linux_old1/drivers/crypto/s5p-sss.c

920 lines
22 KiB
C
Raw Normal View History

/*
* Cryptographic API.
*
* Support for Samsung S5PV210 HW acceleration.
*
* Copyright (C) 2011 NetUP Inc. All rights reserved.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as published
* by the Free Software Foundation.
*
*/
#include <linux/clk.h>
#include <linux/crypto.h>
#include <linux/dma-mapping.h>
#include <linux/err.h>
#include <linux/errno.h>
#include <linux/init.h>
#include <linux/interrupt.h>
#include <linux/io.h>
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/of.h>
#include <linux/platform_device.h>
#include <linux/scatterlist.h>
#include <crypto/ctr.h>
#include <crypto/aes.h>
#include <crypto/algapi.h>
#include <crypto/scatterwalk.h>
#define _SBF(s, v) ((v) << (s))
/* Feed control registers */
#define SSS_REG_FCINTSTAT 0x0000
#define SSS_FCINTSTAT_BRDMAINT BIT(3)
#define SSS_FCINTSTAT_BTDMAINT BIT(2)
#define SSS_FCINTSTAT_HRDMAINT BIT(1)
#define SSS_FCINTSTAT_PKDMAINT BIT(0)
#define SSS_REG_FCINTENSET 0x0004
#define SSS_FCINTENSET_BRDMAINTENSET BIT(3)
#define SSS_FCINTENSET_BTDMAINTENSET BIT(2)
#define SSS_FCINTENSET_HRDMAINTENSET BIT(1)
#define SSS_FCINTENSET_PKDMAINTENSET BIT(0)
#define SSS_REG_FCINTENCLR 0x0008
#define SSS_FCINTENCLR_BRDMAINTENCLR BIT(3)
#define SSS_FCINTENCLR_BTDMAINTENCLR BIT(2)
#define SSS_FCINTENCLR_HRDMAINTENCLR BIT(1)
#define SSS_FCINTENCLR_PKDMAINTENCLR BIT(0)
#define SSS_REG_FCINTPEND 0x000C
#define SSS_FCINTPEND_BRDMAINTP BIT(3)
#define SSS_FCINTPEND_BTDMAINTP BIT(2)
#define SSS_FCINTPEND_HRDMAINTP BIT(1)
#define SSS_FCINTPEND_PKDMAINTP BIT(0)
#define SSS_REG_FCFIFOSTAT 0x0010
#define SSS_FCFIFOSTAT_BRFIFOFUL BIT(7)
#define SSS_FCFIFOSTAT_BRFIFOEMP BIT(6)
#define SSS_FCFIFOSTAT_BTFIFOFUL BIT(5)
#define SSS_FCFIFOSTAT_BTFIFOEMP BIT(4)
#define SSS_FCFIFOSTAT_HRFIFOFUL BIT(3)
#define SSS_FCFIFOSTAT_HRFIFOEMP BIT(2)
#define SSS_FCFIFOSTAT_PKFIFOFUL BIT(1)
#define SSS_FCFIFOSTAT_PKFIFOEMP BIT(0)
#define SSS_REG_FCFIFOCTRL 0x0014
#define SSS_FCFIFOCTRL_DESSEL BIT(2)
#define SSS_HASHIN_INDEPENDENT _SBF(0, 0x00)
#define SSS_HASHIN_CIPHER_INPUT _SBF(0, 0x01)
#define SSS_HASHIN_CIPHER_OUTPUT _SBF(0, 0x02)
#define SSS_REG_FCBRDMAS 0x0020
#define SSS_REG_FCBRDMAL 0x0024
#define SSS_REG_FCBRDMAC 0x0028
#define SSS_FCBRDMAC_BYTESWAP BIT(1)
#define SSS_FCBRDMAC_FLUSH BIT(0)
#define SSS_REG_FCBTDMAS 0x0030
#define SSS_REG_FCBTDMAL 0x0034
#define SSS_REG_FCBTDMAC 0x0038
#define SSS_FCBTDMAC_BYTESWAP BIT(1)
#define SSS_FCBTDMAC_FLUSH BIT(0)
#define SSS_REG_FCHRDMAS 0x0040
#define SSS_REG_FCHRDMAL 0x0044
#define SSS_REG_FCHRDMAC 0x0048
#define SSS_FCHRDMAC_BYTESWAP BIT(1)
#define SSS_FCHRDMAC_FLUSH BIT(0)
#define SSS_REG_FCPKDMAS 0x0050
#define SSS_REG_FCPKDMAL 0x0054
#define SSS_REG_FCPKDMAC 0x0058
#define SSS_FCPKDMAC_BYTESWAP BIT(3)
#define SSS_FCPKDMAC_DESCEND BIT(2)
#define SSS_FCPKDMAC_TRANSMIT BIT(1)
#define SSS_FCPKDMAC_FLUSH BIT(0)
#define SSS_REG_FCPKDMAO 0x005C
/* AES registers */
#define SSS_REG_AES_CONTROL 0x00
#define SSS_AES_BYTESWAP_DI BIT(11)
#define SSS_AES_BYTESWAP_DO BIT(10)
#define SSS_AES_BYTESWAP_IV BIT(9)
#define SSS_AES_BYTESWAP_CNT BIT(8)
#define SSS_AES_BYTESWAP_KEY BIT(7)
#define SSS_AES_KEY_CHANGE_MODE BIT(6)
#define SSS_AES_KEY_SIZE_128 _SBF(4, 0x00)
#define SSS_AES_KEY_SIZE_192 _SBF(4, 0x01)
#define SSS_AES_KEY_SIZE_256 _SBF(4, 0x02)
#define SSS_AES_FIFO_MODE BIT(3)
#define SSS_AES_CHAIN_MODE_ECB _SBF(1, 0x00)
#define SSS_AES_CHAIN_MODE_CBC _SBF(1, 0x01)
#define SSS_AES_CHAIN_MODE_CTR _SBF(1, 0x02)
#define SSS_AES_MODE_DECRYPT BIT(0)
#define SSS_REG_AES_STATUS 0x04
#define SSS_AES_BUSY BIT(2)
#define SSS_AES_INPUT_READY BIT(1)
#define SSS_AES_OUTPUT_READY BIT(0)
#define SSS_REG_AES_IN_DATA(s) (0x10 + (s << 2))
#define SSS_REG_AES_OUT_DATA(s) (0x20 + (s << 2))
#define SSS_REG_AES_IV_DATA(s) (0x30 + (s << 2))
#define SSS_REG_AES_CNT_DATA(s) (0x40 + (s << 2))
#define SSS_REG_AES_KEY_DATA(s) (0x80 + (s << 2))
#define SSS_REG(dev, reg) ((dev)->ioaddr + (SSS_REG_##reg))
#define SSS_READ(dev, reg) __raw_readl(SSS_REG(dev, reg))
#define SSS_WRITE(dev, reg, val) __raw_writel((val), SSS_REG(dev, reg))
#define SSS_AES_REG(dev, reg) ((dev)->aes_ioaddr + SSS_REG_##reg)
#define SSS_AES_WRITE(dev, reg, val) __raw_writel((val), \
SSS_AES_REG(dev, reg))
/* HW engine modes */
#define FLAGS_AES_DECRYPT BIT(0)
#define FLAGS_AES_MODE_MASK _SBF(1, 0x03)
#define FLAGS_AES_CBC _SBF(1, 0x01)
#define FLAGS_AES_CTR _SBF(1, 0x02)
#define AES_KEY_LEN 16
#define CRYPTO_QUEUE_LEN 1
/**
* struct samsung_aes_variant - platform specific SSS driver data
* @aes_offset: AES register offset from SSS module's base.
