[PATCH 06/13] async_tx: add support for asynchronous GF multiplication
From: Dan Williams
Date: Wed Mar 18 2009 - 15:22:50 EST
[ Based on an original patch by Yuri Tikhonov ]
This adds support for doing asynchronous GF multiplication by adding
four additional functions to the async_tx API:
async_pq() does simultaneous XOR of sources and XOR of sources
GF-multiplied by given coefficients.
async_pq_zero_sum() checks if results of calculations match given
ones.
async_gen_syndrome() does simultaneous XOR and R/S syndrome of sources.
async_syndrome_zerosum() checks if results of XOR/syndrome calculation
matches given ones.
Latter two functions just use async_pq() with the appropriate coefficients
in asynchronous case but have significant optimizations in the synchronous
case.
When a request is made to run async_pq against more than the hardware
maximum number of supported sources we need to reuse the previous
generated P and Q values as sources into the next operation. Care must
be taken to remove Q from P' and P from Q'. For example to perform a 5
source pq op with hardware that only supports 4 sources at a time the
following approach is taken:
p, q = PQ(src0, src1, src2, src3, COEF({01}, {02}, {04}, {08}))
p', q' = PQ(p, q, q, src4, COEF({00}, {01}, {00}, {10}))
p' = p + q + q + src4 = p + src4
q' = {00}*p + {01}*q + {00}*q + {10}*src4 = q + {10}*src4
Note: 4 is the minimum acceptable maxpq otherwise we punt to
synchronous-software path.
The DMA_PREP_CONTINUE flag indicates to the driver to reuse p and q as
sources (in the above manner) and fill the remaining slots up to maxpq
with the new sources/coeffecients.
In the zero_sum case decrement max_pq by two to account for the implied
additional p and q sources.
Note: Some devices can natively support P+Q continuation and can skip
this extra work. Devices with this capability can advertise it with
dma_set_maxpq. It is up to each driver how the DMA_PREP_CONTINUE flag
is honored.
Signed-off-by: Yuri Tikhonov <yur@xxxxxxxxxxx>
Signed-off-by: Ilya Yanok <yanok@xxxxxxxxxxx>
Signed-off-by: Dan Williams <dan.j.williams@xxxxxxxxx>
---
arch/arm/mach-iop13xx/setup.c | 2
crypto/async_tx/Kconfig | 4
crypto/async_tx/Makefile | 1
crypto/async_tx/async_pq.c | 590 +++++++++++++++++++++++++++++++++++++++++
crypto/async_tx/async_xor.c | 2
drivers/dma/iop-adma.c | 2
include/linux/async_tx.h | 33 ++
include/linux/dmaengine.h | 58 ++++
8 files changed, 681 insertions(+), 11 deletions(-)
create mode 100644 crypto/async_tx/async_pq.c
diff --git a/arch/arm/mach-iop13xx/setup.c b/arch/arm/mach-iop13xx/setup.c
index cfd4d2e..3846482 100644
--- a/arch/arm/mach-iop13xx/setup.c
+++ b/arch/arm/mach-iop13xx/setup.c
@@ -506,7 +506,7 @@ void __init iop13xx_platform_init(void)
dma_cap_set(DMA_MEMSET, plat_data->cap_mask);
dma_cap_set(DMA_MEMCPY_CRC32C, plat_data->cap_mask);
dma_cap_set(DMA_INTERRUPT, plat_data->cap_mask);
- dma_cap_set(DMA_PQ_XOR, plat_data->cap_mask);
+ dma_cap_set(DMA_PQ, plat_data->cap_mask);
dma_cap_set(DMA_PQ_UPDATE, plat_data->cap_mask);
dma_cap_set(DMA_PQ_ZERO_SUM, plat_data->cap_mask);
break;
diff --git a/crypto/async_tx/Kconfig b/crypto/async_tx/Kconfig
index d8fb391..cb6d731 100644
--- a/crypto/async_tx/Kconfig
+++ b/crypto/async_tx/Kconfig
@@ -14,3 +14,7 @@ config ASYNC_MEMSET
tristate
select ASYNC_CORE
+config ASYNC_PQ
+ tristate
+ select ASYNC_CORE
+
diff --git a/crypto/async_tx/Makefile b/crypto/async_tx/Makefile
index 27baa7d..1b99265 100644
--- a/crypto/async_tx/Makefile
+++ b/crypto/async_tx/Makefile
@@ -2,3 +2,4 @@ obj-$(CONFIG_ASYNC_CORE) += async_tx.o
obj-$(CONFIG_ASYNC_MEMCPY) += async_memcpy.o
obj-$(CONFIG_ASYNC_MEMSET) += async_memset.o
obj-$(CONFIG_ASYNC_XOR) += async_xor.o
+obj-$(CONFIG_ASYNC_PQ) += async_pq.o
diff --git a/crypto/async_tx/async_pq.c b/crypto/async_tx/async_pq.c
new file mode 100644
index 0000000..da47a29
--- /dev/null
+++ b/crypto/async_tx/async_pq.c
@@ -0,0 +1,590 @@
+/*
+ * Copyright(c) 2007 Yuri Tikhonov <yur@xxxxxxxxxxx>
+ * Copyright(c) 2009 Intel Corporation
+ *
+ * Developed for DENX Software Engineering GmbH
+ *
+ * Asynchronous GF-XOR calculations ASYNC_TX API.
+ *
+ * based on async_xor.c code written by:
+ * Dan Williams <dan.j.williams@xxxxxxxxx>
+ *
+ * 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.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc., 59
+ * Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * The full GNU General Public License is included in this distribution in the
+ * file called COPYING.
+ */
+#include <linux/kernel.h>
+#include <linux/interrupt.h>
+#include <linux/dma-mapping.h>
+#include <linux/raid/pq.h>
+#include <linux/async_tx.h>
+
+/**
+ * spare_pages - synchronous zero sum result buffers
+ *
+ * Protected by spare_lock
+ */
+static struct page *spare_pages[2];
+static spinlock_t spare_lock;
+
+/* scribble - space to hold throwaway P buffer for synchronous gen_syndrome */
+static struct page *scribble;
+
+static bool is_raid6_zero_block(void *p)
+{
+ return p == (void *) raid6_empty_zero_page;
+}
+
+/**
+ * do_async_pq - asynchronously calculate P and/or Q
+ */
+static __async_inline struct dma_async_tx_descriptor *
+do_async_pq(struct dma_chan *chan, struct page **blocks, unsigned char *scfs,
+ unsigned int offset, int src_cnt, size_t len,
+ enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback cb_fn, void *cb_param)
+{
+ struct dma_device *dma = chan->device;
+ dma_addr_t dma_dest[2], dma_src[src_cnt];
+ struct dma_async_tx_descriptor *tx = NULL;
+ dma_async_tx_callback _cb_fn;
+ void *_cb_param;
+ unsigned char *scf = NULL;
+ int i, src_off = 0;
+ unsigned short pq_src_cnt;
+ enum async_tx_flags async_flags;
+ enum dma_ctrl_flags dma_flags = 0;
+ int idx;
+ u8 coefs[src_cnt];
+
+ /* DMAs use destinations as sources, so use BIDIRECTIONAL mapping */
+ if (blocks[src_cnt])
+ dma_dest[0] = dma_map_page(dma->dev, blocks[src_cnt],
+ offset, len, DMA_BIDIRECTIONAL);
+ else
+ dma_flags |= DMA_PREP_PQ_DISABLE_P;
+ if (blocks[src_cnt+1])
+ dma_dest[1] = dma_map_page(dma->dev, blocks[src_cnt+1],
+ offset, len, DMA_BIDIRECTIONAL);
+ else
+ dma_flags |= DMA_PREP_PQ_DISABLE_Q;
+
+ /* convert source addresses being careful to collapse 'zero'
+ * sources and update the coefficients accordingly
+ */
+ for (i = 0, idx = 0; i < src_cnt; i++) {
+ if (is_raid6_zero_block(blocks[i]))
+ continue;
+ dma_src[idx] = dma_map_page(dma->dev, blocks[i],
+ offset, len, DMA_TO_DEVICE);
+ coefs[idx] = scfs[i];
+ idx++;
+ }
+ src_cnt = idx;
+
+ while (src_cnt > 0) {
+ async_flags = flags;
+ pq_src_cnt = min(src_cnt, dma_maxpq(dma, flags));
+ /* if we are submitting additional pqs, leave the chain open,
+ * clear the callback parameters, and leave the destination
+ * buffers mapped
+ */
+ if (src_cnt > pq_src_cnt) {
+ async_flags &= ~ASYNC_TX_ACK;
+ dma_flags |= DMA_COMPL_SKIP_DEST_UNMAP;
+ _cb_fn = NULL;
+ _cb_param = NULL;
+ } else {
+ _cb_fn = cb_fn;
+ _cb_param = cb_param;
+ }
+ if (_cb_fn)
+ dma_flags |= DMA_PREP_INTERRUPT;
+ if (scfs)
+ scf = &scfs[src_off];
+
+ /* Since we have clobbered the src_list we are committed
+ * to doing this asynchronously. Drivers force forward
+ * progress in case they can not provide a descriptor
+ */
+ tx = dma->device_prep_dma_pq(chan, dma_dest,
+ &dma_src[src_off], pq_src_cnt,
+ scf, len, dma_flags);
+ if (unlikely(!tx))
+ async_tx_quiesce(&depend_tx);
+
+ /* spin wait for the preceeding transactions to complete */
+ while (unlikely(!tx)) {
+ dma_async_issue_pending(chan);
+ tx = dma->device_prep_dma_pq(chan, dma_dest,
+ &dma_src[src_off], pq_src_cnt,
+ scf, len, dma_flags);
+ }
+
+ async_tx_submit(chan, tx, async_flags, depend_tx,
+ _cb_fn, _cb_param);
+
+ depend_tx = tx;
+ flags |= ASYNC_TX_DEP_ACK;
+
+ /* drop completed sources */
+ src_cnt -= pq_src_cnt;
+ src_off += pq_src_cnt;
+
+ dma_flags |= DMA_PREP_CONTINUE;
+ }
+
+ return tx;
+}
+
+/**
+ * do_sync_pq - synchronously calculate P and Q
+ */
+static void
