> On Wed, Jan 24, 2024 at 02:08:44PM +0530, Akhil R wrote:
> >
> > +static void tegra_sha_init_fallback(struct tegra_sha_ctx *ctx, const char
> *algname)
> > +{
> > + ctx->fallback_tfm = crypto_alloc_ahash(algname, 0, CRYPTO_ALG_ASYNC |
> > + CRYPTO_ALG_NEED_FALLBACK);
> > +
> > + if (IS_ERR(ctx->fallback_tfm)) {
> > + dev_warn(ctx->se->dev, "failed to allocate fallback for %s %ld\n",
> > + algname, PTR_ERR(ctx->fallback_tfm));
> > + ctx->fallback_tfm = NULL;
> > + }
> > +}
>
> This should check that the fallback state size is smaller than
> that of tegra. As otherwise the fallback export/import will break.
Okay. Got it. Will update.
> > +static int tegra_sha_import(struct ahash_request *req, const void *in)
> > +{
> > + struct tegra_sha_reqctx *rctx = ahash_request_ctx(req);
> > + struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
> > + struct tegra_sha_ctx *ctx = crypto_ahash_ctx(tfm);
> > + int i;
> > +
> > + if (ctx->fallback)
> > + return tegra_sha_fallback_import(req, in);
> > +
> > + memcpy(rctx, in, sizeof(*rctx));
> > +
> > + /* Paste all intermediate results */
> > + for (i = 0; i < HASH_RESULT_REG_COUNT; i++)
> > + writel(rctx->result[i],
> > + ctx->se->base + ctx->se->hw->regs->result + (i * 4));
>
> What happens when multiple requests of the same tfm import at
> the same time? Normally we don't actually touch the hardware
> in the import function. Instead, all the hard work happens at
> the end of the update function, which moves hardware state into
> the request object.
>
> The import/export function then simply copies the request object
> state to the in/out buffer.
Understood the issue. But I feel it will be a bit overburden for the update()
to copy/paste these for every call. Let me explore more on the hardware and
come back with a better approach.
Thanks for the comments.
Regards,
Akhil
