[PATCH] makedumpfile/arm64: Add '--mem-usage' support
Bhupesh Sharma
bhsharma at redhat.com
Sun Mar 4 19:41:12 PST 2018
Hello Masaki,
Thanks for your reply.
On Fri, Mar 2, 2018 at 11:01 AM, Masaki Tachibana
<mas-tachibana at vf.jp.nec.com> wrote:
> Hi Bhupesh,
>
> Sorry for the late reply.
> And thank you for your patch.
> I have some questions. Please answer me.
> - Have you succeeded --mem-usage on ppc64, s390x ?
Yes, --mem-usage works fine on both ppc64 and s390x RHEL systems
for me. I tested the same on several ppc64 and s390x machines.
> - By your patch, makedumpfile behaves like this;
> 1.Gets an address of _stext from a vmlinux file.
> 2.Checks how many upper bits are 1 in the address.
> 3.Determines va_bits and info->page_offset by the check.
> Isn't there any other method to get page_offset without a vmlinux ?
Ok, let me give some background here:
On ARM64 platforms the VA_BITS supported by a linux kernel run can be
selected by setting 'ARM64_VA_BITS_*' (please see [1]) config options.
Now, to determine the 'info->page_offset' in arm64 makedumpfile
context ('arch/arm64.c')
we need to determine the VA_BITS which was selected by the underlying
linux kernel.
Now there are several ways to determine the VA_BITS:
(a). Read the CONFIG flags from the user space using something like:
- Create a 'running.config' which will contain the configuration of
the running linux kernel
and grep the VA_BITS from 'running.config' :
# cat /proc/config.gz | gunzip > running.config
- However this is only possible if running linux kernel was configured
to have '/proc/config.gz'
So, this is probably not a good option.
(b). Read '_stext' symbol and calculate the 'va_bits' and
'info->page_offset' using the upper bits are 1 in the address.
There are a couple of ways to do the same via makedumpfile code:
- Use the 'vmlinux' file, which this version of the patch does.
- Use the '/proc/kallsyms' file, which is also possible and I have a
patch ready for this approach as well.
The '/proc/kallsyms' file approach is better in the following aspects:
- We don't need to pass the 'vmlinux' file path separately while
invoking '--mem-usage' option for makedumpfile.
- It also helps the arm64 KASLR makedumpfile implementation (which I
am currently working on and will send out a patch to address the same
soon), as the '_stext' symbol will be randomized and hence cannot be
properly read from the 'vmlinux' file.
If you agree, I can send a new version which reads the '_stext' symbol
from '/proc/kallsyms' and works fine on the arm64 platforms I have
tested it on (both with KASLR turned on and off)
[1]. https://elixir.bootlin.com/linux/v4.9/source/arch/arm64/Kconfig#L518
Regards,
Bhupesh
> Thanks
> Tachibana
>
>> -----Original Message-----
>> From: kexec [mailto:kexec-bounces at lists.infradead.org] On Behalf Of Bhupesh SHARMA
>> Sent: Thursday, February 22, 2018 3:58 AM
>> To: Tachibana Masaki() <mas-tachibana at vf.jp.nec.com>
>> Cc: kexec at lists.infradead.org; Bhupesh Sharma <bhsharma at redhat.com>; Hayashi Masahiko()
>> <mas-hayashi at tg.jp.nec.com>
>> Subject: Re: [PATCH] makedumpfile/arm64: Add '--mem-usage' support
>>
>> Hi Masaki Tachibana,
>>
>> On Tue, Feb 20, 2018 at 4:42 PM, Masaki Tachibana
>> <mas-tachibana at vf.jp.nec.com> wrote:
>> > Hi Bhupesh,
>> >
>> > Sorry for the late reply.
>> > I'll reply by the end the next week.
>>
>> Sure. Thanks for your mail.
>>
>> Regards,
>> Bhupesh
>>
>>
>> > Thanks
>> > tachibana
>> >
>> >> -----Original Message-----
>> >> From: Bhupesh Sharma [mailto:bhsharma at redhat.com]
>> >> Sent: Tuesday, February 20, 2018 1:56 PM
>> >> To: kexec at lists.infradead.org
>> >> Cc: Bhupesh Sharma <bhsharma at redhat.com>; Tachibana Masaki() <mas-tachibana at vf.jp.nec.com>; Nakayama Takuya(
>> >> ) <tak-nakayama at tg.jp.nec.com>; Nishimura Daisuke() <dai-nishimura at rc.jp.nec.com>
>> >> Subject: Re: [PATCH] makedumpfile/arm64: Add '--mem-usage' support
>> >>
>> >> Hello,
>> >>
>> >> On Fri, Feb 9, 2018 at 3:06 PM, Bhupesh Sharma <bhsharma at redhat.com> wrote:
>> >> > Its good to have the makedumpfile '--mem-usage' support
>> >> > for arm64 architecture as well, as it allows one to see the page numbers
>> >> > of current system (1st kernel) in different use.
