Updates for turris-mox-rwtm driver

PATCH Superseded Marek Behún 4 participants 37 messages Jun 13, 2024 – Jun 17, 2024 Arm/cz.nic turris support

lore

Share & tools

lore

Marek Behún

author maintainer

2024-06-13 16:10 UTC #0 lore raw

v2 cover Updates for turris-mox-rwtm driver

Hi Gregory, Arnd, Andy et al.

this is v2 of series of fixes, refactors and cleanups for the turris-mox-rwtm driver.

v1 can be found at:
https://patchwork.kernel.org/project/linux-soc/list/?series=861215

Changes since v1:
- complete dropping of driver's own kobject type

Marek

Marek Behún (17):
firmware: turris-mox-rwtm: Do not complete if there are no waiters
firmware: turris-mox-rwtm: Fix checking return value of
wait_for_completion_timeout()
firmware: turris-mox-rwtm: Use macro constant instead of hardcoded

firmware: turris-mox-rwtm: Use ETH_ALEN instead of hardcoded 6
firmware: turris-mox-rwtm: Use the boolean type where appropriate
firmware: turris-mox-rwtm: Hide signature related constants behind
macros
firmware: turris-mox-rwtm: Fix driver includes
firmware: turris-mox-rwtm: Don't create own kobject type
firmware: turris-mox-rwtm: Simplify debugfs code
firmware: turris-mox-rwtm: Convert rest to devm_* and get rid of
driver .remove()
firmware: turris-mox-rwtm: Use dev_err_probe() where possible
firmware: turris-mox-rwtm: Initialize completion before mailbox
firmware: turris-mox-rwtm: Drop redundant device pointer
firmware: turris-mox-rwtm: Use devm_mutex_init() instead of
mutex_init()
firmware: turris-mox-rwtm: Use container_of() instead of hwrng .priv
member
firmware: turris-mox-rwtm: Use EOPNOTSUPP instead of ENOSYS
firmware: turris-mox-rwtm: Deduplicate command execution code

 drivers/firmware/turris-mox-rwtm.c | 381 ++++++++++++-----------------
 1 file changed, 155 insertions(+), 226 deletions(-)

Signature · 1 line

2.44.2

Marek Behún

author maintainer

2024-06-13 16:10 UTC #1 lore raw

PATCH v2 1/17 firmware: turris-mox-rwtm: Do not complete if there are no waiters

Do not complete the command done completion if there are no waiters. This can happen if a wait_for_completion timed out or was interrupted.

Fixes 389711b37493 ("firmware: Add Turris Mox rWTM firmware driver")

Signed-off-by Marek Behún <[email protected]>

---
 drivers/firmware/turris-mox-rwtm.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

drivers/firmware/turris-mox-rwtm.c +4 -1

index 31d962cdd6eb..f1f9160c4195 100644
@@ -2,7 +2,7 @@
 /*
  * Turris Mox rWTM firmware driver
  *
- * Copyright (C) 2019 Marek Behún <[email protected]>
+ * Copyright (C) 2019, 2024 Marek Behún <[email protected]>
  */
 
 #include <linux/armada-37xx-rwtm-mailbox.h>

@@ -174,6 +174,9 @@ static void mox_rwtm_rx_callback(struct mbox_client *cl, void *data)
 	struct mox_rwtm *rwtm = dev_get_drvdata(cl->dev);
 	struct armada_37xx_rwtm_rx_msg *msg = data;
 
+	if (completion_done(&rwtm->cmd_done))
+		return;
+
 	rwtm->reply = *msg;
 	complete(&rwtm->cmd_done);
 }

Signature · 1 line

2.44.2

Marek Behún

author maintainer

2024-06-13 16:10 UTC #2 lore raw

PATCH v2 2/17 firmware: turris-mox-rwtm: Fix checking return value of wait_for_completion_timeout()

In reply to Marek Behún

The wait_for_completion_timeout() function returns 0 if timed out, and a positive value if completed. Fix the usage of this function.

