Re: [PATCH 0/4] btrfs: initial subpage support for zoned devices

[Qu Wenruo <quwenruo.btrfs@gmx.com>]

在 2024/3/4 15:40, Neal Gompa 写道:
> On Sun, Mar 3, 2024 at 10:13 PM Qu Wenruo <wqu@suse.com> wrote:
>>
>> Ping?
>>
>> I know this is a very niche scenario (subpage + zoned), and the change
>> itself looks very scary, but the change should be safe for non-subpage
>> routine (as the new lock all delalloc ranges covers the page would still
>> at most lock one delalloc range for normal cases).
>>
>> Furthermore without this series, there seems be no proper way to support
>> subpage + zoned, unless we do a much larger change to merge
>> extent_writepage_io() into run_delalloc_range() (which I believe it's
>> still needed, but can be done in the future).
>>
>
> On the contrary, I don't think this is a niche scenario at all. Quite
> the opposite: I expect this to be a *very* common scenario because we
> will see AArch64 systems increasingly rely on subpage because 16K
> AArch64 Linux is used on two very popular platforms: Apple Silicon
> Macs (Fedora Asahi Remix) and Raspberry Pi 5 (Raspbian/RPi OS).

In fact, the above 2 platforms further prove this is still a very niche
combination, at least for consumer hardware.

Apple Silicons, you know it's the usual anti-repair and anti-customer
Apple, there is no way to add a zoned device natively, and if one goes a
convertor/dongle, I strongly doubt if a SATA/SAS to USB convertor
supports APPEND operation correctly.

It's possible to go thunderbolt -> PCIE -> U.2/NVME to attach a ZNS
device, but I strongly doubt if any Asahi Linux user has such hardware
to go in the first place.

It's the same for RPI5, I really appreciate the performance improvement
since RPI4, but the IO is even worse than Apple, and not really get any
better even in RPI5.

For my environment, it's indeed aarch64, but with a better board with a
lot of more IOs (RK3588, 4x PCIE3.0 + 1x PCIE2.0 + 1x PCIE2.0), but it's
still not ideal, and I have to go tcmu-runner to emulate a zoned HDD.

Unfortunately, we're still in the wild west of subpage world.

Thanks,
Qu
>
> We *need* this series, but I do not have the hardware to stress this
> patch set, unfortunately.
>
> The code otherwise looks reasonable to me, though.
>
> Acked-by: Neal Gompa <neal@gompa.dev>
>
>