Re: [PATCH v4 3/3] crypto: s5p: add AES support for Exynos5433

From: Krzysztof Kozlowski
Date: Fri Mar 01 2019 - 05:12:36 EST


On Fri, 1 Mar 2019 at 11:07, Ard Biesheuvel <ard.biesheuvel@xxxxxxxxxx> wrote:
>
> On Fri, 1 Mar 2019 at 10:56, Krzysztof Kozlowski <krzk@xxxxxxxxxx> wrote:
> >
> > On Fri, 22 Feb 2019 at 13:22, Kamil Konieczny
> > <k.konieczny@xxxxxxxxxxxxxxxxxxx> wrote:
> > >
> > > Add AES crypto HW acceleration for Exynos5433, with the help of SlimSSS IP.
> > >
> > > Reviewed-by: Krzysztof Kozlowski <krzk@xxxxxxxxxx>
> > > Signed-off-by: Kamil Konieczny <k.konieczny@xxxxxxxxxxxxxxxxxxx>
> > > ---
> > > drivers/crypto/s5p-sss.c | 50 ++++++++++++++++++++++++++++++++++++----
> > > 1 file changed, 46 insertions(+), 4 deletions(-)
> > > diff --git a/drivers/crypto/s5p-sss.c b/drivers/crypto/s5p-sss.c
> > > index 0064be0e3941..3f45cc5cb94a 100644
> > > --- a/drivers/crypto/s5p-sss.c
> > > +++ b/drivers/crypto/s5p-sss.c
> > > @@ -232,6 +232,7 @@
> > > * struct samsung_aes_variant - platform specific SSS driver data
> > > * @aes_offset: AES register offset from SSS module's base.
> > > * @hash_offset: HASH register offset from SSS module's base.
> > > + * @clk_names: names of clocks needed to run SSS IP
> > > *
> > > * Specifies platform specific configuration of SSS module.
> > > * Note: A structure for driver specific platform data is used for future
> > > @@ -240,6 +241,7 @@
> > > struct samsung_aes_variant {
> > > unsigned int aes_offset;
> > > unsigned int hash_offset;
> > > + const char *clk_names[];
> > > };
> > >
> > > struct s5p_aes_reqctx {
> > > @@ -296,6 +298,7 @@ struct s5p_aes_ctx {
> > > struct s5p_aes_dev {
> > > struct device *dev;
> > > struct clk *clk;
> > > + struct clk *pclk;
> > > void __iomem *ioaddr;
> > > void __iomem *aes_ioaddr;
> > > int irq_fc;
> > > @@ -384,11 +387,19 @@ struct s5p_hash_ctx {
> > > static const struct samsung_aes_variant s5p_aes_data = {
> > > .aes_offset = 0x4000,
> > > .hash_offset = 0x6000,
> > > + .clk_names = { "secss", },
> > > };
> > >
> > > static const struct samsung_aes_variant exynos_aes_data = {
> > > .aes_offset = 0x200,
> > > .hash_offset = 0x400,
> > > + .clk_names = { "secss", },
> > > +};
> > > +
> > > +static const struct samsung_aes_variant exynos5433_slim_aes_data = {
> > > + .aes_offset = 0x400,
> > > + .hash_offset = 0x800,
> > > + .clk_names = { "pclk", "aclk", },
> > > };
> > >
> > > static const struct of_device_id s5p_sss_dt_match[] = {
> > > @@ -400,6 +411,10 @@ static const struct of_device_id s5p_sss_dt_match[] = {
> > > .compatible = "samsung,exynos4210-secss",
> > > .data = &exynos_aes_data,
> > > },
> > > + {
> > > + .compatible = "samsung,exynos5433-slim-sss",
> > > + .data = &exynos5433_slim_aes_data,
> > > + },
> > > { },
> > > };
> > > MODULE_DEVICE_TABLE(of, s5p_sss_dt_match);
> > > @@ -2208,18 +2223,39 @@ static int s5p_aes_probe(struct platform_device *pdev)
> > > return PTR_ERR(pdata->ioaddr);
> > > }
> > >
> > > - pdata->clk = devm_clk_get(dev, "secss");
> > > + pdata->clk = devm_clk_get(dev, variant->clk_names[0]);
> > > if (IS_ERR(pdata->clk)) {
> > > - dev_err(dev, "failed to find secss clock source\n");
> > > + dev_err(dev, "failed to find secss clock %s\n",
> > > + variant->clk_names[0]);
> > > return -ENOENT;
> > > }
> > >
> > > err = clk_prepare_enable(pdata->clk);
> > > if (err < 0) {
> > > - dev_err(dev, "Enabling SSS clk failed, err %d\n", err);
> > > + dev_err(dev, "Enabling clock %s failed, err %d\n",
> > > + variant->clk_names[0], err);
> > > return err;
> > > }
> > >
> > > + if (variant->clk_names[1]) {
> > > + pdata->pclk = devm_clk_get(dev, variant->clk_names[1]);
> >
> > I see on today's linux next:
> > [ 6.391256] Unable to handle kernel paging request at virtual address 00004000
> > on all my Exynos boards (U3, XU, XU3, HC1).
> > I did not do a bisect yet but it might be connected with this commit.
> >
> > The call trace is:
> > (strcmp) from [<c0774014>] (of_property_match_string+0x58/0xd0)
> > (of_property_match_string) from [<c04d3d2c>] (__of_clk_get_by_name+0x5c/0x130)
> > (__of_clk_get_by_name) from [<c04d3fa0>] (clk_get+0x34/0x70)
> > (clk_get) from [<c04d3680>] (devm_clk_get+0x38/0x74)
> > (devm_clk_get) from [<c075c248>] (s5p_aes_probe+0xd4/0x4a0)
> > (s5p_aes_probe) from [<c059dbc4>] (platform_drv_probe+0x6c/0xa4)
> > (platform_drv_probe) from [<c059b4c8>] (really_probe+0x280/0x414)
> > (really_probe) from [<c059b810>] (driver_probe_device+0x78/0x1c0)
> > (driver_probe_device) from [<c059bbb8>] (device_driver_attach+0x58/0x60)
> > (device_driver_attach) from [<c059bcbc>] (__driver_attach+0xfc/0x160)
> > (__driver_attach) from [<c059928c>] (bus_for_each_dev+0x68/0xb4)
> > (bus_for_each_dev) from [<c059a5c4>] (bus_add_driver+0x158/0x214)
> > (bus_add_driver) from [<c059cb94>] (driver_register+0x78/0x110)
> > (driver_register) from [<c01031f8>] (do_one_initcall+0x8c/0x404)
> >
> > Full log:
> > https://krzk.eu/#/builders/21/builds/1153/steps/12/logs/serial0
> >
> > The s5p_aes_probe address points here.
> >
>
> It seems that dereferencing clk_names[1] of struct samsung_aes_variant
> exynos_aes_data
>
> > > + .clk_names = { "secss", },
>
> returns the first member of
>
> > > static const struct samsung_aes_variant s5p_aes_data = {
> > > .aes_offset = 0x4000,
>
> since the array has only one entry in this case, and so the second,
> non-existing entry overlaps with the next data item in the object
> file.
>
> So the fix would be to use
>
> .clk_names = { "secss", NULL },
>
> instead (or define the array to have 2 members)

Ah yes, the size of the array is not fixed during definition. Probably
having explicit size would be less error-prone in the future (if
someone adds third clock).

Best regards,
Krzysztof