*
* Specifies platform specific configuration of SSS module.
* Note: A structure for driver specific platform data is used for future
* expansion of its usage.
*/
struct samsung_aes_variant {
unsigned int aes_offset;
};
struct s5p_aes_reqctx {
unsigned long mode;
};
struct s5p_aes_ctx {
struct s5p_aes_dev *dev;
uint8_t aes_key[AES_MAX_KEY_SIZE];
uint8_t nonce[CTR_RFC3686_NONCE_SIZE];
int keylen;
};
struct s5p_aes_dev {
struct device *dev;
struct clk *clk;
void __iomem *ioaddr;
void __iomem *aes_ioaddr;
int irq_fc;
struct ablkcipher_request *req;
struct s5p_aes_ctx *ctx;
struct scatterlist *sg_src;
struct scatterlist *sg_dst;
/* In case of unaligned access: */
struct scatterlist *sg_src_cpy;
struct scatterlist *sg_dst_cpy;
struct tasklet_struct tasklet;
struct crypto_queue queue;
bool busy;
spinlock_t lock;
};
static struct s5p_aes_dev *s5p_dev;
static const struct samsung_aes_variant s5p_aes_data = {
.aes_offset = 0x4000,
};
static const struct samsung_aes_variant exynos_aes_data = {
.aes_offset = 0x200,
};
static const struct of_device_id s5p_sss_dt_match[] = {
{
.compatible = "samsung,s5pv210-secss",
.data = &s5p_aes_data,
},
{
.compatible = "samsung,exynos4210-secss",
.data = &exynos_aes_data,
},
{ },
};
MODULE_DEVICE_TABLE(of, s5p_sss_dt_match);
static inline struct samsung_aes_variant *find_s5p_sss_version
(struct platform_device *pdev)
{
if (IS_ENABLED(CONFIG_OF) && (pdev->dev.of_node)) {
const struct of_device_id *match;
match = of_match_node(s5p_sss_dt_match,
pdev->dev.of_node);
return (struct samsung_aes_variant *)match->data;
}
return (struct samsung_aes_variant *)
platform_get_device_id(pdev)->driver_data;
}
static void s5p_set_dma_indata(struct s5p_aes_dev *dev, struct scatterlist *sg)
{
SSS_WRITE(dev, FCBRDMAS, sg_dma_address(sg));
SSS_WRITE(dev, FCBRDMAL, sg_dma_len(sg));
}
static void s5p_set_dma_outdata(struct s5p_aes_dev *dev, struct scatterlist *sg)
{
SSS_WRITE(dev, FCBTDMAS, sg_dma_address(sg));
SSS_WRITE(dev, FCBTDMAL, sg_dma_len(sg));
}
static void s5p_free_sg_cpy(struct s5p_aes_dev *dev, struct scatterlist **sg)
{
int len;
if (!*sg)
return;
len = ALIGN(dev->req->nbytes, AES_BLOCK_SIZE);
free_pages((unsigned long)sg_virt(*sg), get_order(len));
kfree(*sg);
*sg = NULL;
}
static void s5p_sg_copy_buf(void *buf, struct scatterlist *sg,
unsigned int nbytes, int out)
{
struct scatter_walk walk;
if (!nbytes)
return;
scatterwalk_start(&walk, sg);
scatterwalk_copychunks(buf, &walk, nbytes, out);
scatterwalk_done(&walk, out, 0);
}
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
static void s5p_sg_done(struct s5p_aes_dev *dev)
{
if (dev->sg_dst_cpy) {
dev_dbg(dev->dev,
"Copying %d bytes of output data back to original place\n",
dev->req->nbytes);
s5p_sg_copy_buf(sg_virt(dev->sg_dst_cpy), dev->req->dst,
dev->req->nbytes, 1);
}
s5p_free_sg_cpy(dev, &dev->sg_src_cpy);
s5p_free_sg_cpy(dev, &dev->sg_dst_cpy);
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
}
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
/* Calls the completion. Cannot be called with dev->lock hold. */
static void s5p_aes_complete(struct s5p_aes_dev *dev, int err)
{
dev->req->base.complete(&dev->req->base, err);
}
static void s5p_unset_outdata(struct s5p_aes_dev *dev)
{
dma_unmap_sg(dev->dev, dev->sg_dst, 1, DMA_FROM_DEVICE);
}
static void s5p_unset_indata(struct s5p_aes_dev *dev)
{
dma_unmap_sg(dev->dev, dev->sg_src, 1, DMA_TO_DEVICE);
}
static int s5p_make_sg_cpy(struct s5p_aes_dev *dev, struct scatterlist *src,
struct scatterlist **dst)
{
void *pages;
int len;
*dst = kmalloc(sizeof(**dst), GFP_ATOMIC);
if (!*dst)
return -ENOMEM;
len = ALIGN(dev->req->nbytes, AES_BLOCK_SIZE);
pages = (void *)__get_free_pages(GFP_ATOMIC, get_order(len));
if (!pages) {
kfree(*dst);
*dst = NULL;
return -ENOMEM;
}
s5p_sg_copy_buf(pages, src, dev->req->nbytes, 0);
sg_init_table(*dst, 1);
sg_set_buf(*dst, pages, len);
return 0;
}
static int s5p_set_outdata(struct s5p_aes_dev *dev, struct scatterlist *sg)
{
int err;
if (!sg->length) {
err = -EINVAL;
goto exit;
}
err = dma_map_sg(dev->dev, sg, 1, DMA_FROM_DEVICE);
if (!err) {
err = -ENOMEM;
goto exit;
}
dev->sg_dst = sg;
err = 0;
exit:
return err;
}
static int s5p_set_indata(struct s5p_aes_dev *dev, struct scatterlist *sg)
{
int err;
if (!sg->length) {
err = -EINVAL;
goto exit;
}
err = dma_map_sg(dev->dev, sg, 1, DMA_TO_DEVICE);
if (!err) {
err = -ENOMEM;
goto exit;
}
dev->sg_src = sg;
err = 0;
exit:
return err;
}
crypto: s5p-sss - Fix missed interrupts when working with 8 kB blocks The tcrypt testing module on Exynos5422-based Odroid XU3/4 board failed on testing 8 kB size blocks: $ sudo modprobe tcrypt sec=1 mode=500 testing speed of async ecb(aes) (ecb-aes-s5p) encryption test 0 (128 bit key, 16 byte blocks): 21971 operations in 1 seconds (351536 bytes) test 1 (128 bit key, 64 byte blocks): 21731 operations in 1 seconds (1390784 bytes) test 2 (128 bit key, 256 byte blocks): 21932 operations in 1 seconds (5614592 bytes) test 3 (128 bit key, 1024 byte blocks): 21685 operations in 1 seconds (22205440 bytes) test 4 (128 bit key, 8192 byte blocks): This was caused by a race issue of missed BRDMA_DONE ("Block cipher Receiving DMA") interrupt. Device starts processing the data in DMA mode immediately after setting length of DMA block: receiving (FCBRDMAL) or transmitting (FCBTDMAL). The driver sets these lengths from interrupt handler through s5p_set_dma_indata() function (or xxx_setdata()). However the interrupt handler was first dealing with receive buffer (dma-unmap old, dma-map new, set receive block length which starts the operation), then with transmit buffer and finally was clearing pending interrupts (FCINTPEND). Because of the time window between setting receive buffer length and clearing pending interrupts, the operation on receive buffer could end already and driver would