80000-0x00000005ffffffff]
> percpu: Embedded 8 pages/cpu s27648 r8192 d29696 u65536
> Built 1 zonelists, mobility grouping on. Total pages: 2070535
> Kernel command line: ro panic=5 domdadm root=/dev/md1 console=srm mminit_debug=4
> Dentry cache hash table entries: 2097152 (order: 11, 16777216 bytes, linear)
> Inode-cache hash table entries: 1048576 (order: 10, 8388608 bytes, linear)
> Sorting __ex_table...
> mem auto-init: stack:off, heap alloc:off, heap free:off
> BUG: Bad page state in process swapper pfn:2ffc3f
> page:fffffc000ecf0fc0 refcount:0 mapcount:1 mapping:0000000000000000 index:0x0
> flags: 0x0()
> raw: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> raw: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> page dumped because: nonzero mapcount
> Modules linked in:
> CPU: 0 PID: 0 Comm: swapper Not tainted 5.7.0-03841-gfa3354e4ea39-dirty #26
> fffffc0001b5bd68 fffffc0001b5be80 fffffc00011cd148 fffffc000ecf0fc0
> fffffc00019803df fffffc0001b5be80 fffffc00011ce340 fffffc000ecf0fc0
> 0000000000000000 fffffc0001b5be80 fffffc0001b482c0 fffffc00027d6618
> fffffc00027da7d0 00000000002ff966 0000000000000000 fffffc0001b5be80
> fffffc00011d1abc fffffc000ecf0fc0 fffffc0002d00000 fffffc0001b5be80
> fffffc0001b2350c 0000000000300000 fffffc0001b48298 fffffc0001b482c0
> Trace:
> [<fffffc00011cd148>] bad_page+0x168/0x1b0
> [<fffffc00011ce340>] free_pcp_prepare+0x1e0/0x290
> [<fffffc00011d1abc>] free_unref_page+0x2c/0xa0
> [<fffffc00014ee5f0>] cmp_ex_sort+0x0/0x30
> [<fffffc00014ee5f0>] cmp_ex_sort+0x0/0x30
> [<fffffc000101001c>] _stext+0x1c/0x20
...
> Memory: 16496504K/16760768K available (8698K kernel code, 12790K rwdata, 2544K rodata, 304K init, 915K bss, 256576K reserved, 0K cma-reserved)
--
Sincerely yours,
Mike.
On Mon, Jul 26, 2021 at 11:43 PM Mike Rapoport <[email protected]> wrote:
>
> On Mon, Jul 26, 2021 at 02:23:20PM -0700, Matt Turner wrote:
> > On Mon, Jul 26, 2021 at 1:06 PM Mike Rapoport <[email protected]> wrote:
> > >
> > > Hi Matt,
> > >
> > > On Mon, Jul 26, 2021 at 12:27:50PM -0700, Matt Turner wrote:
> > > > Reply-To:
> > > >
> > > > Hi Mike!
> > > >
> > > > Since commit fa3354e4ea39 (mm: free_area_init: use maximal zone PFNs rather
> > > > than zone sizes), I get the following BUG on Alpha (an AlphaServer ES47 Marvel)
> > > > and loading userspace leads to a segfault:
> > > >
> > > > (I didn't notice this for a long time because of other unrelated regressions,
> > > > the pandemic, changing jobs, ...)
