On 24/09/24 09:10, Tao Liu wrote:
Hi Aditya & Lianbo,
Thanks for your review on this ppc fix. Recently I'm working on
another high priority task, so I may not respond to your comments in
time, sorry for that.
On Tue, Sep 24, 2024 at 2:28 AM Aditya Gupta <adityag(a)linux.ibm.com> wrote:
> <...snip...>
> How about this approach ?
>
> I feel this might be more dependable, as the file command also uses
> this, as well as fadump (PowerPC firmware-assisted dump) in the kernel.
>
>> As a sidenote, there is still some issue either in patch #1 or #3
>> causing all secondary CPUs (where swapper task is running) to show all
>> threads as '_end'
> What do you think ?
Yes, I like the idea of extracting the abi version through vmlinux.
Thanks a lot for your approach and code, but I haven't tried your code
yet. I will test it when my current task finishes.
Sure, take your time :)
Thanks Tao,
Aditya Gupta
> Thanks,
> Tao Liu
>> Either way what you chose, I do realise the idea to utilise the
>> 'std r2, STK_GOT(r1)' in such a way, was interesting and smart.
>>
>> static bool
>> is_ppc64_elf_abi_v2(void)
>> {
>> const char *kernel_file = pc->namelist;
>> int fd, swap;
>> char eheader[BUFSIZE];
>> Elf64_Ehdr *elf64;
>>
>> if ((fd = open(kernel_file, O_RDONLY)) < 0) {
>> error(INFO, "%s: %s\n", kernel_file,
strerror(errno));
>> return FALSE;
>> }
>> if (read(fd, eheader, BUFSIZE) != BUFSIZE) {
>> error(INFO, "%s: %s\n", kernel_file, strerror(errno));
>> close(fd);
>> return FALSE;
>> }
>> close(fd);
>>
>> if (!STRNEQ(eheader, ELFMAG) || eheader[EI_VERSION] != EV_CURRENT)
>> return FALSE;
>>
>> elf64 = (Elf64_Ehdr *)&eheader[0];
>>
>> swap = (((eheader[EI_DATA] == ELFDATA2LSB) &&
>> (__BYTE_ORDER == __BIG_ENDIAN)) ||
>> ((eheader[EI_DATA] == ELFDATA2MSB) &&
>> (__BYTE_ORDER == __LITTLE_ENDIAN)));
>>
>> /*
>> * The ABI version is stored in E_FLAGS in the ELF header, atleast for
>> * ppc64 binaries.
>> *
>> * These logic is also used by the 'file' command, which states
this in
>> * its magic file:
>> *
>> * >18 leshort 20 PowerPC or cisco
4500,
>> * >18 leshort 21 64-bit PowerPC or
cisco 7500,
>> * >>48 lelong 0 Unspecified or
Power ELF V1 ABI,
>> * >>48 lelong 1 Power ELF V1 ABI,
>> * >>48 lelong 2 OpenPOWER ELF V2
ABI,
>> *
>> * The '>18' above means, data at 18 (0x12), which is
basically
>> * 'elf64->e_machine', which will be 20 for PPC, and 21 for
PPC64
>> *
>> * Similarly, '>>48' is offset 48 (0x30) in the file,
which is
>> * basically 'elf64->e_flags', which has the ELF ABI version
>> */
>> if (elf64->e_flags == 2) {
>> /* ABI v2 */
>> return TRUE;
>> } else if (elf64->e_flags == 1) {
>> /* ABI v1 */
>> return FALSE;
>> }
>>
>> error(WARNING, "Unstable elf_abi v1/v2 detection.\n");
>> return FALSE;
>> }
>>
>> Also, we can those values in the ELF header:
>>
>> $ hexdump vmlinux-little-endian
>> 0000000 457f 464c 0102 0001 0000 0000 0000 0000
>> 0000010 0002 0015 (= 21 = E_MACHINE) 0001 0000 0000 0000 0000 c000
>> 0000020 0040 0000 0000 0000 8bf8 0040 0000 0000
>> 0000030 0002 (= E_FLAGS = ABI) 0000 0040 0038 0002 0040 0027 0026
>> 0000040 0001 0000 0007 0000 0000 0001 0000 0000
>>
>> Also, I see 'pc->namelist' is basically the kernel filename, is that
>> correct ? Can that be renamed to kernel_filename or something ?
