Re: [PATCH v5 1/4] crypto: aspeed: Add ACRY RSA driver

From: Geert Uytterhoeven
Date: Wed Feb 22 2023 - 09:41:38 EST


Hi Neal,

On Wed, Jan 4, 2023 at 2:37 AM Neal Liu <neal_liu@xxxxxxxxxxxxxx> wrote:
> ACRY Engine is designed to accelerate the throughput of
> ECDSA/RSA signature and verification.
>
> This patch aims to add ACRY RSA engine driver for hardware
> acceleration.
>
> Signed-off-by: Neal Liu <neal_liu@xxxxxxxxxxxxxx>

Thanks for your patch, which is now commit 2f1cf4e50c956f88 ("crypto:
aspeed - Add ACRY RSA driver").

> --- /dev/null
> +++ b/drivers/crypto/aspeed/aspeed-acry.c

> +static int aspeed_acry_probe(struct platform_device *pdev)
> +{
> + struct aspeed_acry_dev *acry_dev;
> + struct device *dev = &pdev->dev;
> + struct resource *res;
> + int rc;
> +
> + acry_dev = devm_kzalloc(dev, sizeof(struct aspeed_acry_dev),
> + GFP_KERNEL);
> + if (!acry_dev)
> + return -ENOMEM;
> +
> + acry_dev->dev = dev;
> +
> + platform_set_drvdata(pdev, acry_dev);
> +
> + res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> + acry_dev->regs = devm_ioremap_resource(dev, res);
> + if (IS_ERR(acry_dev->regs))
> + return PTR_ERR(acry_dev->regs);
> +
> + res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
> + acry_dev->acry_sram = devm_ioremap_resource(dev, res);
> + if (IS_ERR(acry_dev->acry_sram))
> + return PTR_ERR(acry_dev->acry_sram);
> +
> + /* Get irq number and register it */
> + acry_dev->irq = platform_get_irq(pdev, 0);
> + if (acry_dev->irq < 0)
> + return -ENXIO;
> +
> + rc = devm_request_irq(dev, acry_dev->irq, aspeed_acry_irq, 0,
> + dev_name(dev), acry_dev);
> + if (rc) {
> + dev_err(dev, "Failed to request irq.\n");
> + return rc;
> + }
> +
> + acry_dev->clk = devm_clk_get_enabled(dev, NULL);
> + if (IS_ERR(acry_dev->clk)) {
> + dev_err(dev, "Failed to get acry clk\n");
> + return PTR_ERR(acry_dev->clk);
> + }
> +
> + acry_dev->ahbc = syscon_regmap_lookup_by_phandle(dev->of_node,
> + "aspeed,ahbc");
> + if (IS_ERR(acry_dev->ahbc)) {
> + dev_err(dev, "Failed to get AHBC regmap\n");
> + return -ENODEV;
> + }
> +
> + /* Initialize crypto hardware engine structure for RSA */
> + acry_dev->crypt_engine_rsa = crypto_engine_alloc_init(dev, true);
> + if (!acry_dev->crypt_engine_rsa) {
> + rc = -ENOMEM;
> + goto clk_exit;
> + }
> +
> + rc = crypto_engine_start(acry_dev->crypt_engine_rsa);
> + if (rc)
> + goto err_engine_rsa_start;
> +
> + tasklet_init(&acry_dev->done_task, aspeed_acry_done_task,
> + (unsigned long)acry_dev);
> +
> + /* Set Data Memory to AHB(CPU) Access Mode */
> + ast_acry_write(acry_dev, ACRY_CMD_DMEM_AHB, ASPEED_ACRY_DMA_CMD);
> +
> + /* Initialize ACRY SRAM index */
> + aspeed_acry_sram_mapping(acry_dev);
> +
> + acry_dev->buf_addr = dmam_alloc_coherent(dev, ASPEED_ACRY_BUFF_SIZE,
> + &acry_dev->buf_dma_addr,
> + GFP_KERNEL);
> + memzero_explicit(acry_dev->buf_addr, ASPEED_ACRY_BUFF_SIZE);

When compile-testing with CONFIG_HAS_DMA=n:

error: argument 1 null where non-null expected [-Werror=nonnull]

The call to dmam_alloc_coherent() lacks error handling, as the returned
address may be NULL.

Moreover, is it safe to allocate this buffer only after the call to
crypto_engine_start()? I.e. could acry_dev->buf_addr be accessed
before, causing a NULL pointer dereference?
Is there any other initialization that should be done earlier?

> +
> + aspeed_acry_register(acry_dev);
> +
> + dev_info(dev, "Aspeed ACRY Accelerator successfully registered\n");
> +
> + return 0;
> +
> +err_engine_rsa_start:
> + crypto_engine_exit(acry_dev->crypt_engine_rsa);
> +clk_exit:
> + clk_disable_unprepare(acry_dev->clk);
> +
> + return rc;
> +}

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds