[PATCH 2/3] mach64: fix image corruption due to reading accelerator registers

[Bartlomiej Zolnierkiewicz]

On 08/27/2018 02:55 PM, Ville Syrjälä wrote:
> On Sat, Aug 25, 2018 at 03:51:52PM -0400, Mikulas Patocka wrote:
>> Reading the registers without waiting for engine idle returns
>> unpredictable values. These unpredictable values result in display
>> corruption - if atyfb_imageblit reads the content of DP_PIX_WIDTH with the
>> bit DP_HOST_TRIPLE_EN set (from previous invocation), the driver would
>> never ever clear the bit, resulting in display corruption.
>>
>> We don't want to wait for idle because it would degrade performance, so
>> this patch modifies the driver so that it never reads accelerator
>> registers.
>>
>> HOST_CNTL doesn't have to be read, we can just write it with
>> HOST_BYTE_ALIGN because no other part of the driver cares if
>> HOST_BYTE_ALIGN is set.
>>
>> DP_PIX_WIDTH is written in the functions atyfb_copyarea and atyfb_fillrect
>> with the default value and in atyfb_imageblit with the value set according
>> to the source image data.
>>
>> Signed-off-by: Mikulas Patocka <mpatocka at redhat.com>
>> Cc: stable at vger.kernel.org
>>
>> ---
>>  drivers/video/fbdev/aty/mach64_accel.c |   22 +++++++++-------------
>>  1 file changed, 9 insertions(+), 13 deletions(-)
>>
>> Index: linux-stable/drivers/video/fbdev/aty/mach64_accel.c
>> ===================================================================
>> --- linux-stable.orig/drivers/video/fbdev/aty/mach64_accel.c	2018-08-24 19:51:34.000000000 +0200
>> +++ linux-stable/drivers/video/fbdev/aty/mach64_accel.c	2018-08-24 20:28:55.000000000 +0200
>> @@ -127,7 +127,7 @@ void aty_init_engine(struct atyfb_par *p
>>  
>>  	/* set host attributes */
>>  	wait_for_fifo(13, par);
>> -	aty_st_le32(HOST_CNTL, 0, par);
>> +	aty_st_le32(HOST_CNTL, HOST_BYTE_ALIGN, par);
>>  
>>  	/* set pattern attributes */
>>  	aty_st_le32(PAT_REG0, 0, par);
>> @@ -233,7 +233,8 @@ void atyfb_copyarea(struct fb_info *info
>>  		rotation = rotation24bpp(dx, direction);
>>  	}
>>  
>> -	wait_for_fifo(4, par);
>> +	wait_for_fifo(5, par);
>> +	aty_st_le32(DP_PIX_WIDTH, par->crtc.dp_pix_width, par);
>>  	aty_st_le32(DP_SRC, FRGD_SRC_BLIT, par);
>>  	aty_st_le32(SRC_Y_X, (sx << 16) | sy, par);
>>  	aty_st_le32(SRC_HEIGHT1_WIDTH1, (width << 16) | area->height, par);
>> @@ -269,7 +270,8 @@ void atyfb_fillrect(struct fb_info *info
>>  		rotation = rotation24bpp(dx, DST_X_LEFT_TO_RIGHT);
>>  	}
>>  
>> -	wait_for_fifo(3, par);
>> +	wait_for_fifo(4, par);
>> +	aty_st_le32(DP_PIX_WIDTH, par->crtc.dp_pix_width, par);
>>  	aty_st_le32(DP_FRGD_CLR, color, par);
>>  	aty_st_le32(DP_SRC,
>>  		    BKGD_SRC_BKGD_CLR | FRGD_SRC_FRGD_CLR | MONO_SRC_ONE,
>> @@ -284,7 +286,7 @@ void atyfb_imageblit(struct fb_info *inf
>>  {
>>  	struct atyfb_par *par = (struct atyfb_par *) info->par;
>>  	u32 src_bytes, dx = image->dx, dy = image->dy, width = image->width;
>> -	u32 pix_width_save, pix_width, host_cntl, rotation = 0, src, mix;
>> +	u32 pix_width, rotation = 0, src, mix;
>>  
>>  	if (par->asleep)
>>  		return;
>> @@ -296,8 +298,7 @@ void atyfb_imageblit(struct fb_info *inf
>>  		return;
>>  	}
>>  
>> -	pix_width = pix_width_save = aty_ld_le32(DP_PIX_WIDTH, par);
>> -	host_cntl = aty_ld_le32(HOST_CNTL, par) | HOST_BYTE_ALIGN;
>> +	pix_width = par->crtc.dp_pix_width;
>>  
>>  	switch (image->depth) {
>>  	case 1:
>> @@ -370,12 +371,11 @@ void atyfb_imageblit(struct fb_info *inf
>>  		mix = FRGD_MIX_D_XOR_S | BKGD_MIX_D;
>>  	}
>>  
>> -	wait_for_fifo(6, par);
>> -	aty_st_le32(DP_WRITE_MASK, 0xFFFFFFFF, par);
> 
> Looks like init_engine() sets this one for us. So dropping should be ok.
> 
>> +	wait_for_fifo(5, par);
>>  	aty_st_le32(DP_PIX_WIDTH, pix_width, par);
>>  	aty_st_le32(DP_MIX, mix, par);
>>  	aty_st_le32(DP_SRC, src, par);
>> -	aty_st_le32(HOST_CNTL, host_cntl, par);
>> +	aty_st_le32(HOST_CNTL, HOST_BYTE_ALIGN, par);
> 
> Presumably we could drop this as well since it never changes.
> 
> Either way
> Reviewed-by: Ville Syrjälä <syrjala at sci.fi>

Patch queued for 4.20, thanks.

Best regards,
--
Bartlomiej Zolnierkiewicz
Samsung R&D Institute Poland
Samsung Electronics