Re: [PATCHv3 1/4] zram: introduce compressing backend abstraction

[Minchan Kim]

Hello Sergey,

On Wed, Feb 12, 2014 at 01:21:03AM +0300, Sergey Senozhatsky wrote:
> ZRAM performs direct LZO compression algorithm calls, making it the one
> and only option. Introduce abstract struct zram_comp in order to support
> multiple compression algorithms. struct zram_comp defines the following
> set of compressing backend operations:
>         .create
>         .destroy
>         .compress
>         .decompress
> 
> Schematically zram write() usually contains the following steps:
> 0) preparation (decompression of partioal IO, etc.)
> 1) lock buffer_lock mutex (protects meta compress buffers)
> 2) compress (using meta compress buffers)
> 3) alloc and map zs_pool object
> 4) copy compressed data (from meta compress buffers) to object allocated by 3)
> 5) free previous pool page, assign a new one
> 6) unlock buffer_lock mutex
> 
> As we can see, compressing buffers must remain untouched from 1) to 4),
> because, otherwise, concurrent write() can overwrite data. At the same
> time, zram_meta must be aware of a) specific compression algorithm
> memory requirements and b) necessary locking to protect compression
> buffers. Besides, zram holds buffer_lock almost through the whole write()
> function, making parallel compression impossible. To remove requirement
> a) new struct zcomp_workmem introduced, which contain buffers required by
> compression algorithm. While struct zram_comp implements workmem handling
> and locking by means of get() and put() semantics and removes requirement
> b) from zram meta. zram_comp ->create() and ->destroy(), respectively,
> allocate and deallocate algorithm specific zcomp_workmem `private' buffer.
> 
> Every zram_comp has a list of idle workmem buffers (at least 1 workmem),
> spinlock to protect idle list and wait queue, making it possible to perform
> parallel compressions. Each time zram issues a zcomp_workmem_get() call, the
> following set of operations performed:
> - spin lock buffer_lock
> - if idle list is not empty, remove workmem from idle list, spin
>   unlock and return workmem pointer
> - if idle list is empty, current adds itself to wait queue. it will be
>   awaken by zcomp_workmem_put() caller.
> 
> zcomp_workmem_put():
> - spin lock buffer_lock
> - add workmem to idle list
> - spin unlock, wake up sleeper (if any)
> 
> In other words, zcomp_workmem_get() turns caller into exclusive user of workmem
> and zcomp_workem_put() makes a particular workmem available.
> 
> Usage examples.
> 
> To initialize compressing backend:
> 	comp = zcomp_create(NAME) /* NAME e.g. lzo */
> 
> which initialises compressing backend if requested algorithm is supported.
> 
> Compress:
> 	wm = zcomp_workmem_get(comp)
> 	zcomp_compress(comp, wm, src, src_len, &dst_len)
> 	[..] /* copy compressed data */
> 	zcomp_workmem_put(comp, wm)
> 
> Decompress:
> 	zcomp_decompress(comp, src, src_len, dst, &dst_len);
> 
> Free compessing backend and its workmem:
> 	zcomp_destroy(comp)
> 
> Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@xxxxxxxxx>
> ---
>  drivers/block/zram/zcomp_lzo.c |  33 +++++++++
>  drivers/block/zram/zram_comp.c | 147 +++++++++++++++++++++++++++++++++++++++++
>  drivers/block/zram/zram_comp.h |  58 ++++++++++++++++
>  3 files changed, 238 insertions(+)
>  create mode 100644 drivers/block/zram/zcomp_lzo.c
>  create mode 100644 drivers/block/zram/zram_comp.c
>  create mode 100644 drivers/block/zram/zram_comp.h
> 
> diff --git a/drivers/block/zram/zcomp_lzo.c b/drivers/block/zram/zcomp_lzo.c
> new file mode 100644
> index 0000000..bbde74e
> --- /dev/null
> +++ b/drivers/block/zram/zcomp_lzo.c
> @@ -0,0 +1,33 @@
> +/*
> + * Copyright (C) 2014 Sergey Senozhatsky.
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU General Public License
> + * as published by the Free Software Foundation; either version
> + * 2 of the License, or (at your option) any later version.
> + */
> +
> +#include <linux/kernel.h>
> +#include <linux/slab.h>
> +#include <linux/lzo.h>
> +
> +#include "zram_comp.h"
> +
> +static void * lzo_create(void)
> +{
> +	return kmalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL);
> +}
> +
> +static void lzo_destroy(void *workmem)
> +{
> +	kfree(workmem);
> +}

It's okay to name "workmem" for lzo because lzo needs only work buffer to work
but let's think about zlib. It needs z_streamp which has many feilds
to work as well as work memory. That's why I don't like "workmem" name which
is very specific. Anyway, it's okay for lzo backend now.