>> >> >
>> >> > Using this we can know how many pages are dumpable when different
>> >> > dump_level is specified.
>> >> >
>> >> > Normally for x86_64, makedumpfile analyzes the 'System Ram' and
>> >> > 'kernel text' program segment of /proc/kcore excluding
>> >> > the crashkernel range, then calculates the page number of different
>> >> > kind per vmcoreinfo.
>> >> >
>> >> > We use the similar logic for arm64, but in addition make the '--mem-usage'
>> >> > usage dependent on the VMLINUX file being passed. This is done to allow
>> >> > information like VA_BITS being determined from kernel symbol like
>> >> > _stext. This allows us to get the VA_BITS before 'set_kcore_vmcoreinfo()'
>> >> > is called.
>> >> >
>> >> > Also I have validated the '--mem-usage' makedumpfile option on several
>> >> > ppc64/ppc64le and s390x machines, so update the makedumpfile.8
>> >> > documentation to indicate that '--mem-usage' option is supported
>> >> > not only on x86_64, but also on ppc64, s390x and arm64.
>> >> >
>> >> > After this patch, when using the '--mem-usage' option with makedumpfile,
>> >> > we get the correct information about the different pages. For e.g.
>> >> > here is an output from my arm64 board:
>> >> >
>> >> > TYPE PAGES EXCLUDABLE DESCRIPTION
>> >> > ----------------------------------------------------------------------
>> >> > ZERO 49524 yes Pages filled with zero
>> >> > NON_PRI_CACHE 15143 yes Cache pages without private flag
>> >> > PRI_CACHE 29147 yes Cache pages with private flag
>> >> > USER 3684 yes User process pages
>> >> > FREE 1450569 yes Free pages
>> >> > KERN_DATA 14243 no Dumpable kernel data
>> >> >
>> >> > page size: 65536
>> >> > Total pages on system: 1562310
>> >> > Total size on system: 102387548160 Byte
>> >> >
>> >> > Cc: Masaki Tachibana <mas-tachibana at vf.jp.nec.com>
>> >> > Cc: Takuya Nakayama <tak-nakayama at tg.jp.nec.com>
>> >> > Cc: Daisuke Nishimura <dai-nishimura at rc.jp.nec.com>
>> >> > Signed-off-by: Bhupesh Sharma <bhsharma at redhat.com>
>> >>
>> >> Ping. Any review comments on this?
>> >>
>> >> Regards,
>> >> Bhupesh
>> >>
>> >> > ---
>> >> > arch/arm64.c | 51 ++++++++++++++++++++++++++++++++++++++++++++++++---
>> >> > makedumpfile.8 | 11 +++++++++--
>> >> > makedumpfile.c | 25 +++++++++++++++++++++++--
>> >> > makedumpfile.h | 1 +
>> >> > 4 files changed, 81 insertions(+), 7 deletions(-)
>> >> >
>> >> > diff --git a/arch/arm64.c b/arch/arm64.c
>> >> > index 25d7a1f4db98..91f113f6447c 100644
>> >> > --- a/arch/arm64.c
>> >> > +++ b/arch/arm64.c
>> >> > @@ -48,6 +48,12 @@ static unsigned long kimage_voffset;
>> >> > #define SZ_64K (64 * 1024)
>> >> > #define SZ_128M (128 * 1024 * 1024)
>> >> >
>> >> > +#define PAGE_OFFSET_36 ((0xffffffffffffffffUL) << 36)
>> >> > +#define PAGE_OFFSET_39 ((0xffffffffffffffffUL) << 39)
>> >> > +#define PAGE_OFFSET_42 ((0xffffffffffffffffUL) << 42)
>> >> > +#define PAGE_OFFSET_47 ((0xffffffffffffffffUL) << 47)
>> >> > +#define PAGE_OFFSET_48 ((0xffffffffffffffffUL) << 48)