> > >
> > > I suspect there will be more surprises down the road :)
> > >
> > > > BUG: Bad page state in process swapper pfn:2ffc53
> > > > page:fffffc000ecf14c0 refcount:0 mapcount:1 mapping:0000000000000000 index:0x0
> > > > flags: 0x0()
> > > > raw: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> > > > raw: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> > > > page dumped because: nonzero mapcount Modules linked in:
> > > > CPU: 0 PID: 0 Comm: swapper Not tainted 5.7.0-03841-gfa3354e4ea39-dirty #26
> > > > fffffc0001b5bd68 fffffc0001b5be80 fffffc00011cd148 fffffc000ecf14c0
> > > > fffffc00019803df fffffc0001b5be80 fffffc00011ce340 fffffc000ecf14c0
> > > > 0000000000000000 fffffc0001b5be80 fffffc0001b482c0 fffffc00027d6618
> > > > fffffc00027da7d0 00000000002ff97a 0000000000000000 fffffc0001b5be80
> > > > fffffc00011d1abc fffffc000ecf14c0 fffffc0002d00000 fffffc0001b5be80
> > > > fffffc0001b2350c 0000000000300000 fffffc0001b48298 fffffc0001b482c0
> > > > Trace:
> > > > [<fffffc00011cd148>] bad_page+0x168/0x1b0
> > > > [<fffffc00011ce340>] free_pcp_prepare+0x1e0/0x290
> > > > [<fffffc00011d1abc>] free_unref_page+0x2c/0xa0
> > > > [<fffffc00014ee5f0>] cmp_ex_sort+0x0/0x30
> > > > [<fffffc00014ee5f0>] cmp_ex_sort+0x0/0x30
> > > > [<fffffc000101001c>] _stext+0x1c/0x20
> > > >
> > > > I haven't tried reproducing this on other machines or QEMU, but I'd be glad to
> > > > if that helps.
> > >
> > > If it's reproducible on QEMU I can debug it locally.
> > >
> > > > Any ideas?
> > >
> > > It seems like memory map is not properly initialized. Can you enable
> > > CONFIG_DEBUG_MEMORY_INIT and add mminit_debug=4 to the command line. The
> > > interesting part of the log would be before "Memory: xK/yK available ..."
> > > line.
> > >
> > > Hopefully it'll give some clues.
> >
> > Sure thing. Please find attached.
>
> > aboot: loading uncompressed vmlinuz-5.7.0-03841-gfa3354e4ea39-dirty...
> > aboot: loading compressed vmlinuz-5.7.0-03841-gfa3354e4ea39-dirty...
> > aboot: PHDR 0 vaddr 0xfffffc0001010000 offset 0xc0 size 0x17c5ae0
> > aboot: bss at 0xfffffc00027d5ae0, size 0xe4ea0
> > aboot: zero-filling 937632 bytes at 0xfffffc00027d5ae0
> > aboot: loading initrd (5965252 bytes/5825 blocks) at 0xfffffc05ff2cc000
> > aboot: starting kernel vmlinuz-5.7.0-03841-gfa3354e4ea39-dirty with arguments ro panic=5 domdadm root=/dev/md1 console=srm mminit_debug=4
> > Linux version 5.7.0-03841-gfa3354e4ea39-dirty (mattst88@ivybridge) (gcc version 11.1.0 (Gentoo 11.1.0-r2 p3), GNU ld (Gentoo 2.36.1 p3) 2.36.1) #26 SMP Sun Jul 25 18:20:06 PDT 2021
> > printk: bootconsole [srm0] enabled
> > Booting on Marvel variation Marvel/EV7 using machine vector MARVEL/EV7 from SRM
> > Major Options: SMP EV67 VERBOSE_MCHECK DEBUG_SPINLOCK MAGIC_SYSRQ
> > Command line: ro panic=5 domdadm root=/dev/md1 console=srm mminit_debug=4
> > memcluster 0, usage 1, start 0, end 1984
> > memcluster 1, usage 0, start 1984, end 1048576
> > memcluster 2, usage 1, start 2097152, end 2097224
> > memcluster 3, usage 0, start 2097224, end 3145728
> > Initial ramdisk at: 0x(____ptrval____) (5965252 bytes)
> > Found an IO7 at PID 0
> > Initializing IO7 at PID 0
> > FIXME: disabling master aborts
> > FIXME: disabling master aborts
> > FIXME: disabling master aborts
> > FIXME: disabling master aborts
> > SMP: 2 CPUs probed -- cpu_present_mask = 3
> > Zone ranges:
> > DMA [mem 0x0000000000f80000-0x00000fffffffdfff]
> > Normal empty
> > Movable zone start for each node
> > Early memory node ranges
> > node 0: [mem 0x0000000000f80000-0x00000001ffffffff]
> > node 0: [mem 0x0000000400090000-0x00000005ffffffff]
>
> I think that the issue is that memory marked as used in memcluster is never
> added to memblock and it skews node/zone sizing calculations.