miss new interrupt. User manual for Exynos5422 confirms in example code that setting DMA block lengths should be the last operation. The tcrypt hang could be also observed in following blocked-task dmesg: INFO: task modprobe:258 blocked for more than 120 seconds. Not tainted 4.6.0-rc4-next-20160419-00005-g9eac8b7b7753-dirty #42 "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. modprobe D c06b09d8 0 258 256 0x00000000 [<c06b09d8>] (__schedule) from [<c06b0f24>] (schedule+0x40/0xac) [<c06b0f24>] (schedule) from [<c06b49f8>] (schedule_timeout+0x124/0x178) [<c06b49f8>] (schedule_timeout) from [<c06b17fc>] (wait_for_common+0xb8/0x144) [<c06b17fc>] (wait_for_common) from [<bf0013b8>] (test_acipher_speed+0x49c/0x740 [tcrypt]) [<bf0013b8>] (test_acipher_speed [tcrypt]) from [<bf003e8c>] (do_test+0x2240/0x30ec [tcrypt]) [<bf003e8c>] (do_test [tcrypt]) from [<bf008048>] (tcrypt_mod_init+0x48/0xa4 [tcrypt]) [<bf008048>] (tcrypt_mod_init [tcrypt]) from [<c010177c>] (do_one_initcall+0x3c/0x16c) [<c010177c>] (do_one_initcall) from [<c0191ff0>] (do_init_module+0x5c/0x1ac) [<c0191ff0>] (do_init_module) from [<c0185610>] (load_module+0x1a30/0x1d08) [<c0185610>] (load_module) from [<c0185ab0>] (SyS_finit_module+0x8c/0x98) [<c0185ab0>] (SyS_finit_module) from [<c01078c0>] (ret_fast_syscall+0x0/0x3c) Fixes: a49e490c7a8a ("crypto: s5p-sss - add S5PV210 advanced crypto engine support") Cc: <stable@vger.kernel.org> Signed-off-by: Krzysztof Kozlowski <k.kozlowski@samsung.com> Tested-by: Marek Szyprowski <m.szyprowski@samsung.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2016-04-22 20:15:23 +08:00
/*
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
* Returns -ERRNO on error (mapping of new data failed).
* On success returns:
* - 0 if there is no more data,
* - 1 if new transmitting (output) data is ready and its address+length
* have to be written to device (by calling s5p_set_dma_outdata()).
crypto: s5p-sss - Fix missed interrupts when working with 8 kB blocks The tcrypt testing module on Exynos5422-based Odroid XU3/4 board failed on testing 8 kB size blocks: $ sudo modprobe tcrypt sec=1 mode=500 testing speed of async ecb(aes) (ecb-aes-s5p) encryption test 0 (128 bit key, 16 byte blocks): 21971 operations in 1 seconds (351536 bytes) test 1 (128 bit key, 64 byte blocks): 21731 operations in 1 seconds (1390784 bytes) test 2 (128 bit key, 256 byte blocks): 21932 operations in 1 seconds (5614592 bytes) test 3 (128 bit key, 1024 byte blocks): 21685 operations in 1 seconds (22205440 bytes) test 4 (128 bit key, 8192 byte blocks): This was caused by a race issue of missed BRDMA_DONE ("Block cipher Receiving DMA") interrupt. Device starts processing the data in DMA mode immediately after setting length of DMA block: receiving (FCBRDMAL) or transmitting (FCBTDMAL). The driver sets these lengths from interrupt handler through s5p_set_dma_indata() function (or xxx_setdata()). However the interrupt handler was first dealing with receive buffer (dma-unmap old, dma-map new, set receive block length which starts the operation), then with transmit buffer and finally was clearing pending interrupts (FCINTPEND). Because of the time window between setting receive buffer length and clearing pending interrupts, the operation on receive buffer could end already and driver would miss new interrupt. User manual for Exynos5422 confirms in example code that setting DMA block lengths should be the last operation. The tcrypt hang could be also observed in following blocked-task dmesg: INFO: task modprobe:258 blocked for more than 120 seconds. Not tainted 4.6.0-rc4-next-20160419-00005-g9eac8b7b7753-dirty #42 "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. modprobe D c06b09d8 0 258 256 0x00000000 [<c06b09d8>] (__schedule) from [<c06b0f24>] (schedule+0x40/0xac) [<c06b0f24>] (schedule) from [<c06b49f8>] (schedule_timeout+0x124/0x178) [<c06b49f8>] (schedule_timeout) from [<c06b17fc>] (wait_for_common+0xb8/0x144) [<c06b17fc>] (wait_for_common) from [<bf0013b8>] (test_acipher_speed+0x49c/0x740 [tcrypt]) [<bf0013b8>] (test_acipher_speed [tcrypt]) from [<bf003e8c>] (do_test+0x2240/0x30ec [tcrypt]) [<bf003e8c>] (do_test [tcrypt]) from [<bf008048>] (tcrypt_mod_init+0x48/0xa4 [tcrypt]) [<bf008048>] (tcrypt_mod_init [tcrypt]) from [<c010177c>] (do_one_initcall+0x3c/0x16c) [<c010177c>] (do_one_initcall) from [<c0191ff0>] (do_init_module+0x5c/0x1ac) [<c0191ff0>] (do_init_module) from [<c0185610>] (load_module+0x1a30/0x1d08) [<c0185610>] (load_module) from [<c0185ab0>] (SyS_finit_module+0x8c/0x98) [<c0185ab0>] (SyS_finit_module) from [<c01078c0>] (ret_fast_syscall+0x0/0x3c) Fixes: a49e490c7a8a ("crypto: s5p-sss - add S5PV210 advanced crypto engine support") Cc: <stable@vger.kernel.org> Signed-off-by: Krzysztof Kozlowski <k.kozlowski@samsung.com> Tested-by: Marek Szyprowski <m.szyprowski@samsung.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2016-04-22 20:15:23 +08:00
*/
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
static int s5p_aes_tx(struct s5p_aes_dev *dev)
{
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
int ret = 0;
s5p_unset_outdata(dev);
if (!sg_is_last(dev->sg_dst)) {
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
ret = s5p_set_outdata(dev, sg_next(dev->sg_dst));
if (!ret)
ret = 1;
}
