hwrng: cctrng - introduce Arm CryptoCell driver
authorHadar Gat <hadar.gat@arm.com>
Fri, 27 Mar 2020 06:10:22 +0000 (09:10 +0300)
committerHerbert Xu <herbert@gondor.apana.org.au>
Thu, 16 Apr 2020 06:49:20 +0000 (16:49 +1000)
Introduce low level Arm CryptoCell TRNG HW support.

Signed-off-by: Hadar Gat <hadar.gat@arm.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
drivers/char/hw_random/Kconfig
drivers/char/hw_random/Makefile
drivers/char/hw_random/cctrng.c [new file with mode: 0644]
drivers/char/hw_random/cctrng.h [new file with mode: 0644]

index 9bc46da..848f26f 100644 (file)
@@ -474,6 +474,18 @@ config HW_RANDOM_KEYSTONE
        help
          This option enables Keystone's hardware random generator.
 
+config HW_RANDOM_CCTRNG
+       tristate "Arm CryptoCell True Random Number Generator support"
+       default HW_RANDOM
+       help
+         This driver provides support for the True Random Number
+         Generator available in Arm TrustZone CryptoCell.
+
+         To compile this driver as a module, choose M here: the module
+         will be called cctrng.
+
+         If unsure, say Y.
+
 endif # HW_RANDOM
 
 config UML_RANDOM
index a7801b4..2c67247 100644 (file)
@@ -41,3 +41,4 @@ obj-$(CONFIG_HW_RANDOM_S390) += s390-trng.o
 obj-$(CONFIG_HW_RANDOM_KEYSTONE) += ks-sa-rng.o
 obj-$(CONFIG_HW_RANDOM_OPTEE) += optee-rng.o
 obj-$(CONFIG_HW_RANDOM_NPCM) += npcm-rng.o
+obj-$(CONFIG_HW_RANDOM_CCTRNG) += cctrng.o
diff --git a/drivers/char/hw_random/cctrng.c b/drivers/char/hw_random/cctrng.c
new file mode 100644 (file)
index 0000000..bdcd562
--- /dev/null
@@ -0,0 +1,736 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (C) 2019-2020 ARM Limited or its affiliates. */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/clk.h>
+#include <linux/hw_random.h>
+#include <linux/io.h>
+#include <linux/platform_device.h>
+#include <linux/pm_runtime.h>
+#include <linux/interrupt.h>
+#include <linux/irqreturn.h>
+#include <linux/workqueue.h>
+#include <linux/circ_buf.h>
+#include <linux/completion.h>
+#include <linux/of.h>
+#include <linux/bitfield.h>
+
+#include "cctrng.h"
+
+#define CC_REG_LOW(name)  (name ## _BIT_SHIFT)
+#define CC_REG_HIGH(name) (CC_REG_LOW(name) + name ## _BIT_SIZE - 1)
+#define CC_GENMASK(name)  GENMASK(CC_REG_HIGH(name), CC_REG_LOW(name))
+
+#define CC_REG_FLD_GET(reg_name, fld_name, reg_val)     \
+       (FIELD_GET(CC_GENMASK(CC_ ## reg_name ## _ ## fld_name), reg_val))
+
+#define CC_HW_RESET_LOOP_COUNT 10
+#define CC_TRNG_SUSPEND_TIMEOUT 3000
+
+/* data circular buffer in words must be:
+ *  - of a power-of-2 size (limitation of circ_buf.h macros)
+ *  - at least 6, the size generated in the EHR according to HW implementation
+ */
+#define CCTRNG_DATA_BUF_WORDS 32
+
+/* The timeout for the TRNG operation should be calculated with the formula:
+ * Timeout = EHR_NUM * VN_COEFF * EHR_LENGTH * SAMPLE_CNT * SCALE_VALUE
+ * while:
+ *  - SAMPLE_CNT is input value from the characterisation process
+ *  - all the rest are constants
+ */
+#define EHR_NUM 1
+#define VN_COEFF 4
+#define EHR_LENGTH CC_TRNG_EHR_IN_BITS
+#define SCALE_VALUE 2
+#define CCTRNG_TIMEOUT(smpl_cnt) \
+       (EHR_NUM * VN_COEFF * EHR_LENGTH * smpl_cnt * SCALE_VALUE)
+
+struct cctrng_drvdata {
+       struct platform_device *pdev;
+       void __iomem *cc_base;
+       struct clk *clk;
+       struct hwrng rng;
+       u32 active_rosc;
+       /* Sampling interval for each ring oscillator:
+        * count of ring oscillator cycles between consecutive bits sampling.
