nvtzvault: take lock before close session

- need to take lock to avoid race conditions when accessing
  mailbox.
- don't use spinlock in irq context as irq is not shared
  and only one request is processed at a time.
- remove debug print.

Bug 5225204
Bug 5275408

Change-Id: I2de5a07d06568012d6557d3a32126478b6ddaf15
Signed-off-by: Nagaraj P N <nagarajp@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3356286
(cherry picked from commit 521e75b7c3cbc5f4724a8a0f3c11c3a2940383b2)
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3364010
GVS: buildbot_gerritrpt <buildbot_gerritrpt@nvidia.com>
Reviewed-by: Sandeep Trasi <strasi@nvidia.com>
Reviewed-by: Leo Chiu <lchiu@nvidia.com>
Reviewed-by: svcacv <svcacv@nvidia.com>

drivers/nvtzvault/nvtzvault-main.c

@@ -198,7 +198,7 @@ static int nvtzvault_open_session(struct nvtzvault_ctx *ctx,
 	ret = nvtzvault_tee_translate_saerror_to_syserror(
 			resp_hdr.result);
 	if (ret != 0) {
-		NVTZVAULT_ERR("%s: SA returned error: %d(%x)\n", __func__, resp_hdr.result,
+		NVTZVAULT_ERR("%s: SA returned error: %d(0x%x)\n", __func__, resp_hdr.result,
 				resp_hdr.result);
 		return ret;
 	}
@@ -284,7 +284,7 @@ static int nvtzvault_invoke_cmd(struct nvtzvault_ctx *ctx,
 	ret = nvtzvault_tee_translate_saerror_to_syserror(
 			resp_hdr.result);
 	if (ret != 0) {
-		NVTZVAULT_ERR("%s: SA returned error: %d(%x)\n", __func__, resp_hdr.result,
+		NVTZVAULT_ERR("%s: SA returned error: %d(0x%x)\n", __func__, resp_hdr.result,
 				resp_hdr.result);
 		return ret;
 	}
@@ -360,7 +360,7 @@ static int nvtzvault_close_session(struct nvtzvault_ctx *ctx,
 		// Only clear session if close was successful
 		set_session_closed(ctx, close_session_ctl->session_id);
 	} else {
-		NVTZVAULT_ERR("%s: SA returned error: %d(%x)\n", __func__, resp_hdr.result,
+		NVTZVAULT_ERR("%s: SA returned error: %d(0x%x)\n", __func__, resp_hdr.result,
 				resp_hdr.result);
 	}
 
@@ -406,13 +406,14 @@ static int nvtzvault_ta_dev_release(struct inode *inode, struct file *filp)
 	struct nvtzvault_close_session_ctl close_session_ctl;
 
 	if (atomic_read(&g_nvtzvault_dev.total_active_session_count) > 0) {
+		mutex_lock(&g_nvtzvault_dev.lock);
 		for (uint32_t i = 0; i < NVTZVAULT_MAX_SESSIONS; i++) {
 			if (is_session_open(ctx, i)) {
-				NVTZVAULT_ERR("%s: closing session %u\n", __func__, i);
 				close_session_ctl.session_id = i;
 				nvtzvault_close_session(ctx, &close_session_ctl);
 			}
 		}
+		mutex_unlock(&g_nvtzvault_dev.lock);
 	}
 
 	kfree(ctx);
@@ -434,13 +435,14 @@ static long nvtzvault_ta_dev_ioctl(struct file *filp, unsigned int ioctl_num, un
 		return -EPERM;
 	}
 
+	mutex_lock(&g_nvtzvault_dev.lock);
+
 	if (atomic_read(&g_nvtzvault_dev.in_suspend_state)) {
 		NVTZVAULT_ERR("%s(): device is in suspend state\n", __func__);
-		return -EBUSY;
+		ret = -EBUSY;
+		goto release_lock;
 	}
 
