sea_to_user sefltest failure

sea_to_user sefltest failure

[Sebastian Ott]

Hi,

sea_to_user fails for me with:

Random seed: 0x6b8b4567
==== Test Assertion Failure ====
   include/kvm_syscalls.h:58: mem != MAP_FAILED
   pid=4923 tid=4923 errno=12 - Cannot allocate memory
      1  0x0000000000405afb: __kvm_mmap at kvm_syscalls.h:58 (discriminator 3)
      2   (inlined by) kvm_mmap at kvm_syscalls.h:65 (discriminator 3)
      3   (inlined by) vm_mem_add at kvm_util.c:1036 (discriminator 3)
      4  0x0000000000402377: vm_create_with_sea_handler at sea_to_user.c:278
      5   (inlined by) main at sea_to_user.c:324
      6  0x0000ffff8d0a621b: ?? ??:0
      7  0x0000ffff8d0a62fb: ?? ??:0
      8  0x000000000040282f: _start at ??:?
   mmap() failed, rc: -1 errno: 12 (Cannot allocate memory)

this could be related to my config?
# CONFIG_CGROUP_HUGETLB is not set
CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE=y
CONFIG_HAVE_ARCH_HUGE_VMAP=y
CONFIG_HAVE_ARCH_HUGE_VMALLOC=y
CONFIG_ARCH_WANT_HUGE_PMD_SHARE=y
CONFIG_ARCH_ENABLE_HUGEPAGE_MIGRATION=y
# CONFIG_PERSISTENT_HUGE_ZERO_FOLIO is not set
CONFIG_TRANSPARENT_HUGEPAGE=y
# CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS is not set
CONFIG_TRANSPARENT_HUGEPAGE_MADVISE=y
# CONFIG_TRANSPARENT_HUGEPAGE_NEVER is not set
CONFIG_TRANSPARENT_HUGEPAGE_SHMEM_HUGE_NEVER=y
# CONFIG_TRANSPARENT_HUGEPAGE_SHMEM_HUGE_ALWAYS is not set
# CONFIG_TRANSPARENT_HUGEPAGE_SHMEM_HUGE_WITHIN_SIZE is not set
# CONFIG_TRANSPARENT_HUGEPAGE_SHMEM_HUGE_ADVISE is not set
CONFIG_TRANSPARENT_HUGEPAGE_TMPFS_HUGE_NEVER=y
# CONFIG_TRANSPARENT_HUGEPAGE_TMPFS_HUGE_ALWAYS is not set
# CONFIG_TRANSPARENT_HUGEPAGE_TMPFS_HUGE_WITHIN_SIZE is not set
# CONFIG_TRANSPARENT_HUGEPAGE_TMPFS_HUGE_ADVISE is not set
CONFIG_PGTABLE_HAS_HUGE_LEAVES=y
CONFIG_ARCH_SUPPORTS_HUGE_PFNMAP=y
CONFIG_ARCH_SUPPORTS_HUGETLBFS=y
CONFIG_HUGETLBFS=y
CONFIG_HUGETLB_PAGE=y
CONFIG_HUGETLB_PMD_PAGE_TABLE_SHARING=y

Looking at the code smth like this would skip the test:
get_backing_src_pagesz(VM_MEM_SRC_ANONYMOUS_HUGETLB);

..or is this not worth the effort and my config is too weird?

Thanks,
Sebastian

Re: sea_to_user sefltest failure

[Sebastian Ott]

On Thu, 11 Dec 2025, Jiaqi Yan wrote:
> CONFIGs seem alright to me. Do you boot kernel with cmdline options like "default_hugepagesz=1G hugepagesz=1G hugepages=64", or dynamically set up
> huge pages via "echo 64 > /sys/kernel/mm/hugepages/hugepages-1048576kB/nr_hugepages"?

