Skip to contentSkip to conversation

reduce large stack usage

PATCH Arnd Bergmann 3 participants 3 messages Mar 2, 2003 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw
Quoted 2 lines · Randy.Dunlap

This patch to 2.5.63 reduces stack usage in generate_default_upcase() from 0x3d4 bytes to just noise (on x86).

Since you are already fixing stack usage bugs in several places, maybe you are interested in this list. I have it from building 2.5.63 with allyesconfig on s390x with gcc-2.95.3. The twofish one is obviously broken, and I suspect huft_build/inflate_dynamic are the cause of the crashes I'm seeing during unpacking of initramfs.

nfs4 and reiserfs look suspicious as well, but I've never used either of them on s390x, so I can't tell if they are likely to cause real problems.

        Arnd <><

8832 twofish_setkey
1696 dohash
1680 huft_build
1680 huft_build
1544 aes_encrypt
1512 inflate_dynamic
1496 inflate_dynamic
1448 device_new_if
1360 inflate_fixed
1360 inflate_fixed
1344 br_ioctl_device
1312 elf_core_dump
1264 sctp_hash_digest
1256 aes_decrypt
1248 gss_pipe_downcall
1248 befs_warning
1248 befs_error
1248 befs_debug
1232 ciGetLeafPrefixKey
1216 nfs4_proc_rename
1192 nfs4_proc_link
1184 root_nfs_name
1184 conmode_default
1136 elf_core_dump
1128 nfsd4_proc_compound
1112 generate_default_upcase
1072 nlmclnt_proc
1016 do_open
1000 reiserfs_rename
992 reiserfs_delete_solid_item
992 nfs4_proc_symlink
992 nfs4_proc_mknod
992 nfs4_proc_mkdir
968 nlmclnt_reclaim
936 reiserfs_delete_item
912 reiserfs_cut_from_item
896 extract_entropy
888 sha512_transform
872 udf_add_entry
840 reiserfs_insert_item
832 tcp_v4_conn_request
832 reiserfs_paste_into_item
824 nfs4_proc_lookup
824 hfs_cat_move
816 udf_load_pvoldesc
816 semctl_main
792 sys_shmctl

generated from:
objdump --disassemble vmlinux |

        grep '\(^0\|aghi.%r15\)' |
        sed -e 's/^0.*<\(.*\)>:/\1/g' -e 's/^ .*-/: /g' |
        while read a b ; do
                if [ $a = : ] ; then
                        echo $b $name
                fi
                name=$a
        done |
        sort -rn |
        head -n 40
#2 lore raw
Quoted 2 lines · Arnd Bergmann

broken, and I suspect huft_build/inflate_dynamic are the cause of the crashes I'm seeing during unpacking of initramfs.

I'm also seeing crashes when unpacking the initramfs if I enlarge it so it takes more than a few (32kb) windows. I get crashes in a copy_from_user deep down the callchain when it is doing the sys_write to a file in ramfs.

Signature · 1 line
Anders Gustafsson - [email protected] - http://0x63.nu/
#3 lore raw
Quoted 3 lines · Anders Gustafsson

I'm also seeing crashes when unpacking the initramfs if I enlarge it so it takes more than a few (32kb) windows. I get crashes in a copy_from_user deep down the callchain when it is doing the sys_write to a file in ramfs.

That's slightly different from what I am experiencing on s390x. Here, I get a panic 'length error' after initramfs is unzipped. It always seems like one byte less is counted than actually is inside the archive.
OTOH, the result of a stack overflow is usually undefined, so it might also be this problem.

	Arnd <><