From: DJ Delorie <dj@redhat.com>
To: Szabolcs Nagy <szabolcs.nagy@arm.com>
Cc: libc-alpha@sourceware.org
Subject: Re: [PATCH 3/6] malloc: Use chunk2rawmem throughout
Date: Tue, 23 Mar 2021 16:25:25 -0400 [thread overview]
Message-ID: <xn35wlk47e.fsf@rhel8.vm> (raw)
In-Reply-To: <e24c5ca639c47eca08574b51f613762ff9338a1a.1616155129.git.szabolcs.nagy@arm.com> (message from Szabolcs Nagy on Fri, 19 Mar 2021 13:26:49 +0000)
Szabolcs Nagy <szabolcs.nagy@arm.com> writes:
> The difference between chunk2mem and chunk2rawmem is that the latter
> does not get the memory tag for the returned pointer. It turns out
> chunk2rawmem almost always works:
Given that these two macros are identical on non-aarch64 systems, I'm
going to gloss over the "is it tagged correctly" question since you
folks can just test it, and other targets won't care ;-)
LGTM
Reviewed-by: DJ Delorie <dj@redhat.com>
> sysmalloc: Returns untagged memory.
> _int_malloc: Returns untagged memory.
> _int_free: Takes untagged memory.
> _int_memalign: Returns untagged memory.
> _int_realloc: Takes and returns tagged memory.
We should probably put this information in comments at each function
implementation too, but at least it's in the source files :-)
> - newmem = chunk2mem (newp);
> + newmem = tag_at (chunk2rawmem (newp));
Ok.
> headers have distinct tags. Converting fully from one to the other
> involves extracting the tag at the other address and creating a
> suitable pointer using it. That can be quite expensive. There are
> - many occasions, though when the pointer will not be dereferenced
> - (for example, because we only want to assert that the pointer is
> - correctly aligned). In these cases it is more efficient not
> - to extract the tag, since the answer will be the same either way.
> - chunk2rawmem() can be used in these cases.
> - */
> + cases when the pointers are not dereferenced (for example only used
> + for alignment check) so the tags are not relevant, and there are
> + cases when user data is not tagged distinctly from malloc headers
> + (user data is untagged because tagging is done late in malloc and
> + early in free). User memory tagging across internal interfaces:
> +
> + sysmalloc: Returns untagged memory.
> + _int_malloc: Returns untagged memory.
> + _int_free: Takes untagged memory.
> + _int_memalign: Returns untagged memory.
> + _int_memalign: Returns untagged memory.
> + _mid_memalign: Returns tagged memory.
> + _int_realloc: Takes and returns tagged memory.
> +*/
Ok.
> -/* Convert a user mem pointer to a chunk address without correcting
> +/* Convert a chunk address to a user mem pointer without correcting
> the tag. */
> #define chunk2rawmem(p) ((void*)((char*)(p) + CHUNK_HDR_SZ))
Heh. Ok.
> #define misaligned_chunk(p) \
> - ((uintptr_t)(MALLOC_ALIGNMENT == CHUNK_HDR_SZ ? (p) : chunk2mem (p)) \
> + ((uintptr_t)(MALLOC_ALIGNMENT == CHUNK_HDR_SZ ? (p) : chunk2rawmem (p)) \
> & MALLOC_ALIGN_MASK)
Ok.
> - return chunk2mem (p);
> + return chunk2rawmem (p);
Ok.
> - return chunk2mem (p);
> + return chunk2rawmem (p);
Ok.
> - uintptr_t mem = (uintptr_t) chunk2mem(p);
> + uintptr_t mem = (uintptr_t) chunk2rawmem(p);
Ok.
> - tcache_entry *e = (tcache_entry *) chunk2mem (chunk);
> + tcache_entry *e = (tcache_entry *) chunk2rawmem (chunk);
Ok.
> - void *newmem = chunk2mem (newp);
> + void *newmem = tag_at (chunk2rawmem (newp));
Ok.
> - void *p = chunk2mem (victim);
> + void *p = chunk2rawmem (victim);
Ok.
> - void *p = chunk2mem (victim);
> + void *p = chunk2rawmem (victim);
Ok.
> - void *p = chunk2mem (victim);
> + void *p = chunk2rawmem (victim);
Ok.
> - void *p = chunk2mem (victim);
> + void *p = chunk2rawmem (victim);
Ok.
> - void *p = chunk2mem (victim);
> + void *p = chunk2rawmem (victim);
Ok.
> - void *p = chunk2mem (victim);
> + void *p = chunk2rawmem (victim);
Ok.
> - void *p = chunk2mem (victim);
> + void *p = chunk2rawmem (victim);
Ok.
> - tcache_entry *e = (tcache_entry *) chunk2mem (p);
> + tcache_entry *e = (tcache_entry *) chunk2rawmem (p);
Ok.
> - free_perturb (chunk2mem(p), size - CHUNK_HDR_SZ);
> + free_perturb (chunk2rawmem(p), size - CHUNK_HDR_SZ);
Ok.
> - free_perturb (chunk2mem(p), size - CHUNK_HDR_SZ);
> + free_perturb (chunk2rawmem(p), size - CHUNK_HDR_SZ);
Ok.
> - return chunk2mem (newp);
> + return chunk2rawmem (newp);
Ok.
> - return chunk2mem (p);
> + return chunk2rawmem (p);
Ok.
next prev parent reply other threads:[~2021-03-23 20:25 UTC|newest]
Thread overview: 13+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-03-19 13:25 [PATCH 0/6] malloc: more memory tagging optimizations Szabolcs Nagy
2021-03-19 13:26 ` [PATCH 1/6] malloc: Use memsize instead of CHUNK_AVAILABLE_SIZE Szabolcs Nagy
2021-03-23 20:01 ` DJ Delorie
2021-03-19 13:26 ` [PATCH 2/6] malloc: Use different tag after mremap Szabolcs Nagy
2021-03-23 20:03 ` DJ Delorie
2021-03-19 13:26 ` [PATCH 3/6] malloc: Use chunk2rawmem throughout Szabolcs Nagy
2021-03-23 20:25 ` DJ Delorie [this message]
2021-03-19 13:27 ` [PATCH 4/6] malloc: Rename chunk2rawmem Szabolcs Nagy
2021-03-23 20:43 ` DJ Delorie
2021-03-19 13:27 ` [PATCH 5/6] malloc: Remove unnecessary tagging around _mid_memalign Szabolcs Nagy
2021-03-23 20:44 ` DJ Delorie
2021-03-19 13:27 ` [PATCH 6/6] malloc: Ensure mtag code path in checked_request2size is cold Szabolcs Nagy
2021-03-23 20:46 ` DJ Delorie
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=xn35wlk47e.fsf@rhel8.vm \
--to=dj@redhat.com \
--cc=libc-alpha@sourceware.org \
--cc=szabolcs.nagy@arm.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 read-only IMAP folder(s) and NNTP newsgroup(s).