Fixes 389711b37493 ("firmware: Add Turris Mox rWTM firmware driver")

Fixes 2eab59cf0d20 ("firmware: turris-mox-rwtm: fail probing when firmware does not support hwrng")

Signed-off-by Marek Behún <[email protected]>

Reviewed-by Ilpo Järvinen <[email protected]>

---
 drivers/firmware/turris-mox-rwtm.c | 15 ++++++---------
 1 file changed, 6 insertions(+), 9 deletions(-)

drivers/firmware/turris-mox-rwtm.c +6 -9

index f1f9160c4195..3f4758e03c81 100644
@@ -202,9 +202,8 @@ static int mox_get_board_info(struct mox_rwtm *rwtm)
 	if (ret < 0)
 		return ret;
 
-	ret = wait_for_completion_timeout(&rwtm->cmd_done, HZ / 2);
-	if (ret < 0)
-		return ret;
+	if (!wait_for_completion_timeout(&rwtm->cmd_done, HZ / 2))
+		return -ETIMEDOUT;
 
 	ret = mox_get_status(MBOX_CMD_BOARD_INFO, reply->retval);
 	if (ret == -ENODATA) {

@@ -238,9 +237,8 @@ static int mox_get_board_info(struct mox_rwtm *rwtm)
 	if (ret < 0)
 		return ret;
 
-	ret = wait_for_completion_timeout(&rwtm->cmd_done, HZ / 2);
-	if (ret < 0)
-		return ret;
+	if (!wait_for_completion_timeout(&rwtm->cmd_done, HZ / 2))
+		return -ETIMEDOUT;
 
 	ret = mox_get_status(MBOX_CMD_ECDSA_PUB_KEY, reply->retval);
 	if (ret == -ENODATA) {

@@ -277,9 +275,8 @@ static int check_get_random_support(struct mox_rwtm *rwtm)
 	if (ret < 0)
 		return ret;
 
-	ret = wait_for_completion_timeout(&rwtm->cmd_done, HZ / 2);
-	if (ret < 0)
-		return ret;
+	if (!wait_for_completion_timeout(&rwtm->cmd_done, HZ / 2))
+		return -ETIMEDOUT;
 
 	return mox_get_status(MBOX_CMD_GET_RANDOM, rwtm->reply.retval);
 }

Signature · 1 line

2.44.2

Marek Behún

author maintainer

2024-06-13 16:10 UTC #3 lore raw

PATCH v2 3/17 firmware: turris-mox-rwtm: Use macro constant instead of hardcoded 4096

In reply to Marek Behún

The 4096 bytes limit in mox_hwrng_read() is due to the DMA buffer being allocated to one PAGE_SIZE bytes. Use new local macro constant RWTM_DMA_BUFFER_SIZE at allocation time and when used in mox_hwrng_read().

Signed-off-by Marek Behún <[email protected]>

---
 drivers/firmware/turris-mox-rwtm.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

drivers/firmware/turris-mox-rwtm.c +6 -4

index 3f4758e03c81..b8deb13aed98 100644
@@ -19,6 +19,8 @@
 
 #define DRIVER_NAME		"turris-mox-rwtm"
 
+#define RWTM_DMA_BUFFER_SIZE	PAGE_SIZE
+
 /*
  * The macros and constants below come from Turris Mox's rWTM firmware code.
  * This firmware is open source and it's sources can be found at

@@ -287,8 +289,8 @@ static int mox_hwrng_read(struct hwrng *rng, void *data, size_t max, bool wait)
 	struct armada_37xx_rwtm_tx_msg msg;
 	int ret;
 
-	if (max > 4096)
-		max = 4096;
+	if (max > RWTM_DMA_BUFFER_SIZE)
+		max = RWTM_DMA_BUFFER_SIZE;
 
 	msg.command = MBOX_CMD_GET_RANDOM;
 	msg.args[0] = 1;

