RE: [RFC PATCH V2 3/4] crypto: Add Xilinx SHA3 driver

[Kalyani Akula]

Hi Corentin,

Sorry for the delayed response.
Please find my responses inline.

> -----Original Message-----
> From: Corentin Labbe <clabbe.montjoie@xxxxxxxxx>
> Sent: Thursday, January 10, 2019 7:28 PM
> To: Kalyani Akula <kalyania@xxxxxxxxxx>
> Cc: herbert@xxxxxxxxxxxxxxxxxxx; davem@xxxxxxxxxxxxx; linux-
> crypto@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; Kalyani Akula
> <kalyania@xxxxxxxxxx>; Sarat Chand Savitala <saratcha@xxxxxxxxxx>
> Subject: Re: [RFC PATCH V2 3/4] crypto: Add Xilinx SHA3 driver
> 
> On Wed, Jan 09, 2019 at 02:56:24PM +0530, Kalyani Akula wrote:
> > This patch adds SHA3 driver support for the Xilinx ZynqMP SoC.
> >
> > Signed-off-by: Kalyani Akula <kalyani.akula@xxxxxxxxxx>
> > ---
> >  drivers/crypto/Kconfig      |   10 ++
> >  drivers/crypto/Makefile     |    1 +
> >  drivers/crypto/zynqmp-sha.c |  305
> > +++++++++++++++++++++++++++++++++++++++++++
> >  3 files changed, 316 insertions(+), 0 deletions(-)  create mode
> > 100644 drivers/crypto/zynqmp-sha.c
> 
> Hello
> 
> I have some comment below
> 
> > +static int zynqmp_sha_update(struct ahash_request *req) {
> > +	const struct zynqmp_eemi_ops *eemi_ops =
> zynqmp_pm_get_eemi_ops();
> > +	struct zynqmp_sha_ctx *tctx = crypto_tfm_ctx(req->base.tfm);
> > +	struct zynqmp_sha_dev *dd = tctx->dd;
> > +	size_t dma_size = req->nbytes;
> > +	dma_addr_t dma_addr;
> > +	char *kbuf;
> > +	int ret;
> > +
> > +	if (!req->nbytes)
> > +		return 0;
> > +
> > +	if (!eemi_ops || !eemi_ops->sha_hash)
> > +		return -ENOTSUPP;
> > +
> > +	kbuf = dma_alloc_coherent(dd->dev, dma_size, &dma_addr,
> GFP_KERNEL);
> > +	if (!kbuf)
> > +		return -ENOMEM;
> > +
> > +	scatterwalk_map_and_copy(kbuf, req->src, 0, req->nbytes, 0);
> > +	 __flush_cache_user_range((unsigned long)kbuf,
> > +				  (unsigned long)kbuf + dma_size);
> 
> Since kbuf is in dma coherent memory, I dont understand why do you flush
> cache.
[kalyani] Agree, not required will fix in next version.
> 
> > +	ret = eemi_ops->sha_hash(dma_addr, req->nbytes,
> ZYNQMP_SHA3_UPDATE);
> > +	dma_free_coherent(dd->dev, dma_size, kbuf, dma_addr);
> 
> Since your update function does not return/write any result, I suppose that
> the resulat is kept in hardware, so what happen if two hash process happen
> in parallel ?
[kalyani] yes, the result will be stored in SHA3 engine.
Firmware don't have the support to process 2 hash requests in parallel using IPI
> Furthermore, how do you have tested import/export function ?
> Anyway, I am sure they are totally buggy.
> 
[kalyani] I will verify and fix in the next version
> > +
> > +	return ret;
> > +}
> > +
> > +static int zynqmp_sha_final(struct ahash_request *req) {
> > +	const struct zynqmp_eemi_ops *eemi_ops =
> zynqmp_pm_get_eemi_ops();
> > +	struct zynqmp_sha_ctx *tctx = crypto_tfm_ctx(req->base.tfm);
> > +	struct zynqmp_sha_dev *dd = tctx->dd;
> > +	size_t dma_size = SHA384_DIGEST_SIZE;
> > +	dma_addr_t dma_addr;
> > +	char *kbuf;
> > +	int ret;
> > +
> > +	if (!eemi_ops || !eemi_ops->sha_hash)
> > +		return -ENOTSUPP;
> 
> You can do this check at probe time.
[kalyani] Yes, will fix in next version
> 
> > +static int zynqmp_sha_finup(struct ahash_request *req) {
> > +	zynqmp_sha_update(req);
> > +	zynqmp_sha_final(req);
> > +
> > +	return 0;
> > +}
> > +
> > +static int zynqmp_sha_digest(struct ahash_request *req) {
> > +	zynqmp_sha_init(req);
> > +	zynqmp_sha_update(req);
> > +	zynqmp_sha_final(req);
> > +
> > +	return 0;
> > +}
> 
> So you ignore the return value of
> zynqmp_sha_init/zynqmp_sha_update/zynqmp_sha_final().
> 
> > +static int zynqmp_sha_probe(struct platform_device *pdev) {
> > +	struct zynqmp_sha_dev *sha_dd;
> > +	struct device *dev = &pdev->dev;
> > +	int err;
> > +
> > +	sha_dd = devm_kzalloc(&pdev->dev, sizeof(*sha_dd),
> GFP_KERNEL);
> > +	if (!sha_dd)
> > +		return -ENOMEM;
> > +
> > +	sha_dd->dev = dev;
> > +	platform_set_drvdata(pdev, sha_dd);
> > +	INIT_LIST_HEAD(&sha_dd->list);
> > +	spin_lock_init(&sha_dd->lock);
> > +	crypto_init_queue(&sha_dd->queue,
> ZYNQMP_SHA_QUEUE_LENGTH);
> 
> As already said in my last review, why init the queue if you do not use it ?
> 
> 
> > +	spin_lock(&zynqmp_sha.lock);
> > +	list_add_tail(&sha_dd->list, &zynqmp_sha.dev_list);
> > +	spin_unlock(&zynqmp_sha.lock);
> 
> Why do you maintain a device list ?
> Clearly your current driver support only one hardware instance.
[kalyani] True, driver support only one HW instance , will fix in the next version

Regards,
kalyani
> 
> Regards