On Fri, Jan 10, 2025 at 10:44:09AM +0530, Md Sadre Alam wrote:
> This patch fixes a bug introduced in the previous commit where the
> BAM_DESC_CNT_TRSHLD register was conditionally written based on BAM-NDP
> mode. Additionally, it addresses an issue where reading the BAM_REVISION
> register hangs if num-ees is not zero. A check has been added to prevent
> this.
>
> Cc: stable@xxxxxxxxxxxxxxx
> Fixes: 57a7138d0627 ("dmaengine: qcom: bam_dma: Avoid writing unavailable register")
> Reported-by: Georgi Djakov <djakov@xxxxxxxxxx>
> Link: https://lore.kernel.org/lkml/9ef3daa8-cdb1-49f2-8d19-a72d6210ff3a@xxxxxxxxxx/
> Signed-off-by: Md Sadre Alam <quic_mdalam@xxxxxxxxxxx>
> ---
> drivers/dma/qcom/bam_dma.c | 23 ++++++++++++++++-------
> 1 file changed, 16 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/dma/qcom/bam_dma.c b/drivers/dma/qcom/bam_dma.c
> index c14557efd577..2b88b27f2f91 100644
> --- a/drivers/dma/qcom/bam_dma.c
> +++ b/drivers/dma/qcom/bam_dma.c
> @@ -445,11 +445,15 @@ static void bam_reset(struct bam_device *bdev)
> writel_relaxed(val, bam_addr(bdev, 0, BAM_CTRL));
>
> /* set descriptor threshold, start with 4 bytes */
> - if (in_range(bdev->bam_revision, BAM_NDP_REVISION_START,
> - BAM_NDP_REVISION_END))
> + if (!bdev->num_ees && in_range(bdev->bam_revision, BAM_NDP_REVISION_START,
> + BAM_NDP_REVISION_END))
> writel_relaxed(DEFAULT_CNT_THRSHLD,
> bam_addr(bdev, 0, BAM_DESC_CNT_TRSHLD));
>
> + if (bdev->num_ees && !bdev->bam_revision)
> + writel_relaxed(DEFAULT_CNT_THRSHLD, bam_addr(bdev, 0,
> + BAM_DESC_CNT_TRSHLD));
> +
> /* Enable default set of h/w workarounds, ie all except BAM_FULL_PIPE */
> writel_relaxed(BAM_CNFG_BITS_DEFAULT, bam_addr(bdev, 0, BAM_CNFG_BITS));
>
> @@ -1006,10 +1010,14 @@ static void bam_apply_new_config(struct bam_chan *bchan,
> maxburst = bchan->slave.src_maxburst;
> else
> maxburst = bchan->slave.dst_maxburst;
> - if (in_range(bdev->bam_revision, BAM_NDP_REVISION_START,
> - BAM_NDP_REVISION_END))
> + if (!bdev->num_ees && in_range(bdev->bam_revision, BAM_NDP_REVISION_START,
> + BAM_NDP_REVISION_END))
> writel_relaxed(maxburst,
> bam_addr(bdev, 0, BAM_DESC_CNT_TRSHLD));
> +
> + if (bdev->num_ees && !bdev->bam_revision)
> + writel_relaxed(DEFAULT_CNT_THRSHLD, bam_addr(bdev, 0,
> + BAM_DESC_CNT_TRSHLD));
I guess you meant writel_relaxed(maxburst, ...) here?
This patch is quite confusing. We shouldn't duplicate the register
writes here just to have different handling for if (bdev->num_ees) and
if (!bdev->num_ees).
Also, num-ees is unrelated to the question if the BAM is BAM-NDP or
BAM-Lite. Typically we specify qcom,num-ees in the device tree for a BAM
if the BAM is either:
- Controlled remotely (= powered on and initialized outside of Linux)
This is the case for the SLIMbus BAM Georgi mentioned.
- Powered remotely (= powered on outside of Linux, but must be
initialized inside Linux)
Reading BAM_REVISION in these cases will hang in bam_init(), because we
cannot guarantee the BAM is already powered on when the bam_dma driver
is being loaded in Linux. We need to delay reading the register until
the BAM is up.
Given that these writes happen only for the !bdev->controlled_remotely
case, you could fix this more cleanly by reading the BAM revision inside
bam_reset().
Thanks,
Stephan