+do_sync_pq(struct page **blocks, unsigned char *scfs, unsigned int offset,
+ int src_cnt, size_t len, enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback cb_fn, void *cb_param)
+{
+ u8 *p = NULL;
+ u8 *q = NULL;
+ u8 *ptrs[src_cnt];
+ int d, z;
+ u8 wd, wq, wp;
+
+ /* address convert inputs */
+ if (blocks[src_cnt])
+ p = (u8 *)(page_address(blocks[src_cnt]) + offset);
+ if (blocks[src_cnt+1])
+ q = (u8 *)(page_address(blocks[src_cnt+1]) + offset);
+ for (z = 0; z < src_cnt; z++) {
+ if (is_raid6_zero_block(blocks[z]))
+ ptrs[z] = (void *) blocks[z];
+ else
+ ptrs[z] = (u8 *)(page_address(blocks[z]) + offset);
+ }
+
+ for (d = 0; d < len; d++) {
+ wq = wp = ptrs[0][d];
+ for (z = 1; z < src_cnt; z++) {
+ wd = ptrs[z][d];
+ wp ^= wd;
+ wq ^= raid6_gfmul[scfs[z]][wd];
+ }
+ if (p)
+ p[d] = wp;
+ if (q)
+ q[d] = wq;
+ }
+
+ async_tx_sync_epilog(cb_fn, cb_param);
+}
+
+/**
+ * async_pq - attempt to do XOR and Galois calculations in parallel using
+ * a dma engine.
+ * @blocks: source block array from 0 to (src_cnt-1) with the p destination
+ * at blocks[src_cnt] and q at blocks[src_cnt + 1]. Only one of two
+ * destinations may be present (another then has to be set to NULL).
+ * NOTE: client code must assume the contents of this array are destroyed
+ * @offset: offset in pages to start transaction
+ * @src_cnt: number of source pages
+ * @scfs: array of source coefficients used in GF-multiplication
+ * @len: length in bytes
+ * @flags: ASYNC_TX_ACK, ASYNC_TX_DEP_ACK, ASYNC_TX_ASYNC_ONLY
+ * @depend_tx: depends on the result of this transaction.
+ * @cb_fn: function to call when the operation completes
+ * @cb_param: parameter to pass to the callback routine
+ */
+struct dma_async_tx_descriptor *
+async_pq(struct page **blocks, unsigned int offset, int src_cnt,
+ unsigned char *scfs, size_t len, enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback cb_fn, void *cb_param)
+{
+ struct dma_chan *chan = async_tx_find_channel(depend_tx, DMA_PQ,
+ &blocks[src_cnt], 2,
+ blocks, src_cnt, len);
+ struct dma_device *device = chan ? chan->device : NULL;
+ struct dma_async_tx_descriptor *tx = NULL;
+ bool do_async = false;
+
+ if (device && (src_cnt <= dma_maxpq(device, 0) ||
+ dma_maxpq(device, DMA_PREP_CONTINUE) > 0))
+ do_async = true;
+
+ if (!device && (flags & ASYNC_TX_ASYNC_ONLY))
+ return NULL;
+
+ if (do_async) {
+ /* run pq asynchronously */
+ tx = do_async_pq(chan, blocks, scfs, offset, src_cnt, len,
+ flags, depend_tx, cb_fn, cb_param);
+ } else {
+ /* run pq synchronously */
+ if (!blocks[src_cnt+1]) { /* only p requested, just xor */
+ flags |= ASYNC_TX_XOR_ZERO_DST;
+ return async_xor(blocks[src_cnt], blocks, offset,
+ src_cnt, len, flags, depend_tx,
+ cb_fn, cb_param);
+ }
+
+ /* wait for any prerequisite operations */
+ async_tx_quiesce(&depend_tx);
+
+ do_sync_pq(blocks, scfs, offset, src_cnt, len, flags,
+ depend_tx, cb_fn, cb_param);
+ }
+
+ return tx;
+}
+EXPORT_SYMBOL_GPL(async_pq);
+
+/**
+ * do_sync_gen_syndrome - synchronously calculate P (xor) and Q (Reed-Solomon
+ * code)
+ */
+static void
+do_sync_gen_syndrome(struct page **blocks, unsigned int offset, int src_cnt,
+ size_t len, enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback cb_fn, void *cb_param)
+{
+ int i;
+ void *tsrc[src_cnt+2];
+
+ for (i = 0; i < src_cnt + 2; i++) {
+ if (is_raid6_zero_block(blocks[i]))
+ tsrc[i] = (void *) blocks[i];
+ else
+ tsrc[i] = page_address(blocks[i]) + offset;
+ }
+
+ raid6_call.gen_syndrome(i, len, tsrc);
+
+ async_tx_sync_epilog(cb_fn, cb_param);
+}
+
+/**
+ * async_gen_syndrome - attempt to generate P (xor) and Q (Reed-Solomon code)
+ * with a dma engine for a given set of blocks. This routine assumes a
+ * field of GF(2^8) with a primitive polynomial of 0x11d and a generator
+ * of {02}.
