Re: [PATCH] crypto: marvell/cesa - Avoid empty transfer descriptor

[Herbert Xu <herbert@gondor.apana.org.au>]

On Sat, May 10, 2025 at 01:14:22PM +0200, Corentin Labbe wrote:
>
> This is the git diff result to be sure I didnt miss a patch: http://kernel.montjoie.ovh/cesa.debug.diff

Thanks! I think we have a smoking gun:

[   45.700298] mv_cesa_dma_step: 1 0xc7011440 0x9256040 "0x9256020"
[   45.706141] mv_cesa_ahash_req_cleanup: 0 0xc93b9c00
[   45.711996] mv_cesa_int: 1 0x4ea1 0x80
[   45.716875] mv_cesa_int: 0 0x4ea1 0x80
[   45.720627] mv_cesa_tdma_process: 1 "0x9256020"
[   45.724380] mv_cesa_tdma_process: 0 0x9256140
[   45.728757] mv_cesa_ahash_complete: 1 0xc7011400
[   45.733112] mv_cesa_ahash_complete: 0 0xc7011200
[   45.737741] mv_cesa_tdma_process: 1 0 0xc7011400
[   45.742364] mv_cesa_tdma_process: 0 0 0xc7011200
[   45.746994] mv_cesa_ahash_req_cleanup: 1 0xc7011400
[   45.751614] mv_cesa_ahash_req_cleanup: 0 0xc7011200
[   45.756635] mv_cesa_ahash_queue_req: 0 0xc93b9c00
[   45.766104] mv_cesa_dma_step: 0 0xc93b9c40 "0x9256020" 0x9256000
[   45.771972] alg: ahash: mv-sha1 test failed (wrong result) on test vector 3, cfg="init+update+update+final two even splits"

The descriptor 0x9256020 was just freed by engine 1, and it's still
the current pointer of engine 1.  It was then immediately reused by
engine 0 starting a new chain.  It's conceivable that engine 1 then
somehow starts executing on it at the same time.

Cheers,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt