Re: [PATCH] perf symbols: Don't split kernel symbols when using kallsyms

From: Jiri Olsa
Date: Mon Sep 30 2019 - 05:26:28 EST

Next message: Oleksij Rempel: "[PATCH v1] net: phy: at803x: add ar9331 support"
Previous message: Vitaly Kuznetsov: "Re: [PATCH v2 6/8] KVM: x86: Fold 'enum kvm_ex_reg' definitions into 'enum kvm_reg'"
In reply to: Wei Li: "[PATCH] perf symbols: Don't split kernel symbols when using kallsyms"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On Sun, Sep 29, 2019 at 08:34:25PM +0800, Wei Li wrote:
> If there is ovelapping of vmlinux addresses with modules on a system,
> the kernel symbols will be split to unique DSOs([kernel].N) when using
> /proc/kallsyms, introduced by commit 2e538c4a1847 ("perf tools: Improve
> kernel/modules symbol lookup").
>
> When doing perf probe/report on such a system, it will fail in finding
> symbol, as the symbol split process is after the map searching, and the
> created maps can't be matched with name in kernel_get_module_map().

hi,
could you please write the whole failing code path/stack?

>
> I think the split is confusing and not so useful when using [kernel].N
> instead of the original ELF section names here. So remove the split
> process when using /proc/kallsyms only.

I haven't seens this code for some time, so I might be missing
something, but we currently create new maps/dso for the overlapping
parts like:

[kernel]
[mod1]
[kernel].1
[mod2]
[kernel].2
[mod3]
[kernel].3

IIUC your change wont create new map/dso pairs and will dend up with:

[kernel]
[mod1]
[mod2]
[mod3]

and map_groups__fixup_end call will fix [kernel].end to [mod1].start

then what happens when you want to resolve symbol from [kernel].3 range ?