+ * @blocks: source block array ordered from 0..src_cnt-1 with the P destination
+ * at blocks[src_cnt] and Q at blocks[src_cnt + 1]. Only one of two
+ * destinations may be present (another then has to be set to NULL). Some
+ * raid6 schemes calculate the syndrome over all disks with P and Q set to
+ * zero. In this case we catch 'zero' blocks with is_raid6_zero_block()
+ * so we can drop them in the async case, or skip the page_address()
+ * conversion in the sync case.
+ * NOTE: client code must assume the contents of this array are destroyed
+ * @offset: offset in pages to start transaction
+ * @src_cnt: number of source pages: 2 < src_cnt <= 255
+ * @len: length of blocks in bytes
+ * @flags: ASYNC_TX_ACK, ASYNC_TX_DEP_ACK, ASYNC_TX_ASYNC_ONLY
+ * @depend_tx: P+Q operation depends on the result of this transaction.
+ * @cb_fn: function to call when P+Q generation completes
+ * @cb_param: parameter to pass to the callback routine
+ */
+struct dma_async_tx_descriptor *
+async_gen_syndrome(struct page **blocks, unsigned int offset, int src_cnt,
+ size_t len, enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback cb_fn, void *cb_param)
+{
+ struct dma_chan *chan = async_tx_find_channel(depend_tx, DMA_PQ,
+ &blocks[src_cnt], 2,
+ blocks, src_cnt, len);
+ struct dma_device *device = chan ? chan->device : NULL;
+ struct dma_async_tx_descriptor *tx = NULL;
+ bool do_async = false;
+
+ BUG_ON(src_cnt > 255 || (!blocks[src_cnt] && !blocks[src_cnt+1]));
+
+ if (device && (src_cnt <= dma_maxpq(device, 0) ||
+ dma_maxpq(device, DMA_PREP_CONTINUE) > 0))
+ do_async = true;
+
+ if (!do_async && (flags & ASYNC_TX_ASYNC_ONLY))
+ return NULL;
+
+ if (do_async) {
+ /* run the p+q asynchronously */
+ tx = do_async_pq(chan, blocks, (uint8_t *)raid6_gfexp,
+ offset, src_cnt, len, flags, depend_tx,
+ cb_fn, cb_param);
+ } else {
+ /* run the pq synchronously */
+ /* wait for any prerequisite operations */
+ async_tx_quiesce(&depend_tx);
+
+ if (!blocks[src_cnt])
+ blocks[src_cnt] = scribble;
+ if (!blocks[src_cnt+1])
+ blocks[src_cnt+1] = scribble;
+ do_sync_gen_syndrome(blocks, offset, src_cnt, len, flags,
+ depend_tx, cb_fn, cb_param);
+ }
+
+ return tx;
+}
+EXPORT_SYMBOL_GPL(async_gen_syndrome);
+
+static __async_inline enum dma_ctrl_flags
+__pq_zero_sum_map_pages(dma_addr_t *dma, int src_cnt, struct device *dev,
+ struct page **blocks, unsigned int offset, size_t len)
+{
+ enum dma_ctrl_flags flags = 0;
+ int i;
+
+ if (!blocks[src_cnt])
+ flags |= DMA_PREP_PQ_DISABLE_P;
+ if (!blocks[src_cnt+1])
+ flags |= DMA_PREP_PQ_DISABLE_Q;
+ for (i = 0; i < src_cnt + 2; i++)
+ if (likely(blocks[i])) {
+ dma[i] = dma_map_page(dev, blocks[i], offset, len,
+ DMA_TO_DEVICE);
+ BUG_ON(is_raid6_zero_block(blocks[i]));
+ }
+ return flags;
+}
+
+/**
+ * async_pq_zero_sum - attempt a PQ parities check with a dma engine.
+ * @blocks: array of source pages. The 0..src_cnt-1 are the sources, the
+ * src_cnt and src_cnt+1 are the P and Q destinations to check, resp.
+ * Only one of two destinations may be present.
