From: Jessica Clarke <jrtc27@jrtc27.com>
To: Junio C Hamano <gitster@pobox.com>
Cc: git@vger.kernel.org
Subject: Re: [PATCH v2] mem-pool: Don't assume uintmax_t is aligned enough for all types
Date: Sun, 23 Jan 2022 20:23:53 +0000 [thread overview]
Message-ID: <FD089F0E-C826-4A8E-BC16-F763E256B410@jrtc27.com> (raw)
In-Reply-To: <xmqqsfte8awj.fsf@gitster.g>
On 23 Jan 2022, at 20:17, Junio C Hamano <gitster@pobox.com> wrote:
>
> Jessica Clarke <jrtc27@jrtc27.com> writes:
>
>> Currently mem_pool_alloc uses sizeof(uintmax_t) as a proxy for what
>> should be _Alignof(max_align_t) in C11. On most architectures this is
>
> Lose "Currently", as the present tense describes the status quo, the
> shape of the problematic code we have today that wants improvement
> by the proposed patch.
Do you want a v3 or is that something you'll amend on git-am?
>> sufficient (though on m68k it is in fact overly strict, since the
>> de-facto ABI, which differs from the specified System V ABI, has the
>> maximum alignment of all types as 2 bytes), but on CHERI, and thus Arm's
>> Morello prototype, it is insufficient for any type that stores a
>> pointer, which must be aligned to 128 bits (on 64-bit architectures
>> extended with CHERI), whilst uintmax_t is a 64-bit integer.
>
> OK.
>
>> Fix this by introducing our own approximation for max_align_t and a
>> means to compute _Alignof it without relying on C11. Currently this
>> union only contains uintmax_t and void *, but more types can be added as
>> needed.
>
> Nicely described.
>
>> +/*
>> + * The inner union is an approximation for C11's max_align_t, and the
>> + * struct + offsetof computes _Alignof. This can all just be replaced
>> + * with _Alignof(max_align_t) if/when C11 is part of the baseline.
>> + *
>> + * Add more types to the union if the current set is insufficient.
>> + */
>> +struct git_max_alignment {
>> + char unalign;
>> + union {
>> + uintmax_t max_align_uintmax;
>> + void *max_align_pointer;
>> + } aligned;
>> +};
>> +#define GIT_MAX_ALIGNMENT offsetof(struct git_max_alignment, aligned)
>> +
>
> The original computed the alignment requirement for uintmax_t as
> sizeof(uintmax_t), not as
>
> offsetof(struct {
> char unalign;
> union { uintmax_t i; } aligned;
> }, aligned)
>
> because if you have an array of a type, each element of it must be
> aligned appropriately already for that type, without the unalignment
> the outer struct enforces. I wonder if your complex offsetof is
> equivalent to sizeof(union { uintmax_t u; void *p; })?
>
> IOW, in this struct:
>
> struct max_alignment_helper {
> char unalign;
> union {
> uintmax_t uintmax_t_unused;
> void *pointer_unused;
> } u[2];
> } s;
>
> both s.u[0] and s.u[1] must be properly aligned, so wouldn't the
> alignment requirement for the union type, which can be used to hold
> a single value of either uintmax_t or a poinhter, be the distance
> between these two array elements, i.e. sizeof(s.u[0])?
>
> To put it differently in yet another way, wouldn't it simplify down
> to this?
>
> union max_alignment_helper {
> uintmax_t uintmax_t_unused;
> void *pointer_unused;
> };
> #define GIT_MAX_ALIGNMENT sizeof(union max_alignment_helper);
>
> I am not saying that the "a forcibly unaligned union in a struct" is
> a bad/wrong way to express what you want to achieve. I just do not
> know if there is a reason to choose it over a seemingly simpler
> sizeof(that union) without the outer struct and unalign member.
So, sizeof(X) does not always equal _Alignof(X), even for primitive
types, _Alignof need only be a factor of sizeof. The two are the same
on most architectures, and is a sensible ABI, but the exception is the
m68k case I was referring to above. On m68k, sizeof(long long) == 8,
but _Alignof(long long) == 2 (yes this is a real pain point of its ABI;
in particular int is only 2-byte aligned, but futex(2) explicitly
requires 4-byte alignment). So using sizeof definitely gets you
something sufficiently aligned, but can waste space. This doesn’t
affect CHERI/Morello, all our implementations keep sizeof == _Alignof,
but as I was changing this code I felt I should use the more precise
construct.
