1 files changed, 531 insertions, 0 deletions

diff --git a/drivers/crypto/nx/nx-842.c b/drivers/crypto/nx/nx-842.c
new file mode 100644
index 000000000..d94e25df5
--- /dev/null
+++ b/drivers/crypto/nx/nx-842.c
@@ -0,0 +1,531 @@
+/*
+ * Cryptographic API for the NX-842 hardware compression.
+ *
+ * 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.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * Copyright (C) IBM Corporation, 2011-2015
+ *
+ * Designer of the Power data compression engine:
+ *   Bulent Abali <abali@us.ibm.com>
+ *
+ * Original Authors: Robert Jennings <rcj@linux.vnet.ibm.com>
+ *                   Seth Jennings <sjenning@linux.vnet.ibm.com>
+ *
+ * Rewrite: Dan Streetman <ddstreet@ieee.org>
+ *
+ * This is an interface to the NX-842 compression hardware in PowerPC
+ * processors.  Most of the complexity of this drvier is due to the fact that
+ * the NX-842 compression hardware requires the input and output data buffers
+ * to be specifically aligned, to be a specific multiple in length, and within
+ * specific minimum and maximum lengths.  Those restrictions, provided by the
+ * nx-842 driver via nx842_constraints, mean this driver must use bounce
+ * buffers and headers to correct misaligned in or out buffers, and to split
+ * input buffers that are too large.
+ *
+ * This driver will fall back to software decompression if the hardware
+ * decompression fails, so this driver's decompression should never fail as
+ * long as the provided compressed buffer is valid.  Any compressed buffer
+ * created by this driver will have a header (except ones where the input
+ * perfectly matches the constraints); so users of this driver cannot simply
+ * pass a compressed buffer created by this driver over to the 842 software
+ * decompression library.  Instead, users must use this driver to decompress;
+ * if the hardware fails or is unavailable, the compressed buffer will be
+ * parsed and the header removed, and the raw 842 buffer(s) passed to the 842
+ * software decompression library.
+ *
+ * This does not fall back to software compression, however, since the caller
+ * of this function is specifically requesting hardware compression; if the
+ * hardware compression fails, the caller can fall back to software
+ * compression, and the raw 842 compressed buffer that the software compressor
+ * creates can be passed to this driver for hardware decompression; any
+ * buffer without our specific header magic is assumed to be a raw 842 buffer
+ * and passed directly to the hardware.  Note that the software compression
+ * library will produce a compressed buffer that is incompatible with the
+ * hardware decompressor if the original input buffer length is not a multiple
+ * of 8; if such a compressed buffer is passed to this driver for
+ * decompression, the hardware will reject it and this driver will then pass
+ * it over to the software library for decompression.
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/vmalloc.h>
+#include <linux/sw842.h>
+#include <linux/spinlock.h>
+
+#include "nx-842.h"
+
+/* The first 5 bits of this magic are 0x1f, which is an invalid 842 5-bit
+ * template (see lib/842/842.h), so this magic number will never appear at
+ * the start of a raw 842 compressed buffer.  That is important, as any buffer
+ * passed to us without this magic is assumed to be a raw 842 compressed
+ * buffer, and passed directly to the hardware to decompress.