@@ -479,8 +481,8 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 		return -ENOMEM;
 
 	rwtm->dev = dev;
-	rwtm->buf = dmam_alloc_coherent(dev, PAGE_SIZE, &rwtm->buf_phys,
-					GFP_KERNEL);
+	rwtm->buf = dmam_alloc_coherent(dev, RWTM_DMA_BUFFER_SIZE,
+					&rwtm->buf_phys, GFP_KERNEL);
 	if (!rwtm->buf)
 		return -ENOMEM;

Signature · 1 line

2.44.2

Ilpo Järvinen

Reviewed-by on v2 3/17

2024-06-13 16:22 UTC #4 lore raw

Re: [PATCH v2 03/17] firmware: turris-mox-rwtm: Use macro constant instead of hardcoded 4096

Quoted 34 lines · Marek Behún

Signed-off-by Marek Behún <[email protected]>

---
 drivers/firmware/turris-mox-rwtm.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

drivers/firmware/turris-mox-rwtm.c +4 -2

index 3f4758e03c81..b8deb13aed98 100644
@@ -19,6 +19,8 @@
 
 #define DRIVER_NAME		"turris-mox-rwtm"
 
+#define RWTM_DMA_BUFFER_SIZE	PAGE_SIZE
+
 /*
  * The macros and constants below come from Turris Mox's rWTM firmware code.
  * This firmware is open source and it's sources can be found at

@@ -287,8 +289,8 @@ static int mox_hwrng_read(struct hwrng *rng, void *data, size_t max, bool wait)
 	struct armada_37xx_rwtm_tx_msg msg;
 	int ret;
 
-	if (max > 4096)
-		max = 4096;
+	if (max > RWTM_DMA_BUFFER_SIZE)
+		max = RWTM_DMA_BUFFER_SIZE;
 
 	msg.command = MBOX_CMD_GET_RANDOM;
 	msg.args[0] = 1;

drivers/firmware/turris-mox-rwtm.c static int turris_mox_rwtm_probe(struct platform_device *pdev) · line 487

Quoted 9 lines

drivers/firmware/turris-mox-rwtm.c +2 -2

@@ -479,8 +481,8 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 		return -ENOMEM;
 
 	rwtm->dev = dev;
-	rwtm->buf = dmam_alloc_coherent(dev, PAGE_SIZE, &rwtm->buf_phys,
-					GFP_KERNEL);
+	rwtm->buf = dmam_alloc_coherent(dev, RWTM_DMA_BUFFER_SIZE,
+					&rwtm->buf_phys, GFP_KERNEL);
 	if (!rwtm->buf)
 		return -ENOMEM;

Reviewed-by Ilpo Järvinen <[email protected]>

Signature · 1 line

i.

Andy Shevchenko

2024-06-13 17:52 UTC #5 lore raw

Re: [PATCH v2 03/17] firmware: turris-mox-rwtm: Use macro constant instead of hardcoded 4096

In reply to Marek Behún

Quoted 4 lines · Marek Behún

The 4096 bytes limit in mox_hwrng_read() is due to the DMA buffer being allocated to one PAGE_SIZE bytes. Use new local macro constant RWTM_DMA_BUFFER_SIZE at allocation time and when used in mox_hwrng_read().

...

Quoted 1 line

+#define RWTM_DMA_BUFFER_SIZE PAGE_SIZE

Is it guaranteed to work on any possible PAGE_SIZE settings? We have some architectures that may have it quite different to 4k. Even if this driver will never be run in such an environment, strictly speaking it might not be a good idea to replace 4096 with PAGE_SIZE.

...

drivers/firmware/turris-mox-rwtm.c static int mox_hwrng_read(struct hwrng *rng, void *data, size_t max, bool wait) · line 293

Quoted 2 lines

drivers/firmware/turris-mox-rwtm.c +2 -0

+       if (max > RWTM_DMA_BUFFER_SIZE)
+               max = RWTM_DMA_BUFFER_SIZE;

You probably want max() from minmax.c.