thanks,
jirka

>
> On my arm32 system:
>
> V2R7C00_HI1212-OSN9800 ~/liwei # ./perf probe -v -a printk
> probe-definition(0): printk
> symbol:printk file:(null) line:0 offset:0 return:0 lazy:(null)
> 0 arguments
> map_groups__set_modules_path_dir: cannot open /lib/modules/5.3.0-rc5+ dir
> Problems setting modules path maps, continuing anyway...
> No kprobe blacklist support, ignored
> Failed to open cache(-1): /root/.debug/[kernel.kallsyms]/8eb36727183e955c790f0c7feb22d8306be7ce99/probes
> Cache open error: -1
> Looking at the vmlinux_path (8 entries long)
> symsrc__init: cannot get elf header.
> Could not open debuginfo. Try to use symbols.
> Looking at the vmlinux_path (8 entries long)
> symsrc__init: cannot get elf header.
> Failed to open /proc/kcore. Note /proc/kcore requires CAP_SYS_RAWIO capability to access.
> Using /proc/kallsyms for symbols
> Failed to find symbol printk in kernel
> Error: Failed to add events. Reason: No such file or directory (Code: -2)
>
> V2R7C00_HI1212-OSN9800 ~/liwei # ./perf probe -F
> vector_addrexcptn
> vector_dabt
> vector_fiq
> vector_irq
> vector_pabt
> vector_rst
> vector_und
>
> V2R7C00_HI1212-OSN9800 ~/liwei # sort -u /proc/kallsyms | head -n 10
> 01df0000 t __vectors_start
> 01df1000 t __stubs_start
> 01df1004 t vector_rst
> 01df1020 t vector_irq
> 01df10a0 t vector_dabt
> 01df1120 t vector_pabt
> 01df11a0 t vector_und
> 01df1220 t vector_addrexcptn
> 01df1240 T vector_fiq
> bf000000 t $a [eth]
>
> V2R7C00_HI1212-OSN9800 ~/liwei # sort -u /proc/kallsyms | grep _stext -B 10
> c1e04000 t swapper_pg_dir
> c1e08000 T stext
> c1e08000 t _text
> c1e080b8 t __create_page_tables
> c1e081bc t __fixup_smp
> c1e08224 t __fixup_smp_on_up
> c1e08240 t __fixup_pv_table
> c1e082b0 t __vet_atags
> c1e08300 T __turn_mmu_on
> c1e08300 t __idmap_text_start
> c1e08300 t _stext
>
> V2R7C00_HI1212-OSN9800 ~/liwei # ./perf record -a sleep 2
> Couldn't synthesize bpf events.
> [ perf record: Woken up 1 times to write data ]
> [ perf record: Captured and wrote 0.024 MB perf.data (140 samples) ]
> V2R7C00_HI1212-OSN9800 ~/liwei # ./perf report
> # To display the perf.data header info, please use --header/--header-only options.
> #
> #
> # Total Lost Samples: 0
> #
> # Samples: 140 of event 'cycles'
> # Event count (approx.): 9286809
> #
> # Overhead Command Shared Object Symbol
> # ........ ....... ................. .................
> #
> 15.89% swapper [unknown] [k] 0xc1e10b2c
> 7.83% swapper [unknown] [k] 0xc1e303f4
> 6.66% swapper [unknown] [k] 0xc1e84d58
> 6.28% sleep ld-2.27.so [.] do_lookup_x
> 6.07% swapper [unknown] [k] 0xc23f59dc
> 3.19% swapper [unknown] [k] 0xc23f0bfc
> 3.18% swapper [unknown] [k] 0xc2194d18
> 3.18% sleep libc-2.27.so [.] _dl_addr
> 3.17% sleep [unknown] [k] 0xc1e23454
> 3.16% sleep ld-2.27.so [.] check_match
> 3.12% sleep ld-2.27.so [.] strcmp
> 3.12% swapper [unknown] [k] 0xc1e52ce8
> 3.10% sleep [unknown] [k] 0xc1f738a4
> 3.08% sleep [unknown] [k] 0xc21a1ef8
> 3.05% sleep [unknown] [k] 0xc1f7109c
> 3.02% sleep [unknown] [k] 0xc1f59520
> 2.99% sleep [unknown] [k] 0xc23f54d4
> ...
>
> Signed-off-by: Wei Li <liwei391@xxxxxxxxxx>
> ---
> tools/perf/util/symbol.c | 54 ++++++----------------------------------
> 1 file changed, 7 insertions(+), 47 deletions(-)
>
> diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c
> index 4efde7879474..1c0d6d0e4ed1 100644
> --- a/tools/perf/util/symbol.c
> +++ b/tools/perf/util/symbol.c
> @@ -743,9 +743,7 @@ static int map_groups__split_kallsyms_for_kcore(struct map_groups *kmaps, struct
> }
>
> /*
> - * Split the symbols into maps, making sure there are no overlaps, i.e. the
> - * kernel range is broken in several maps, named [kernel].N, as we don't have
> - * the original ELF section names vmlinux have.
> + * Remove module and useless symbols from the map derived from /proc/kallsyms.
> */
> static int map_groups__split_kallsyms(struct map_groups *kmaps, struct dso *dso, u64 delta,
> struct map *initial_map)
> @@ -753,10 +751,9 @@ static int map_groups__split_kallsyms(struct map_groups *kmaps, struct dso *dso,
> struct machine *machine;
> struct map *curr_map = initial_map;
> struct symbol *pos;
> - int count = 0, moved = 0;
> + int moved = 0;
> struct rb_root_cached *root = &dso->symbols;
> struct rb_node *next = rb_first_cached(root);
> - int kernel_range = 0;
> bool x86_64;
>
> if (!kmaps)
> @@ -823,57 +820,20 @@ static int map_groups__split_kallsyms(struct map_groups *kmaps, struct dso *dso,
> * symbols at this point.
> */
> goto discard_symbol;
> - } else if (curr_map != initial_map) {
> - char dso_name[PATH_MAX];
> - struct dso *ndso;
> -
> + } else {
> + curr_map = initial_map;
> if (delta) {
> /* Kernel was relocated at boot time */
> pos->start -= delta;
> pos->end -= delta;
> }
> -
> - if (count == 0) {
> - curr_map = initial_map;
> - goto add_symbol;
> - }
> -
> - if (dso->kernel == DSO_TYPE_GUEST_KERNEL)
> - snprintf(dso_name, sizeof(dso_name),
> - "[guest.kernel].%d",
> - kernel_range++);
> - else
> - snprintf(dso_name, sizeof(dso_name),
> - "[kernel].%d",
> - kernel_range++);
> -
> - ndso = dso__new(dso_name);
> - if (ndso == NULL)
> - return -1;
> -
> - ndso->kernel = dso->kernel;
> -
> - curr_map = map__new2(pos->start, ndso);
> - if (curr_map == NULL) {
> - dso__put(ndso);
> - return -1;
> - }
> -
> - curr_map->map_ip = curr_map->unmap_ip = identity__map_ip;
> - map_groups__insert(kmaps, curr_map);
> - ++kernel_range;
> - } else if (delta) {
> - /* Kernel was relocated at boot time */
> - pos->start -= delta;
> - pos->end -= delta;
> }
> -add_symbol:
> +
> if (curr_map != initial_map) {
> rb_erase_cached(&pos->rb_node, root);
> symbols__insert(&curr_map->dso->symbols, pos);
> ++moved;
> - } else
> - ++count;
> + }
>
> continue;
> discard_symbol:
> @@ -887,7 +847,7 @@ static int map_groups__split_kallsyms(struct map_groups *kmaps, struct dso *dso,
> dso__set_loaded(curr_map->dso);
> }
>
> - return count + moved;
> + return moved;
> }
>
> bool symbol__restricted_filename(const char *filename,
> --
> 2.17.1
>

Next message: Oleksij Rempel: "[PATCH v1] net: phy: at803x: add ar9331 support"
Previous message: Vitaly Kuznetsov: "Re: [PATCH v2 6/8] KVM: x86: Fold 'enum kvm_ex_reg' definitions into 'enum kvm_reg'"
In reply to: Wei Li: "[PATCH] perf symbols: Don't split kernel symbols when using kallsyms"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]