Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Sat, 19 Sep 2020 04:02:12 +0100
From:   Matthew Wilcox <willy@infradead.org>
To:     Arvind Sankar <nivedita@alum.mit.edu>
Cc:     Linus Torvalds <torvalds@linux-foundation.org>,
        "Gustavo A. R. Silva" <gustavoars@kernel.org>,
        Dennis Zhou <dennis@kernel.org>, Tejun Heo <tj@kernel.org>,
        Christoph Lameter <cl@linux.com>,
        Linux-MM <linux-mm@kvack.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Kees Cook <keescook@chromium.org>
Subject: Re: [GIT PULL] percpu fix for v5.9-rc6
Message-ID: <20200919030212.GK32101@casper.infradead.org>
References: <CAHk-=wg=vvSf3M9O1VkwyYB4D4W6XS2AHVpQn6hEQY+usWrKGg@mail.gmail.com>
 <20200918200252.GH32101@casper.infradead.org>
 <CAHk-=wiNjJGhAMBwYixwkADpNharvcuOG-AMCdii1q_Xo_Ky_A@mail.gmail.com>
 <20200918202909.GA2946008@rani.riverdale.lan>
 <CAHk-=wh-ryuY7KBNWr1n+kgQ5_CHB3-X+od-djBV4W-1kQFokA@mail.gmail.com>
 <20200918210050.GA2953017@rani.riverdale.lan>
 <CAHk-=wgyKF9vnac3mw6v-Bo5D8X-rcrkF=BsZ2jX+OveGkGgBw@mail.gmail.com>
 <20200918223957.GA2964553@rani.riverdale.lan>
 <CAHk-=wg+3Poqh_HQ93nPMWOXLQHZhvYNuwScoQ-WaYWuriLYAg@mail.gmail.com>
 <20200919025336.GA3008405@rani.riverdale.lan>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200919025336.GA3008405@rani.riverdale.lan>
Precedence: bulk

On Fri, Sep 18, 2020 at 10:53:36PM -0400, Arvind Sankar wrote:
> I think you can't do this in standard C. It's a GCC extension.
> 
> 	A structure containing a flexible array member, or a union
> 	containing such a structure (possibly recursively), may not be a
> 	member of a structure or an element of an array. (However, these
> 	uses are permitted by GCC as extensions.)

I actually have a patch in the works which wants to do this.

 struct pagevec {
-       unsigned char nr;
-       bool percpu_pvec_drained;
-       struct page *pages[PAGEVEC_SIZE];
+       union {
+               struct {
+                       unsigned char sz;
+                       unsigned char nr;
+                       bool percpu_pvec_drained;
+                       struct page *pages[];
+               };
+               void *__p[PAGEVEC_SIZE + 1];
+       };
 };

I don't think ANSI C permits this, but it's useful to be able to declare
a pagevec on the stack and be guaranteed to get enough memory to hold
a useful sized array of pointers (as well as be able to dynamically
allocate a larger pagevec for the cases which want such a thing).

We could certainly split pagevec into a variable length array version
and have a struct stack_pagevec which had the extra padding, but that
involves changing a lot more code.