+ */
+#define NX842_CRYPTO_MAGIC	(0xf842)
+#define NX842_CRYPTO_HEADER_SIZE(g)				\
+	(sizeof(struct nx842_crypto_header) +			\
+	 sizeof(struct nx842_crypto_header_group) * (g))
+#define NX842_CRYPTO_HEADER_MAX_SIZE				\
+	NX842_CRYPTO_HEADER_SIZE(NX842_CRYPTO_GROUP_MAX)
+
+/* bounce buffer size */
+#define BOUNCE_BUFFER_ORDER	(2)
+#define BOUNCE_BUFFER_SIZE					\
+	((unsigned int)(PAGE_SIZE << BOUNCE_BUFFER_ORDER))
+
+/* try longer on comp because we can fallback to sw decomp if hw is busy */
+#define COMP_BUSY_TIMEOUT	(250) /* ms */
+#define DECOMP_BUSY_TIMEOUT	(50) /* ms */
+
+struct nx842_crypto_param {
+	u8 *in;
+	unsigned int iremain;
+	u8 *out;
+	unsigned int oremain;
+	unsigned int ototal;
+};
+
+static int update_param(struct nx842_crypto_param *p,
+			unsigned int slen, unsigned int dlen)
+{
+	if (p->iremain < slen)
+		return -EOVERFLOW;
+	if (p->oremain < dlen)
+		return -ENOSPC;
+
+	p->in += slen;
+	p->iremain -= slen;
+	p->out += dlen;
+	p->oremain -= dlen;
+	p->ototal += dlen;
+
+	return 0;
+}
+
+int nx842_crypto_init(struct crypto_tfm *tfm, struct nx842_driver *driver)
+{
+	struct nx842_crypto_ctx *ctx = crypto_tfm_ctx(tfm);
+
+	spin_lock_init(&ctx->lock);
+	ctx->driver = driver;
+	ctx->wmem = kmalloc(driver->workmem_size, GFP_KERNEL);
+	ctx->sbounce = (u8 *)__get_free_pages(GFP_KERNEL, BOUNCE_BUFFER_ORDER);
+	ctx->dbounce = (u8 *)__get_free_pages(GFP_KERNEL, BOUNCE_BUFFER_ORDER);
+	if (!ctx->wmem || !ctx->sbounce || !ctx->dbounce) {
+		kfree(ctx->wmem);
+		free_page((unsigned long)ctx->sbounce);
+		free_page((unsigned long)ctx->dbounce);
+		return -ENOMEM;
+	}
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(nx842_crypto_init);
+
+void nx842_crypto_exit(struct crypto_tfm *tfm)
+{
+	struct nx842_crypto_ctx *ctx = crypto_tfm_ctx(tfm);
+
+	kfree(ctx->wmem);
+	free_page((unsigned long)ctx->sbounce);
+	free_page((unsigned long)ctx->dbounce);
+}
+EXPORT_SYMBOL_GPL(nx842_crypto_exit);
+
+static void check_constraints(struct nx842_constraints *c)
+{
+	/* limit maximum, to always have enough bounce buffer to decompress */
+	if (c->maximum > BOUNCE_BUFFER_SIZE)
+		c->maximum = BOUNCE_BUFFER_SIZE;
+}
+
+static int nx842_crypto_add_header(struct nx842_crypto_header *hdr, u8 *buf)
+{
+	int s = NX842_CRYPTO_HEADER_SIZE(hdr->groups);
+
+	/* compress should have added space for header */
+	if (s > be16_to_cpu(hdr->group[0].padding)) {
+		pr_err("Internal error: no space for header\n");
+		return -EINVAL;
+	}
+
+	memcpy(buf, hdr, s);
+
+	print_hex_dump_debug("header ", DUMP_PREFIX_OFFSET, 16, 1, buf, s, 0);
+
+	return 0;
+}
+
+static int compress(struct nx842_crypto_ctx *ctx,
+		    struct nx842_crypto_param *p,
+		    struct nx842_crypto_header_group *g,
+		    struct nx842_constraints *c,
+		    u16 *ignore,
+		    unsigned int hdrsize)
+{
+	unsigned int slen = p->iremain, dlen = p->oremain, tmplen;
+	unsigned int adj_slen = slen;
+	u8 *src = p->in, *dst = p->out;
+	int ret, dskip = 0;
+	ktime_t timeout;
+
+	if (p->iremain == 0)
+		return -EOVERFLOW;
+
