From: Yonghong Song <yhs@fb.com>
To: Arnaldo Carvalho de Melo <arnaldo.melo@gmail.com>,
<dwarves@vger.kernel.org>
Cc: Alexei Starovoitov <ast@kernel.org>,
Andrii Nakryiko <andrii@kernel.org>,
Bill Wendling <morbo@google.com>, <bpf@vger.kernel.org>,
<kernel-team@fb.com>
Subject: Re: [PATCH dwarves v2 3/3] dwarf_loader: permit merging all dwarf cu's for clang lto built binary
Date: Sun, 28 Mar 2021 13:18:19 -0700 [thread overview]
Message-ID: <72580b46-3bf8-ce91-de61-d1c1a68b01ad@fb.com> (raw)
In-Reply-To: <20210328061702.1956720-1-yhs@fb.com>
On 3/27/21 11:17 PM, Yonghong Song wrote:
> For vmlinux built with clang thin-lto or lto, there exist
> cross cu type references. For example, the below can happen:
> compile unit 1:
> tag 10: type A
> compile unit 2:
> ...
> refer to type A (tag 10 in compile unit 1)
> I only checked a few but have seen type A may be a simple type
> like "unsigned char" or a complex type like an array of base types.
>
> To resolve this issue, the tag DW_AT_producer of the first
> DW_TAG_compile_unit is checked. If the binary is built
> with clang lto, all debuginfo dwarf cu's will be merged
> into one pahole cu which will resolve the above
> cross-cu tag reference issue. To test whether a binary
> is built with clang lto or not, The "clang version"
> and "-flto" will be checked against DW_AT_producer string
> for the first 5 debuginfo cu's. The reason is that
> a few linux files disabled lto for various reasons.
>
> Merging cu's will create a single cu with lots of types, tags
> and functions. For example with clang thin-lto built vmlinux,
> I saw 9M entries in types table, 5.2M in tags table. The
> below are pahole wallclock time for different hashbits:
> command line: time pahole -J vmlinux
> # of hashbits wallclock time in seconds
> 15 460
> 16 255
> 17 131
> 18 97
> 19 75
> 20 69
> 21 64
> 22 62
> 23 58
> 24 64
>
> The problem is with hashtags__find(), esp. the loop
> uint32_t bucket = hashtags__fn(id);
> const struct hlist_head *head = hashtable + bucket;
> hlist_for_each_entry(tpos, pos, head, hash_node) {
> if (tpos->id == id)
> return tpos;
> }
>
> Say we have 9M types and (1 << 15) buckets, that means each bucket
> will have roughly 64 elements. So each lookup will traverse
> the loop 32 iterations on average.
>
> If we have 1 << 21 buckets, then each buckets will have 4 elements,
> and the average number of loop iterations for hashtags__find()
> will be 2.
>
> Note that the number of hashbits 24 makes performance worse
> than 23. The reason could be that 23 hashbits can cover 8M
> buckets (close to 9M for the number of entries in types table).
> Higher number of hash bits allocates more memory and becomes
> less cache efficient compared to 23 hashbits.
>
> This patch picks # of hashbits 21 as the starting value
> and will try to allocate memory based on that, if memory
> allocation fails, we will go with less hashbits until
> we reach hashbits 15 which is the default for
> non merge-cu case.
>
> Signed-off-by: Yonghong Song <yhs@fb.com>
> ---
> dwarf_loader.c | 120 +++++++++++++++++++++++++++++++++++++++++++++++++
> 1 file changed, 120 insertions(+)
>
> diff --git a/dwarf_loader.c b/dwarf_loader.c
> index aa6372a..1e63ca9 100644
> --- a/dwarf_loader.c
> +++ b/dwarf_loader.c
> @@ -51,6 +51,7 @@ struct strings *strings;
> #endif
>
> static uint32_t hashtags__bits = 15;
> +static uint32_t max_hashtags__bits = 21;
>
> static uint32_t hashtags__fn(Dwarf_Off key)
> {
> @@ -2484,6 +2485,115 @@ static int cus__load_debug_types(struct cus *cus, struct conf_load *conf,
> return 0;
> }
>
> +static bool cus__merging_cu(Dwarf *dw)
> +{
> + uint8_t pointer_size, offset_size;
> + Dwarf_Off off = 0, noff;
> + size_t cuhl;
> + int cnt = 0;
> +
> + /*
> + * Just checking the first cu is not enough.