+        * Value of 0 indicates non-valid rosc
+        */
+       u32 smpl_ratio[CC_TRNG_NUM_OF_ROSCS];
+
+       u32 data_buf[CCTRNG_DATA_BUF_WORDS];
+       struct circ_buf circ;
+       struct work_struct compwork;
+       struct work_struct startwork;
+
+       /* pending_hw - 1 when HW is pending, 0 when it is idle */
+       atomic_t pending_hw;
+
+       /* protects against multiple concurrent consumers of data_buf */
+       spinlock_t read_lock;
+};
+
+
+/* functions for write/read CC registers */
+static inline void cc_iowrite(struct cctrng_drvdata *drvdata, u32 reg, u32 val)
+{
+       iowrite32(val, (drvdata->cc_base + reg));
+}
+static inline u32 cc_ioread(struct cctrng_drvdata *drvdata, u32 reg)
+{
+       return ioread32(drvdata->cc_base + reg);
+}
+
+
+static int cc_trng_pm_get(struct device *dev)
+{
+       int rc = 0;
+
+       rc = pm_runtime_get_sync(dev);
+
+       /* pm_runtime_get_sync() can return 1 as a valid return code */
+       return (rc == 1 ? 0 : rc);
+}
+
+static void cc_trng_pm_put_suspend(struct device *dev)
+{
+       int rc = 0;
+
+       pm_runtime_mark_last_busy(dev);
+       rc = pm_runtime_put_autosuspend(dev);
+       if (rc)
+               dev_err(dev, "pm_runtime_put_autosuspend returned %x\n", rc);
+}
+
+static int cc_trng_pm_init(struct cctrng_drvdata *drvdata)
+{
+       struct device *dev = &(drvdata->pdev->dev);
+
+       /* must be before the enabling to avoid redundant suspending */
+       pm_runtime_set_autosuspend_delay(dev, CC_TRNG_SUSPEND_TIMEOUT);
+       pm_runtime_use_autosuspend(dev);
+       /* set us as active - note we won't do PM ops until cc_trng_pm_go()! */
+       return pm_runtime_set_active(dev);
+}
+
+static void cc_trng_pm_go(struct cctrng_drvdata *drvdata)
+{
+       struct device *dev = &(drvdata->pdev->dev);
+
+       /* enable the PM module*/
+       pm_runtime_enable(dev);
+}
+
+static void cc_trng_pm_fini(struct cctrng_drvdata *drvdata)
+{
+       struct device *dev = &(drvdata->pdev->dev);
+
+       pm_runtime_disable(dev);
+}
+
+
+static inline int cc_trng_parse_sampling_ratio(struct cctrng_drvdata *drvdata)
+{
+       struct device *dev = &(drvdata->pdev->dev);
+       struct device_node *np = drvdata->pdev->dev.of_node;
+       int rc;
+       int i;
+       /* ret will be set to 0 if at least one rosc has (sampling ratio > 0) */
+       int ret = -EINVAL;
+
+       rc = of_property_read_u32_array(np, "arm,rosc-ratio",
+                                       drvdata->smpl_ratio,
+                                       CC_TRNG_NUM_OF_ROSCS);
+       if (rc) {
+               /* arm,rosc-ratio was not found in device tree */
+               return rc;
+       }
+
+       /* verify that at least one rosc has (sampling ratio > 0) */
+       for (i = 0; i < CC_TRNG_NUM_OF_ROSCS; ++i) {
+               dev_dbg(dev, "rosc %d sampling ratio %u",
+                       i, drvdata->smpl_ratio[i]);
+
+               if (drvdata->smpl_ratio[i] > 0)
+                       ret = 0;
+       }
+
+       return ret;
+}
+
+static int cc_trng_change_rosc(struct cctrng_drvdata *drvdata)
+{
+       struct device *dev = &(drvdata->pdev->dev);
+
+       dev_dbg(dev, "cctrng change rosc (was %d)\n", drvdata->active_rosc);
+       drvdata->active_rosc += 1;
+
+       while (drvdata->active_rosc < CC_TRNG_NUM_OF_ROSCS) {
+               if (drvdata->smpl_ratio[drvdata->active_rosc] > 0)
+                       return 0;
+
+               drvdata->active_rosc += 1;
+       }
+       return -EINVAL;
+}
+
+