Jess
> Other than that, looks OK to me. Especially the parts that use the
> macro look correctly converted.
>
> Thanks.
>
>> @@ -69,9 +85,9 @@ void *mem_pool_alloc(struct mem_pool *pool, size_t len)
>> struct mp_block *p = NULL;
>> void *r;
>>
>> - /* round up to a 'uintmax_t' alignment */
>> - if (len & (sizeof(uintmax_t) - 1))
>> - len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
>> + /* round up to a 'GIT_MAX_ALIGNMENT' alignment */
>> + if (len & (GIT_MAX_ALIGNMENT - 1))
>> + len += GIT_MAX_ALIGNMENT - (len & (GIT_MAX_ALIGNMENT - 1));
>>
>> if (pool->mp_block &&
>> pool->mp_block->end - pool->mp_block->next_free >= len)
>
>
>> /*
>> * Allocate a new mp_block and insert it after the block specified in
>> * `insert_after`. If `insert_after` is NULL, then insert block at the
>> @@ -69,9 +85,9 @@ void *mem_pool_alloc(struct mem_pool *pool, size_t len)
>> struct mp_block *p = NULL;
>> void *r;
>>
>> - /* round up to a 'uintmax_t' alignment */
>> - if (len & (sizeof(uintmax_t) - 1))
>> - len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
>> + /* round up to a 'GIT_MAX_ALIGNMENT' alignment */
>> + if (len & (GIT_MAX_ALIGNMENT - 1))
>> + len += GIT_MAX_ALIGNMENT - (len & (GIT_MAX_ALIGNMENT - 1));
>>
>> if (pool->mp_block &&
>> pool->mp_block->end - pool->mp_block->next_free >= len)
next prev parent reply other threads:[~2022-01-23 20:24 UTC|newest]
Thread overview: 32+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-01-05 13:23 [PATCH] Properly align memory allocations and temporary buffers Jessica Clarke
2022-01-06 21:46 ` Taylor Blau
2022-01-06 21:56 ` Jessica Clarke
2022-01-06 22:27 ` Junio C Hamano
2022-01-06 22:56 ` Jessica Clarke
2022-01-07 0:10 ` Junio C Hamano
2022-01-07 0:22 ` Jessica Clarke
2022-01-07 0:31 ` brian m. carlson
2022-01-07 0:39 ` Jessica Clarke
2022-01-07 1:43 ` brian m. carlson
2022-01-07 2:08 ` Jessica Clarke
2022-01-07 2:11 ` Jessica Clarke
2022-01-07 19:30 ` Junio C Hamano
2022-01-07 19:33 ` Jessica Clarke
2022-01-07 20:56 ` René Scharfe
2022-01-07 21:30 ` Junio C Hamano
2022-01-07 23:30 ` René Scharfe
2022-01-08 0:18 ` Elijah Newren
2022-01-06 23:22 ` brian m. carlson
2022-01-06 23:31 ` Jessica Clarke
2022-01-07 14:57 ` Philip Oakley
2022-01-07 16:08 ` René Scharfe
2022-01-07 16:21 ` Jessica Clarke
2022-01-12 13:58 ` Jessica Clarke
2022-01-12 15:47 ` René Scharfe
2022-01-12 15:49 ` Jessica Clarke
2022-01-23 15:24 ` [PATCH v2] mem-pool: Don't assume uintmax_t is aligned enough for all types Jessica Clarke
2022-01-23 20:17 ` Junio C Hamano
2022-01-23 20:23 ` Jessica Clarke [this message]
2022-01-23 20:28 ` Junio C Hamano
2022-01-23 20:33 ` [PATCH v3] " Jessica Clarke
2022-01-24 17:11 ` Junio C Hamano
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
List information: http://vger.kernel.org/majordomo-info.html
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=FD089F0E-C826-4A8E-BC16-F763E256B410@jrtc27.com \
--to=jrtc27@jrtc27.com \
--cc=git@vger.kernel.org \
--cc=gitster@pobox.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.
Code repositories for project(s) associated with this public inbox
https://80x24.org/mirrors/git.git
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).