Re: [PATCH v12 00/12] crypto/dmaengine: qce: introduce BAM locking and use DMA for register I/O
From: Bartosz Golaszewski
Date: Wed Mar 11 2026 - 04:07:54 EST
On Tue, Mar 10, 2026 at 4:44 PM Bartosz Golaszewski
<bartosz.golaszewski@xxxxxxxxxxxxxxxx> wrote:
>
> This iteration is built on top of the v11 RFC with remaining issues
> fixed and the mechanism for communicating the scratchpad address from
> clients to the BAM driver changed from slave config to descriptor
> metadata.
>
> However: during stress-testing I noticed that sometimes a transaction
> would end with an error. The engine was indicating that a write/read to
> the config registers was performed while the engine was busy (bit 17 of
> the STATUS register was set). It turns out that we must not just
> unconditionally append the UNLOCK descriptor to the "issued" queue, we
> must wait for the transaction to end before we queue it so this version
> takes this into account and queues the UNLOCK descriptor from the
> workqueue.
>
> With this all stress tests and benchmarks from cryptsetup work fine.
>
Mani, Stephan: sorry, I forgot to update the cover letter to Cc you.
Doing it now here.
Stephan: I tried to use READ command but it would crash on sm8650, so
I went with WRITE. :(
Bartosz