Re: [PATCH v2] zswap: allow setting default status, compressor and allocator in Kconfig

[Maciej S. Szmigiero]

On 24.10.2019 13:04, Vlastimil Babka wrote:
> On 10/24/19 1:22 AM, Maciej S. Szmigiero wrote:
>> The compressed cache for swap pages (zswap) currently needs from 1 to 3
>> extra kernel command line parameters in order to make it work: it has to be
>> enabled by adding a "zswap.enabled=1" command line parameter and if one
>> wants a different compressor or pool allocator than the default lzo / zbud
>> combination then these choices also need to be specified on the kernel
>> command line in additional parameters.
>>
>> Using a different compressor and allocator for zswap is actually pretty
>> common as guides often recommend using the lz4 / z3fold pair instead of
>> the default one.
>> In such case it is also necessary to remember to enable the appropriate
>> compression algorithm and pool allocator in the kernel config manually.
>>
>> Let's avoid the need for adding these kernel command line parameters and
>> automatically pull in the dependencies for the selected compressor
>> algorithm and pool allocator by adding an appropriate default switches to
>> Kconfig.
>>
>> The default values for these options match what the code was using
>> previously as its defaults.
>>
>> Signed-off-by: Maciej S. Szmigiero <mail@xxxxxxxxxxxxxxxxxxxxx>
>> ---
>> Changes from v1:
>> Rename CONFIG_ZSWAP_DEFAULT_COMP_* to CONFIG_ZSWAP_COMPRESSOR_DEFAULT_*
>> and CONFIG_ZSWAP_DEFAULT_ZPOOL_* to CONFIG_ZSWAP_ZPOOL_DEFAULT_* while
>> dropping the "_NAME" suffix from the final string option in both cases.
>>
>>  mm/Kconfig | 103 ++++++++++++++++++++++++++++++++++++++++++++++++++++-
>>  mm/zswap.c |  26 ++++++++------
>>  2 files changed, 117 insertions(+), 12 deletions(-)
>>
>> diff --git a/mm/Kconfig b/mm/Kconfig
>> index a5dae9a7eb51..267316941324 100644
>> --- a/mm/Kconfig
>> +++ b/mm/Kconfig
>> @@ -525,7 +525,6 @@ config MEM_SOFT_DIRTY
>>  config ZSWAP
>>  	bool "Compressed cache for swap pages (EXPERIMENTAL)"
>>  	depends on FRONTSWAP && CRYPTO=y
>> -	select CRYPTO_LZO
>>  	select ZPOOL
>>  	help
>>  	  A lightweight compressed cache for swap pages.  It takes
>> @@ -541,6 +540,108 @@ config ZSWAP
>>  	  they have not be fully explored on the large set of potential
>>  	  configurations and workloads that exist.
>>  
>> +choice
>> +	prompt "Compressed cache for swap pages default compressor"
>> +	depends on ZSWAP
>> +	default ZSWAP_COMPRESSOR_DEFAULT_LZO
>> +	help
>> +	  Selects the default compression algorithm for the compressed cache
>> +	  for swap pages.
>> +	  If in doubt, select 'LZO'.
> 
> Could it e.g. suggest which one is the fastest and which most space
> efficient?

While even the algorithms themselves (CRYTPO_{DEFLATE,LZO,..}) don't
have any particular descriptions with respect to their performance I
guess we can add a reference to, for example, a recent benchmark of
various in-kernel algorithms (when used for zram) that is available at
https://lwn.net/Articles/751795/.
This way the user will have at least some kind of reference.

It is also worth noting here that similar CONFIG_PSTORE_* and
CONFIG_RD_* options don't have detailed descriptions either.

> Also does this cover all compression algorithms?

Yes.

> Are we going to add new
> options if there are new ones? Wouldn't a string instead of choice be
> sufficient here?

If this is changed to a string prompt we'll lose automatic pulling in of
an appropriate CONFIG_CRYPTO_* dependency.

Other similar options are presented as a choice, too, see for example
CONFIG_KERNEL_{GZIP,BZIP2,LZMA,XZ,LZO,LZ4,UNCOMPRESSED} and
CONFIG_INITRAMFS_COMPRESSION_{NONE,GZIP,BZIP2,LZMA,XZ,LZO,LZ4} (that
maps to an extension string table called CONFIG_INITRAMFS_COMPRESSION).

New compression algorithm being added to the kernel is a rare event,
I also envision that not every new algorithm will need to be presented
as a choice for zswap default.

>> +
>> +choice
>> +	prompt "Compressed cache for swap pages default allocator"
>> +	depends on ZSWAP
>> +	default ZSWAP_ZPOOL_DEFAULT_ZBUD
>> +	help
>> +	  Selects the default allocator for the compressed cache for
>> +	  swap pages.
>> +	  The default is 'zbud' for compatibility, however please do
>> +	  read the description of each of the allocators below before
>> +	  making a right choice.
> 
> It's somewhat unfortunate that the choice options don't include the
> description and one has to go look for it elsewhere.

Could copy the allocator descriptions into these choice options but
since the allocators themselves are literally the next set of options
in Kconfig are their descriptions really worth repeating here?

> Also, shouldn't the list include zsmalloc?

You are right, will add zsmalloc in a respin.

>> +	  The selection made here can be overridden by using the kernel
>> +	  command line 'zswap.zpool=' option.
>> +
>> +config ZSWAP_ZPOOL_DEFAULT_ZBUD
>> +	bool "zbud"
>> +	select ZBUD
>> +	help
>> +	  Use the zbud allocator as the default allocator.
>> +
>> +config ZSWAP_ZPOOL_DEFAULT_Z3FOLD
>> +	bool "z3fold"
>> +	select Z3FOLD
>> +	help
>> +	  Use the z3fold allocator as the default allocator.
>> +endchoice
>> +
>> +config ZSWAP_ZPOOL_DEFAULT
>> +       string
>> +       default "zbud" if ZSWAP_ZPOOL_DEFAULT_ZBUD
>> +       default "z3fold" if ZSWAP_ZPOOL_DEFAULT_Z3FOLD
>> +       default ""
>> +
>> +config ZSWAP_DEFAULT_ON
>> +	bool "Enable the compressed cache for swap pages by default"
>> +	depends on ZSWAP
>> +	help
>> +	  If selected, the compressed cache for swap pages will be enabled
>> +	  at boot, otherwise it will be disabled.
>> +
>> +	  The selection made here can be overridden by using the kernel
>> +	  command line 'zswap.enabled=' option.
>> +
>>  config ZPOOL
>>  	tristate "Common API for compressed memory storage"
>>  	help
>> diff --git a/mm/zswap.c b/mm/zswap.c
>> index 46a322316e52..71795b6f5b71 100644
>> --- a/mm/zswap.c

Thanks,
Maciej