+ * NOTE: client code must assume the contents of this array are destroyed
+ * @offset: offset in pages to start transaction
+ * @src_cnt: number of source pages
+ * @scfs: coefficients to use in GF-multiplications
+ * @len: length in bytes
+ * @pqres: SUM_CHECK_P_RESULT and/or SUM_CHECK_Q_RESULT are set on zero sum fail
+ * @flags: ASYNC_TX_ACK, ASYNC_TX_DEP_ACK
+ * @depend_tx: depends on the result of this transaction.
+ * @cb_fn: function to call when the xor completes
+ * @cb_param: parameter to pass to the callback routine
+ */
+struct dma_async_tx_descriptor *
+async_pq_zero_sum(struct page **blocks, unsigned int offset, int src_cnt,
+ unsigned char *scfs, size_t len, enum sum_check_flags *pqres,
+ enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback cb_fn, void *cb_param)
+{
+ struct dma_chan *chan = async_tx_find_channel(depend_tx,
+ DMA_PQ_ZERO_SUM,
+ &blocks[src_cnt], 2,
+ blocks, src_cnt, len);
+ struct dma_device *device = chan ? chan->device : NULL;
+ struct dma_async_tx_descriptor *tx = NULL;
+ enum dma_ctrl_flags dma_flags = cb_fn ? DMA_PREP_INTERRUPT : 0;
+
+ BUG_ON(src_cnt < 2);
+
+ if (device && src_cnt <= dma_maxpq(device, 0) - 2) {
+ dma_addr_t dma_src[src_cnt + 2];
+
+ dma_flags |= __pq_zero_sum_map_pages(dma_src, src_cnt,
+ device->dev, blocks,
+ offset, len);
+ tx = device->device_prep_dma_pqzero_sum(chan, dma_src, src_cnt,
+ scfs, len, pqres,
+ dma_flags);
+
+ if (unlikely(!tx)) {
+ async_tx_quiesce(&depend_tx);
+
+ while (unlikely(!tx)) {
+ dma_async_issue_pending(chan);
+ tx = device->device_prep_dma_pqzero_sum(chan,
+ dma_src, src_cnt, scfs, len,
+ pqres, dma_flags);
+ }
+ }
+
+ async_tx_submit(chan, tx, flags, depend_tx, cb_fn, cb_param);
+ } else {
+ struct page *pdest = blocks[src_cnt];
+ struct page *qdest = blocks[src_cnt + 1];
+ void *p, *q, *s;
+
+ flags &= ~ASYNC_TX_ACK;
+
+ spin_lock(&spare_lock);
+ blocks[src_cnt] = spare_pages[0];
+ blocks[src_cnt + 1] = spare_pages[1];
+ tx = async_pq(blocks, offset, src_cnt, scfs, len, flags,
+ depend_tx, NULL, NULL);
+ async_tx_quiesce(&tx);
+
+ *pqres = 0;
+ if (pdest) {
+ p = page_address(pdest) + offset;
+ s = page_address(spare_pages[0]) + offset;
+ *pqres |= !!memcmp(p, s, len) << SUM_CHECK_P;
+ }
+
+ if (qdest) {
+ q = page_address(qdest) + offset;
+ s = page_address(spare_pages[1]) + offset;
+ *pqres |= !!memcmp(q, s, len) << SUM_CHECK_Q;
+ }
+ spin_unlock(&spare_lock);
+
+ async_tx_sync_epilog(cb_fn, cb_param);
+ }
+
+ return tx;
+}
+EXPORT_SYMBOL_GPL(async_pq_zero_sum);
+
+/**
+ * async_syndrome_zero_sum - attempt a P (xor) and Q (Reed-Solomon code)
+ * parities check with a dma engine. This routine assumes a field of
+ * GF(2^8) with a primitive polynomial of 0x11d and a generator of {02}.
+ * @blocks: array of source pages. The 0..src_cnt-1 are the sources, the
+ * src_cnt and src_cnt+1 are the P and Q destinations to check, resp.
+ * Only one of two destinations may be present.
+ * NOTE: client code must assume the contents of this array are destroyed
+ * @offset: offset in pages to start transaction
+ * @src_cnt: number of source pages
+ * @len: length in bytes
+ * @pqres: SUM_CHECK_P_RESULT and/or SUM_CHECK_Q_RESULT are set on zero sum fail
+ * @flags: ASYNC_TX_ACK, ASYNC_TX_DEP_ACK
+ * @depend_tx: depends on the result of this transaction.