crypto: s5p-sss - Fix missed interrupts when working with 8 kB blocks The tcrypt testing module on Exynos5422-based Odroid XU3/4 board failed on testing 8 kB size blocks: $ sudo modprobe tcrypt sec=1 mode=500 testing speed of async ecb(aes) (ecb-aes-s5p) encryption test 0 (128 bit key, 16 byte blocks): 21971 operations in 1 seconds (351536 bytes) test 1 (128 bit key, 64 byte blocks): 21731 operations in 1 seconds (1390784 bytes) test 2 (128 bit key, 256 byte blocks): 21932 operations in 1 seconds (5614592 bytes) test 3 (128 bit key, 1024 byte blocks): 21685 operations in 1 seconds (22205440 bytes) test 4 (128 bit key, 8192 byte blocks): This was caused by a race issue of missed BRDMA_DONE ("Block cipher Receiving DMA") interrupt. Device starts processing the data in DMA mode immediately after setting length of DMA block: receiving (FCBRDMAL) or transmitting (FCBTDMAL). The driver sets these lengths from interrupt handler through s5p_set_dma_indata() function (or xxx_setdata()). However the interrupt handler was first dealing with receive buffer (dma-unmap old, dma-map new, set receive block length which starts the operation), then with transmit buffer and finally was clearing pending interrupts (FCINTPEND). Because of the time window between setting receive buffer length and clearing pending interrupts, the operation on receive buffer could end already and driver would miss new interrupt. User manual for Exynos5422 confirms in example code that setting DMA block lengths should be the last operation. The tcrypt hang could be also observed in following blocked-task dmesg: INFO: task modprobe:258 blocked for more than 120 seconds. Not tainted 4.6.0-rc4-next-20160419-00005-g9eac8b7b7753-dirty #42 "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. modprobe D c06b09d8 0 258 256 0x00000000 [<c06b09d8>] (__schedule) from [<c06b0f24>] (schedule+0x40/0xac) [<c06b0f24>] (schedule) from [<c06b49f8>] (schedule_timeout+0x124/0x178) [<c06b49f8>] (schedule_timeout) from [<c06b17fc>] (wait_for_common+0xb8/0x144) [<c06b17fc>] (wait_for_common) from [<bf0013b8>] (test_acipher_speed+0x49c/0x740 [tcrypt]) [<bf0013b8>] (test_acipher_speed [tcrypt]) from [<bf003e8c>] (do_test+0x2240/0x30ec [tcrypt]) [<bf003e8c>] (do_test [tcrypt]) from [<bf008048>] (tcrypt_mod_init+0x48/0xa4 [tcrypt]) [<bf008048>] (tcrypt_mod_init [tcrypt]) from [<c010177c>] (do_one_initcall+0x3c/0x16c) [<c010177c>] (do_one_initcall) from [<c0191ff0>] (do_init_module+0x5c/0x1ac) [<c0191ff0>] (do_init_module) from [<c0185610>] (load_module+0x1a30/0x1d08) [<c0185610>] (load_module) from [<c0185ab0>] (SyS_finit_module+0x8c/0x98) [<c0185ab0>] (SyS_finit_module) from [<c01078c0>] (ret_fast_syscall+0x0/0x3c) Fixes: a49e490c7a8a ("crypto: s5p-sss - add S5PV210 advanced crypto engine support") Cc: <stable@vger.kernel.org> Signed-off-by: Krzysztof Kozlowski <k.kozlowski@samsung.com> Tested-by: Marek Szyprowski <m.szyprowski@samsung.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2016-04-22 20:15:23 +08:00
return ret;
}
crypto: s5p-sss - Fix missed interrupts when working with 8 kB blocks The tcrypt testing module on Exynos5422-based Odroid XU3/4 board failed on testing 8 kB size blocks: $ sudo modprobe tcrypt sec=1 mode=500 testing speed of async ecb(aes) (ecb-aes-s5p) encryption test 0 (128 bit key, 16 byte blocks): 21971 operations in 1 seconds (351536 bytes) test 1 (128 bit key, 64 byte blocks): 21731 operations in 1 seconds (1390784 bytes) test 2 (128 bit key, 256 byte blocks): 21932 operations in 1 seconds (5614592 bytes) test 3 (128 bit key, 1024 byte blocks): 21685 operations in 1 seconds (22205440 bytes) test 4 (128 bit key, 8192 byte blocks): This was caused by a race issue of missed BRDMA_DONE ("Block cipher Receiving DMA") interrupt. Device starts processing the data in DMA mode immediately after setting length of DMA block: receiving (FCBRDMAL) or transmitting (FCBTDMAL). The driver sets these lengths from interrupt handler through s5p_set_dma_indata() function (or xxx_setdata()). However the interrupt handler was first dealing with receive buffer (dma-unmap old, dma-map new, set receive block length which starts the operation), then with transmit buffer and finally was clearing pending interrupts (FCINTPEND). Because of the time window between setting receive buffer length and clearing pending interrupts, the operation on receive buffer could end already and driver would miss new interrupt. User manual for Exynos5422 confirms in example code that setting DMA block lengths should be the last operation. The tcrypt hang could be also observed in following blocked-task dmesg: INFO: task modprobe:258 blocked for more than 120 seconds. Not tainted 4.6.0-rc4-next-20160419-00005-g9eac8b7b7753-dirty #42 "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. modprobe D c06b09d8 0 258 256 0x00000000 [<c06b09d8>] (__schedule) from [<c06b0f24>] (schedule+0x40/0xac) [<c06b0f24>] (schedule) from [<c06b49f8>] (schedule_timeout+0x124/0x178) [<c06b49f8>] (schedule_timeout) from [<c06b17fc>] (wait_for_common+0xb8/0x144) [<c06b17fc>] (wait_for_common) from [<bf0013b8>] (test_acipher_speed+0x49c/0x740 [tcrypt]) [<bf0013b8>] (test_acipher_speed [tcrypt]) from [<bf003e8c>] (do_test+0x2240/0x30ec [tcrypt]) [<bf003e8c>] (do_test [tcrypt]) from [<bf008048>] (tcrypt_mod_init+0x48/0xa4 [tcrypt]) [<bf008048>] (tcrypt_mod_init [tcrypt]) from [<c010177c>] (do_one_initcall+0x3c/0x16c) [<c010177c>] (do_one_initcall) from [<c0191ff0>] (do_init_module+0x5c/0x1ac) [<c0191ff0>] (do_init_module) from [<c0185610>] (load_module+0x1a30/0x1d08) [<c0185610>] (load_module) from [<c0185ab0>] (SyS_finit_module+0x8c/0x98) [<c0185ab0>] (SyS_finit_module) from [<c01078c0>] (ret_fast_syscall+0x0/0x3c) Fixes: a49e490c7a8a ("crypto: s5p-sss - add S5PV210 advanced crypto engine support") Cc: <stable@vger.kernel.org> Signed-off-by: Krzysztof Kozlowski <k.kozlowski@samsung.com> Tested-by: Marek Szyprowski <m.szyprowski@samsung.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2016-04-22 20:15:23 +08:00
/*
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
* Returns -ERRNO on error (mapping of new data failed).
* On success returns:
* - 0 if there is no more data,
* - 1 if new receiving (input) data is ready and its address+length
* have to be written to device (by calling s5p_set_dma_indata()).