Signature · 2 lines

With Best Regards,
Andy Shevchenko

Arnd Bergmann

2024-06-14 05:45 UTC #6 lore raw

Re: [PATCH v2 03/17] firmware: turris-mox-rwtm: Use macro constant instead of hardcoded 4096

Quoted 6 lines · Andy Shevchenko

Quoted 4 lines · Marek Behún
The 4096 bytes limit in mox_hwrng_read() is due to the DMA buffer being allocated to one PAGE_SIZE bytes. Use new local macro constant RWTM_DMA_BUFFER_SIZE at allocation time and when used in mox_hwrng_read().

Quoted 8 lines

...
Quoted 1 line
+#define RWTM_DMA_BUFFER_SIZE PAGE_SIZE
Is it guaranteed to work on any possible PAGE_SIZE settings? We have some architectures that may have it quite different to 4k. Even if this driver will never be run in such an environment, strictly speaking it might not be a good idea to replace 4096 with PAGE_SIZE.

This is a Cortex-A53, so it does support 64KB page size and the driver should be written to work with that even if it's silly to use that configuration in practice.

I did not check if using PAGE_SIZE or SZ_4K is the correct choice for CONFIG_PAGE_SIZE_64KB kernel here.

Arnd

Marek Behún

author maintainer

2024-06-17 10:58 UTC #7 lore raw

Re: [PATCH v2 03/17] firmware: turris-mox-rwtm: Use macro constant instead of hardcoded 4096

In reply to Andy Shevchenko

Quoted 6 lines · On Thu, 13 Jun 2024 19:51:56 +0200 Andy Shevchenko <[email protected]>…

Quoted 4 lines · Marek Behún
The 4096 bytes limit in mox_hwrng_read() is due to the DMA buffer being allocated to one PAGE_SIZE bytes. Use new local macro constant RWTM_DMA_BUFFER_SIZE at allocation time and when used in mox_hwrng_read().

Quoted 8 lines

...
Quoted 1 line
+#define RWTM_DMA_BUFFER_SIZE PAGE_SIZE
Is it guaranteed to work on any possible PAGE_SIZE settings? We have some architectures that may have it quite different to 4k. Even if this driver will never be run in such an environment, strictly speaking it might not be a good idea to replace 4096 with PAGE_SIZE.

It would work with 64K, but I will use SZ_4K from sizes.h.

Quoted 6 lines

...
drivers/firmware/turris-mox-rwtm.c static int mox_hwrng_read(struct hwrng *rng, void *data, size_t max, bool wait) · line 293
Quoted 2 lines
drivers/firmware/turris-mox-rwtm.c +2 -0
+       if (max > RWTM_DMA_BUFFER_SIZE)
+               max = RWTM_DMA_BUFFER_SIZE;  
You probably want max() from minmax.c.

You mean min() :)

Andy Shevchenko

2024-06-17 11:02 UTC #8 lore raw

Re: [PATCH v2 03/17] firmware: turris-mox-rwtm: Use macro constant instead of hardcoded 4096

Quoted 3 lines · Marek Behún

Quoted 1 line · On Thu, 13 Jun 2024 19:51:56 +0200 Andy Shevchenko <[email protected]>…
On Thu, Jun 13, 2024 at 6:10 PM Marek Behún <[email protected]> wrote:

...

Quoted 6 lines

Quoted 4 lines
drivers/firmware/turris-mox-rwtm.c static int mox_hwrng_read(struct hwrng *rng, void *data, size_t max, bool wait) · line 293
Quoted 2 lines
drivers/firmware/turris-mox-rwtm.c +2 -0
+       if (max > RWTM_DMA_BUFFER_SIZE)
+               max = RWTM_DMA_BUFFER_SIZE;
You probably want max() from minmax.c.
You mean min() :)

Right, or even clamp().

Signature · 2 lines

With Best Regards,
Andy Shevchenko

Marek Behún

author maintainer

2024-06-13 16:11 UTC #9 lore raw

PATCH v2 4/17 firmware: turris-mox-rwtm: Use ETH_ALEN instead of hardcoded 6

In reply to Marek Behún

Use the ETH_ALEN macro instead of hardcoded 6 for MAC address length.