+ * @cb_fn: function to call when the xor completes
+ * @cb_param: parameter to pass to the callback routine
+ */
+struct dma_async_tx_descriptor *
+async_syndrome_zero_sum(struct page **blocks, unsigned int offset, int src_cnt,
+ size_t len, enum sum_check_flags *pqres,
+ enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback cb_fn, void *cb_param)
+{
+ struct dma_chan *chan = async_tx_find_channel(depend_tx,
+ DMA_PQ_ZERO_SUM,
+ &blocks[src_cnt], 2,
+ blocks, src_cnt, len);
+ struct dma_device *device = chan ? chan->device : NULL;
+ struct dma_async_tx_descriptor *tx = NULL;
+ enum dma_ctrl_flags dma_flags = cb_fn ? DMA_PREP_INTERRUPT : 0;
+
+ BUG_ON(src_cnt < 2);
+
+ if (device && src_cnt <= dma_maxpq(device, 0) - 2) {
+ dma_addr_t dma_src[src_cnt + 2];
+
+ dma_flags |= __pq_zero_sum_map_pages(dma_src, src_cnt,
+ device->dev, blocks,
+ offset, len);
+ tx = device->device_prep_dma_pqzero_sum(chan, dma_src, src_cnt,
+ (uint8_t *)raid6_gfexp,
+ len, pqres, dma_flags);
+
+ if (unlikely(!tx)) {
+ async_tx_quiesce(&depend_tx);
+ while (unlikely(!tx)) {
+ dma_async_issue_pending(chan);
+ tx = device->device_prep_dma_pqzero_sum(chan,
+ dma_src, src_cnt,
+ (uint8_t *)raid6_gfexp, len,
+ pqres, dma_flags);
+ }
+ }
+
+ async_tx_submit(chan, tx, flags, depend_tx, cb_fn, cb_param);
+ } else {
+ struct page *pdest = blocks[src_cnt];
+ struct page *qdest = blocks[src_cnt + 1];
+ enum async_tx_flags lflags = flags;
+ void *p, *q, *s;
+
+ lflags &= ~ASYNC_TX_ACK;
+
+ spin_lock(&spare_lock);
+ blocks[src_cnt] = spare_pages[0];
+ blocks[src_cnt + 1] = spare_pages[1];
+ tx = async_gen_syndrome(blocks, offset,
+ src_cnt, len, lflags,
+ depend_tx, NULL, NULL);
+ async_tx_quiesce(&tx);
+
+ *pqres = 0;
+ if (pdest) {
+ p = page_address(pdest) + offset;
+ s = page_address(spare_pages[0]) + offset;
+ *pqres |= !!memcmp(p, s, len) << SUM_CHECK_P;
+ }
+
+ if (qdest) {
+ q = page_address(qdest) + offset;
+ s = page_address(spare_pages[1]) + offset;
+ *pqres |= !!memcmp(q, s, len) << SUM_CHECK_Q;
+ }
+ spin_unlock(&spare_lock);
+
+ async_tx_sync_epilog(cb_fn, cb_param);
+ }
+
+ return tx;
+}
+EXPORT_SYMBOL_GPL(async_syndrome_zero_sum);
+
+static void safe_put_page(struct page *p)
+{
+ if (p)
+ put_page(p);
+}
+
+static int __init async_pq_init(void)
+{
+ spin_lock_init(&spare_lock);
+
+ spare_pages[0] = alloc_page(GFP_KERNEL);
+ if (!spare_pages[0])
+ goto abort;
+ spare_pages[1] = alloc_page(GFP_KERNEL);
+ if (!spare_pages[1])
+ goto abort;
+ scribble = alloc_page(GFP_KERNEL);
+ if (!scribble)
+ goto abort;
+ return 0;
+abort:
+ safe_put_page(scribble);
+ safe_put_page(spare_pages[1]);
+ safe_put_page(spare_pages[0]);
+ printk(KERN_ERR "%s: cannot allocate spare!\n", __func__);
+ return -ENOMEM;
+}
+
+static void __exit async_pq_exit(void)
+{
+ safe_put_page(scribble);
+ safe_put_page(spare_pages[1]);
+ safe_put_page(spare_pages[0]);
+}
+
+module_init(async_pq_init);
+module_exit(async_pq_exit);
+
+MODULE_AUTHOR("Yuri Tikhonov <yur@xxxxxxxxxxx>, Dan Williams <dan.j.williams@xxxxxxxxx>");
+MODULE_DESCRIPTION("asynchronous pq/pq-zero-sum api");
+MODULE_LICENSE("GPL");
diff --git a/crypto/async_tx/async_xor.c b/crypto/async_tx/async_xor.c
index d1a084e..fcd6be5 100644
--- a/crypto/async_tx/async_xor.c
+++ b/crypto/async_tx/async_xor.c
@@ -65,7 +65,7 @@ do_async_xor(struct dma_chan *chan, struct page *dest, struct page **src_list,
while (src_cnt) {
async_flags = flags;
dma_flags = 0;
- xor_src_cnt = min(src_cnt, dma->max_xor);
+ xor_src_cnt = min(src_cnt, (int)dma->max_xor);
/* if we are submitting additional xors, leave the chain open,
* clear the callback parameters, and leave the destination
* buffer mapped
diff --git a/drivers/dma/iop-adma.c b/drivers/dma/iop-adma.c
index 16adbe6..8bccf85 100644
--- a/drivers/dma/iop-adma.c
+++ b/drivers/dma/iop-adma.c
@@ -1258,7 +1258,7 @@ static int __devinit iop_adma_probe(struct platform_device *pdev)
dev_printk(KERN_INFO, &pdev->dev, "Intel(R) IOP: "
"( %s%s%s%s%s%s%s%s%s%s)\n",
- dma_has_cap(DMA_PQ_XOR, dma_dev->cap_mask) ? "pq_xor " : "",
+ dma_has_cap(DMA_PQ, dma_dev->cap_mask) ? "pq " : "",
dma_has_cap(DMA_PQ_UPDATE, dma_dev->cap_mask) ? "pq_update " : "",
dma_has_cap(DMA_PQ_ZERO_SUM, dma_dev->cap_mask) ? "pq_zero_sum " : "",
dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "xor " : "",
diff --git a/include/linux/async_tx.h b/include/linux/async_tx.h
index 6370f32..1f10141 100644
--- a/include/linux/async_tx.h
+++ b/include/linux/async_tx.h
@@ -59,12 +59,15 @@ struct dma_chan_ref {
* @ASYNC_TX_ACK: immediately ack the descriptor, precludes setting up a
* dependency chain
* @ASYNC_TX_DEP_ACK: ack the dependency descriptor. Useful for chaining.