crypto: s5p-sss - Fix missed interrupts when working with 8 kB blocks The tcrypt testing module on Exynos5422-based Odroid XU3/4 board failed on testing 8 kB size blocks: $ sudo modprobe tcrypt sec=1 mode=500 testing speed of async ecb(aes) (ecb-aes-s5p) encryption test 0 (128 bit key, 16 byte blocks): 21971 operations in 1 seconds (351536 bytes) test 1 (128 bit key, 64 byte blocks): 21731 operations in 1 seconds (1390784 bytes) test 2 (128 bit key, 256 byte blocks): 21932 operations in 1 seconds (5614592 bytes) test 3 (128 bit key, 1024 byte blocks): 21685 operations in 1 seconds (22205440 bytes) test 4 (128 bit key, 8192 byte blocks): This was caused by a race issue of missed BRDMA_DONE ("Block cipher Receiving DMA") interrupt. Device starts processing the data in DMA mode immediately after setting length of DMA block: receiving (FCBRDMAL) or transmitting (FCBTDMAL). The driver sets these lengths from interrupt handler through s5p_set_dma_indata() function (or xxx_setdata()). However the interrupt handler was first dealing with receive buffer (dma-unmap old, dma-map new, set receive block length which starts the operation), then with transmit buffer and finally was clearing pending interrupts (FCINTPEND). Because of the time window between setting receive buffer length and clearing pending interrupts, the operation on receive buffer could end already and driver would miss new interrupt. User manual for Exynos5422 confirms in example code that setting DMA block lengths should be the last operation. The tcrypt hang could be also observed in following blocked-task dmesg: INFO: task modprobe:258 blocked for more than 120 seconds. Not tainted 4.6.0-rc4-next-20160419-00005-g9eac8b7b7753-dirty #42 "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. modprobe D c06b09d8 0 258 256 0x00000000 [<c06b09d8>] (__schedule) from [<c06b0f24>] (schedule+0x40/0xac) [<c06b0f24>] (schedule) from [<c06b49f8>] (schedule_timeout+0x124/0x178) [<c06b49f8>] (schedule_timeout) from [<c06b17fc>] (wait_for_common+0xb8/0x144) [<c06b17fc>] (wait_for_common) from [<bf0013b8>] (test_acipher_speed+0x49c/0x740 [tcrypt]) [<bf0013b8>] (test_acipher_speed [tcrypt]) from [<bf003e8c>] (do_test+0x2240/0x30ec [tcrypt]) [<bf003e8c>] (do_test [tcrypt]) from [<bf008048>] (tcrypt_mod_init+0x48/0xa4 [tcrypt]) [<bf008048>] (tcrypt_mod_init [tcrypt]) from [<c010177c>] (do_one_initcall+0x3c/0x16c) [<c010177c>] (do_one_initcall) from [<c0191ff0>] (do_init_module+0x5c/0x1ac) [<c0191ff0>] (do_init_module) from [<c0185610>] (load_module+0x1a30/0x1d08) [<c0185610>] (load_module) from [<c0185ab0>] (SyS_finit_module+0x8c/0x98) [<c0185ab0>] (SyS_finit_module) from [<c01078c0>] (ret_fast_syscall+0x0/0x3c) Fixes: a49e490c7a8a ("crypto: s5p-sss - add S5PV210 advanced crypto engine support") Cc: <stable@vger.kernel.org> Signed-off-by: Krzysztof Kozlowski <k.kozlowski@samsung.com> Tested-by: Marek Szyprowski <m.szyprowski@samsung.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2016-04-22 20:15:23 +08:00
*/
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
static int s5p_aes_rx(struct s5p_aes_dev *dev/*, bool *set_dma*/)
{
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
int ret = 0;
s5p_unset_indata(dev);
if (!sg_is_last(dev->sg_src)) {
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
ret = s5p_set_indata(dev, sg_next(dev->sg_src));
if (!ret)
ret = 1;
}
crypto: s5p-sss - Fix missed interrupts when working with 8 kB blocks The tcrypt testing module on Exynos5422-based Odroid XU3/4 board failed on testing 8 kB size blocks: $ sudo modprobe tcrypt sec=1 mode=500 testing speed of async ecb(aes) (ecb-aes-s5p) encryption test 0 (128 bit key, 16 byte blocks): 21971 operations in 1 seconds (351536 bytes) test 1 (128 bit key, 64 byte blocks): 21731 operations in 1 seconds (1390784 bytes) test 2 (128 bit key, 256 byte blocks): 21932 operations in 1 seconds (5614592 bytes) test 3 (128 bit key, 1024 byte blocks): 21685 operations in 1 seconds (22205440 bytes) test 4 (128 bit key, 8192 byte blocks): This was caused by a race issue of missed BRDMA_DONE ("Block cipher Receiving DMA") interrupt. Device starts processing the data in DMA mode immediately after setting length of DMA block: receiving (FCBRDMAL) or transmitting (FCBTDMAL). The driver sets these lengths from interrupt handler through s5p_set_dma_indata() function (or xxx_setdata()). However the interrupt handler was first dealing with receive buffer (dma-unmap old, dma-map new, set receive block length which starts the operation), then with transmit buffer and finally was clearing pending interrupts (FCINTPEND). Because of the time window between setting receive buffer length and clearing pending interrupts, the operation on receive buffer could end already and driver would miss new interrupt. User manual for Exynos5422 confirms in example code that setting DMA block lengths should be the last operation. The tcrypt hang could be also observed in following blocked-task dmesg: INFO: task modprobe:258 blocked for more than 120 seconds. Not tainted 4.6.0-rc4-next-20160419-00005-g9eac8b7b7753-dirty #42 "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. modprobe D c06b09d8 0 258 256 0x00000000 [<c06b09d8>] (__schedule) from [<c06b0f24>] (schedule+0x40/0xac) [<c06b0f24>] (schedule) from [<c06b49f8>] (schedule_timeout+0x124/0x178) [<c06b49f8>] (schedule_timeout) from [<c06b17fc>] (wait_for_common+0xb8/0x144) [<c06b17fc>] (wait_for_common) from [<bf0013b8>] (test_acipher_speed+0x49c/0x740 [tcrypt]) [<bf0013b8>] (test_acipher_speed [tcrypt]) from [<bf003e8c>] (do_test+0x2240/0x30ec [tcrypt]) [<bf003e8c>] (do_test [tcrypt]) from [<bf008048>] (tcrypt_mod_init+0x48/0xa4 [tcrypt]) [<bf008048>] (tcrypt_mod_init [tcrypt]) from [<c010177c>] (do_one_initcall+0x3c/0x16c) [<c010177c>] (do_one_initcall) from [<c0191ff0>] (do_init_module+0x5c/0x1ac) [<c0191ff0>] (do_init_module) from [<c0185610>] (load_module+0x1a30/0x1d08) [<c0185610>] (load_module) from [<c0185ab0>] (SyS_finit_module+0x8c/0x98) [<c0185ab0>] (SyS_finit_module) from [<c01078c0>] (ret_fast_syscall+0x0/0x3c) Fixes: a49e490c7a8a ("crypto: s5p-sss - add S5PV210 advanced crypto engine support") Cc: <stable@vger.kernel.org> Signed-off-by: Krzysztof Kozlowski <k.kozlowski@samsung.com> Tested-by: Marek Szyprowski <m.szyprowski@samsung.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2016-04-22 20:15:23 +08:00
return ret;
}
static irqreturn_t s5p_aes_interrupt(int irq, void *dev_id)
{
struct platform_device *pdev = dev_id;
struct s5p_aes_dev *dev = platform_get_drvdata(pdev);
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
int err_dma_tx = 0;
int err_dma_rx = 0;
bool tx_end = false;
unsigned long flags;
uint32_t status;
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
int err;
spin_lock_irqsave(&dev->lock, flags);
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
/*
* Handle rx or tx interrupt. If there is still data (scatterlist did not
* reach end), then map next scatterlist entry.
* In case of such mapping error, s5p_aes_complete() should be called.
*
* If there is no more data in tx scatter list, call s5p_aes_complete()
* and schedule new tasklet.
*/
status = SSS_READ(dev, FCINTSTAT);
if (status & SSS_FCINTSTAT_BRDMAINT)
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
err_dma_rx = s5p_aes_rx(dev);
if (status & SSS_FCINTSTAT_BTDMAINT) {
if (sg_is_last(dev->sg_dst))
tx_end = true;
err_dma_tx = s5p_aes_tx(dev);
}
SSS_WRITE(dev, FCINTPEND, status);
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
if (err_dma_rx < 0) {
err = err_dma_rx;
goto error;
}
if (err_dma_tx < 0) {
err = err_dma_tx;
goto error;
}
if (tx_end) {
s5p_sg_done(dev);
spin_unlock_irqrestore(&dev->lock, flags);
s5p_aes_complete(dev, 0);
/* Device is still busy */
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
tasklet_schedule(&dev->tasklet);
} else {
/*
* Writing length of DMA block (either receiving or
* transmitting) will start the operation immediately, so this
* should be done at the end (even after clearing pending
* interrupts to not miss the interrupt).