+static void cc_trng_enable_rnd_source(struct cctrng_drvdata *drvdata)
+{
+       u32 max_cycles;
+
+       /* Set watchdog threshold to maximal allowed time (in CPU cycles) */
+       max_cycles = CCTRNG_TIMEOUT(drvdata->smpl_ratio[drvdata->active_rosc]);
+       cc_iowrite(drvdata, CC_RNG_WATCHDOG_VAL_REG_OFFSET, max_cycles);
+
+       /* enable the RND source */
+       cc_iowrite(drvdata, CC_RND_SOURCE_ENABLE_REG_OFFSET, 0x1);
+
+       /* unmask RNG interrupts */
+       cc_iowrite(drvdata, CC_RNG_IMR_REG_OFFSET, (u32)~CC_RNG_INT_MASK);
+}
+
+
+/* increase circular data buffer index (head/tail) */
+static inline void circ_idx_inc(int *idx, int bytes)
+{
+       *idx += (bytes + 3) >> 2;
+       *idx &= (CCTRNG_DATA_BUF_WORDS - 1);
+}
+
+static inline size_t circ_buf_space(struct cctrng_drvdata *drvdata)
+{
+       return CIRC_SPACE(drvdata->circ.head,
+                         drvdata->circ.tail, CCTRNG_DATA_BUF_WORDS);
+
+}
+
+static int cctrng_read(struct hwrng *rng, void *data, size_t max, bool wait)
+{
+       /* current implementation ignores "wait" */
+
+       struct cctrng_drvdata *drvdata = (struct cctrng_drvdata *)rng->priv;
+       struct device *dev = &(drvdata->pdev->dev);
+       u32 *buf = (u32 *)drvdata->circ.buf;
+       size_t copied = 0;
+       size_t cnt_w;
+       size_t size;
+       size_t left;
+
+       if (!spin_trylock(&drvdata->read_lock)) {
+               /* concurrent consumers from data_buf cannot be served */
+               dev_dbg_ratelimited(dev, "unable to hold lock\n");
+               return 0;
+       }
+
+       /* copy till end of data buffer (without wrap back) */
+       cnt_w = CIRC_CNT_TO_END(drvdata->circ.head,
+                               drvdata->circ.tail, CCTRNG_DATA_BUF_WORDS);
+       size = min((cnt_w<<2), max);
+       memcpy(data, &(buf[drvdata->circ.tail]), size);
+       copied = size;
+       circ_idx_inc(&drvdata->circ.tail, size);
+       /* copy rest of data in data buffer */
+       left = max - copied;
+       if (left > 0) {
+               cnt_w = CIRC_CNT(drvdata->circ.head,
+                                drvdata->circ.tail, CCTRNG_DATA_BUF_WORDS);
+               size = min((cnt_w<<2), left);
+               memcpy(data, &(buf[drvdata->circ.tail]), size);
+               copied += size;
+               circ_idx_inc(&drvdata->circ.tail, size);
+       }
+
+       spin_unlock(&drvdata->read_lock);
+
+       if (circ_buf_space(drvdata) >= CC_TRNG_EHR_IN_WORDS) {
+               if (atomic_cmpxchg(&drvdata->pending_hw, 0, 1) == 0) {
+                       /* re-check space in buffer to avoid potential race */
+                       if (circ_buf_space(drvdata) >= CC_TRNG_EHR_IN_WORDS) {
+                               /* increment device's usage counter */
+                               int rc = cc_trng_pm_get(dev);
+
+                               if (rc) {
+                                       dev_err(dev,
+                                               "cc_trng_pm_get returned %x\n",
+                                               rc);
+                                       return rc;
+                               }
+
+                               /* schedule execution of deferred work handler
+                                * for filling of data buffer
+                                */
+                               schedule_work(&drvdata->startwork);
+                       } else {
+                               atomic_set(&drvdata->pending_hw, 0);
+                       }
+               }
+       }
+
