From: David Gibson <david-xT8FGy+AXnRB3Ne2BGzF6laj5H9X9Tb+@public.gmane.org>
To: Bin Meng <bmeng.cn-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
Cc: devicetree-compiler-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
Subject: Re: [PATCH] libfdt: Fix the undefined behavior in fdt_num_mem_rsv()
Date: Thu, 1 Apr 2021 14:25:08 +1100 [thread overview]
Message-ID: <YGU9FH3WccMBrfl1@yekko.fritz.box> (raw)
In-Reply-To: <20210330095645.515056-1-bmeng.cn-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
[-- Attachment #1: Type: text/plain, Size: 3164 bytes --]
On Tue, Mar 30, 2021 at 05:56:45PM +0800, Bin Meng wrote:
> With LLVM 10.0.0+, the following codes in fdt_num_mem_rsv() does not
> work any more for an fdt that is at address 0:
>
> for (i = 0; (re = fdt_mem_rsv(fdt, i)) != NULL; i++) {
> if (fdt64_ld_(&re->size) == 0)
> return i;
> }
>
> Due to LLVM's optimization engine utilizing a UB in C, the following
> code pattern:
>
> if ((pointer + offset) != NULL)
>
> is transformed into:
>
> if (pointer != NULL)
>
> because if pointer is NULL and offset is non-zero, the result of
> (pointer + offset) is UB. So LLVM is free to exploit such UB to
> perform some optimization.
>
> In this case, fdt_mem_rsv() gets inlined and returns (pointer + offset).
> And LLVM in turns emits codes to check fdt against NULL, which won't
> work for fdt at address 0.
I don't think this really fixes anything. It might fool LLVM into
doing what you need right now, but I don't see any reason to expect it
will keep doing so.
IIUC, you're saying that the specific problem is that adding a
non-zero offset to a NULL pointer is UB, which happens inside
fdt_mem_rsv_() if n != 0. But with your patch, that UB still exists..
>
> Signed-off-by: Bin Meng <bmeng.cn-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
> ---
>
> libfdt/fdt_ro.c | 17 +++++++++++++----
> 1 file changed, 13 insertions(+), 4 deletions(-)
>
> diff --git a/libfdt/fdt_ro.c b/libfdt/fdt_ro.c
> index 17584da..4db4013 100644
> --- a/libfdt/fdt_ro.c
> +++ b/libfdt/fdt_ro.c
> @@ -157,18 +157,26 @@ int fdt_generate_phandle(const void *fdt, uint32_t *phandle)
> return 0;
> }
>
> -static const struct fdt_reserve_entry *fdt_mem_rsv(const void *fdt, int n)
> +static bool fdt_is_mem_rsv(const void *fdt, int n)
> {
> unsigned int offset = n * sizeof(struct fdt_reserve_entry);
> unsigned int absoffset = fdt_off_mem_rsvmap(fdt) + offset;
>
> if (!can_assume(VALID_INPUT)) {
> if (absoffset < fdt_off_mem_rsvmap(fdt))
> - return NULL;
> + return false;
> if (absoffset > fdt_totalsize(fdt) -
> sizeof(struct fdt_reserve_entry))
> - return NULL;
> + return false;
> }
> +
> + return true;
> +}
> +
> +static const struct fdt_reserve_entry *fdt_mem_rsv(const void *fdt, int n)
> +{
> + if (!fdt_is_mem_rsv(fdt, n))
> + return NULL;
> return fdt_mem_rsv_(fdt, n);
> }
>
> @@ -191,7 +199,8 @@ int fdt_num_mem_rsv(const void *fdt)
> int i;
> const struct fdt_reserve_entry *re;
>
> - for (i = 0; (re = fdt_mem_rsv(fdt, i)) != NULL; i++) {
> + for (i = 0; fdt_is_mem_rsv(fdt, i); i++) {
> + re = fdt_mem_rsv_(fdt, i);
.. here ^^.
Basically if your compiled is going to optimized based on (NULL +
something) being UB, and the NULL pointer is address 0, that's
fundamentally incompatible with storing a device tree at address 0.
> if (fdt64_ld_(&re->size) == 0)
> return i;
> }
--
David Gibson | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_
| _way_ _around_!
http://www.ozlabs.org/~dgibson
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]
next prev parent reply other threads:[~2021-04-01 3:25 UTC|newest]
Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-03-30 9:56 [PATCH] libfdt: Fix the undefined behavior in fdt_num_mem_rsv() Bin Meng
[not found] ` <20210330095645.515056-1-bmeng.cn-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2021-04-01 3:25 ` David Gibson [this message]
[not found] ` <YGU9FH3WccMBrfl1-l+x2Y8Cxqc4e6aEkudXLsA@public.gmane.org>
2021-04-02 4:52 ` Bin Meng
[not found] ` <CAEUhbmVsszFcUC8tEYmJ4djPvyDmXspDg-gb9rYZpm8zWwxc1g-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
2021-04-08 2:54 ` David Gibson
[not found] ` <YG5wZa/ti6klXEZQ-l+x2Y8Cxqc4e6aEkudXLsA@public.gmane.org>
2021-04-08 3:21 ` Bin Meng
[not found] ` <CAEUhbmWeTpRHPdqMtcNWdF-ORJ0OXcMi1YJeN+KUpbkqGKvFGg-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
2021-04-20 1:30 ` David Gibson
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=YGU9FH3WccMBrfl1@yekko.fritz.box \
--to=david-xt8fgy+axnrb3ne2bgzf6laj5h9x9tb+@public.gmane.org \
--cc=bmeng.cn-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org \
--cc=devicetree-compiler-u79uwXL29TY76Z2rM5mHXA@public.gmane.org \
/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 read-only IMAP folder(s) and NNTP newsgroup(s).