Signed-off-by Marek Behún <[email protected]>

Reviewed-by Ilpo Järvinen <[email protected]>

---
 drivers/firmware/turris-mox-rwtm.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

drivers/firmware/turris-mox-rwtm.c +2 -1

index b8deb13aed98..3a7969d15d27 100644
@@ -10,6 +10,7 @@
 #include <linux/debugfs.h>
 #include <linux/dma-mapping.h>
 #include <linux/hw_random.h>
+#include <linux/if_ether.h>
 #include <linux/mailbox_client.h>
 #include <linux/module.h>
 #include <linux/mutex.h>

@@ -67,7 +68,7 @@ struct mox_rwtm {
 	int has_board_info;
 	u64 serial_number;
 	int board_version, ram_size;
-	u8 mac_address1[6], mac_address2[6];
+	u8 mac_address1[ETH_ALEN], mac_address2[ETH_ALEN];
 
 	/* public key burned in eFuse */
 	int has_pubkey;

Signature · 1 line

2.44.2

Marek Behún

author maintainer

2024-06-13 16:11 UTC #10 lore raw

PATCH v2 5/17 firmware: turris-mox-rwtm: Use the boolean type where appropriate

In reply to Marek Behún

Use the boolean type for has_board_info, has_pubkey and last_sig_done members of the driver's private structure.

Signed-off-by Marek Behún <[email protected]>

---
 drivers/firmware/turris-mox-rwtm.c | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)

drivers/firmware/turris-mox-rwtm.c +8 -7

index 3a7969d15d27..703b05c404e4 100644
@@ -17,6 +17,7 @@
 #include <linux/of.h>
 #include <linux/platform_device.h>
 #include <linux/slab.h>
+#include <linux/types.h>
 
 #define DRIVER_NAME		"turris-mox-rwtm"

@@ -65,13 +66,13 @@ struct mox_rwtm {
 	struct completion cmd_done;
 
 	/* board information */
-	int has_board_info;
+	bool has_board_info;
 	u64 serial_number;
 	int board_version, ram_size;
 	u8 mac_address1[ETH_ALEN], mac_address2[ETH_ALEN];
 
 	/* public key burned in eFuse */
-	int has_pubkey;
+	bool has_pubkey;
 	u8 pubkey[135];
 
 #ifdef CONFIG_DEBUG_FS

@@ -83,7 +84,7 @@ struct mox_rwtm {
 	 */
 	struct dentry *debugfs_root;
 	u32 last_sig[34];
-	int last_sig_done;
+	bool last_sig_done;
 #endif
 };

@@ -227,7 +228,7 @@ static int mox_get_board_info(struct mox_rwtm *rwtm)
 				  reply->status[5]);
 		reply_to_mac_addr(rwtm->mac_address2, reply->status[6],
 				  reply->status[7]);
-		rwtm->has_board_info = 1;
+		rwtm->has_board_info = true;
 
 		pr_info("Turris Mox serial number %016llX\n",
 			rwtm->serial_number);

@@ -254,7 +255,7 @@ static int mox_get_board_info(struct mox_rwtm *rwtm)
 	} else {
 		u32 *s = reply->status;
 
-		rwtm->has_pubkey = 1;
+		rwtm->has_pubkey = true;
 		sprintf(rwtm->pubkey,
 			"%06x%08x%08x%08x%08x%08x%08x%08x%08x%08x%08x%08x%08x%08x%08x%08x%08x",
 			ret, s[0], s[1], s[2], s[3], s[4], s[5], s[6], s[7],

@@ -351,7 +352,7 @@ static ssize_t do_sign_read(struct file *file, char __user *buf, size_t len,
 
 	/* 2 arrays of 17 32-bit words are 136 bytes */
 	ret = simple_read_from_buffer(buf, len, ppos, rwtm->last_sig, 136);
-	rwtm->last_sig_done = 0;
+	rwtm->last_sig_done = false;
 
 	return ret;
 }

@@ -417,7 +418,7 @@ static ssize_t do_sign_write(struct file *file, const char __user *buf,
 	 */
 	memcpy(rwtm->last_sig, rwtm->buf + 68, 136);
 	cpu_to_be32_array(rwtm->last_sig, rwtm->last_sig, 34);
-	rwtm->last_sig_done = 1;
+	rwtm->last_sig_done = true;
 
 	mutex_unlock(&rwtm->busy);
 	return len;