+       return copied;
+}
+
+static void cc_trng_hw_trigger(struct cctrng_drvdata *drvdata)
+{
+       u32 tmp_smpl_cnt = 0;
+       struct device *dev = &(drvdata->pdev->dev);
+
+       dev_dbg(dev, "cctrng hw trigger.\n");
+
+       /* enable the HW RND clock */
+       cc_iowrite(drvdata, CC_RNG_CLK_ENABLE_REG_OFFSET, 0x1);
+
+       /* do software reset */
+       cc_iowrite(drvdata, CC_RNG_SW_RESET_REG_OFFSET, 0x1);
+       /* in order to verify that the reset has completed,
+        * the sample count need to be verified
+        */
+       do {
+               /* enable the HW RND clock   */
+               cc_iowrite(drvdata, CC_RNG_CLK_ENABLE_REG_OFFSET, 0x1);
+
+               /* set sampling ratio (rng_clocks) between consecutive bits */
+               cc_iowrite(drvdata, CC_SAMPLE_CNT1_REG_OFFSET,
+                          drvdata->smpl_ratio[drvdata->active_rosc]);
+
+               /* read the sampling ratio  */
+               tmp_smpl_cnt = cc_ioread(drvdata, CC_SAMPLE_CNT1_REG_OFFSET);
+
+       } while (tmp_smpl_cnt != drvdata->smpl_ratio[drvdata->active_rosc]);
+
+       /* disable the RND source for setting new parameters in HW */
+       cc_iowrite(drvdata, CC_RND_SOURCE_ENABLE_REG_OFFSET, 0);
+
+       cc_iowrite(drvdata, CC_RNG_ICR_REG_OFFSET, 0xFFFFFFFF);
+
+       cc_iowrite(drvdata, CC_TRNG_CONFIG_REG_OFFSET, drvdata->active_rosc);
+
+       /* Debug Control register: set to 0 - no bypasses */
+       cc_iowrite(drvdata, CC_TRNG_DEBUG_CONTROL_REG_OFFSET, 0);
+
+       cc_trng_enable_rnd_source(drvdata);
+}
+
+void cc_trng_compwork_handler(struct work_struct *w)
+{
+       u32 isr = 0;
+       u32 ehr_valid = 0;
+       struct cctrng_drvdata *drvdata =
+                       container_of(w, struct cctrng_drvdata, compwork);
+       struct device *dev = &(drvdata->pdev->dev);
+       int i;
+
+       /* stop DMA and the RNG source */
+       cc_iowrite(drvdata, CC_RNG_DMA_ENABLE_REG_OFFSET, 0);
+       cc_iowrite(drvdata, CC_RND_SOURCE_ENABLE_REG_OFFSET, 0);
+
+       /* read RNG_ISR and check for errors */
+       isr = cc_ioread(drvdata, CC_RNG_ISR_REG_OFFSET);
+       ehr_valid = CC_REG_FLD_GET(RNG_ISR, EHR_VALID, isr);
+       dev_dbg(dev, "Got RNG_ISR=0x%08X (EHR_VALID=%u)\n", isr, ehr_valid);
+
+#ifdef CONFIG_CRYPTO_FIPS
+       if (CC_REG_FLD_GET(RNG_ISR, CRNGT_ERR, isr) && fips_enabled) {
+               fips_fail_notify();
+               /* FIPS error is fatal */
+               panic("Got HW CRNGT error while fips is enabled!\n");
+       }
+#endif
+
+       /* Clear all pending RNG interrupts */
+       cc_iowrite(drvdata, CC_RNG_ICR_REG_OFFSET, isr);
+
+
+       if (!ehr_valid) {
+               /* in case of AUTOCORR/TIMEOUT error, try the next ROSC */
+               if (CC_REG_FLD_GET(RNG_ISR, AUTOCORR_ERR, isr) ||
+                               CC_REG_FLD_GET(RNG_ISR, WATCHDOG, isr)) {
+                       dev_dbg(dev, "cctrng autocorr/timeout error.\n");
+                       goto next_rosc;
+               }
+
+               /* in case of VN error, ignore it */
+       }
+
+       /* read EHR data from registers */
+       for (i = 0; i < CC_TRNG_EHR_IN_WORDS; i++) {
+               /* calc word ptr in data_buf */
+               u32 *buf = (u32 *)drvdata->circ.buf;
+
+               buf[drvdata->circ.head] = cc_ioread(drvdata,
+                               CC_EHR_DATA_0_REG_OFFSET + (i*sizeof(u32)));
+
+               /* EHR_DATA registers are cleared on read. In case 0 value was
+                * returned, restart the entropy collection.
