Re: [PATCH] sparc64: fix hugetlb for sun4u

[John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>]

Hi Anthony,

On Tue, 2025-07-15 at 18:24 -0700, Anthony Yznaga wrote:
> An attempt to exercise sparc hugetlb code in a sun4u-based guest
> running under qemu results in the guest hanging due to being stuck
> in a trap loop. This is due to invalid hugetlb TTEs being installed
> that do not have the expected _PAGE_PMD_HUGE and page size bits set.
> Although the breakage has gone apparently unnoticed for several years,
> fix it now so there is the option to exercise sparc hugetlb code under
> qemu. This can be useful because sun4v support in qemu does not support
> linux guests currently and sun4v-based hardware resources may not be
> readily available.
> 
> Fix tested with a 6.15.2 and 6.16-rc6 kernels by running libhugetlbfs
> tests on a qemu guest running Debian 13.
> 
> Fixes: c7d9f77d33a7 ("sparc64: Multi-page size support")
> Cc: stable@vger.kernel.org
> Signed-off-by: Anthony Yznaga <anthony.yznaga@oracle.com>
> ---
>  arch/sparc/mm/hugetlbpage.c | 20 ++++++++++++++++++++
>  1 file changed, 20 insertions(+)
> 
> diff --git a/arch/sparc/mm/hugetlbpage.c b/arch/sparc/mm/hugetlbpage.c
> index 80504148d8a5..2048b5c42ca8 100644
> --- a/arch/sparc/mm/hugetlbpage.c
> +++ b/arch/sparc/mm/hugetlbpage.c
> @@ -22,6 +22,26 @@
>  
>  static pte_t sun4u_hugepage_shift_to_tte(pte_t entry, unsigned int shift)
>  {
> +	unsigned long hugepage_size = _PAGE_SZ4MB_4U;
> +
> +	pte_val(entry) = pte_val(entry) & ~_PAGE_SZALL_4U;
> +
> +	switch (shift) {
> +	case HPAGE_256MB_SHIFT:
> +		hugepage_size = _PAGE_SZ256MB_4U;
> +		pte_val(entry) |= _PAGE_PMD_HUGE;
> +		break;
> +	case HPAGE_SHIFT:
> +		pte_val(entry) |= _PAGE_PMD_HUGE;
> +		break;
> +	case HPAGE_64K_SHIFT:
> +		hugepage_size = _PAGE_SZ64K_4U;
> +		break;
> +	default:
> +		WARN_ONCE(1, "unsupported hugepage shift=%u\n", shift);
> +	}
> +
> +	pte_val(entry) = pte_val(entry) | hugepage_size;
>  	return entry;
>  }
>  

I have compiled a 4.19.325 kernel with the patched applied on top and booted
that kernel onto an UltraSPARC IIIi machine. I cannot test a current kernel
at the moment as newer kernels do not run stable on sun4u machines at the
moment.

I will try to bisect this problem within the next days. Currently, I don't have
serial access to this machine, so I cannot easily reboot it when the kernel
crashes but this should be resolved within the next days.

Here is the output from /proc/cpuinfo:

root@raverin:/home/glaubitz/libhugetlbfs# cat /proc/cpuinfo 
cpu             : TI UltraSparc IIIi (Jalapeno)
fpu             : UltraSparc IIIi integrated FPU
pmu             : ultra3i
prom            : OBP 4.22.33 2007/06/18 12:45
type            : sun4u
ncpus probed    : 1
ncpus active    : 1
D$ parity tl1   : 0
I$ parity tl1   : 0
Cpu0ClkTck      : 000000005995f5c0
cpucaps         : flush,stbar,swap,muldiv,v9,ultra3,mul32,div32,v8plus,vis,vis2
MMU Type        : Cheetah+
MMU PGSZs       : 8K,64K,512K,4MB
root@raverin:/home/glaubitz/libhugetlbfs#

I allocated 25 huge pages:

root@raverin:/home/glaubitz/libhugetlbfs# echo 25 > /proc/sys/vm/nr_hugepages
root@raverin:/home/glaubitz/libhugetlbfs# hugeadm --pool-list
      Size  Minimum  Current  Maximum  Default
     65536        0        0        0         
   8388608       25       25       25        *
 268435456        0        0        0         
2147483648        0        0        0         
root@raverin:/home/glaubitz/libhugetlbfs#