>> >> > +
>> >> > #define pgd_val(x) ((x).pgd)
>> >> > #define pud_val(x) (pgd_val((x).pgd))
>> >> > #define pmd_val(x) (pud_val((x).pud))
>> >> > @@ -140,8 +146,6 @@ pud_offset(pgd_t *pgda, pgd_t *pgdv, unsigned long vaddr)
>> >> >
>> >> > static int calculate_plat_config(void)
>> >> > {
>> >> > - va_bits = NUMBER(VA_BITS);
>> >> > -
>> >> > /* derive pgtable_level as per arch/arm64/Kconfig */
>> >> > if ((PAGESIZE() == SZ_16K && va_bits == 36) ||
>> >> > (PAGESIZE() == SZ_64K && va_bits == 42)) {
>> >> > @@ -188,7 +192,6 @@ get_machdep_info_arm64(void)
>> >> > kimage_voffset = NUMBER(kimage_voffset);
>> >> > info->max_physmem_bits = PHYS_MASK_SHIFT;
>> >> > info->section_size_bits = SECTIONS_SIZE_BITS;
>> >> > - info->page_offset = 0xffffffffffffffffUL << (va_bits - 1);
>> >> >
>> >> > DEBUG_MSG("kimage_voffset : %lx\n", kimage_voffset);
>> >> > DEBUG_MSG("max_physmem_bits : %lx\n", info->max_physmem_bits);
>> >> > @@ -219,6 +222,48 @@ get_xen_info_arm64(void)
>> >> > int
>> >> > get_versiondep_info_arm64(void)
>> >> > {
>> >> > + unsigned long long stext;
>> >> > +
>> >> > + /* We can read the _stext symbol from vmlinux and determine the
>> >> > + * VA_BITS and page_offset.
>> >> > + */
>> >> > +
>> >> > + /* Open the vmlinux file */
>> >> > + open_kernel_file();
>> >> > + set_dwarf_debuginfo("vmlinux", NULL,
>> >> > + info->name_vmlinux, info->fd_vmlinux);
>> >> > +
>> >> > + if (!get_symbol_info())
>> >> > + return FALSE;
>> >> > +
>> >> > + /* Get the '_stext' symbol */
>> >> > + if (SYMBOL(_stext) == NOT_FOUND_SYMBOL) {
>> >> > + ERRMSG("Can't get the symbol of _stext.\n");
>> >> > + return FALSE;
>> >> > + } else {
>> >> > + stext = SYMBOL(_stext);
>> >> > + }
>> >> > +
>> >> > + /* Derive va_bits as per arch/arm64/Kconfig */
>> >> > + if ((stext & PAGE_OFFSET_36) == PAGE_OFFSET_36) {
>> >> > + va_bits = 36;
>> >> > + } else if ((stext & PAGE_OFFSET_39) == PAGE_OFFSET_39) {
>> >> > + va_bits = 39;
>> >> > + } else if ((stext & PAGE_OFFSET_42) == PAGE_OFFSET_42) {
>> >> > + va_bits = 42;
>> >> > + } else if ((stext & PAGE_OFFSET_47) == PAGE_OFFSET_47) {
>> >> > + va_bits = 47;
>> >> > + } else if ((stext & PAGE_OFFSET_48) == PAGE_OFFSET_48) {
>> >> > + va_bits = 48;
>> >> > + } else {
>> >> > + ERRMSG("Cannot find a proper _stext for calculating VA_BITS\n");
>> >> > + return FALSE;
>> >> > + }
>> >> > +
>> >> > + info->page_offset = (0xffffffffffffffffUL) << (va_bits - 1);
>> >> > +
>> >> > + DEBUG_MSG("page_offset=%lx, va_bits=%d\n", info->page_offset, va_bits);
>> >> > +
>> >> > return TRUE;
>> >> > }
>> >> >
>> >> > diff --git a/makedumpfile.8 b/makedumpfile.8
>> >> > index 15db7947d62f..be9620035316 100644
>> >> > --- a/makedumpfile.8
>> >> > +++ b/makedumpfile.8
>> >> > @@ -593,7 +593,7 @@ last cleared on the crashed kernel, through "dmesg --clear" for example.
>> >> >
>> >> > .TP
>> >> > \fB\-\-mem-usage\fR
>> >> > -This option is only for x86_64.
>> >> > +This option is currently supported on x86_64, arm64, ppc64 and s390x.