Neither of these. When I do the test is skipped:
# echo 64 > /sys/kernel/mm/hugepages/hugepages-1048576kB/nr_hugepages
# ./arm64/sea_to_user
Random seed: 0x6b8b4567
# Mapped 0x40000 pages: gva=0x80000000 to gpa=0xff80000000
# Before EINJect: data=0xbaadcafe
# EINJ_GVA=0x81234bad, einj_gpa=0xff81234bad, einj_hva=0xffff41234bad, 
einj_hpa=0x80241234bad
ok 1 # SKIP EINJ module probably not loaded?sh: line 1: 
/sys/kernel/debug/apei/einj/error_type: No such file or directory
Bail out! Failed to write EINJ entry: No such file or directory (2)
# 1 skipped test(s) detected. Consider enabling relevant config options to 
improve coverage.
# Planned tests != run tests (0 != 1)
# Totals: pass:0 fail:0 xfail:0 xpass:0 skip:1 error:0

Re: sea_to_user sefltest failure

[Jiaqi Yan]

On Thu, Dec 11, 2025 at 10:11 AM Sebastian Ott <sebott@redhat.com> wrote:
>
> On Thu, 11 Dec 2025, Jiaqi Yan wrote:
> > CONFIGs seem alright to me. Do you boot kernel with cmdline options like "default_hugepagesz=1G hugepagesz=1G hugepages=64", or dynamically set up
> > huge pages via "echo 64 > /sys/kernel/mm/hugepages/hugepages-1048576kB/nr_hugepages"?
>
> Neither of these. When I do the test is skipped:
> # echo 64 > /sys/kernel/mm/hugepages/hugepages-1048576kB/nr_hugepages
> # ./arm64/sea_to_user
> Random seed: 0x6b8b4567
> # Mapped 0x40000 pages: gva=0x80000000 to gpa=0xff80000000
> # Before EINJect: data=0xbaadcafe
> # EINJ_GVA=0x81234bad, einj_gpa=0xff81234bad, einj_hva=0xffff41234bad,
> einj_hpa=0x80241234bad
> ok 1 # SKIP EINJ module probably not loaded?sh: line 1:
> /sys/kernel/debug/apei/einj/error_type: No such file or directory
> Bail out! Failed to write EINJ entry: No such file or directory (2)

Looks like EINJ is not available on your test machine. You will need
to check several things:
1. does your firmware have EINJ support? kernel should log something
like "ACPI: EINJ populated" at boot time, but it doesn't guarantee the
EINJ support, just suggesting it may support.
2. if you are sure about the firmware, did you build kernel with
CONFIG_ACPI_APEI_EINJ=m?
3. if CONFIG_ACPI_APEI_EINJ=m, please also "modprobe einj". If
modprobe failed or /sys/kernel/debug/apei/einj/error_type still
doesn't show up, firmware probably doesn't support EINJ at all.

I will probably add more comments at the top of the file to clarify things.

> # 1 skipped test(s) detected. Consider enabling relevant config options to
> improve coverage.
> # Planned tests != run tests (0 != 1)
> # Totals: pass:0 fail:0 xfail:0 xpass:0 skip:1 error:0
>

Re: sea_to_user sefltest failure

[Gavin Shan]

On 12/12/25 4:11 AM, Sebastian Ott wrote:
> On Thu, 11 Dec 2025, Jiaqi Yan wrote:
>> CONFIGs seem alright to me. Do you boot kernel with cmdline options like "default_hugepagesz=1G hugepagesz=1G hugepages=64", or dynamically set up
>> huge pages via "echo 64 > /sys/kernel/mm/hugepages/hugepages-1048576kB/nr_hugepages"?
> 
> Neither of these. When I do the test is skipped:
> # echo 64 > /sys/kernel/mm/hugepages/hugepages-1048576kB/nr_hugepages

The backing memory is tied to 1GB hugetlb pages which are available on 4KB
and 16KB host, but not on 64KB host. So I guess 512MB hugetlb pages shall
be used on 64KB host?

> # ./arm64/sea_to_user
> Random seed: 0x6b8b4567
> # Mapped 0x40000 pages: gva=0x80000000 to gpa=0xff80000000
> # Before EINJect: data=0xbaadcafe
> # EINJ_GVA=0x81234bad, einj_gpa=0xff81234bad, einj_hva=0xffff41234bad, einj_hpa=0x80241234bad
> ok 1 # SKIP EINJ module probably not loaded?sh: line 1: /sys/kernel/debug/apei/einj/error_type: No such file or directory
> Bail out! Failed to write EINJ entry: No such file or directory (2)
> # 1 skipped test(s) detected. Consider enabling relevant config options to improve coverage.
> # Planned tests != run tests (0 != 1)
> # Totals: pass:0 fail:0 xfail:0 xpass:0 skip:1 error:0
> 

On one of my machines where RAS feature used to work, the test case passed
on 4KB host.

host$ cat /proc/1/smaps | grep KernelPageSize | head -n 1
KernelPageSize:        4 kB
host$ modprobe einj
host$ echo 4 > /sys/kernel/mm/hugepages/hugepages-1048576kB/nr_hugepages

host$ ./sea_to_user
Random seed: 0x6b8b4567
# Mapped 0x40000 pages: gva=0x80000000 to gpa=0xff80000000
# Before EINJect: data=0xbaadcafe
# EINJ_GVA=0x81234bad, einj_gpa=0xff81234bad, einj_hva=0xffff41234bad, einj_hpa=0x400081234bad
# echo 0x10 > /sys/kernel/debug/apei/einj/error_type - done
# echo 0x2 > /sys/kernel/debug/apei/einj/flags - done
# echo 0x400081234bad > /sys/kernel/debug/apei/einj/param1 - done
# echo 0xffffffffffffffff > /sys/kernel/debug/apei/einj/param2 - done
# echo 0x1 > /sys/kernel/debug/apei/einj/notrigger - done
# echo 0x1 > /sys/kernel/debug/apei/einj/error_inject - done
# Memory UER EINJected
# SIGBUS (7) received, dumping siginfo...
# si_signo=7, si_errno=0, si_code=4, si_addr=0xffff41234000
ok 1 # SKIP SEA is claimed by host APEI

Thanks,
Gavin

Re: sea_to_user sefltest failure

[Marc Zyngier]

On Thu, 11 Dec 2025 18:04:48 +0000,
Jiaqi Yan <jiaqiyan@google.com> wrote:
> 
> [1  <text/plain; UTF-8 (quoted-printable)>]
> Hi Sebastian,
> 
> CONFIGs seem alright to me. Do you boot kernel with cmdline options like
> "default_hugepagesz=1G hugepagesz=1G hugepages=64", or dynamically set up
> huge pages via "echo 64 >
> /sys/kernel/mm/hugepages/hugepages-1048576kB/nr_hugepages"?

I don't think this is irrelevant. The whole thing seems to have some
logic flaws, see the extensive report from Zenghui[1] as a reply to
your series.

	M.

[1] https://lore.kernel.org/r/3061f5f8-cef0-b7b1-c4de-f2ceea29af9a@huawei.com

-- 
Without deviation from the norm, progress is not possible.

Re: sea_to_user sefltest failure

[Jiaqi Yan]

On Thu, Dec 11, 2025 at 10:19 AM Marc Zyngier <maz@kernel.org> wrote:
>
> On Thu, 11 Dec 2025 18:04:48 +0000,
> Jiaqi Yan <jiaqiyan@google.com> wrote:
> >
> > [1  <text/plain; UTF-8 (quoted-printable)>]
> > Hi Sebastian,
> >
> > CONFIGs seem alright to me. Do you boot kernel with cmdline options like
> > "default_hugepagesz=1G hugepagesz=1G hugepages=64", or dynamically set up
> > huge pages via "echo 64 >
> > /sys/kernel/mm/hugepages/hugepages-1048576kB/nr_hugepages"?
>
> I don't think this is irrelevant. The whole thing seems to have some
> logic flaws, see the extensive report from Zenghui[1] as a reply to
> your series.

Perhaps my test is too coupled with my test environment. I will work
with Zhenghui to improve the test for people.

>
>         M.
>
> [1] https://lore.kernel.org/r/3061f5f8-cef0-b7b1-c4de-f2ceea29af9a@huawei.com
>
> --
> Without deviation from the norm, progress is not possible.