> +
> +extern struct zram_comp_backend zcomp_lzo;
> +struct zram_comp_backend zcomp_lzo = {
> +	.compress = lzo1x_1_compress,
> +	.decompress = lzo1x_decompress_safe,
> +	.create = lzo_create,
> +	.destroy = lzo_destroy,
> +	.name = "lzo",
> +};
> diff --git a/drivers/block/zram/zram_comp.c b/drivers/block/zram/zram_comp.c
> new file mode 100644
> index 0000000..139a468
> --- /dev/null
> +++ b/drivers/block/zram/zram_comp.c
> @@ -0,0 +1,147 @@
> +/*
> + * Copyright (C) 2014 Sergey Senozhatsky.
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU General Public License
> + * as published by the Free Software Foundation; either version
> + * 2 of the License, or (at your option) any later version.
> + */
> +
> +#include <linux/kernel.h>
> +#include <linux/string.h>
> +#include <linux/slab.h>
> +#include <linux/vmalloc.h>
> +#include <linux/wait.h>
> +#include <linux/sched.h>
> +
> +#include "zram_comp.h"
> +
> +extern struct zram_comp_backend zcomp_lzo;
> +
> +static void workmem_free(struct zram_comp *comp, struct zcomp_workmem *workmem)
> +{
> +	comp->backend->destroy(workmem->private);
> +	free_pages((unsigned long)workmem->buffer, 1);
> +	kfree(workmem);
> +}
> +
> +/* allocate new workmem structure with ->mem of requested size,
> + * return NULL on error */
> +static struct zcomp_workmem *workmem_alloc(struct zram_comp *comp)
> +{
> +	struct zcomp_workmem *workmem = kmalloc(sizeof(*workmem), GFP_NOFS);
> +	if (!workmem)
> +		return NULL;
> +
> +	INIT_LIST_HEAD(&workmem->list);
> +	/* algorithm specific working memory buffer */
> +	workmem->private = comp->backend->create();
> +	/* allocate 2 pages. 1 for compressed data, plus 1 extra for the
> +	 * case when compressed size is larger than the original one. */
> +	workmem->buffer = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 1);
> +	if (!workmem->private || !workmem->buffer)
> +		goto fail;
> +
> +	return workmem;
> +fail:
> +	workmem_free(comp, workmem);
> +	return NULL;
> +}

Hmm, you introduced workmem again. :(
As I said above, workmem is too specific.
It's only valid for lzo and lz4 so I don't want to expose workmem out of zcomp
and that's why I suggested zcomp_strm or zcomp_stream.
It seems "stream" is more general term for compressor.

> +
> +/* get existing idle workmem or wait until other process release
> + * (workmem_put()) one for us */
> +struct zcomp_workmem *zcomp_workmem_get(struct zram_comp *comp)
> +{
> +	struct zcomp_workmem *wm;
> +retry:
> +	spin_lock(&comp->buffer_lock);
> +	if (list_empty(&comp->idle_workmem)) {
> +		spin_unlock(&comp->buffer_lock);
> +		wait_event(comp->workmem_wait,
> +				!list_empty(&comp->idle_workmem));
> +		goto retry;
> +	}
> +
> +	wm = list_entry(comp->idle_workmem.next,
> +			struct zcomp_workmem, list);
> +	list_del(&wm->list);
> +	spin_unlock(&comp->buffer_lock);
> +	return wm;
> +}
> +
> +/* add workmem back to idle list and wake up waiter (if any) */
> +void zcomp_workmem_put(struct zram_comp *comp,
> +		struct zcomp_workmem *workmem)
> +{
> +	spin_lock(&comp->buffer_lock);
> +	list_add_tail(&workmem->list, &comp->idle_workmem);

Why do you implement queue instead of stack?

> +	spin_unlock(&comp->buffer_lock);
> +
> +	if (waitqueue_active(&comp->workmem_wait))

Possible losting wakeup?

> +		wake_up(&comp->workmem_wait);
> +}
> +
> +int zcomp_compress(struct zram_comp *comp, struct zcomp_workmem *workmem,
> +		const unsigned char *src, size_t src_len, size_t *dst_len)
> +{

As I said, I don't like zcomp_workmem.
Please use zcomp_strm and let's make naming clear.
Something is zcomp and sometime is zram_comp.
I prefer zcomp but it depends on you.

> +	return comp->backend->compress(src, src_len, workmem->buffer,
> +			dst_len, workmem->private);
> +}
> +
> +int zcomp_decompress(struct zram_comp *comp, const unsigned char *src,
> +		size_t src_len, unsigned char *dst, size_t *dst_len)
> +{
> +	return comp->backend->decompress(src, src_len, dst, dst_len);
> +}
> +
> +/* free allocated workmem buffers and zram_comp */
> +void zcomp_destroy(struct zram_comp *comp)
> +{
> +	struct zcomp_workmem *wm;
> +	while (!list_empty(&comp->idle_workmem)) {
> +		wm = list_entry(comp->idle_workmem.next,
> +				struct zcomp_workmem, list);
> +		list_del(&wm->list);
> +		workmem_free(comp, wm);
> +	}
> +	kfree(comp);
> +}
> +
> +static struct zram_comp_backend *find_backend(const char *compress)
> +{
> +	if (sysfs_streq(compress, "lzo"))
> +		return &zcomp_lzo;
> +	return NULL;
> +}
> +
> +/* search available compressors for requested algorithm.

Please correct comment style.

> + * allocate new zram_comp and initialize it. return NULL
> + * if requested algorithm is not supported or in case
> + * of init error */
> +struct zram_comp *zcomp_create(const char *compress)
> +{
> +	struct zram_comp *comp;
> +	struct zram_comp_backend *backend;
> +	struct zcomp_workmem *wm;
> +
> +	backend = find_backend(compress);
> +	if (!backend)
> +		return NULL;
> +
> +	comp = kzalloc(sizeof(struct zram_comp), GFP_KERNEL);
> +	if (!comp)
> +		return NULL;
> +
> +	comp->backend = backend;
> +	spin_lock_init(&comp->buffer_lock);
> +	INIT_LIST_HEAD(&comp->idle_workmem);
> +	init_waitqueue_head(&comp->workmem_wait);
> +
> +	wm = workmem_alloc(comp);
> +	if (!wm) {
> +		zcomp_destroy(comp);
> +		return NULL;
> +	}
> +	list_add_tail(&wm->list, &comp->idle_workmem);
> +	return comp;
> +}
> diff --git a/drivers/block/zram/zram_comp.h b/drivers/block/zram/zram_comp.h
> new file mode 100644
> index 0000000..90355e8
> --- /dev/null
> +++ b/drivers/block/zram/zram_comp.h
> @@ -0,0 +1,58 @@
> +/*
> + * Copyright (C) 2014 Sergey Senozhatsky.
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU General Public License
> + * as published by the Free Software Foundation; either version
> + * 2 of the License, or (at your option) any later version.
> + */
> +
> +#ifndef _ZRAM_COMP_H_
> +#define _ZRAM_COMP_H_
> +
> +#include <linux/types.h>
> +#include <linux/spinlock.h>
> +
> +struct zcomp_workmem {
> +	void *buffer;     /* compression/decompression buffer */
> +	void *private;      /* algorithm workmem */
> +	struct list_head list;
> +};
> +
> +/* static compression backend */
> +struct zram_comp_backend {
> +	int (*compress)(const unsigned char *src, size_t src_len,
> +			unsigned char *dst, size_t *dst_len, void *wrkmem);

zram works based on PAGE so I like compress_page so maybe we couldn't
reduce a argument "src_len".

> +
> +	int (*decompress)(const unsigned char *src, size_t src_len,
> +			unsigned char *dst, size_t *dst_len);
> +
> +	void *(*create)(void);
> +	void (*destroy)(void *workmem);
> +
> +	const char *name;
> +};
> +
> +/* dynamic per-device compression frontend */
> +struct zram_comp {
> +	/* protect workmem list */
> +	spinlock_t buffer_lock;
> +	/* list of available workmems */
> +	struct list_head idle_workmem;
> +	wait_queue_head_t workmem_wait;
> +	struct zram_comp_backend *backend;
> +};
> +
> +struct zram_comp *zcomp_create(const char *comp);
> +void zcomp_destroy(struct zram_comp *comp);
> +
> +struct zcomp_workmem *zcomp_workmem_get(struct zram_comp *comp);
> +void zcomp_workmem_put(struct zram_comp *comp,
> +		struct zcomp_workmem *workmem);
> +
> +int zcomp_compress(struct zram_comp *comp, struct zcomp_workmem *workmem,
> +		const unsigned char *src, size_t src_len, size_t *dst_len);
> +
> +int zcomp_decompress(struct zram_comp *comp, const unsigned char *src,
> +		size_t src_len, unsigned char *dst, size_t *dst_len);
> +#endif /* _ZRAM_COMP_H_ */
> -- 
> 1.9.0.rc3.244.g3497008
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

-- 
Kind regards,
Minchan Kim
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/