Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [RESEND PATCH v4 1/8] bitops: Introduce the for_each_set_clump
 macro
To:     William Breathitt Gray <vilhelm.gray@gmail.com>,
        linus.walleij@linaro.org, akpm@linux-foundation.org
Cc:     linux-gpio@vger.kernel.org, linux-arch@vger.kernel.org,
        linux-kernel@vger.kernel.org, andriy.shevchenko@linux.intel.com,
        Arnd Bergmann <arnd@arndb.de>
References: <cover.1538441919.git.vilhelm.gray@gmail.com>
 <f5250b9c78377e476667be1b616c86cf573b57dc.1538441919.git.vilhelm.gray@gmail.com>
From:   Rasmus Villemoes <linux@rasmusvillemoes.dk>
Message-ID: <40ecad49-2797-0d30-b52d-a2e6838dc1ab@rasmusvillemoes.dk>
Date:   Tue, 2 Oct 2018 09:42:48 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101
 Thunderbird/52.9.1
MIME-Version: 1.0
In-Reply-To: <f5250b9c78377e476667be1b616c86cf573b57dc.1538441919.git.vilhelm.gray@gmail.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 2018-10-02 03:13, William Breathitt Gray wrote:
> This macro iterates for each group of bits (clump) with set bits, within
> a bitmap memory region. For each iteration, "clump" is set to the found
> clump index, "index" is set to the word index of the bitmap containing
> the found clump, and "offset" is set to the bit offset of the found
> clump within the respective bitmap word.

I can't say I'm a fan. It seems rather clumsy and ad-hoc - though I do
see how it matches the code you replace in drivers/gpio/. When I
initially read the cover letter, I assumed that one would get a sequence
of 4-bit values, but one has to dig the actual value out of the bitmap
afterwards using all of index, offset and a mask computed from clump_size.

> +
> +/**
> + * find_next_clump - find next clump with set bits in a memory region
> + * @index: location to store bitmap word index of found clump
> + * @offset: bits offset of the found clump within the respective bitmap word
> + * @bits: address to base the search on
> + * @size: bitmap size in number of clumps

That's a rather inconvenient unit, no? And rather easy to get wrong, I
can easily see people passing nbits instead.

I think you could reduce the number of arguments to this helper and the
macro, while getting rid of some confusion: Drop index and offset, let
clump_index be start_index and measured in bit positions (like
find_next_bit et al), and let the return value also be a bit position.
And instead of index and offset, have another unsigned long* output
parameter that gives the actual value at [return value:return
value+clump_size]. IOW, I think the prototype should be close to
find_next_bit, except that in case of "clumps", there's an extra piece
of information to return.

> + * @clump_index: clump index at which to start searching
> + * @clump_size: clump size in bits
> + *
> + * Returns the clump index for the next clump with set bits; the respective
> + * bitmap word index is stored at the location pointed by @index, and the bits
> + * offset of the found clump within the respective bitmap word is stored at the
> + * location pointed by @offset. If no bits are set, returns @size.
> + */
> +size_t find_next_clump(size_t *const index, unsigned int *const offset,
> +		       const unsigned long *const bits, const size_t size,
> +		       const size_t clump_index, const unsigned int clump_size)
> +{
> +	size_t i;
> +	unsigned int bits_offset;
> +	unsigned long word_mask;
> +	const unsigned long clump_mask = GENMASK(clump_size - 1, 0);
> +
> +	for (i = clump_index; i < size; i++) {
> +		bits_offset = i * clump_size;
> +
> +		*index = BIT_WORD(bits_offset);
> +		*offset = bits_offset % BITS_PER_LONG;
> +
> +		word_mask = bits[*index] & (clump_mask << *offset);
> +		if (!word_mask)
> +			continue;

The cover letter says

  The clump_size argument can be an arbitrary number of bits and is not
  required to be a multiple of 2.

by which I assume you mean "power of 2", but either way, the above code
does not seem to take into account the case where bits_offset +
clump_size straddles a word boundary, so it wouldn't work for a
clump_size that does not divide BITS_PER_LONG.

May I suggest another approach:

unsigned long bitmap_get_value(const unsigned long *bitmap, unsigned
start, unsigned width): Get the value of bitmap[start:start+width] for
1<=width<=BITS_PER_LONG (it's up to the caller to ensure this is within
the defined region). That can almost be an inline

bitmap_get_value(const unsigned long *bitmap, unsigned start, unsigned
width)
{
  unsigned idx = BIT_WORD(start);
  unsigned offset = start % BITS_PER_LONG;
  unsigned long lower = bitmap[idx] >> offset;
  unsigned long upper = offset <= BITS_PER_LONG - width ? 0 :
bitmap[idx+1] << (BITS_PER_LONG - offset);
  return (lower | upper) & GENMASK(width-1, 0)
}

Then you can implement the for_each_set_clump by a (IMO) more readable

  for (i = 0, start = 0; i < num_ports; i++, start += gpio_reg_size) {
    word_mask = bitmap_get_value(mask, start, gpio_reg_size);
    if (!word_mask)
      continue;
    ...
  }

Or, if you do want find_next_clump/for_each_set_clump, that can be
implemented in terms of this.

Rasmus