+                */
+               if (buf[drvdata->circ.head] == 0) {
+                       dev_dbg(dev, "Got 0 value in EHR. active_rosc %u\n",
+                               drvdata->active_rosc);
+                       goto next_rosc;
+               }
+
+               circ_idx_inc(&drvdata->circ.head, 1<<2);
+       }
+
+       atomic_set(&drvdata->pending_hw, 0);
+
+       /* continue to fill data buffer if needed */
+       if (circ_buf_space(drvdata) >= CC_TRNG_EHR_IN_WORDS) {
+               if (atomic_cmpxchg(&drvdata->pending_hw, 0, 1) == 0) {
+                       /* Re-enable rnd source */
+                       cc_trng_enable_rnd_source(drvdata);
+                       return;
+               }
+       }
+
+       cc_trng_pm_put_suspend(dev);
+
+       dev_dbg(dev, "compwork handler done\n");
+       return;
+
+next_rosc:
+       if ((circ_buf_space(drvdata) >= CC_TRNG_EHR_IN_WORDS) &&
+                       (cc_trng_change_rosc(drvdata) == 0)) {
+               /* trigger trng hw with next rosc */
+               cc_trng_hw_trigger(drvdata);
+       } else {
+               atomic_set(&drvdata->pending_hw, 0);
+               cc_trng_pm_put_suspend(dev);
+       }
+}
+
+static irqreturn_t cc_isr(int irq, void *dev_id)
+{
+       struct cctrng_drvdata *drvdata = (struct cctrng_drvdata *)dev_id;
+       struct device *dev = &(drvdata->pdev->dev);
+       u32 irr;
+
+       /* if driver suspended return, probably shared interrupt */
+       if (pm_runtime_suspended(dev))
+               return IRQ_NONE;
+
+       /* read the interrupt status */
+       irr = cc_ioread(drvdata, CC_HOST_RGF_IRR_REG_OFFSET);
+       dev_dbg(dev, "Got IRR=0x%08X\n", irr);
+
+       if (irr == 0) /* Probably shared interrupt line */
+               return IRQ_NONE;
+
+       /* clear interrupt - must be before processing events */
+       cc_iowrite(drvdata, CC_HOST_RGF_ICR_REG_OFFSET, irr);
+
+       /* RNG interrupt - most probable */
+       if (irr & CC_HOST_RNG_IRQ_MASK) {
+               /* Mask RNG interrupts - will be unmasked in deferred work */
+               cc_iowrite(drvdata, CC_RNG_IMR_REG_OFFSET, 0xFFFFFFFF);
+
+               /* We clear RNG interrupt here,
+                * to avoid it from firing as we'll unmask RNG interrupts.
+                */
+               cc_iowrite(drvdata, CC_HOST_RGF_ICR_REG_OFFSET,
+                          CC_HOST_RNG_IRQ_MASK);
+
+               irr &= ~CC_HOST_RNG_IRQ_MASK;
+
+               /* schedule execution of deferred work handler */
+               schedule_work(&drvdata->compwork);
+       }
+
+       if (irr) {
+               dev_dbg_ratelimited(dev,
+                               "IRR includes unknown cause bits (0x%08X)\n",
+                               irr);
+               /* Just warning */
+       }
+
+       return IRQ_HANDLED;
+}
+
+void cc_trng_startwork_handler(struct work_struct *w)
+{
+       struct cctrng_drvdata *drvdata =
+                       container_of(w, struct cctrng_drvdata, startwork);
+
+       drvdata->active_rosc = 0;
+       cc_trng_hw_trigger(drvdata);
+}
+
+
+static int cc_trng_clk_init(struct cctrng_drvdata *drvdata)
+{
+       struct clk *clk;
+       struct device *dev = &(drvdata->pdev->dev);
+       int rc = 0;
+
+       clk = devm_clk_get_optional(dev, NULL);
+       if (IS_ERR(clk)) {
+               if (PTR_ERR(clk) != -EPROBE_DEFER)
+                       dev_err(dev, "Error getting clock: %pe\n", clk);
+               return PTR_ERR(clk);
+       }
+       drvdata->clk = clk;
+
+       rc = clk_prepare_enable(drvdata->clk);
+       if (rc) {
+               dev_err(dev, "Failed to enable clock\n");
+               return rc;
+       }
+
+       return 0;
+}
+
+static void cc_trng_clk_fini(struct cctrng_drvdata *drvdata)
+{
+       clk_disable_unprepare(drvdata->clk);
+}
+
+
+static int cctrng_probe(struct platform_device *pdev)
+{
+       struct resource *req_mem_cc_regs = NULL;
+       struct cctrng_drvdata *drvdata;
+       struct device *dev = &pdev->dev;
+       int rc = 0;
+       u32 val;
+       int irq;
+
+       drvdata = devm_kzalloc(dev, sizeof(*drvdata), GFP_KERNEL);
+       if (!drvdata)
+               return -ENOMEM;
+
+       drvdata->rng.name = devm_kstrdup(dev, dev_name(dev), GFP_KERNEL);
+       if (!drvdata->rng.name)
+               return -ENOMEM;
+
+       drvdata->rng.read = cctrng_read;
+       drvdata->rng.priv = (unsigned long)drvdata;
+       drvdata->rng.quality = CC_TRNG_QUALITY;
+
+       platform_set_drvdata(pdev, drvdata);
+       drvdata->pdev = pdev;
+
+       drvdata->circ.buf = (char *)drvdata->data_buf;
+
+       /* Get device resources */
+       /* First CC registers space */
+       req_mem_cc_regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       /* Map registers space */
+       drvdata->cc_base = devm_ioremap_resource(dev, req_mem_cc_regs);
+       if (IS_ERR(drvdata->cc_base)) {
+               dev_err(dev, "Failed to ioremap registers");
+               return PTR_ERR(drvdata->cc_base);
+       }
+
+       dev_dbg(dev, "Got MEM resource (%s): %pR\n", req_mem_cc_regs->name,
+               req_mem_cc_regs);
+       dev_dbg(dev, "CC registers mapped from %pa to 0x%p\n",
+               &req_mem_cc_regs->start, drvdata->cc_base);
+
+       /* Then IRQ */
+       irq = platform_get_irq(pdev, 0);
+       if (irq < 0) {
+               dev_err(dev, "Failed getting IRQ resource\n");
+               return irq;
+       }
+
+       /* parse sampling rate from device tree */
+       rc = cc_trng_parse_sampling_ratio(drvdata);
+       if (rc) {
+               dev_err(dev, "Failed to get legal sampling ratio for rosc\n");
+               return rc;
+       }
+
+       rc = cc_trng_clk_init(drvdata);
+       if (rc) {
+               dev_err(dev, "cc_trng_clk_init failed\n");
+               return rc;
+       }
+
+       INIT_WORK(&drvdata->compwork, cc_trng_compwork_handler);
+       INIT_WORK(&drvdata->startwork, cc_trng_startwork_handler);
+       spin_lock_init(&drvdata->read_lock);
+
+       /* register the driver isr function */
+       rc = devm_request_irq(dev, irq, cc_isr, IRQF_SHARED, "cctrng", drvdata);
+       if (rc) {
+               dev_err(dev, "Could not register to interrupt %d\n", irq);
+               goto post_clk_err;
+       }
+       dev_dbg(dev, "Registered to IRQ: %d\n", irq);
+
+       /* Clear all pending interrupts */
+       val = cc_ioread(drvdata, CC_HOST_RGF_IRR_REG_OFFSET);
+       dev_dbg(dev, "IRR=0x%08X\n", val);
+       cc_iowrite(drvdata, CC_HOST_RGF_ICR_REG_OFFSET, val);
+
+       /* unmask HOST RNG interrupt */
+       cc_iowrite(drvdata, CC_HOST_RGF_IMR_REG_OFFSET,
+                  cc_ioread(drvdata, CC_HOST_RGF_IMR_REG_OFFSET) &
+                  ~CC_HOST_RNG_IRQ_MASK);
+
+       /* init PM */
+       rc = cc_trng_pm_init(drvdata);
+       if (rc) {
+               dev_err(dev, "cc_trng_pm_init failed\n");
+               goto post_clk_err;
+       }
+
+       /* increment device's usage counter */
+       rc = cc_trng_pm_get(dev);
+       if (rc) {
+               dev_err(dev, "cc_trng_pm_get returned %x\n", rc);
+               goto post_pm_err;
+       }
+
+       /* set pending_hw to verify that HW won't be triggered from read */
+       atomic_set(&drvdata->pending_hw, 1);
+
+       /* registration of the hwrng device */
+       rc = hwrng_register(&drvdata->rng);
+       if (rc) {
+               dev_err(dev, "Could not register hwrng device.\n");
+               goto post_pm_err;
+       }
+
+       /* trigger HW to start generate data */
+       drvdata->active_rosc = 0;
+       cc_trng_hw_trigger(drvdata);
+
+       /* All set, we can allow auto-suspend */
+       cc_trng_pm_go(drvdata);
+
+       dev_info(dev, "ARM cctrng device initialized\n");
+
+       return 0;
+
+post_pm_err:
+       cc_trng_pm_fini(drvdata);
+
+post_clk_err:
+       cc_trng_clk_fini(drvdata);
+
+       return rc;
+}
+
+static int cctrng_remove(struct platform_device *pdev)
+{
+       struct cctrng_drvdata *drvdata = platform_get_drvdata(pdev);
+       struct device *dev = &pdev->dev;
+
+       dev_dbg(dev, "Releasing cctrng resources...\n");
+
+       hwrng_unregister(&drvdata->rng);
+
+       cc_trng_pm_fini(drvdata);
+
+       cc_trng_clk_fini(drvdata);
+
+       dev_info(dev, "ARM cctrng device terminated\n");
+
+       return 0;
+}
+
+static int __maybe_unused cctrng_suspend(struct device *dev)
+{
+       struct cctrng_drvdata *drvdata = dev_get_drvdata(dev);
+
+       dev_dbg(dev, "set HOST_POWER_DOWN_EN\n");
+       cc_iowrite(drvdata, CC_HOST_POWER_DOWN_EN_REG_OFFSET,
+                       POWER_DOWN_ENABLE);
+
+       clk_disable_unprepare(drvdata->clk);
+
+       return 0;
+}
+
+static bool cctrng_wait_for_reset_completion(struct cctrng_drvdata *drvdata)
+{
+       unsigned int val;
+       unsigned int i;
+
+       for (i = 0; i < CC_HW_RESET_LOOP_COUNT; i++) {
+               /* in cc7x3 NVM_IS_IDLE indicates that CC reset is
+                *  completed and device is fully functional
+                */
+               val = cc_ioread(drvdata, CC_NVM_IS_IDLE_REG_OFFSET);
+               if (val & BIT(CC_NVM_IS_IDLE_VALUE_BIT_SHIFT)) {
+                       /* hw indicate reset completed */
+                       return true;
+               }
+               /* allow scheduling other process on the processor */
+               schedule();
+       }
+       /* reset not completed */
+       return false;
+}
+
+static int __maybe_unused cctrng_resume(struct device *dev)
+{
+       struct cctrng_drvdata *drvdata = dev_get_drvdata(dev);
+       int rc;
+
+       dev_dbg(dev, "unset HOST_POWER_DOWN_EN\n");
+       /* Enables the device source clk */
+       rc = clk_prepare_enable(drvdata->clk);
+       if (rc) {
+               dev_err(dev, "failed getting clock back on. We're toast.\n");
+               return rc;
+       }
+
+       /* wait for Cryptocell reset completion */
+       if (!cctrng_wait_for_reset_completion(drvdata)) {
+               dev_err(dev, "Cryptocell reset not completed");
+               return -EBUSY;
+       }
+
+       /* unmask HOST RNG interrupt */
+       cc_iowrite(drvdata, CC_HOST_RGF_IMR_REG_OFFSET,
+                  cc_ioread(drvdata, CC_HOST_RGF_IMR_REG_OFFSET) &
+                  ~CC_HOST_RNG_IRQ_MASK);
+
+       cc_iowrite(drvdata, CC_HOST_POWER_DOWN_EN_REG_OFFSET,
+                  POWER_DOWN_DISABLE);
+
+       return 0;
+}
+
+static UNIVERSAL_DEV_PM_OPS(cctrng_pm, cctrng_suspend, cctrng_resume, NULL);
+
+static const struct of_device_id arm_cctrng_dt_match[] = {
+       { .compatible = "arm,cryptocell-713-trng", },
+       { .compatible = "arm,cryptocell-703-trng", },
+       {},
+};
+MODULE_DEVICE_TABLE(of, arm_cctrng_dt_match);
+
+static struct platform_driver cctrng_driver = {
+       .driver = {
+               .name = "cctrng",
+               .of_match_table = arm_cctrng_dt_match,
+               .pm = &cctrng_pm,
+       },
+       .probe = cctrng_probe,
+       .remove = cctrng_remove,
+};
+
+static int __init cctrng_mod_init(void)
+{
+       /* Compile time assertion checks */