Thanks, this patch fixes it. With the patch applied, I see
Zone ranges:
DMA [mem 0x0000000000000000-0x00000fffffffdfff]
Normal empty
Movable zone start for each node
Early memory node ranges
node 0: [mem 0x0000000000000000-0x00000001ffffffff]
node 0: [mem 0x0000000400000000-0x00000005ffffffff]
Initmem setup node 0 [mem 0x0000000000000000-0x00000005ffffffff]
If you want to send me this patch with your S-o-b I'll take it through
my alpha git tree.
Thanks Mike!
Matt
On Tue, Jul 27, 2021 at 12:24:26PM -0700, Matt Turner wrote:
> On Mon, Jul 26, 2021 at 11:43 PM Mike Rapoport <[email protected]> wrote:
> >
> > On Mon, Jul 26, 2021 at 02:23:20PM -0700, Matt Turner wrote:
> > > On Mon, Jul 26, 2021 at 1:06 PM Mike Rapoport <[email protected]> wrote:
> > > >
> > > > Hi Matt,
> > > >
> > > > On Mon, Jul 26, 2021 at 12:27:50PM -0700, Matt Turner wrote:
> > > > > Reply-To:
> > > > >
> > > > > Hi Mike!
> > > > >
> > > > > Since commit fa3354e4ea39 (mm: free_area_init: use maximal zone PFNs rather
> > > > > than zone sizes), I get the following BUG on Alpha (an AlphaServer ES47 Marvel)
> > > > > and loading userspace leads to a segfault:
> > > > >
> > > > > (I didn't notice this for a long time because of other unrelated regressions,
> > > > > the pandemic, changing jobs, ...)
> > > >
> > > > I suspect there will be more surprises down the road :)
> > > >
> > > > > BUG: Bad page state in process swapper pfn:2ffc53
> > > > > page:fffffc000ecf14c0 refcount:0 mapcount:1 mapping:0000000000000000 index:0x0
> > > > > flags: 0x0()
> > > > > raw: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> > > > > raw: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> > > > > page dumped because: nonzero mapcount Modules linked in:
> > > > > CPU: 0 PID: 0 Comm: swapper Not tainted 5.7.0-03841-gfa3354e4ea39-dirty #26
> > > > > fffffc0001b5bd68 fffffc0001b5be80 fffffc00011cd148 fffffc000ecf14c0
> > > > > fffffc00019803df fffffc0001b5be80 fffffc00011ce340 fffffc000ecf14c0
> > > > > 0000000000000000 fffffc0001b5be80 fffffc0001b482c0 fffffc00027d6618
> > > > > fffffc00027da7d0 00000000002ff97a 0000000000000000 fffffc0001b5be80
> > > > > fffffc00011d1abc fffffc000ecf14c0 fffffc0002d00000 fffffc0001b5be80
> > > > > fffffc0001b2350c 0000000000300000 fffffc0001b48298 fffffc0001b482c0
> > > > > Trace:
> > > > > [<fffffc00011cd148>] bad_page+0x168/0x1b0
> > > > > [<fffffc00011ce340>] free_pcp_prepare+0x1e0/0x290
> > > > > [<fffffc00011d1abc>] free_unref_page+0x2c/0xa0
> > > > > [<fffffc00014ee5f0>] cmp_ex_sort+0x0/0x30
> > > > > [<fffffc00014ee5f0>] cmp_ex_sort+0x0/0x30
> > > > > [<fffffc000101001c>] _stext+0x1c/0x20
> > > > >
> > > > > I haven't tried reproducing this on other machines or QEMU, but I'd be glad to
> > > > > if that helps.
> > > >
> > > > If it's reproducible on QEMU I can debug it locally.
> > > >
> > > > > Any ideas?
> > > >
> > > > It seems like memory map is not properly initialized. Can you enable
> > > > CONFIG_DEBUG_MEMORY_INIT and add mminit_debug=4 to the command line. The
> > > > interesting part of the log would be before "Memory: xK/yK available ..."
> > > > line.
> > > >
> > > > Hopefully it'll give some clues.
> > >
> > > Sure thing. Please find attached.
> >
> > > aboot: loading uncompressed vmlinuz-5.7.0-03841-gfa3354e4ea39-dirty...
> > > aboot: loading compressed vmlinuz-5.7.0-03841-gfa3354e4ea39-dirty...