-	mutex_lock(&g_nvtzvault_dev.lock);
-
 	switch (ioctl_num) {
 	case NVTZVAULT_IOCTL_OPEN_SESSION:
 		open_session_ctl = kzalloc(sizeof(*open_session_ctl), GFP_KERNEL);
@@ -825,11 +827,17 @@ static int nvtzvault_suspend(struct device *dev)
 	/* Add print to log in nvlog buffer  */
 	dev_err(dev, "%s start\n", __func__);
 
-	if (atomic_read(&g_nvtzvault_dev.total_active_session_count) > 0)
+	mutex_lock(&g_nvtzvault_dev.lock);
+
+	if (atomic_read(&g_nvtzvault_dev.total_active_session_count) > 0) {
+		mutex_unlock(&g_nvtzvault_dev.lock);
 		return -EBUSY;
+	}
 
 	atomic_set(&g_nvtzvault_dev.in_suspend_state, 1);
 
+	mutex_unlock(&g_nvtzvault_dev.lock);
+
 	/* Add print to log in nvlog buffer  */
 	dev_err(dev, "%s done\n", __func__);
 

drivers/nvtzvault/oesp-mailbox.c

@@ -65,7 +65,6 @@ struct mbox_ctx {
 	uint64_t hpse_carveout_size;
 	void *oesp_mbox_reg_base_va;
 	uint64_t oesp_mbox_reg_size;
-	spinlock_t lock;
 	struct mbox_resp resp;
 } g_mbox_ctx;
 
@@ -166,7 +165,6 @@ static irqreturn_t tegra_hpse_irq_handler(int irq, void *dev_id)
 	volatile u32 reg_val;
 	u8 *oesp_reg_mem_ptr = g_mbox_ctx.oesp_mbox_reg_base_va;
 	volatile struct mbox_resp *resp = &g_mbox_ctx.resp;
-	unsigned long flags;
 
 	/* Read PSC control register to check interrupt status */
 	reg_val = readl(oesp_reg_mem_ptr + PSC_CTRL_REG_OFFSET);
@@ -175,13 +173,14 @@ static irqreturn_t tegra_hpse_irq_handler(int irq, void *dev_id)
 	if ((reg_val & MBOX_OUT_VALID) == 0U)
 		return IRQ_NONE;
 
-	spin_lock_irqsave(&g_mbox_ctx.lock, flags);
-
 	/* Read response registers */
 	resp->task_opcode = readl(oesp_reg_mem_ptr + RESP_OPCODE_OFFSET);
 	resp->format_flag = readl(oesp_reg_mem_ptr + RESP_FORMAT_FLAG_OFFSET);
 	resp->status = readl(oesp_reg_mem_ptr + RESP_STATUS_OFFSET);
 
+	/* Ensure register read is complete before acknowledging response */
+	rmb();
+
 	/* Set MBOX_OUT_DONE to acknowledge response */
 	reg_val = readl(oesp_reg_mem_ptr + EXT_CTRL_OFFSET);
 	reg_val |= MBOX_OUT_DONE;
@@ -190,8 +189,6 @@ static irqreturn_t tegra_hpse_irq_handler(int irq, void *dev_id)
 	/* Signal completion to waiting thread */
 	complete(&mbox_completion);
 
-	spin_unlock_irqrestore(&g_mbox_ctx.lock, flags);
-
 	return IRQ_HANDLED;
 }
 
@@ -277,8 +274,6 @@ int32_t oesp_mailbox_init(struct platform_device *pdev)
 	g_mbox_ctx.oesp_mbox_reg_base_va = oesp_mbox_reg_base_va;
 	g_mbox_ctx.oesp_mbox_reg_size = oesp_mbox_reg_size;
 
-	spin_lock_init(&g_mbox_ctx.lock);
-
 	/* Get IRQ from tegra-hpse node */
 	irq = platform_get_irq(pdev, 0);
 	if (irq < 0) {