Signature · 1 line

2.44.2

Marek Behún

author maintainer

2024-06-13 16:11 UTC #11 lore raw

PATCH v2 6/17 firmware: turris-mox-rwtm: Hide signature related constants behind macros

In reply to Marek Behún

Hide signature generation related constants behind macros instead of hardcoding the values.

Use SHA512_DIGEST_SIZE from crypto/sha2.h instead of hardcoded 64 as the message size.

Signed-off-by Marek Behún <[email protected]>

---
 drivers/firmware/turris-mox-rwtm.c | 34 +++++++++++++++++++-----------
 1 file changed, 22 insertions(+), 12 deletions(-)

drivers/firmware/turris-mox-rwtm.c +22 ~~-12~~

index 703b05c404e4..fb263f9bbab9 100644
@@ -5,6 +5,7 @@
  * Copyright (C) 2019, 2024 Marek Behún <[email protected]>
  */
 
+#include <crypto/sha2.h>
 #include <linux/armada-37xx-rwtm-mailbox.h>
 #include <linux/completion.h>
 #include <linux/debugfs.h>

@@ -29,6 +30,12 @@
  * https://gitlab.labs.nic.cz/turris/mox-boot-builder/tree/master/wtmi.
  */
 
+#define MOX_ECC_NUMBER_WORDS	17
+#define MOX_ECC_NUMBER_LEN	(MOX_ECC_NUMBER_WORDS * sizeof(u32))
+
+#define MOX_ECC_SIGNATURE_WORDS	(2 * MOX_ECC_NUMBER_WORDS)
+#define MOX_ECC_SIGNATURE_LEN	(MOX_ECC_NUMBER_WORDS * sizeof(u32))
+
 #define MBOX_STS_SUCCESS	(0 << 30)
 #define MBOX_STS_FAIL		(1 << 30)
 #define MBOX_STS_BADCMD		(2 << 30)

@@ -83,7 +90,7 @@ struct mox_rwtm {
 	 * from userspace.
 	 */
 	struct dentry *debugfs_root;
-	u32 last_sig[34];
+	u32 last_sig[MOX_ECC_SIGNATURE_WORDS];
 	bool last_sig_done;
 #endif
 };

@@ -344,14 +351,15 @@ static ssize_t do_sign_read(struct file *file, char __user *buf, size_t len,
 	if (*ppos != 0)
 		return 0;
 
-	if (len < 136)
+	if (len < MOX_ECC_SIGNATURE_LEN)
 		return -EINVAL;
 
 	if (!rwtm->last_sig_done)
 		return -ENODATA;
 
 	/* 2 arrays of 17 32-bit words are 136 bytes */
-	ret = simple_read_from_buffer(buf, len, ppos, rwtm->last_sig, 136);
+	ret = simple_read_from_buffer(buf, len, ppos, rwtm->last_sig,
+				      MOX_ECC_SIGNATURE_LEN);
 	rwtm->last_sig_done = false;
 
 	return ret;

@@ -366,8 +374,7 @@ static ssize_t do_sign_write(struct file *file, const char __user *buf,
 	loff_t dummy = 0;
 	ssize_t ret;
 
-	/* the input is a SHA-512 hash, so exactly 64 bytes have to be read */
-	if (len != 64)
+	if (len != SHA512_DIGEST_SIZE)
 		return -EINVAL;
 