*/
if (err_dma_tx == 1)
s5p_set_dma_outdata(dev, dev->sg_dst);
if (err_dma_rx == 1)
s5p_set_dma_indata(dev, dev->sg_src);
crypto: s5p-sss - Fix missed interrupts when working with 8 kB blocks The tcrypt testing module on Exynos5422-based Odroid XU3/4 board failed on testing 8 kB size blocks: $ sudo modprobe tcrypt sec=1 mode=500 testing speed of async ecb(aes) (ecb-aes-s5p) encryption test 0 (128 bit key, 16 byte blocks): 21971 operations in 1 seconds (351536 bytes) test 1 (128 bit key, 64 byte blocks): 21731 operations in 1 seconds (1390784 bytes) test 2 (128 bit key, 256 byte blocks): 21932 operations in 1 seconds (5614592 bytes) test 3 (128 bit key, 1024 byte blocks): 21685 operations in 1 seconds (22205440 bytes) test 4 (128 bit key, 8192 byte blocks): This was caused by a race issue of missed BRDMA_DONE ("Block cipher Receiving DMA") interrupt. Device starts processing the data in DMA mode immediately after setting length of DMA block: receiving (FCBRDMAL) or transmitting (FCBTDMAL). The driver sets these lengths from interrupt handler through s5p_set_dma_indata() function (or xxx_setdata()). However the interrupt handler was first dealing with receive buffer (dma-unmap old, dma-map new, set receive block length which starts the operation), then with transmit buffer and finally was clearing pending interrupts (FCINTPEND). Because of the time window between setting receive buffer length and clearing pending interrupts, the operation on receive buffer could end already and driver would miss new interrupt. User manual for Exynos5422 confirms in example code that setting DMA block lengths should be the last operation. The tcrypt hang could be also observed in following blocked-task dmesg: INFO: task modprobe:258 blocked for more than 120 seconds. Not tainted 4.6.0-rc4-next-20160419-00005-g9eac8b7b7753-dirty #42 "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. modprobe D c06b09d8 0 258 256 0x00000000 [<c06b09d8>] (__schedule) from [<c06b0f24>] (schedule+0x40/0xac) [<c06b0f24>] (schedule) from [<c06b49f8>] (schedule_timeout+0x124/0x178) [<c06b49f8>] (schedule_timeout) from [<c06b17fc>] (wait_for_common+0xb8/0x144) [<c06b17fc>] (wait_for_common) from [<bf0013b8>] (test_acipher_speed+0x49c/0x740 [tcrypt]) [<bf0013b8>] (test_acipher_speed [tcrypt]) from [<bf003e8c>] (do_test+0x2240/0x30ec [tcrypt]) [<bf003e8c>] (do_test [tcrypt]) from [<bf008048>] (tcrypt_mod_init+0x48/0xa4 [tcrypt]) [<bf008048>] (tcrypt_mod_init [tcrypt]) from [<c010177c>] (do_one_initcall+0x3c/0x16c) [<c010177c>] (do_one_initcall) from [<c0191ff0>] (do_init_module+0x5c/0x1ac) [<c0191ff0>] (do_init_module) from [<c0185610>] (load_module+0x1a30/0x1d08) [<c0185610>] (load_module) from [<c0185ab0>] (SyS_finit_module+0x8c/0x98) [<c0185ab0>] (SyS_finit_module) from [<c01078c0>] (ret_fast_syscall+0x0/0x3c) Fixes: a49e490c7a8a ("crypto: s5p-sss - add S5PV210 advanced crypto engine support") Cc: <stable@vger.kernel.org> Signed-off-by: Krzysztof Kozlowski <k.kozlowski@samsung.com> Tested-by: Marek Szyprowski <m.szyprowski@samsung.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2016-04-22 20:15:23 +08:00
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
spin_unlock_irqrestore(&dev->lock, flags);
}
return IRQ_HANDLED;
error:
s5p_sg_done(dev);
dev->busy = false;
spin_unlock_irqrestore(&dev->lock, flags);
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
s5p_aes_complete(dev, err);
return IRQ_HANDLED;
}
static void s5p_set_aes(struct s5p_aes_dev *dev,
uint8_t *key, uint8_t *iv, unsigned int keylen)
{
void __iomem *keystart;
if (iv)
memcpy_toio(dev->aes_ioaddr + SSS_REG_AES_IV_DATA(0), iv, 0x10);
if (keylen == AES_KEYSIZE_256)
keystart = dev->aes_ioaddr + SSS_REG_AES_KEY_DATA(0);
else if (keylen == AES_KEYSIZE_192)
keystart = dev->aes_ioaddr + SSS_REG_AES_KEY_DATA(2);
else
keystart = dev->aes_ioaddr + SSS_REG_AES_KEY_DATA(4);
memcpy_toio(keystart, key, keylen);
}
static bool s5p_is_sg_aligned(struct scatterlist *sg)
{
while (sg) {
if (!IS_ALIGNED(sg->length, AES_BLOCK_SIZE))
return false;
sg = sg_next(sg);
}
return true;
}
static int s5p_set_indata_start(struct s5p_aes_dev *dev,
struct ablkcipher_request *req)
{
struct scatterlist *sg;
int err;
dev->sg_src_cpy = NULL;
sg = req->src;
if (!s5p_is_sg_aligned(sg)) {
dev_dbg(dev->dev,
"At least one unaligned source scatter list, making a copy\n");
err = s5p_make_sg_cpy(dev, sg, &dev->sg_src_cpy);
if (err)
return err;
sg = dev->sg_src_cpy;
}
err = s5p_set_indata(dev, sg);
if (err) {
s5p_free_sg_cpy(dev, &dev->sg_src_cpy);
return err;
}
return 0;
}
static int s5p_set_outdata_start(struct s5p_aes_dev *dev,
struct ablkcipher_request *req)
{
struct scatterlist *sg;
int err;
dev->sg_dst_cpy = NULL;
sg = req->dst;
if (!s5p_is_sg_aligned(sg)) {
dev_dbg(dev->dev,
"At least one unaligned dest scatter list, making a copy\n");
err = s5p_make_sg_cpy(dev, sg, &dev->sg_dst_cpy);
if (err)
return err;
sg = dev->sg_dst_cpy;
}
err = s5p_set_outdata(dev, sg);
if (err) {
s5p_free_sg_cpy(dev, &dev->sg_dst_cpy);
return err;
}
return 0;
}
static void s5p_aes_crypt_start(struct s5p_aes_dev *dev, unsigned long mode)
{
struct ablkcipher_request *req = dev->req;
uint32_t aes_control;
unsigned long flags;
int err;
aes_control = SSS_AES_KEY_CHANGE_MODE;
if (mode & FLAGS_AES_DECRYPT)
aes_control |= SSS_AES_MODE_DECRYPT;
if ((mode & FLAGS_AES_MODE_MASK) == FLAGS_AES_CBC)
aes_control |= SSS_AES_CHAIN_MODE_CBC;