>> >> > This option is used to show the page numbers of current system in different
>> >> > use. It should be executed in 1st kernel. By the help of this, user can know
>> >> > how many pages is dumpable when different dump_level is specified. It analyzes
>> >> > @@ -601,12 +601,19 @@ the 'System Ram' and 'kernel text' program segment of /proc/kcore excluding
>> >> > the crashkernel range, then calculates the page number of different kind per
>> >> > vmcoreinfo. So currently /proc/kcore need be specified explicitly.
>> >> >
>> >> > +For arm64, path to vmlinux file should be specified as well.
>> >> > +
>> >> > .br
>> >> > -.B Example:
>> >> > +.B Example (for architectures other than arm64):
>> >> > .br
>> >> > # makedumpfile \-\-mem-usage /proc/kcore
>> >> > +
>> >> > +.br
>> >> > +.B Example (for arm64 architecture):
>> >> > .br
>> >> >
>> >> > +# makedumpfile \-\-mem-usage vmlinux /proc/kcore
>> >> > +.br
>> >> >
>> >> > .TP
>> >> > \fB\-\-diskset=VMCORE\fR
>> >> > diff --git a/makedumpfile.c b/makedumpfile.c
>> >> > index ed138d339d9a..b38b5000aa74 100644
>> >> > --- a/makedumpfile.c
>> >> > +++ b/makedumpfile.c
>> >> > @@ -11090,7 +11090,14 @@ static struct option longopts[] = {
>> >> > {"cyclic-buffer", required_argument, NULL, OPT_CYCLIC_BUFFER},
>> >> > {"eppic", required_argument, NULL, OPT_EPPIC},
>> >> > {"non-mmap", no_argument, NULL, OPT_NON_MMAP},
>> >> > +#ifdef __aarch64__
>> >> > + /* VMLINUX file is required for aarch64 for get
>> >> > + * the symbols required to calculate va_bits.
>> >> > + */
>> >> > + {"mem-usage", required_argument, NULL, OPT_MEM_USAGE},
>> >> > +#else
>> >> > {"mem-usage", no_argument, NULL, OPT_MEM_USAGE},
>> >> > +#endif
>> >> > {"splitblock-size", required_argument, NULL, OPT_SPLITBLOCK_SIZE},
>> >> > {"work-dir", required_argument, NULL, OPT_WORKING_DIR},
>> >> > {"num-threads", required_argument, NULL, OPT_NUM_THREADS},
>> >> > @@ -11201,8 +11208,22 @@ main(int argc, char *argv[])
>> >> > info->flag_partial_dmesg = 1;
>> >> > break;
>> >> > case OPT_MEM_USAGE:
>> >> > - info->flag_mem_usage = 1;
>> >> > - break;
>> >> > + info->flag_mem_usage = 1;
>> >> > +#ifdef __aarch64__
>> >> > + /* VMLINUX file is required for aarch64 for get
>> >> > + * the symbols required to calculate va_bits and
>> >> > + * it should be the 1st command parameter being
>> >> > + * specified.
>> >> > + */
>> >> > + if (strcmp(optarg, "/proc/kcore") == 0) {
>> >> > + MSG("vmlinux path should be 1st commandline parameter with --mem-usage option.\n");
>> >> > + goto out;
>> >> > + }
>> >> > + else {
>> >> > + info->name_vmlinux = optarg;
>> >> > + }
>> >> > +#endif
>> >> > + break;
>> >> > case OPT_COMPRESS_SNAPPY:
>> >> > info->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
>> >> > break;
>> >> > diff --git a/makedumpfile.h b/makedumpfile.h
>> >> > index 01eece231475..f65d91870b73 100644
>> >> > --- a/makedumpfile.h
>> >> > +++ b/makedumpfile.h
>> >> > @@ -2308,6 +2308,7 @@ struct elf_prstatus {
>> >> > /*
>> >> > * Function Prototype.
>> >> > */
>> >> > +int open_kernel_file(void);
>> >> > mdf_pfn_t get_num_dumpable_cyclic(void);
>> >> > mdf_pfn_t get_num_dumpable_cyclic_withsplit(void);
>> >> > int get_loads_dumpfile_cyclic(void);
>> >> > --
>> >> > 2.7.4
>> >> >
>> >
>> > _______________________________________________
>> > kexec mailing list
>> > kexec at lists.infradead.org
>> > http://lists.infradead.org/mailman/listinfo/kexec
>>
>> _______________________________________________
>> kexec mailing list
>> kexec at lists.infradead.org
>> http://lists.infradead.org/mailman/listinfo/kexec
>
>
More information about the kexec
mailing list