+	if (p->oremain == 0 || hdrsize + c->minimum > dlen)
+		return -ENOSPC;
+
+	if (slen % c->multiple)
+		adj_slen = round_up(slen, c->multiple);
+	if (slen < c->minimum)
+		adj_slen = c->minimum;
+	if (slen > c->maximum)
+		adj_slen = slen = c->maximum;
+	if (adj_slen > slen || (u64)src % c->alignment) {
+		adj_slen = min(adj_slen, BOUNCE_BUFFER_SIZE);
+		slen = min(slen, BOUNCE_BUFFER_SIZE);
+		if (adj_slen > slen)
+			memset(ctx->sbounce + slen, 0, adj_slen - slen);
+		memcpy(ctx->sbounce, src, slen);
+		src = ctx->sbounce;
+		slen = adj_slen;
+		pr_debug("using comp sbounce buffer, len %x\n", slen);
+	}
+
+	dst += hdrsize;
+	dlen -= hdrsize;
+
+	if ((u64)dst % c->alignment) {
+		dskip = (int)(PTR_ALIGN(dst, c->alignment) - dst);
+		dst += dskip;
+		dlen -= dskip;
+	}
+	if (dlen % c->multiple)
+		dlen = round_down(dlen, c->multiple);
+	if (dlen < c->minimum) {
+nospc:
+		dst = ctx->dbounce;
+		dlen = min(p->oremain, BOUNCE_BUFFER_SIZE);
+		dlen = round_down(dlen, c->multiple);
+		dskip = 0;
+		pr_debug("using comp dbounce buffer, len %x\n", dlen);
+	}
+	if (dlen > c->maximum)
+		dlen = c->maximum;
+
+	tmplen = dlen;
+	timeout = ktime_add_ms(ktime_get(), COMP_BUSY_TIMEOUT);
+	do {
+		dlen = tmplen; /* reset dlen, if we're retrying */
+		ret = ctx->driver->compress(src, slen, dst, &dlen, ctx->wmem);
+		/* possibly we should reduce the slen here, instead of
+		 * retrying with the dbounce buffer?
+		 */
+		if (ret == -ENOSPC && dst != ctx->dbounce)
+			goto nospc;
+	} while (ret == -EBUSY && ktime_before(ktime_get(), timeout));
+	if (ret)
+		return ret;
+
+	dskip += hdrsize;
+
+	if (dst == ctx->dbounce)
+		memcpy(p->out + dskip, dst, dlen);
+
+	g->padding = cpu_to_be16(dskip);
+	g->compressed_length = cpu_to_be32(dlen);
+	g->uncompressed_length = cpu_to_be32(slen);
+
+	if (p->iremain < slen) {
+		*ignore = slen - p->iremain;
+		slen = p->iremain;
+	}
+
+	pr_debug("compress slen %x ignore %x dlen %x padding %x\n",
+		 slen, *ignore, dlen, dskip);
+
+	return update_param(p, slen, dskip + dlen);
+}
+
+int nx842_crypto_compress(struct crypto_tfm *tfm,
+			  const u8 *src, unsigned int slen,
+			  u8 *dst, unsigned int *dlen)
+{
+	struct nx842_crypto_ctx *ctx = crypto_tfm_ctx(tfm);
+	struct nx842_crypto_header *hdr = &ctx->header;
+	struct nx842_crypto_param p;
+	struct nx842_constraints c = *ctx->driver->constraints;
+	unsigned int groups, hdrsize, h;
+	int ret, n;
+	bool add_header;
+	u16 ignore = 0;
+
+	check_constraints(&c);
+
+	p.in = (u8 *)src;
+	p.iremain = slen;
+	p.out = dst;
+	p.oremain = *dlen;
+	p.ototal = 0;
+
+	*dlen = 0;
+
+	groups = min_t(unsigned int, NX842_CRYPTO_GROUP_MAX,
+		       DIV_ROUND_UP(p.iremain, c.maximum));
+	hdrsize = NX842_CRYPTO_HEADER_SIZE(groups);
+
+	spin_lock_bh(&ctx->lock);
+
+	/* skip adding header if the buffers meet all constraints */
+	add_header = (p.iremain % c.multiple	||
+		      p.iremain < c.minimum	||
+		      p.iremain > c.maximum	||
+		      (u64)p.in % c.alignment	||
+		      p.oremain % c.multiple	||
+		      p.oremain < c.minimum	||