 	/* if last result is not zero user has not read that information yet */

@@ -388,17 +395,18 @@ static ssize_t do_sign_write(struct file *file, const char __user *buf,
 	 *   3. Address of the buffer where ECDSA signature value S shall be
 	 *      stored by the rWTM firmware.
 	 */
-	memset(rwtm->buf, 0, 4);
-	ret = simple_write_to_buffer(rwtm->buf + 4, 64, &dummy, buf, len);
+	memset(rwtm->buf, 0, sizeof(u32));
+	ret = simple_write_to_buffer(rwtm->buf + sizeof(u32),
+				     SHA512_DIGEST_SIZE, &dummy, buf, len);
 	if (ret < 0)
 		goto unlock_mutex;
-	be32_to_cpu_array(rwtm->buf, rwtm->buf, 17);
+	be32_to_cpu_array(rwtm->buf, rwtm->buf, MOX_ECC_NUMBER_WORDS);
 
 	msg.command = MBOX_CMD_SIGN;
 	msg.args[0] = 1;
 	msg.args[1] = rwtm->buf_phys;
-	msg.args[2] = rwtm->buf_phys + 68;
-	msg.args[3] = rwtm->buf_phys + 2 * 68;
+	msg.args[2] = rwtm->buf_phys + MOX_ECC_NUMBER_LEN;
+	msg.args[3] = rwtm->buf_phys + 2 * MOX_ECC_NUMBER_LEN;
 	ret = mbox_send_message(rwtm->mbox, &msg);
 	if (ret < 0)
 		goto unlock_mutex;

@@ -416,8 +424,10 @@ static ssize_t do_sign_write(struct file *file, const char __user *buf,
 	 * computed by the rWTM firmware and convert their words from
 	 * LE to BE.
 	 */
-	memcpy(rwtm->last_sig, rwtm->buf + 68, 136);
-	cpu_to_be32_array(rwtm->last_sig, rwtm->last_sig, 34);
+	memcpy(rwtm->last_sig, rwtm->buf + MOX_ECC_NUMBER_LEN,
+	       MOX_ECC_SIGNATURE_LEN);
+	cpu_to_be32_array(rwtm->last_sig, rwtm->last_sig,
+			  MOX_ECC_SIGNATURE_WORDS);
 	rwtm->last_sig_done = true;
 
 	mutex_unlock(&rwtm->busy);

Signature · 1 line

2.44.2

Marek Behún

author maintainer

2024-06-13 16:11 UTC #12 lore raw

PATCH v2 7/17 firmware: turris-mox-rwtm: Fix driver includes

In reply to Marek Behún

Drop including of.h, include several other headers that are used but not included directly.

Signed-off-by Marek Behún <[email protected]>

---
 drivers/firmware/turris-mox-rwtm.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

drivers/firmware/turris-mox-rwtm.c +6 -1

index fb263f9bbab9..75a98386a2b0 100644
@@ -8,16 +8,21 @@
 #include <crypto/sha2.h>
 #include <linux/armada-37xx-rwtm-mailbox.h>
 #include <linux/completion.h>
+#include <linux/container_of.h>
 #include <linux/debugfs.h>
+#include <linux/device.h>
 #include <linux/dma-mapping.h>
+#include <linux/err.h>
+#include <linux/fs.h>
 #include <linux/hw_random.h>
 #include <linux/if_ether.h>
+#include <linux/kobject.h>
 #include <linux/mailbox_client.h>
 #include <linux/module.h>
 #include <linux/mutex.h>
-#include <linux/of.h>
 #include <linux/platform_device.h>
 #include <linux/slab.h>
+#include <linux/sysfs.h>
 #include <linux/types.h>
 
 #define DRIVER_NAME		"turris-mox-rwtm"

Signature · 1 line

2.44.2

Marek Behún

author maintainer

2024-06-13 16:11 UTC #13 lore raw

PATCH v2 8/17 firmware: turris-mox-rwtm: Don't create own kobject type

In reply to Marek Behún

In order to create attribute files in /sys/firmware/turris-mox-rwtm, this driver creates it's own kobject type.

Simplify this by dropping this own kobject creation, and instead creating standard device attribute files.

For backwards compatibility with sysfs ABI, create a symlink /sys/firmware/turris-mox-rwtm, pointing to this device's sysfs directory.