> > > aboot: PHDR 0 vaddr 0xfffffc0001010000 offset 0xc0 size 0x17c5ae0
> > > aboot: bss at 0xfffffc00027d5ae0, size 0xe4ea0
> > > aboot: zero-filling 937632 bytes at 0xfffffc00027d5ae0
> > > aboot: loading initrd (5965252 bytes/5825 blocks) at 0xfffffc05ff2cc000
> > > aboot: starting kernel vmlinuz-5.7.0-03841-gfa3354e4ea39-dirty with arguments ro panic=5 domdadm root=/dev/md1 console=srm mminit_debug=4
> > > Linux version 5.7.0-03841-gfa3354e4ea39-dirty (mattst88@ivybridge) (gcc version 11.1.0 (Gentoo 11.1.0-r2 p3), GNU ld (Gentoo 2.36.1 p3) 2.36.1) #26 SMP Sun Jul 25 18:20:06 PDT 2021
> > > printk: bootconsole [srm0] enabled
> > > Booting on Marvel variation Marvel/EV7 using machine vector MARVEL/EV7 from SRM
> > > Major Options: SMP EV67 VERBOSE_MCHECK DEBUG_SPINLOCK MAGIC_SYSRQ
> > > Command line: ro panic=5 domdadm root=/dev/md1 console=srm mminit_debug=4
> > > memcluster 0, usage 1, start 0, end 1984
> > > memcluster 1, usage 0, start 1984, end 1048576
> > > memcluster 2, usage 1, start 2097152, end 2097224
> > > memcluster 3, usage 0, start 2097224, end 3145728
> > > Initial ramdisk at: 0x(____ptrval____) (5965252 bytes)
> > > Found an IO7 at PID 0
> > > Initializing IO7 at PID 0
> > > FIXME: disabling master aborts
> > > FIXME: disabling master aborts
> > > FIXME: disabling master aborts
> > > FIXME: disabling master aborts
> > > SMP: 2 CPUs probed -- cpu_present_mask = 3
> > > Zone ranges:
> > > DMA [mem 0x0000000000f80000-0x00000fffffffdfff]
> > > Normal empty
> > > Movable zone start for each node
> > > Early memory node ranges
> > > node 0: [mem 0x0000000000f80000-0x00000001ffffffff]
> > > node 0: [mem 0x0000000400090000-0x00000005ffffffff]
> >
> > I think that the issue is that memory marked as used in memcluster is never
> > added to memblock and it skews node/zone sizing calculations.
>
> Thanks, this patch fixes it. With the patch applied, I see
>
> Zone ranges:
> DMA [mem 0x0000000000000000-0x00000fffffffdfff]
> Normal empty
> Movable zone start for each node
> Early memory node ranges
> node 0: [mem 0x0000000000000000-0x00000001ffffffff]
> node 0: [mem 0x0000000400000000-0x00000005ffffffff]
> Initmem setup node 0 [mem 0x0000000000000000-0x00000005ffffffff]
>
> If you want to send me this patch with your S-o-b I'll take it through
> my alpha git tree.
The patch is on its way :)
Now I'm really curios how commit e7793e53901b ("arc: update comment about
HIGHMEM implementation") will work out.
> Thanks Mike!
> Matt
--
Sincerely yours,
Mike.
On Tue, Jul 27, 2021 at 1:41 PM Mike Rapoport <[email protected]> wrote:
>
> On Tue, Jul 27, 2021 at 12:24:26PM -0700, Matt Turner wrote:
> > On Mon, Jul 26, 2021 at 11:43 PM Mike Rapoport <[email protected]> wrote:
> > >
> > > On Mon, Jul 26, 2021 at 02:23:20PM -0700, Matt Turner wrote:
> > > > On Mon, Jul 26, 2021 at 1:06 PM Mike Rapoport <[email protected]> wrote:
> > > > >
> > > > > Hi Matt,
> > > > >
> > > > > On Mon, Jul 26, 2021 at 12:27:50PM -0700, Matt Turner wrote:
> > > > > > Reply-To:
> > > > > >
> > > > > > Hi Mike!
> > > > > >
> > > > > > Since commit fa3354e4ea39 (mm: free_area_init: use maximal zone PFNs rather
> > > > > > than zone sizes), I get the following BUG on Alpha (an AlphaServer ES47 Marvel)
> > > > > > and loading userspace leads to a segfault:
> > > > > >
> > > > > > (I didn't notice this for a long time because of other unrelated regressions,
> > > > > > the pandemic, changing jobs, ...)