+		      p.oremain > c.maximum	||
+		      (u64)p.out % c.alignment);
+
+	hdr->magic = cpu_to_be16(NX842_CRYPTO_MAGIC);
+	hdr->groups = 0;
+	hdr->ignore = 0;
+
+	while (p.iremain > 0) {
+		n = hdr->groups++;
+		ret = -ENOSPC;
+		if (hdr->groups > NX842_CRYPTO_GROUP_MAX)
+			goto unlock;
+
+		/* header goes before first group */
+		h = !n && add_header ? hdrsize : 0;
+
+		if (ignore)
+			pr_warn("internal error, ignore is set %x\n", ignore);
+
+		ret = compress(ctx, &p, &hdr->group[n], &c, &ignore, h);
+		if (ret)
+			goto unlock;
+	}
+
+	if (!add_header && hdr->groups > 1) {
+		pr_err("Internal error: No header but multiple groups\n");
+		ret = -EINVAL;
+		goto unlock;
+	}
+
+	/* ignore indicates the input stream needed to be padded */
+	hdr->ignore = cpu_to_be16(ignore);
+	if (ignore)
+		pr_debug("marked %d bytes as ignore\n", ignore);
+
+	if (add_header)
+		ret = nx842_crypto_add_header(hdr, dst);
+	if (ret)
+		goto unlock;
+
+	*dlen = p.ototal;
+
+	pr_debug("compress total slen %x dlen %x\n", slen, *dlen);
+
+unlock:
+	spin_unlock_bh(&ctx->lock);
+	return ret;
+}
+EXPORT_SYMBOL_GPL(nx842_crypto_compress);
+
+static int decompress(struct nx842_crypto_ctx *ctx,
+		      struct nx842_crypto_param *p,
+		      struct nx842_crypto_header_group *g,
+		      struct nx842_constraints *c,
+		      u16 ignore)
+{
+	unsigned int slen = be32_to_cpu(g->compressed_length);
+	unsigned int required_len = be32_to_cpu(g->uncompressed_length);
+	unsigned int dlen = p->oremain, tmplen;
+	unsigned int adj_slen = slen;
+	u8 *src = p->in, *dst = p->out;
+	u16 padding = be16_to_cpu(g->padding);
+	int ret, spadding = 0, dpadding = 0;
+	ktime_t timeout;
+
+	if (!slen || !required_len)
+		return -EINVAL;
+
+	if (p->iremain <= 0 || padding + slen > p->iremain)
+		return -EOVERFLOW;
+
+	if (p->oremain <= 0 || required_len - ignore > p->oremain)
+		return -ENOSPC;
+
+	src += padding;
+
+	if (slen % c->multiple)
+		adj_slen = round_up(slen, c->multiple);
+	if (slen < c->minimum)
+		adj_slen = c->minimum;
+	if (slen > c->maximum)
+		goto usesw;
+	if (slen < adj_slen || (u64)src % c->alignment) {
+		/* we can append padding bytes because the 842 format defines
+		 * an "end" template (see lib/842/842_decompress.c) and will
+		 * ignore any bytes following it.
+		 */
+		if (slen < adj_slen)
+			memset(ctx->sbounce + slen, 0, adj_slen - slen);
+		memcpy(ctx->sbounce, src, slen);
+		src = ctx->sbounce;
+		spadding = adj_slen - slen;
+		slen = adj_slen;
+		pr_debug("using decomp sbounce buffer, len %x\n", slen);
+	}
+
+	if (dlen % c->multiple)
+		dlen = round_down(dlen, c->multiple);
+	if (dlen < required_len || (u64)dst % c->alignment) {
+		dst = ctx->dbounce;
+		dlen = min(required_len, BOUNCE_BUFFER_SIZE);
+		pr_debug("using decomp dbounce buffer, len %x\n", dlen);
+	}
+	if (dlen < c->minimum)
+		goto usesw;
+	if (dlen > c->maximum)
+		dlen = c->maximum;
+
+	tmplen = dlen;
+	timeout = ktime_add_ms(ktime_get(), DECOMP_BUSY_TIMEOUT);
+	do {
+		dlen = tmplen; /* reset dlen, if we're retrying */