else if ((mode & FLAGS_AES_MODE_MASK) == FLAGS_AES_CTR)
aes_control |= SSS_AES_CHAIN_MODE_CTR;
if (dev->ctx->keylen == AES_KEYSIZE_192)
aes_control |= SSS_AES_KEY_SIZE_192;
else if (dev->ctx->keylen == AES_KEYSIZE_256)
aes_control |= SSS_AES_KEY_SIZE_256;
aes_control |= SSS_AES_FIFO_MODE;
/* as a variant it is possible to use byte swapping on DMA side */
aes_control |= SSS_AES_BYTESWAP_DI
| SSS_AES_BYTESWAP_DO
| SSS_AES_BYTESWAP_IV
| SSS_AES_BYTESWAP_KEY
| SSS_AES_BYTESWAP_CNT;
spin_lock_irqsave(&dev->lock, flags);
SSS_WRITE(dev, FCINTENCLR,
SSS_FCINTENCLR_BTDMAINTENCLR | SSS_FCINTENCLR_BRDMAINTENCLR);
SSS_WRITE(dev, FCFIFOCTRL, 0x00);
err = s5p_set_indata_start(dev, req);
if (err)
goto indata_error;
err = s5p_set_outdata_start(dev, req);
if (err)
goto outdata_error;
SSS_AES_WRITE(dev, AES_CONTROL, aes_control);
s5p_set_aes(dev, dev->ctx->aes_key, req->info, dev->ctx->keylen);
s5p_set_dma_indata(dev, dev->sg_src);
s5p_set_dma_outdata(dev, dev->sg_dst);
SSS_WRITE(dev, FCINTENSET,
SSS_FCINTENSET_BTDMAINTENSET | SSS_FCINTENSET_BRDMAINTENSET);
spin_unlock_irqrestore(&dev->lock, flags);
return;
outdata_error:
s5p_unset_indata(dev);
indata_error:
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
s5p_sg_done(dev);
dev->busy = false;
spin_unlock_irqrestore(&dev->lock, flags);
crypto: s5p-sss - Fix spinlock recursion on LRW(AES) Running TCRYPT with LRW compiled causes spinlock recursion: testing speed of async lrw(aes) (lrw(ecb-aes-s5p)) encryption tcrypt: test 0 (256 bit key, 16 byte blocks): 19007 operations in 1 seconds (304112 bytes) tcrypt: test 1 (256 bit key, 64 byte blocks): 15753 operations in 1 seconds (1008192 bytes) tcrypt: test 2 (256 bit key, 256 byte blocks): 14293 operations in 1 seconds (3659008 bytes) tcrypt: test 3 (256 bit key, 1024 byte blocks): 11906 operations in 1 seconds (12191744 bytes) tcrypt: test 4 (256 bit key, 8192 byte blocks): BUG: spinlock recursion on CPU#1, irq/84-10830000/89  lock: 0xeea99a68, .magic: dead4ead, .owner: irq/84-10830000/89, .owner_cpu: 1 CPU: 1 PID: 89 Comm: irq/84-10830000 Not tainted 4.11.0-rc1-00001-g897ca6d0800d #559 Hardware name: SAMSUNG EXYNOS (Flattened Device Tree) [<c010e1ec>] (unwind_backtrace) from [<c010ae1c>] (show_stack+0x10/0x14) [<c010ae1c>] (show_stack) from [<c03449c0>] (dump_stack+0x78/0x8c) [<c03449c0>] (dump_stack) from [<c015de68>] (do_raw_spin_lock+0x11c/0x120) [<c015de68>] (do_raw_spin_lock) from [<c0720110>] (_raw_spin_lock_irqsave+0x20/0x28) [<c0720110>] (_raw_spin_lock_irqsave) from [<c0572ca0>] (s5p_aes_crypt+0x2c/0xb4) [<c0572ca0>] (s5p_aes_crypt) from [<bf1d8aa4>] (do_encrypt+0x78/0xb0 [lrw]) [<bf1d8aa4>] (do_encrypt [lrw]) from [<bf1d8b00>] (encrypt_done+0x24/0x54 [lrw]) [<bf1d8b00>] (encrypt_done [lrw]) from [<c05732a0>] (s5p_aes_complete+0x60/0xcc) [<c05732a0>] (s5p_aes_complete) from [<c0573440>] (s5p_aes_interrupt+0x134/0x1a0) [<c0573440>] (s5p_aes_interrupt) from [<c01667c4>] (irq_thread_fn+0x1c/0x54) [<c01667c4>] (irq_thread_fn) from [<c0166a98>] (irq_thread+0x12c/0x1e0) [<c0166a98>] (irq_thread) from [<c0136a28>] (kthread+0x108/0x138) [<c0136a28>] (kthread) from [<c0107778>] (ret_from_fork+0x14/0x3c) Interrupt handling routine was calling req->base.complete() under spinlock. In most cases this wasn't fatal but when combined with some of the cipher modes (like LRW) this caused recursion - starting the new encryption (s5p_aes_crypt()) while still holding the spinlock from previous round (s5p_aes_complete()). Beside that, the s5p_aes_interrupt() error handling path could execute two completions in case of error for RX and TX blocks. Rewrite the interrupt handling routine and the completion by: 1. Splitting the operations on scatterlist copies from s5p_aes_complete() into separate s5p_sg_done(). This still should be done under lock. The s5p_aes_complete() now only calls req->base.complete() and it has to be called outside of lock. 2. Moving the s5p_aes_complete() out of spinlock critical sections. In interrupt service routine s5p_aes_interrupts(), it appeared in few places, including error paths inside other functions called from ISR. This code was not so obvious to read so simplify it by putting the s5p_aes_complete() only within ISR level. Reported-by: Nathan Royce <nroycea+kernel@gmail.com> Cc: <stable@vger.kernel.org> # v4.10.x: 07de4bc88c crypto: s5p-sss - Fix completing Cc: <stable@vger.kernel.org> # v4.10.x Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-09 05:14:20 +08:00
s5p_aes_complete(dev, err);
}
static void s5p_tasklet_cb(unsigned long data)
{
struct s5p_aes_dev *dev = (struct s5p_aes_dev *)data;
struct crypto_async_request *async_req, *backlog;
struct s5p_aes_reqctx *reqctx;
unsigned long flags;
spin_lock_irqsave(&dev->lock, flags);
backlog = crypto_get_backlog(&dev->queue);
async_req = crypto_dequeue_request(&dev->queue);
if (!async_req) {
dev->busy = false;
spin_unlock_irqrestore(&dev->lock, flags);
return;
}
spin_unlock_irqrestore(&dev->lock, flags);
if (backlog)
backlog->complete(backlog, -EINPROGRESS);
dev->req = ablkcipher_request_cast(async_req);
dev->ctx = crypto_tfm_ctx(dev->req->base.tfm);
reqctx = ablkcipher_request_ctx(dev->req);
s5p_aes_crypt_start(dev, reqctx->mode);
}
static int s5p_aes_handle_req(struct s5p_aes_dev *dev,
struct ablkcipher_request *req)
{
unsigned long flags;
int err;
spin_lock_irqsave(&dev->lock, flags);
err = ablkcipher_enqueue_request(&dev->queue, req);