> > > > >
> > > > > I suspect there will be more surprises down the road :)
> > > > >
> > > > > > BUG: Bad page state in process swapper pfn:2ffc53
> > > > > > page:fffffc000ecf14c0 refcount:0 mapcount:1 mapping:0000000000000000 index:0x0
> > > > > > flags: 0x0()
> > > > > > raw: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> > > > > > raw: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> > > > > > page dumped because: nonzero mapcount Modules linked in:
> > > > > > CPU: 0 PID: 0 Comm: swapper Not tainted 5.7.0-03841-gfa3354e4ea39-dirty #26
> > > > > > fffffc0001b5bd68 fffffc0001b5be80 fffffc00011cd148 fffffc000ecf14c0
> > > > > > fffffc00019803df fffffc0001b5be80 fffffc00011ce340 fffffc000ecf14c0
> > > > > > 0000000000000000 fffffc0001b5be80 fffffc0001b482c0 fffffc00027d6618
> > > > > > fffffc00027da7d0 00000000002ff97a 0000000000000000 fffffc0001b5be80
> > > > > > fffffc00011d1abc fffffc000ecf14c0 fffffc0002d00000 fffffc0001b5be80
> > > > > > fffffc0001b2350c 0000000000300000 fffffc0001b48298 fffffc0001b482c0
> > > > > > Trace:
> > > > > > [<fffffc00011cd148>] bad_page+0x168/0x1b0
> > > > > > [<fffffc00011ce340>] free_pcp_prepare+0x1e0/0x290
> > > > > > [<fffffc00011d1abc>] free_unref_page+0x2c/0xa0
> > > > > > [<fffffc00014ee5f0>] cmp_ex_sort+0x0/0x30
> > > > > > [<fffffc00014ee5f0>] cmp_ex_sort+0x0/0x30
> > > > > > [<fffffc000101001c>] _stext+0x1c/0x20
> > > > > >
> > > > > > I haven't tried reproducing this on other machines or QEMU, but I'd be glad to
> > > > > > if that helps.
> > > > >
> > > > > If it's reproducible on QEMU I can debug it locally.
> > > > >
> > > > > > Any ideas?
> > > > >
> > > > > It seems like memory map is not properly initialized. Can you enable
> > > > > CONFIG_DEBUG_MEMORY_INIT and add mminit_debug=4 to the command line. The
> > > > > interesting part of the log would be before "Memory: xK/yK available ..."
> > > > > line.
> > > > >
> > > > > Hopefully it'll give some clues.
> > > >
> > > > Sure thing. Please find attached.
> > >
> > > > aboot: loading uncompressed vmlinuz-5.7.0-03841-gfa3354e4ea39-dirty...
> > > > aboot: loading compressed vmlinuz-5.7.0-03841-gfa3354e4ea39-dirty...
> > > > aboot: PHDR 0 vaddr 0xfffffc0001010000 offset 0xc0 size 0x17c5ae0
> > > > aboot: bss at 0xfffffc00027d5ae0, size 0xe4ea0
> > > > aboot: zero-filling 937632 bytes at 0xfffffc00027d5ae0
> > > > aboot: loading initrd (5965252 bytes/5825 blocks) at 0xfffffc05ff2cc000
> > > > aboot: starting kernel vmlinuz-5.7.0-03841-gfa3354e4ea39-dirty with arguments ro panic=5 domdadm root=/dev/md1 console=srm mminit_debug=4
> > > > Linux version 5.7.0-03841-gfa3354e4ea39-dirty (mattst88@ivybridge) (gcc version 11.1.0 (Gentoo 11.1.0-r2 p3), GNU ld (Gentoo 2.36.1 p3) 2.36.1) #26 SMP Sun Jul 25 18:20:06 PDT 2021
> > > > printk: bootconsole [srm0] enabled
> > > > Booting on Marvel variation Marvel/EV7 using machine vector MARVEL/EV7 from SRM
> > > > Major Options: SMP EV67 VERBOSE_MCHECK DEBUG_SPINLOCK MAGIC_SYSRQ
> > > > Command line: ro panic=5 domdadm root=/dev/md1 console=srm mminit_debug=4
> > > > memcluster 0, usage 1, start 0, end 1984
> > > > memcluster 1, usage 0, start 1984, end 1048576
> > > > memcluster 2, usage 1, start 2097152, end 2097224
> > > > memcluster 3, usage 0, start 2097224, end 3145728
> > > > Initial ramdisk at: 0x(____ptrval____) (5965252 bytes)
