linux-kernel - Re: [PATCH] dmaengine: qcom-bam: fix circular buffer handling

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <f4522b78-b406-954c-57b7-923e6ab31f96@codeaurora.org>
Date:   Tue, 18 Jun 2019 12:43:35 +0530
From:   Sricharan R <sricharan@...eaurora.org>
To:     Srinivas Kandagatla <srinivas.kandagatla@...aro.org>,
        vkoul@...nel.org
Cc:     dmaengine@...r.kernel.org, linux-kernel@...r.kernel.org,
        linux-arm-msm@...r.kernel.org
Subject: Re: [PATCH] dmaengine: qcom-bam: fix circular buffer handling

Hi Srini,

On 6/14/2019 7:50 PM, Srinivas Kandagatla wrote:
> For some reason arguments to most of the circular buffers
> macros are used in reverse, tail is used for head and vice versa.
> 
> This leads to bam thinking that there is an extra descriptor at the
> end and leading to retransmitting descriptor which was not scheduled
> by any driver. This happens after MAX_DESCRIPTORS (4096) are scheduled
> and done, so most of the drivers would not notice this, unless they are
> heavily using bam dma. Originally found this issue while testing
> SoundWire over SlimBus on DB845c which uses DMA very heavily for
> read/writes.
> 
> Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@...aro.org>
> ---
>  drivers/dma/qcom/bam_dma.c | 9 ++++-----
>  1 file changed, 4 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/dma/qcom/bam_dma.c b/drivers/dma/qcom/bam_dma.c
> index cb860cb53c27..43d7b0a9713a 100644
> --- a/drivers/dma/qcom/bam_dma.c
> +++ b/drivers/dma/qcom/bam_dma.c
> @@ -350,8 +350,8 @@ static const struct reg_offset_data bam_v1_7_reg_info[] = {
>  #define BAM_DESC_FIFO_SIZE	SZ_32K
>  #define MAX_DESCRIPTORS (BAM_DESC_FIFO_SIZE / sizeof(struct bam_desc_hw) - 1)
>  #define BAM_FIFO_SIZE	(SZ_32K - 8)
> -#define IS_BUSY(chan)	(CIRC_SPACE(bchan->tail, bchan->head,\
> -			 MAX_DESCRIPTORS + 1) == 0)
> +#define IS_BUSY(chan)	(CIRC_SPACE(bchan->head, bchan->tail,\
> +			 MAX_DESCRIPTORS) == 0)
>  
>  struct bam_chan {
>  	struct virt_dma_chan vc;
> @@ -806,7 +806,7 @@ static u32 process_channel_irqs(struct bam_device *bdev)
>  		offset /= sizeof(struct bam_desc_hw);
>  
>  		/* Number of bytes available to read */
> -		avail = CIRC_CNT(offset, bchan->head, MAX_DESCRIPTORS + 1);
> +		avail = CIRC_CNT(bchan->head, offset, MAX_DESCRIPTORS);
>
 one question, so MAX_DESCRIPTORS is already a mask,
    #define MAX_DESCRIPTORS (BAM_DESC_FIFO_SIZE / sizeof(struct bam_desc_hw) - 1)

 CIRC_CNT/SPACE macros also does a size - 1, so would it not be a problem if we
 just pass MAX_DESCRIPTORS ?

Regards,
 Sricharan
  
>  		list_for_each_entry_safe(async_desc, tmp,
>  					 &bchan->desc_list, desc_node) {
> @@ -997,8 +997,7 @@ static void bam_start_dma(struct bam_chan *bchan)
>  			bam_apply_new_config(bchan, async_desc->dir);
>  
>  		desc = async_desc->curr_desc;
> -		avail = CIRC_SPACE(bchan->tail, bchan->head,
> -				   MAX_DESCRIPTORS + 1);
> +		avail = CIRC_SPACE(bchan->head, bchan->tail, MAX_DESCRIPTORS);
>  
>  		if (async_desc->num_desc > avail)
>  			async_desc->xfer_len = avail;
> 

-- 
"QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation