[PATCH] dmaengine: pl330: do not generate unaligned access

[Robin Murphy]

On 07/12/16 13:17, Vladimir Murzin wrote:
> When PL330 is used with !MMU the following fault is seen:
> 
> Unhandled fault: alignment exception (0x801) at 0x8f26a002
> Internal error: : 801 [#1] ARM
> Modules linked in:
> CPU: 0 PID: 640 Comm: dma0chan0-copy0 Not tainted 4.8.0-6a82063-clean+ #1600
> Hardware name: ARM-Versatile Express
> task: 8f1baa80 task.stack: 8e6fe000
> PC is at _setup_req+0x4c/0x350
> LR is at 0x8f2cbc00
> pc : [<801ea538>]    lr : [<8f2cbc00>]    psr: 60000093
> sp : 8e6ffdc0  ip : 00000000  fp : 00000000
> r10: 00000000  r9 : 8f2cba10  r8 : 8f2cbc00
> r7 : 80000013  r6 : 8f21a050  r5 : 8f21a000  r4 : 8f2ac800
> r3 : 8e6ffe18  r2 : 00944251  r1 : ffffffbc  r0 : 8f26a000
> Flags: nZCv  IRQs off  FIQs on  Mode SVC_32  ISA ARM  Segment none
> Control: 00c5387c
> Process dma0chan0-copy0 (pid: 640, stack limit = 0x8e6fe210)
> Stack: (0x8e6ffdc0 to 0x8e700000)
> fdc0: 00000001 60000093 00000000 8f2cba10 8f26a000 00000004 8f0ae000 8f2cbc00
> fde0: 8f0ae000 8f2ac800 8f21a000 8f21a050 80000013 8f2cbc00 8f2cba10 00000000
> fe00: 60000093 801ebca0 8e6ffe18 000013ff 40000093 00000000 00944251 8f2ac800
> fe20: a0000013 8f2b1320 00001986 00000000 00000001 000013ff 8f1e4f00 8f2cba10
> fe40: 8e6fff6c 801e9044 00000003 00000000 fef98c80 002faf07 8e6ffe7c 00000000
> fe60: 00000002 00000000 00001986 8f1f158d 8f1e4f00 80568de4 00000002 00000000
> fe80: 00001986 8f1f53ff 40000001 80580500 8f1f158d 8001e00c 00000000 cfdfdfdf
> fea0: fdae2a25 00000001 00000004 8e6fe000 00000008 00000010 00000000 00000005
> fec0: 8f2b1330 8f2b1334 8e6ffe80 8e6ffe8c 00001986 00000000 8f21a014 00000001
> fee0: 8e6ffe60 8e6ffe78 00000002 00000000 000013ff 00000001 80568de4 8f1e8018
> ff00: 0000158d 8055ec30 00000001 803f6b00 00001986 8f2cba10 fdae2a25 00000001
> ff20: 8f1baca8 8e6fff24 8e6fff24 00000000 8e6fff24 ac6f3037 00000000 00000000
> ff40: 00000000 8e6fe000 8f1e4f40 00000000 8f1e4f40 8f1e4f00 801e84ec 00000000
> ff60: 00000000 00000000 00000000 80031714 dfdfdfcf 00000000 dfdfdfcf 8f1e4f00
> ff80: 00000000 8e6fff84 8e6fff84 00000000 8e6fff90 8e6fff90 8e6fffac 8f1e4f40
> ffa0: 80031640 00000000 00000000 8000f548 00000000 00000000 00000000 00000000
> ffc0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
> ffe0: 00000000 00000000 00000000 00000000 00000013 00000000 dfdfdfcf cfdfdfdf
> [<801ea538>] (_setup_req) from [<801ebca0>] (pl330_tasklet+0x41c/0x490)
> [<801ebca0>] (pl330_tasklet) from [<801e9044>] (dmatest_func+0xb58/0x149c)
> [<801e9044>] (dmatest_func) from [<80031714>] (kthread+0xd4/0xec)
> [<80031714>] (kthread) from [<8000f548>] (ret_from_fork+0x14/0x2c)
> Code: e3a03001 e3e01043 e5c03001 e59d3048 (e5802002)
> 
> This happens because _emit_{ADDH,MOV,GO) accessing to unaligned data
> while writing to buffer. Fix it with writing to buffer byte by byte.
> 
> Cc: Robin Murphy <robin.murphy at arm.com>
> Cc: Vinod Koul <vinod.koul at intel.com>
> Signed-off-by: Vladimir Murzin <vladimir.murzin at arm.com>

Reviewed-by: Robin Murphy <robin.murphy at arm.com>

I don't have a NOMMU setup, but I checked arm64 (both little- and
big-endian) sees no regressions:

Tested-by: Robin Murphy <robin.murphy at arm.com>

Cheers,
Robin.

> ---
>  drivers/dma/pl330.c |   15 ++++++++++-----
>  1 file changed, 10 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/dma/pl330.c b/drivers/dma/pl330.c
> index 030fe05..e4c1ba7 100644
> --- a/drivers/dma/pl330.c
> +++ b/drivers/dma/pl330.c
> @@ -570,7 +570,8 @@ static inline u32 _emit_ADDH(unsigned dry_run, u8 buf[],
>  
>  	buf[0] = CMD_DMAADDH;
>  	buf[0] |= (da << 1);
> -	*((__le16 *)&buf[1]) = cpu_to_le16(val);
> +	buf[1] = val;
> +	buf[2] = val >> 8;
>  
>  	PL330_DBGCMD_DUMP(SZ_DMAADDH, "\tDMAADDH %s %u\n",
>  		da == 1 ? "DA" : "SA", val);
> @@ -724,7 +725,10 @@ static inline u32 _emit_MOV(unsigned dry_run, u8 buf[],
>  
>  	buf[0] = CMD_DMAMOV;
>  	buf[1] = dst;
> -	*((__le32 *)&buf[2]) = cpu_to_le32(val);
> +	buf[2] = val;
> +	buf[3] = val >> 8;
> +	buf[4] = val >> 16;
> +	buf[5] = val >> 24;
>  
>  	PL330_DBGCMD_DUMP(SZ_DMAMOV, "\tDMAMOV %s 0x%x\n",
>  		dst == SAR ? "SAR" : (dst == DAR ? "DAR" : "CCR"), val);
> @@ -899,10 +903,11 @@ static inline u32 _emit_GO(unsigned dry_run, u8 buf[],
>  
>  	buf[0] = CMD_DMAGO;
>  	buf[0] |= (ns << 1);
> -
>  	buf[1] = chan & 0x7;
> -
> -	*((__le32 *)&buf[2]) = cpu_to_le32(addr);
> +	buf[2] = addr;
> +	buf[3] = addr >> 8;
> +	buf[4] = addr >> 16;
> +	buf[5] = addr >> 24;
>  
>  	return SZ_DMAGO;
>  }
>