> > > > Found an IO7 at PID 0
> > > > Initializing IO7 at PID 0
> > > > FIXME: disabling master aborts
> > > > FIXME: disabling master aborts
> > > > FIXME: disabling master aborts
> > > > FIXME: disabling master aborts
> > > > SMP: 2 CPUs probed -- cpu_present_mask = 3
> > > > Zone ranges:
> > > > DMA [mem 0x0000000000f80000-0x00000fffffffdfff]
> > > > Normal empty
> > > > Movable zone start for each node
> > > > Early memory node ranges
> > > > node 0: [mem 0x0000000000f80000-0x00000001ffffffff]
> > > > node 0: [mem 0x0000000400090000-0x00000005ffffffff]
> > >
> > > I think that the issue is that memory marked as used in memcluster is never
> > > added to memblock and it skews node/zone sizing calculations.
> >
> > Thanks, this patch fixes it. With the patch applied, I see
> >
> > Zone ranges:
> > DMA [mem 0x0000000000000000-0x00000fffffffdfff]
> > Normal empty
> > Movable zone start for each node
> > Early memory node ranges
> > node 0: [mem 0x0000000000000000-0x00000001ffffffff]
> > node 0: [mem 0x0000000400000000-0x00000005ffffffff]
> > Initmem setup node 0 [mem 0x0000000000000000-0x00000005ffffffff]
> >
> > If you want to send me this patch with your S-o-b I'll take it through
> > my alpha git tree.
>
> The patch is on its way :)
Thanks a bunch, Mike, for the patch and for all you do!
> Now I'm really curios how commit e7793e53901b ("arc: update comment about
> HIGHMEM implementation") will work out.
You probably mean fdb7d9b7acd0 ("alpha: remove DISCONTIGMEM and
NUMA"). Works well so far!
Maybe in some alternative universe where I have infinite free time
(and my own power plant!) I'll get a second AlphaServer ES47 with the
enormous "hose" [1] to link the two and I'll poke you for guidance on
restoring NUMA support :)
Thanks again!
[1] https://www.ebay.com/itm/302471226394
On Fri, Jul 30, 2021 at 12:53:58PM -0700, Matt Turner wrote:
> On Tue, Jul 27, 2021 at 1:41 PM Mike Rapoport <[email protected]> wrote:
> > On Tue, Jul 27, 2021 at 12:24:26PM -0700, Matt Turner wrote:
> > >
> > > Thanks, this patch fixes it. With the patch applied, I see
> > >
> > > Zone ranges:
> > > DMA [mem 0x0000000000000000-0x00000fffffffdfff]
> > > Normal empty
> > > Movable zone start for each node
> > > Early memory node ranges
> > > node 0: [mem 0x0000000000000000-0x00000001ffffffff]
> > > node 0: [mem 0x0000000400000000-0x00000005ffffffff]
> > > Initmem setup node 0 [mem 0x0000000000000000-0x00000005ffffffff]
> > >
> > > If you want to send me this patch with your S-o-b I'll take it through
> > > my alpha git tree.
> >
> > The patch is on its way :)
>
> Thanks a bunch, Mike, for the patch and for all you do!
>
> > Now I'm really curios how commit e7793e53901b ("arc: update comment about
> > HIGHMEM implementation") will work out.
Oh, sure, I've copied the wrong line from the log :)
> You probably mean fdb7d9b7acd0 ("alpha: remove DISCONTIGMEM and
> NUMA"). Works well so far!
>
> Maybe in some alternative universe where I have infinite free time
> (and my own power plant!) I'll get a second AlphaServer ES47 with the
> enormous "hose" [1] to link the two and I'll poke you for guidance on
> restoring NUMA support :)
Heh, I'm not sure that restoring NUMA on alpha is really good use for free
time even if it were infinite :)
--
Sincerely yours,
Mike.