> + * In linux, some C files may have LTO is disabled, e.g.,
> + * e242db40be27 x86, vdso: disable LTO only for vDSO
> + * d2dcd3e37475 x86, cpu: disable LTO for cpu.c
> + * Fortunately, disabling LTO for a particular file in a LTO build
> + * is rather an exception. Iterating 5 cu's to check whether
> + * LTO is used or not should be enough.
> + */
> + while (dwarf_nextcu(dw, off, &noff, &cuhl, NULL, &pointer_size,
> + &offset_size) == 0) {
> + Dwarf_Die die_mem;
> + Dwarf_Die *cu_die = dwarf_offdie(dw, off + cuhl, &die_mem);
> +
> + if (cu_die == NULL)
> + break;
> +
> + if (++cnt > 5)
> + break;
> +
> + const char *producer = attr_string(cu_die, DW_AT_producer);
> + if (strstr(producer, "clang version") != NULL &&
> + strstr(producer, "-flto") != NULL)
> + return true;
> +
> + off = noff;
> + }
> +
> + return false;
> +}
> +
> +static int cus__merge_and_process_cu(struct cus *cus, struct conf_load *conf,
> + Dwfl_Module *mod, Dwarf *dw, Elf *elf,
> + const char *filename,
> + const unsigned char *build_id,
> + int build_id_len,
> + struct dwarf_cu *type_dcu)
> +{
> + uint8_t pointer_size, offset_size;
> + struct dwarf_cu *dcu = NULL;
> + Dwarf_Off off = 0, noff;
> + struct cu *cu = NULL;
> + size_t cuhl;
> +
> + while (dwarf_nextcu(dw, off, &noff, &cuhl, NULL, &pointer_size,
> + &offset_size) == 0) {
> + Dwarf_Die die_mem;
> + Dwarf_Die *cu_die = dwarf_offdie(dw, off + cuhl, &die_mem);
> +
> + if (cu_die == NULL)
> + break;
> +
> + if (cu == NULL) {
> + cu = cu__new("", pointer_size, build_id, build_id_len,
> + filename);
> + if (cu == NULL || cu__set_common(cu, conf, mod, elf) != 0)
> + return 1;
This is an oversight. I forgot to change back to DWARF_CB_ABORT after
some experiments. I just sent out v3 to correct this and added the
kernel patch link which is needed to make this auto lto discovery work.
> +
> + dcu = malloc(sizeof(struct dwarf_cu));
> + if (dcu == NULL)
> + return 1;
> +
[...]
prev parent reply other threads:[~2021-03-28 20:19 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-03-28 6:16 [PATCH dwarves v2 0/3] permit merging all dwarf cu's for clang lto built binary Yonghong Song
2021-03-28 6:16 ` [PATCH dwarves v2 1/3] dwarf_loader: permits flexible HASHTAGS__BITS Yonghong Song
2021-03-28 6:16 ` [PATCH dwarves v2 2/3] dwarf_loader: factor out common code to initialize a cu Yonghong Song
2021-03-28 6:17 ` [PATCH dwarves v2 3/3] dwarf_loader: permit merging all dwarf cu's for clang lto built binary Yonghong Song
2021-03-28 20:18 ` Yonghong Song [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=72580b46-3bf8-ce91-de61-d1c1a68b01ad@fb.com \
--to=yhs@fb.com \
--cc=andrii@kernel.org \
--cc=arnaldo.melo@gmail.com \
--cc=ast@kernel.org \
--cc=bpf@vger.kernel.org \
--cc=dwarves@vger.kernel.org \
--cc=kernel-team@fb.com \
--cc=morbo@google.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).