Re: [PATCH v8 3/8] crypto: caam - drop global context pointer and init_done

[Andrey Smirnov <andrew.smirnov@gmail.com>]

On Tue, Mar 17, 2020 at 9:45 AM Horia Geantă <horia.geanta@nxp.com> wrote:
>
> On 3/16/2020 5:01 PM, Andrey Smirnov wrote:
> > diff --git a/drivers/crypto/caam/caamrng.c b/drivers/crypto/caam/caamrng.c
> > index 69a02ac5de54..753625f2b2c0 100644
> > --- a/drivers/crypto/caam/caamrng.c
> > +++ b/drivers/crypto/caam/caamrng.c
> > @@ -70,6 +70,7 @@ struct buf_data {
> >
> >  /* rng per-device context */
> >  struct caam_rng_ctx {
> > +     struct hwrng rng;
> >       struct device *jrdev;
> >       dma_addr_t sh_desc_dma;
> >       u32 sh_desc[DESC_RNG_LEN];
> > @@ -78,13 +79,10 @@ struct caam_rng_ctx {
> >       struct buf_data bufs[2];
> >  };
> [...]
> > +static struct caam_rng_ctx *to_caam_rng_ctx(struct hwrng *r)
> > +{
> > +     return (struct caam_rng_ctx *)r->priv;
> > +}
> [...]
> > -static struct hwrng caam_rng = {
> > -     .name           = "rng-caam",
> > -     .init           = caam_init,
> > -     .cleanup        = caam_cleanup,
> > -     .read           = caam_read,
> > -};
> I would keep this statically allocated, see below.
>
> > @@ -342,18 +332,27 @@ int caam_rng_init(struct device *ctrldev)
> >       if (!rng_inst)
> >               return 0;
> >
> > -     rng_ctx = kmalloc(sizeof(*rng_ctx), GFP_DMA | GFP_KERNEL);
> > -     if (!rng_ctx)
> > +     if (!devres_open_group(ctrldev, caam_rng_init, GFP_KERNEL))
> > +             return -ENOMEM;
> > +
> > +     ctx = devm_kzalloc(ctrldev, sizeof(*ctx), GFP_DMA | GFP_KERNEL);
> > +     if (!ctx)
> >               return -ENOMEM;
> >
> > +     ctx->rng.name    = "rng-caam";
> > +     ctx->rng.init    = caam_init;
> > +     ctx->rng.cleanup = caam_cleanup;
> > +     ctx->rng.read    = caam_read;
> > +     ctx->rng.priv    = (unsigned long)ctx;
> > +
> >       dev_info(ctrldev, "registering rng-caam\n");
> >
> > -     err = hwrng_register(&caam_rng);
> > -     if (!err) {
> > -             init_done = true;
> > -             return err;
> > +     ret = devm_hwrng_register(ctrldev, &ctx->rng);
> Now that hwrng.priv is used to keep driver's private data / caam_rng_ctx,
> and thus container_of() is no longer needed to get from hwrng struct
> to caam_rng_ctx, it's no longer needed to embed struct hwrng
> into caam_rng_ctx.
>

Why do we want this change though? It doesn't allow us to constify
"struct hwrng", don't seem to give any other benefits (maybe I am
missing something?) at the same time creating a new implicit global
variable. I'd rather not do this.

Thanks,
Andrey Smirnov