The testsuite then passed without any problems:

root@raverin:/home/glaubitz/libhugetlbfs# make check
PASS
zero_filesize_segment (8M: 64):
test_root (8M: 64):     PASS
meminfo_nohuge (8M: 64):        PASS
gethugepagesize (8M: 64):       PASS
gethugepagesizes (8M: 64):      PASS
HUGETLB_VERBOSE=1 empty_mounts (8M: 64):        PASS
HUGETLB_VERBOSE=1 large_mounts (8M: 64):        PASS
find_path (8M: 64):     PASS
unlinked_fd (8M: 64):   PASS
readback (8M: 64):      PASS
truncate (8M: 64):      PASS
shared (8M: 64):        PASS
mprotect (8M: 64):      PASS
mlock (8M: 64): PASS
misalign (8M: 64):      PASS
fallocate_basic.sh (8M: 64):    PASS
fallocate_align.sh (8M: 64):    PASS
ptrace-write-hugepage (8M: 64): PASS
icache-hygiene (8M: 64):        PASS
slbpacaflush (8M: 64):  Bad configuration: Atleast online 2 cpus are required
straddle_4GB_static (8M: 64):   PASS
huge_at_4GB_normal_below_static (8M: 64):       FAIL    Wrong address with MAP_FIXED normal
huge_below_4GB_normal_above_static (8M: 64):    PASS
map_high_truncate_2 (8M: 64):   PASS
misaligned_offset (8M: 64):     PASS (inconclusive)
truncate_above_4GB (8M: 64):    PASS
brk_near_huge (8M: 64): Fatal glibc error: malloc.c:2601 (sysmalloc): assertion failed: 
task-size-overrun (8M: 64):     PASS
stack_grow_into_huge (8M: 64):  PASS
corrupt-by-cow-opt (8M: 64):    PASS
noresv-preserve-resv-page (8M: 64):     PASS
noresv-regarded-as-resv (8M: 64):       PASS
readahead_reserve.sh (8M: 64):  PASS
madvise_reserve.sh (8M: 64):    PASS
fadvise_reserve.sh (8M: 64):    PASS
mremap-expand-slice-collision.sh (8M: 64):      PASS
mremap-fixed-normal-near-huge.sh (8M: 64):      PASS
mremap-fixed-huge-near-normal.sh (8M: 64):      PASS
set shmmax limit to 67108864
shm-perms (8M: 64):     PASS
private (8M: 64):       PASS
fork-cow (8M: 64):      PASS
direct (8M: 64):        Bad configuration: Failed to open direct-IO file: Invalid argument
malloc (8M: 64):        PASS
LD_PRELOAD=libhugetlbfs.so HUGETLB_MORECORE=yes malloc (8M: 64):        SKIPPED
LD_PRELOAD=libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_RESTRICT_EXE=unknown:none malloc (8M: 64):      SKIPPED
LD_PRELOAD=libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_RESTRICT_EXE=unknown:malloc malloc (8M: 64):    SKIPPED
malloc_manysmall (8M: 64):      PASS
LD_PRELOAD=libhugetlbfs.so HUGETLB_MORECORE=yes malloc_manysmall (8M: 64):      SKIPPED
GLIBC_TUNABLES=glibc.malloc.tcache_count=0 heapshrink (8M: 64): PASS
GLIBC_TUNABLES=glibc.malloc.tcache_count=0 LD_PRELOAD=libheapshrink.so heapshrink (8M: 64):     PASS
GLIBC_TUNABLES=glibc.malloc.tcache_count=0 LD_PRELOAD=libhugetlbfs.so HUGETLB_MORECORE=yes heapshrink (8M: 64): SKIPPED
GLIBC_TUNABLES=glibc.malloc.tcache_count=0 LD_PRELOAD=libhugetlbfs.so libheapshrink.so HUGETLB_MORECORE=yes heapshrink (8M: 64):        SKIPPED
GLIBC_TUNABLES=glibc.malloc.tcache_count=0 LD_PRELOAD=libheapshrink.so HUGETLB_MORECORE=yes HUGETLB_MORECORE_SHRINK=yes heapshrink (8M: 64):    SKIPPED
GLIBC_TUNABLES=glibc.malloc.tcache_count=0 LD_PRELOAD=libhugetlbfs.so libheapshrink.so HUGETLB_MORECORE=yes HUGETLB_MORECORE_SHRINK=yes heapshrink (8M: 64):    SKIPPED
HUGETLB_VERBOSE=1 HUGETLB_MORECORE=yes heap-overflow (8M: 64):  SKIPPED
HUGETLB_VERBOSE=0 linkhuge_nofd (8M: 64):
LD_PRELOAD=libhugetlbfs.so HUGETLB_VERBOSE=0 linkhuge_nofd (8M: 64):
HUGETLB_VERBOSE=0 xB.linkhuge_nofd (8M: 64):
HUGETLB_VERBOSE=0 xBDT.linkhuge_nofd (8M: 64):
HUGETLB_MINIMAL_COPY=no HUGETLB_VERBOSE=0 xB.linkhuge_nofd (8M: 64):
HUGETLB_MINIMAL_COPY=no HUGETLB_VERBOSE=0 xBDT.linkhuge_nofd (8M: 64):
HUGETLB_ELFMAP=no HUGETLB_VERBOSE=0 xB.linkhuge_nofd (8M: 64):
HUGETLB_ELFMAP=no HUGETLB_VERBOSE=0 xBDT.linkhuge_nofd (8M: 64):
linkhuge (8M: 64):
LD_PRELOAD=libhugetlbfs.so linkhuge (8M: 64):
xB.linkhuge (8M: 64):
xBDT.linkhuge (8M: 64):
HUGETLB_MINIMAL_COPY=no xB.linkhuge (8M: 64):
HUGETLB_MINIMAL_COPY=no xBDT.linkhuge (8M: 64):
HUGETLB_ELFMAP=no xB.linkhuge (8M: 64):
HUGETLB_ELFMAP=no xBDT.linkhuge (8M: 64):
HUGETLB_SHARE=1 xB.linkshare (8M: 64):
HUGETLB_SHARE=1 xBDT.linkshare (8M: 64):
HUGETLB_SHARE=1 xB.linkshare (8M: 64):
HUGETLB_SHARE=1 xBDT.linkshare (8M: 64):
HUGETLB_SHARE=0 xB.linkhuge (8M: 64):
HUGETLB_SHARE=1 xB.linkhuge (8M: 64):
HUGETLB_SHARE=0 xBDT.linkhuge (8M: 64):
HUGETLB_SHARE=1 xBDT.linkhuge (8M: 64):
linkhuge_rw (8M: 64):
HUGETLB_ELFMAP=R linkhuge_rw (8M: 64):
HUGETLB_ELFMAP=W linkhuge_rw (8M: 64):
HUGETLB_ELFMAP=RW linkhuge_rw (8M: 64):
HUGETLB_ELFMAP=no linkhuge_rw (8M: 64):
HUGETLB_MINIMAL_COPY=no HUGETLB_ELFMAP=R linkhuge_rw (8M: 64):
HUGETLB_MINIMAL_COPY=no HUGETLB_ELFMAP=W linkhuge_rw (8M: 64):
HUGETLB_MINIMAL_COPY=no HUGETLB_ELFMAP=RW linkhuge_rw (8M: 64):
HUGETLB_ELFMAP=R HUGETLB_SHARE=0 linkhuge_rw (8M: 64):
HUGETLB_ELFMAP=R HUGETLB_SHARE=1 linkhuge_rw (8M: 64):
HUGETLB_ELFMAP=W HUGETLB_SHARE=0 linkhuge_rw (8M: 64):
HUGETLB_ELFMAP=W HUGETLB_SHARE=1 linkhuge_rw (8M: 64):
HUGETLB_ELFMAP=RW HUGETLB_SHARE=0 linkhuge_rw (8M: 64):
HUGETLB_ELFMAP=RW HUGETLB_SHARE=1 linkhuge_rw (8M: 64):
chunk-overcommit (8M: 64):      PASS
alloc-instantiate-race shared (8M: 64): Bad configuration: Atleast online 2 cpus are required
alloc-instantiate-race private (8M: 64):        Bad configuration: Atleast online 2 cpus are required
truncate_reserve_wraparound (8M: 64):   PASS
truncate_sigbus_versus_oom (8M: 64):    PASS
get_huge_pages (8M: 64):        PASS
shmoverride_linked (8M: 64):    PASS
HUGETLB_SHM=yes shmoverride_linked (8M: 64):    PASS
LD_PRELOAD=libhugetlbfs.so shmoverride_unlinked (8M: 64):       PASS
LD_PRELOAD=libhugetlbfs.so HUGETLB_SHM=yes shmoverride_unlinked (8M: 64):       PASS
quota.sh (8M: 64):      PASS
counters.sh (8M: 64):   PASS
mmap-gettest 10 25 (8M: 64):    PASS
mmap-cow 24 25 (8M: 64):        PASS
set shmmax limit to 209715200
shm-fork 10 12 (8M: 64):        PASS
set shmmax limit to 209715200
shm-fork 10 25 (8M: 64):        PASS
set shmmax limit to 209715200
shm-getraw 25 /dev/full (8M: 64):       PASS
fallocate_stress.sh (8M: 64):   PASS
********** TEST SUMMARY
*                      8M            
*                      32-bit 64-bit 
*     Total testcases:     0    111   
*             Skipped:     0      9   
*                PASS:     0     57   
*                FAIL:     0      1   
*    Killed by signal:     0      1   
*   Bad configuration:     0      4   
*       Expected FAIL:     0      0   
*     Unexpected PASS:     0      0   
*    Test not present:     0     39   
* Strange test result:     0      0   
**********
root@raverin:/home/glaubitz/libhugetlbfs#

Tested-by: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>

Adrian

-- 
 .''`.  John Paul Adrian Glaubitz
: :' :  Debian Developer
`. `'   Physicist
  `-    GPG: 62FF 8A75 84E0 2956 9546  0006 7426 3B37 F5B5 F913