Re: [PATCH 2/3] sbitmap: ammortize cost of clearing bits

From: Jens Axboe
Date: Sun Dec 09 2018 - 01:20:06 EST

On Dec 8, 2018, at 10:51 PM, Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
> Hi,
>> On Thu, Nov 29, 2018 at 06:12:33PM -0700, Jens Axboe wrote:
>> sbitmap maintains a set of words that we use to set and clear bits, with
>> each bit representing a tag for blk-mq. Even though we spread the bits
>> out and maintain a hint cache, one particular bit allocated will end up
>> being cleared in the exact same spot.
>> This introduces batched clearing of bits. Instead of clearing a given
>> bit, the same bit is set in a cleared/free mask instead. If we fail
>> allocating a bit from a given word, then we check the free mask, and
>> batch move those cleared bits at that time. This trades 64 atomic bitops
>> for 2 cmpxchg().
>> In a threaded poll test case, half the overhead of getting and clearing
>> tags is removed with this change. On another poll test case with a
>> single thread, performance is unchanged.
>> Signed-off-by: Jens Axboe <axboe@xxxxxxxxx>
> This patch results in irq lock inversion warnings when trying to boot
> from usb drives. This was observed with qemu boots of aarch64, x86, and
> x86_64 images.

This one is a false positive, was already reported last week. Just need to add some lockdep annotation for it.