>>
>> Thanks,
>> Aditya Gupta
>>
>>> Thanks,
>>>
>>> Aditya Gupta
>>>
>>>> + char buf1[BUFSIZE];
>>>> + char *pos1, *pos2;
>>>> + int errflag = 0;
>>>> + ulong stk_got = 0;
>>>> + static bool ret = false;
>>>> + static bool checked = false;
>>>> +
>>>> + if (checked == true || !symbol_exists("tm_reclaim"))
>>>> + return ret;
>>>> +
>>>> + sprintf(buf1, "x/16i tm_reclaim");
>>>> + open_tmpfile();
>>>> + if (!gdb_pass_through(buf1, pc->tmpfile, GNU_RETURN_ON_ERROR))
>>>> + goto out;
>>>> + checked = true;
>>>> + rewind(pc->tmpfile);
>>>> + while (fgets(buf1, BUFSIZE, pc->tmpfile)) {
>>>> + // "std r2, STK_GOT(r1)" is expected
>>>> + if (strstr(buf1, "std") &&
>>>> + strstr(buf1, "(r1)") &&
>>>> + (pos1 = strstr(buf1, "r2,"))) {
>>>> + pos1 += strlen("r2,");
>>>> + for (pos2 = pos1; *pos2 != '\0' &&
*pos2 != '('; pos2++);
>>>> + *pos2 = '\0';
>>>> + stk_got = stol(pos1, RETURN_ON_ERROR|QUIET,
&errflag);
>>>> + break;
>>>> + }
>>>> + }
>>>> +
>>>> + if (!errflag) {
>>>> + switch (stk_got) {
>>>> + case 24:
>>>> + ret = true;
>>>> + case 40:
>>>> + goto out;
>>>> + }
>>>> + }
>>>> + error(WARNING, "Unstable elf_abi v1/v2 detection.\n");
>>>> +out:
>>>> + close_tmpfile();
>>>> + return ret;
>>>> +}
>>>> /*
>>>> * get the SP and PC values for idle tasks.
>>>> @@ -2834,9 +2880,17 @@ get_ppc64_frame(struct bt_info *bt, ulong *getpc,
ulong *getsp)
>>>> sp = ppc64_get_sp(task);
>>>> if (!INSTACK(sp, bt))
>>>> goto out;
>>>> - readmem(sp+STACK_FRAME_OVERHEAD, KVADDR, ®s,
>>>> - sizeof(struct ppc64_pt_regs),
>>>> - "PPC64 pt_regs", FAULT_ON_ERROR);
>>>> +
>>>> + if (THIS_KERNEL_VERSION >= LINUX(6,2,0) &&
is_ppc64_elf_abi_v2()) {
>>>> + readmem(sp+STACK_SWITCH_FRAME_REGS, KVADDR, ®s,
>>>> + sizeof(struct ppc64_pt_regs),
>>>> + "PPC64 pt_regs", FAULT_ON_ERROR);
>>>> + } else {
>>>> + readmem(sp+STACK_FRAME_OVERHEAD, KVADDR, ®s,
>>>> + sizeof(struct ppc64_pt_regs),
>>>> + "PPC64 pt_regs", FAULT_ON_ERROR);
>>>> + }
>>>> +
>>>> ip = regs.nip;
>>>> closest = closest_symbol(ip);
>>>> if (STREQ(closest, ".__switch_to") || STREQ(closest,
"__switch_to")) {