+ * @ASYNC_TX_ASYNC_ONLY: if set then try to perform operation requested only in
+ * the asynchronous mode. Useful for R6 recovery.
*/
enum async_tx_flags {
ASYNC_TX_XOR_ZERO_DST = (1 << 0),
ASYNC_TX_XOR_DROP_DST = (1 << 1),
- ASYNC_TX_ACK = (1 << 3),
- ASYNC_TX_DEP_ACK = (1 << 4),
+ ASYNC_TX_ACK = (1 << 2),
+ ASYNC_TX_DEP_ACK = (1 << 3),
+ ASYNC_TX_ASYNC_ONLY = (1 << 4),
};
#ifdef CONFIG_DMA_ENGINE
@@ -140,5 +143,31 @@ async_trigger_callback(enum async_tx_flags flags,
struct dma_async_tx_descriptor *depend_tx,
dma_async_tx_callback cb_fn, void *cb_fn_param);
+struct dma_async_tx_descriptor *
+async_pq(struct page **blocks, unsigned int offset, int src_cnt,
+ unsigned char *scfs, size_t len, enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback cb_fn, void *cb_param);
+
+struct dma_async_tx_descriptor *
+async_gen_syndrome(struct page **blocks, unsigned int offset, int src_cnt,
+ size_t len, enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback cb_fn, void *cb_param);
+
+struct dma_async_tx_descriptor *
+async_pq_zero_sum(struct page **blocks, unsigned int offset, int src_cnt,
+ unsigned char *scfs, size_t len, enum sum_check_flags *pqres,
+ enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback cb_fn, void *cb_param);
+
+struct dma_async_tx_descriptor *
+async_syndrome_zero_sum(struct page **blocks, unsigned int offset, int src_cnt,
+ size_t len, enum sum_check_flags *pqres,
+ enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback cb_fn, void *cb_param);
+
void async_tx_quiesce(struct dma_async_tx_descriptor **tx);
#endif /* _ASYNC_TX_H_ */
diff --git a/include/linux/dmaengine.h b/include/linux/dmaengine.h
index cd17392..a7fa966 100644
--- a/include/linux/dmaengine.h
+++ b/include/linux/dmaengine.h
@@ -55,7 +55,7 @@ enum dma_status {
enum dma_transaction_type {
DMA_MEMCPY,
DMA_XOR,
- DMA_PQ_XOR,
+ DMA_PQ,
DMA_DUAL_XOR,
DMA_PQ_UPDATE,
DMA_ZERO_SUM,
@@ -73,20 +73,28 @@ enum dma_transaction_type {
/**
* enum dma_ctrl_flags - DMA flags to augment operation preparation,
- * control completion, and communicate status.
+ * control completion, and communicate status.
* @DMA_PREP_INTERRUPT - trigger an interrupt (callback) upon completion of
- * this transaction
+ * this transaction
* @DMA_CTRL_ACK - the descriptor cannot be reused until the client
- * acknowledges receipt, i.e. has has a chance to establish any
- * dependency chains
+ * acknowledges receipt, i.e. has has a chance to establish any dependency
+ * chains
* @DMA_COMPL_SKIP_SRC_UNMAP - set to disable dma-unmapping the source buffer(s)
* @DMA_COMPL_SKIP_DEST_UNMAP - set to disable dma-unmapping the destination(s)
+ * @DMA_PREP_PQ_DISABLE_P - prevent generation of P while generating Q
+ * @DMA_PREP_PQ_DISABLE_Q - prevent generation of Q while generating P
+ * @DMA_PREP_CONTINUE - indicate to a driver that it is reusing buffers as
+ * sources that were the result of a previous operation, in the case of a PQ
+ * operation it continues the calculation with new sources
*/
enum dma_ctrl_flags {
DMA_PREP_INTERRUPT = (1 << 0),
DMA_CTRL_ACK = (1 << 1),
DMA_COMPL_SKIP_SRC_UNMAP = (1 << 2),
DMA_COMPL_SKIP_DEST_UNMAP = (1 << 3),
+ DMA_PREP_PQ_DISABLE_P = (1 << 4),
+ DMA_PREP_PQ_DISABLE_Q = (1 << 5),
+ DMA_PREP_CONTINUE = (1 << 6),
};
/**
@@ -228,6 +236,7 @@ struct dma_async_tx_descriptor {
* @global_node: list_head for global dma_device_list
* @cap_mask: one or more dma_capability flags
* @max_xor: maximum number of xor sources, 0 if no capability
+ * @max_pq: maximum number of PQ sources and PQ-continue capability
* @dev_id: unique device ID
* @dev: struct device reference for dma mapping api
* @device_alloc_chan_resources: allocate resources and return the
@@ -235,7 +244,9 @@ struct dma_async_tx_descriptor {
* @device_free_chan_resources: release DMA channel's resources
* @device_prep_dma_memcpy: prepares a memcpy operation
* @device_prep_dma_xor: prepares a xor operation
+ * @device_prep_dma_pq: prepares a pq operation
* @device_prep_dma_zero_sum: prepares a zero_sum operation
+ * @device_prep_dma_pqzero_sum: prepares a pqzero_sum operation
* @device_prep_dma_memset: prepares a memset operation
* @device_prep_dma_interrupt: prepares an end of chain interrupt operation
* @device_prep_slave_sg: prepares a slave dma operation
@@ -249,7 +260,9 @@ struct dma_device {
struct list_head channels;
struct list_head global_node;
dma_cap_mask_t cap_mask;
- int max_xor;
+ unsigned short max_xor;
+ unsigned short max_pq;
+ #define DMA_HAS_PQ_CONTINUE (1 << 15)
int dev_id;
struct device *dev;
@@ -263,9 +276,17 @@ struct dma_device {
struct dma_async_tx_descriptor *(*device_prep_dma_xor)(
struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
unsigned int src_cnt, size_t len, unsigned long flags);
+ struct dma_async_tx_descriptor *(*device_prep_dma_pq)(
+ struct dma_chan *chan, dma_addr_t *dst, dma_addr_t *src,
+ unsigned int src_cnt, unsigned char *scf,
+ size_t len, unsigned long flags);
struct dma_async_tx_descriptor *(*device_prep_dma_zero_sum)(
struct dma_chan *chan, dma_addr_t *src, unsigned int src_cnt,
size_t len, enum sum_check_flags *result, unsigned long flags);
+ struct dma_async_tx_descriptor *(*device_prep_dma_pqzero_sum)(
+ struct dma_chan *chan, dma_addr_t *src, unsigned int src_cnt,
+ unsigned char *scf, size_t len, enum sum_check_flags *pqres,
+ unsigned long flags);
struct dma_async_tx_descriptor *(*device_prep_dma_memset)(
struct dma_chan *chan, dma_addr_t dest, int value, size_t len,
unsigned long flags);
@@ -284,6 +305,31 @@ struct dma_device {
void (*device_issue_pending)(struct dma_chan *chan);
};
+static inline void dma_set_maxpq(struct dma_device *dma, int maxpq, int has_pq_continue)
+{
+ dma->max_pq = maxpq;
+ if (has_pq_continue)
+ dma->max_pq |= DMA_HAS_PQ_CONTINUE;
+}
+
+/* dma_maxpq - reduce maxpq in the face of continued operations
+ * @dma - dma device with PQ capability
+ * @flags - to determine if DMA_PREP_CONTINUE is set
+ *
+ * When an engine does not support native continuation we need 3 extra
+ * source slots to reuse P and Q with the following coefficients:
+ * 1/ {00} * P : remove P from Q', but use it as a source for P'
+ * 2/ {01} * Q : use Q to continue Q' calculation
+ * 3/ {00} * Q : subtract Q from P' to cancel (2)
+ */
+static inline int dma_maxpq(struct dma_device *dma, enum dma_ctrl_flags flags)
+{
+ if ((flags & DMA_PREP_CONTINUE) &&
+ (dma->max_pq & DMA_HAS_PQ_CONTINUE) == 0)
+ return dma->max_pq - 3;
+ return dma->max_pq & ~DMA_HAS_PQ_CONTINUE;
+}
+
/* --- public DMA engine API --- */
#ifdef CONFIG_DMA_ENGINE
--
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/