+		ret = ctx->driver->decompress(src, slen, dst, &dlen, ctx->wmem);
+	} while (ret == -EBUSY && ktime_before(ktime_get(), timeout));
+	if (ret) {
+usesw:
+		/* reset everything, sw doesn't have constraints */
+		src = p->in + padding;
+		slen = be32_to_cpu(g->compressed_length);
+		spadding = 0;
+		dst = p->out;
+		dlen = p->oremain;
+		dpadding = 0;
+		if (dlen < required_len) { /* have ignore bytes */
+			dst = ctx->dbounce;
+			dlen = BOUNCE_BUFFER_SIZE;
+		}
+		pr_info_ratelimited("using software 842 decompression\n");
+		ret = sw842_decompress(src, slen, dst, &dlen);
+	}
+	if (ret)
+		return ret;
+
+	slen -= spadding;
+
+	dlen -= ignore;
+	if (ignore)
+		pr_debug("ignoring last %x bytes\n", ignore);
+
+	if (dst == ctx->dbounce)
+		memcpy(p->out, dst, dlen);
+
+	pr_debug("decompress slen %x padding %x dlen %x ignore %x\n",
+		 slen, padding, dlen, ignore);
+
+	return update_param(p, slen + padding, dlen);
+}
+
+int nx842_crypto_decompress(struct crypto_tfm *tfm,
+			    const u8 *src, unsigned int slen,
+			    u8 *dst, unsigned int *dlen)
+{
+	struct nx842_crypto_ctx *ctx = crypto_tfm_ctx(tfm);
+	struct nx842_crypto_header *hdr;
+	struct nx842_crypto_param p;
+	struct nx842_constraints c = *ctx->driver->constraints;
+	int n, ret, hdr_len;
+	u16 ignore = 0;
+
+	check_constraints(&c);
+
+	p.in = (u8 *)src;
+	p.iremain = slen;
+	p.out = dst;
+	p.oremain = *dlen;
+	p.ototal = 0;
+
+	*dlen = 0;
+
+	hdr = (struct nx842_crypto_header *)src;
+
+	spin_lock_bh(&ctx->lock);
+
+	/* If it doesn't start with our header magic number, assume it's a raw
+	 * 842 compressed buffer and pass it directly to the hardware driver
+	 */
+	if (be16_to_cpu(hdr->magic) != NX842_CRYPTO_MAGIC) {
+		struct nx842_crypto_header_group g = {
+			.padding =		0,
+			.compressed_length =	cpu_to_be32(p.iremain),
+			.uncompressed_length =	cpu_to_be32(p.oremain),
+		};
+
+		ret = decompress(ctx, &p, &g, &c, 0);
+		if (ret)
+			goto unlock;
+
+		goto success;
+	}
+
+	if (!hdr->groups) {
+		pr_err("header has no groups\n");
+		ret = -EINVAL;
+		goto unlock;
+	}
+	if (hdr->groups > NX842_CRYPTO_GROUP_MAX) {
+		pr_err("header has too many groups %x, max %x\n",
+		       hdr->groups, NX842_CRYPTO_GROUP_MAX);
+		ret = -EINVAL;
+		goto unlock;
+	}
+
+	hdr_len = NX842_CRYPTO_HEADER_SIZE(hdr->groups);
+	if (hdr_len > slen) {
+		ret = -EOVERFLOW;
+		goto unlock;
+	}
+
+	memcpy(&ctx->header, src, hdr_len);
+	hdr = &ctx->header;
+
+	for (n = 0; n < hdr->groups; n++) {
+		/* ignore applies to last group */
+		if (n + 1 == hdr->groups)
+			ignore = be16_to_cpu(hdr->ignore);
+
+		ret = decompress(ctx, &p, &hdr->group[n], &c, ignore);
+		if (ret)
+			goto unlock;
+	}
+
+success:
+	*dlen = p.ototal;
+
+	pr_debug("decompress total slen %x dlen %x\n", slen, *dlen);
+
+	ret = 0;
+
+unlock:
+	spin_unlock_bh(&ctx->lock);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(nx842_crypto_decompress);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("IBM PowerPC Nest (NX) 842 Hardware Compression Driver");
+MODULE_AUTHOR("Dan Streetman <ddstreet@ieee.org>");