+       BUILD_BUG_ON(CCTRNG_DATA_BUF_WORDS < 6);
+       BUILD_BUG_ON((CCTRNG_DATA_BUF_WORDS & (CCTRNG_DATA_BUF_WORDS-1)) != 0);
+
+       return platform_driver_register(&cctrng_driver);
+}
+module_init(cctrng_mod_init);
+
+static void __exit cctrng_mod_exit(void)
+{
+       platform_driver_unregister(&cctrng_driver);
+}
+module_exit(cctrng_mod_exit);
+
+/* Module description */
+MODULE_DESCRIPTION("ARM CryptoCell TRNG Driver");
+MODULE_AUTHOR("ARM");
+MODULE_LICENSE("GPL v2");
diff --git a/drivers/char/hw_random/cctrng.h b/drivers/char/hw_random/cctrng.h
new file mode 100644 (file)
index 0000000..1f2fde9
--- /dev/null
@@ -0,0 +1,72 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright (C) 2019-2020 ARM Limited or its affiliates. */
+
+#include <linux/bitops.h>
+
+#define POWER_DOWN_ENABLE 0x01
+#define POWER_DOWN_DISABLE 0x00
+
+/* hwrng quality: bits of true entropy per 1024 bits of input */
+#define CC_TRNG_QUALITY        1024
+
+/* CryptoCell TRNG HW definitions */
+#define CC_TRNG_NUM_OF_ROSCS   4
+/* The number of words generated in the entropy holding register (EHR)
+ * 6 words (192 bit) according to HW implementation
+ */
+#define CC_TRNG_EHR_IN_WORDS   6
+#define CC_TRNG_EHR_IN_BITS    (CC_TRNG_EHR_IN_WORDS * BITS_PER_TYPE(u32))
+
+#define CC_HOST_RNG_IRQ_MASK BIT(CC_HOST_RGF_IRR_RNG_INT_BIT_SHIFT)
+
+/* RNG interrupt mask */
+#define CC_RNG_INT_MASK (BIT(CC_RNG_IMR_EHR_VALID_INT_MASK_BIT_SHIFT) | \
+                        BIT(CC_RNG_IMR_AUTOCORR_ERR_INT_MASK_BIT_SHIFT) | \
+                        BIT(CC_RNG_IMR_CRNGT_ERR_INT_MASK_BIT_SHIFT) | \
+                        BIT(CC_RNG_IMR_VN_ERR_INT_MASK_BIT_SHIFT) | \
+                        BIT(CC_RNG_IMR_WATCHDOG_INT_MASK_BIT_SHIFT))
+
+// --------------------------------------
+// BLOCK: RNG
+// --------------------------------------
+#define CC_RNG_IMR_REG_OFFSET  0x0100UL
+#define CC_RNG_IMR_EHR_VALID_INT_MASK_BIT_SHIFT        0x0UL
+#define CC_RNG_IMR_AUTOCORR_ERR_INT_MASK_BIT_SHIFT     0x1UL
+#define CC_RNG_IMR_CRNGT_ERR_INT_MASK_BIT_SHIFT        0x2UL
+#define CC_RNG_IMR_VN_ERR_INT_MASK_BIT_SHIFT   0x3UL
+#define CC_RNG_IMR_WATCHDOG_INT_MASK_BIT_SHIFT 0x4UL
+#define CC_RNG_ISR_REG_OFFSET  0x0104UL
+#define CC_RNG_ISR_EHR_VALID_BIT_SHIFT 0x0UL
+#define CC_RNG_ISR_EHR_VALID_BIT_SIZE  0x1UL
+#define CC_RNG_ISR_AUTOCORR_ERR_BIT_SHIFT      0x1UL
+#define CC_RNG_ISR_AUTOCORR_ERR_BIT_SIZE       0x1UL
+#define CC_RNG_ISR_CRNGT_ERR_BIT_SHIFT 0x2UL
+#define CC_RNG_ISR_CRNGT_ERR_BIT_SIZE  0x1UL
+#define CC_RNG_ISR_WATCHDOG_BIT_SHIFT  0x4UL
+#define CC_RNG_ISR_WATCHDOG_BIT_SIZE   0x1UL
+#define CC_RNG_ICR_REG_OFFSET  0x0108UL
+#define CC_TRNG_CONFIG_REG_OFFSET      0x010CUL
+#define CC_EHR_DATA_0_REG_OFFSET       0x0114UL
+#define CC_RND_SOURCE_ENABLE_REG_OFFSET        0x012CUL
+#define CC_SAMPLE_CNT1_REG_OFFSET      0x0130UL
+#define CC_TRNG_DEBUG_CONTROL_REG_OFFSET       0x0138UL
+#define CC_RNG_SW_RESET_REG_OFFSET     0x0140UL
+#define CC_RNG_CLK_ENABLE_REG_OFFSET   0x01C4UL
+#define CC_RNG_DMA_ENABLE_REG_OFFSET   0x01C8UL
+#define CC_RNG_WATCHDOG_VAL_REG_OFFSET 0x01D8UL
+// --------------------------------------
+// BLOCK: SEC_HOST_RGF
+// --------------------------------------
+#define CC_HOST_RGF_IRR_REG_OFFSET     0x0A00UL
+#define CC_HOST_RGF_IRR_RNG_INT_BIT_SHIFT      0xAUL
+#define CC_HOST_RGF_IMR_REG_OFFSET     0x0A04UL
+#define CC_HOST_RGF_ICR_REG_OFFSET     0x0A08UL
+
+#define CC_HOST_POWER_DOWN_EN_REG_OFFSET       0x0A78UL
+
+// --------------------------------------
+// BLOCK: NVM
+// --------------------------------------
+#define CC_NVM_IS_IDLE_REG_OFFSET      0x0F10UL
+#define CC_NVM_IS_IDLE_VALUE_BIT_SHIFT 0x0UL
+#define CC_NVM_IS_IDLE_VALUE_BIT_SIZE  0x1UL