if (dev->busy) {
spin_unlock_irqrestore(&dev->lock, flags);
goto exit;
}
dev->busy = true;
spin_unlock_irqrestore(&dev->lock, flags);
tasklet_schedule(&dev->tasklet);
exit:
return err;
}
static int s5p_aes_crypt(struct ablkcipher_request *req, unsigned long mode)
{
struct crypto_ablkcipher *tfm = crypto_ablkcipher_reqtfm(req);
struct s5p_aes_reqctx *reqctx = ablkcipher_request_ctx(req);
struct s5p_aes_ctx *ctx = crypto_ablkcipher_ctx(tfm);
struct s5p_aes_dev *dev = ctx->dev;
if (!IS_ALIGNED(req->nbytes, AES_BLOCK_SIZE)) {
dev_err(dev->dev, "request size is not exact amount of AES blocks\n");
return -EINVAL;
}
reqctx->mode = mode;
return s5p_aes_handle_req(dev, req);
}
static int s5p_aes_setkey(struct crypto_ablkcipher *cipher,
const uint8_t *key, unsigned int keylen)
{
struct crypto_tfm *tfm = crypto_ablkcipher_tfm(cipher);
struct s5p_aes_ctx *ctx = crypto_tfm_ctx(tfm);
if (keylen != AES_KEYSIZE_128 &&
keylen != AES_KEYSIZE_192 &&
keylen != AES_KEYSIZE_256)
return -EINVAL;
memcpy(ctx->aes_key, key, keylen);
ctx->keylen = keylen;
return 0;
}
static int s5p_aes_ecb_encrypt(struct ablkcipher_request *req)
{
return s5p_aes_crypt(req, 0);
}
static int s5p_aes_ecb_decrypt(struct ablkcipher_request *req)
{
return s5p_aes_crypt(req, FLAGS_AES_DECRYPT);
}
static int s5p_aes_cbc_encrypt(struct ablkcipher_request *req)
{
return s5p_aes_crypt(req, FLAGS_AES_CBC);
}
static int s5p_aes_cbc_decrypt(struct ablkcipher_request *req)
{
return s5p_aes_crypt(req, FLAGS_AES_DECRYPT | FLAGS_AES_CBC);
}
static int s5p_aes_cra_init(struct crypto_tfm *tfm)
{
struct s5p_aes_ctx *ctx = crypto_tfm_ctx(tfm);
ctx->dev = s5p_dev;
tfm->crt_ablkcipher.reqsize = sizeof(struct s5p_aes_reqctx);
return 0;
}
static struct crypto_alg algs[] = {
{
.cra_name = "ecb(aes)",
.cra_driver_name = "ecb-aes-s5p",
.cra_priority = 100,
.cra_flags = CRYPTO_ALG_TYPE_ABLKCIPHER |
CRYPTO_ALG_ASYNC |
CRYPTO_ALG_KERN_DRIVER_ONLY,
.cra_blocksize = AES_BLOCK_SIZE,
.cra_ctxsize = sizeof(struct s5p_aes_ctx),
.cra_alignmask = 0x0f,
.cra_type = &crypto_ablkcipher_type,
.cra_module = THIS_MODULE,
.cra_init = s5p_aes_cra_init,
.cra_u.ablkcipher = {
.min_keysize = AES_MIN_KEY_SIZE,
.max_keysize = AES_MAX_KEY_SIZE,
.setkey = s5p_aes_setkey,
.encrypt = s5p_aes_ecb_encrypt,
.decrypt = s5p_aes_ecb_decrypt,
}
},
{
.cra_name = "cbc(aes)",
.cra_driver_name = "cbc-aes-s5p",
.cra_priority = 100,
.cra_flags = CRYPTO_ALG_TYPE_ABLKCIPHER |
CRYPTO_ALG_ASYNC |
CRYPTO_ALG_KERN_DRIVER_ONLY,
.cra_blocksize = AES_BLOCK_SIZE,
.cra_ctxsize = sizeof(struct s5p_aes_ctx),
.cra_alignmask = 0x0f,
.cra_type = &crypto_ablkcipher_type,
.cra_module = THIS_MODULE,
.cra_init = s5p_aes_cra_init,
.cra_u.ablkcipher = {
.min_keysize = AES_MIN_KEY_SIZE,
.max_keysize = AES_MAX_KEY_SIZE,
.ivsize = AES_BLOCK_SIZE,
.setkey = s5p_aes_setkey,
.encrypt = s5p_aes_cbc_encrypt,
.decrypt = s5p_aes_cbc_decrypt,
}
},
};
static int s5p_aes_probe(struct platform_device *pdev)
{
struct device *dev = &pdev->dev;
int i, j, err = -ENODEV;
struct samsung_aes_variant *variant;
struct s5p_aes_dev *pdata;
struct resource *res;
if (s5p_dev)
return -EEXIST;
pdata = devm_kzalloc(dev, sizeof(*pdata), GFP_KERNEL);
if (!pdata)
return -ENOMEM;
res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
pdata->ioaddr = devm_ioremap_resource(&pdev->dev, res);
if (IS_ERR(pdata->ioaddr))
return PTR_ERR(pdata->ioaddr);
variant = find_s5p_sss_version(pdev);
pdata->clk = devm_clk_get(dev, "secss");
if (IS_ERR(pdata->clk)) {
dev_err(dev, "failed to find secss clock source\n");
return -ENOENT;
}
err = clk_prepare_enable(pdata->clk);
if (err < 0) {
dev_err(dev, "Enabling SSS clk failed, err %d\n", err);
return err;
}
spin_lock_init(&pdata->lock);
pdata->aes_ioaddr = pdata->ioaddr + variant->aes_offset;
pdata->irq_fc = platform_get_irq(pdev, 0);
if (pdata->irq_fc < 0) {
err = pdata->irq_fc;
dev_warn(dev, "feed control interrupt is not available.\n");
goto err_irq;
}
err = devm_request_threaded_irq(dev, pdata->irq_fc, NULL,
s5p_aes_interrupt, IRQF_ONESHOT,
pdev->name, pdev);
if (err < 0) {
dev_warn(dev, "feed control interrupt is not available.\n");
goto err_irq;
}
pdata->busy = false;
pdata->dev = dev;
platform_set_drvdata(pdev, pdata);
s5p_dev = pdata;
tasklet_init(&pdata->tasklet, s5p_tasklet_cb, (unsigned long)pdata);
crypto_init_queue(&pdata->queue, CRYPTO_QUEUE_LEN);
for (i = 0; i < ARRAY_SIZE(algs); i++) {
err = crypto_register_alg(&algs[i]);
if (err)
goto err_algs;
}
dev_info(dev, "s5p-sss driver registered\n");
return 0;
err_algs:
dev_err(dev, "can't register '%s': %d\n", algs[i].cra_name, err);
for (j = 0; j < i; j++)
crypto_unregister_alg(&algs[j]);
tasklet_kill(&pdata->tasklet);
err_irq:
clk_disable_unprepare(pdata->clk);
s5p_dev = NULL;
return err;
}
static int s5p_aes_remove(struct platform_device *pdev)
{
struct s5p_aes_dev *pdata = platform_get_drvdata(pdev);
int i;
if (!pdata)
return -ENODEV;
for (i = 0; i < ARRAY_SIZE(algs); i++)
crypto_unregister_alg(&algs[i]);
tasklet_kill(&pdata->tasklet);
clk_disable_unprepare(pdata->clk);
s5p_dev = NULL;
return 0;
}
static struct platform_driver s5p_aes_crypto = {
.probe = s5p_aes_probe,
.remove = s5p_aes_remove,
.driver = {
.name = "s5p-secss",
.of_match_table = s5p_sss_dt_match,
},
};
module_platform_driver(s5p_aes_crypto);
MODULE_DESCRIPTION("S5PV210 AES hw acceleration support.");
MODULE_LICENSE("GPL v2");
MODULE_AUTHOR("Vladimir Zapolskiy <vzapolskiy@gmail.com>");