Signed-off-by Marek Behún <[email protected]>

---
 drivers/firmware/turris-mox-rwtm.c | 107 ++++++++---------------------
 1 file changed, 29 insertions(+), 78 deletions(-)

drivers/firmware/turris-mox-rwtm.c +29 ~~-78~~

index 75a98386a2b0..d17dc0679439 100644
@@ -21,7 +21,6 @@
 #include <linux/module.h>
 #include <linux/mutex.h>
 #include <linux/platform_device.h>
-#include <linux/slab.h>
 #include <linux/sysfs.h>
 #include <linux/types.h>

@@ -60,13 +59,10 @@ enum mbox_cmd {
 	MBOX_CMD_OTP_WRITE	= 8,
 };
 
-struct mox_kobject;
-
 struct mox_rwtm {
 	struct device *dev;
 	struct mbox_client mbox_client;
 	struct mbox_chan *mbox;
-	struct mox_kobject *kobj;
 	struct hwrng hwrng;
 
 	struct armada_37xx_rwtm_rx_msg reply;

@@ -100,59 +96,17 @@ struct mox_rwtm {
 #endif
 };
 
-struct mox_kobject {
-	struct kobject kobj;
-	struct mox_rwtm *rwtm;
-};
-
-static inline struct kobject *rwtm_to_kobj(struct mox_rwtm *rwtm)
-{
-	return &rwtm->kobj->kobj;
-}
-
-static inline struct mox_rwtm *to_rwtm(struct kobject *kobj)
-{
-	return container_of(kobj, struct mox_kobject, kobj)->rwtm;
-}
-
-static void mox_kobj_release(struct kobject *kobj)
-{
-	kfree(to_rwtm(kobj)->kobj);
-}
-
-static const struct kobj_type mox_kobj_ktype = {
-	.release	= mox_kobj_release,
-	.sysfs_ops	= &kobj_sysfs_ops,
-};
-
-static int mox_kobj_create(struct mox_rwtm *rwtm)
-{
-	rwtm->kobj = kzalloc(sizeof(*rwtm->kobj), GFP_KERNEL);
-	if (!rwtm->kobj)
-		return -ENOMEM;
-
-	kobject_init(rwtm_to_kobj(rwtm), &mox_kobj_ktype);
-	if (kobject_add(rwtm_to_kobj(rwtm), firmware_kobj, "turris-mox-rwtm")) {
-		kobject_put(rwtm_to_kobj(rwtm));
-		return -ENXIO;
-	}
-
-	rwtm->kobj->rwtm = rwtm;
-
-	return 0;
-}
-
 #define MOX_ATTR_RO(name, format, cat)				\
 static ssize_t							\
-name##_show(struct kobject *kobj, struct kobj_attribute *a,	\
+name##_show(struct device *dev, struct device_attribute *a,	\
 	    char *buf)						\
 {								\
-	struct mox_rwtm *rwtm = to_rwtm(kobj);	\
+	struct mox_rwtm *rwtm = dev_get_drvdata(dev);		\
 	if (!rwtm->has_##cat)					\
 		return -ENODATA;				\
 	return sprintf(buf, format, rwtm->name);		\
 }								\
-static struct kobj_attribute mox_attr_##name = __ATTR_RO(name)
+static DEVICE_ATTR_RO(name)
 
 MOX_ATTR_RO(serial_number, "%016llX\n", board_info);
 MOX_ATTR_RO(board_version, "%i\n", board_info);

@@ -161,6 +115,17 @@ MOX_ATTR_RO(mac_address1, "%pM\n", board_info);
 MOX_ATTR_RO(mac_address2, "%pM\n", board_info);
 MOX_ATTR_RO(pubkey, "%s\n", pubkey);
 
+static struct attribute *turris_mox_rwtm_attrs[] = {
+	&dev_attr_serial_number.attr,
+	&dev_attr_board_version.attr,
+	&dev_attr_ram_size.attr,
+	&dev_attr_mac_address1.attr,
+	&dev_attr_mac_address2.attr,
+	&dev_attr_pubkey.attr,
+	NULL
+};
+ATTRIBUTE_GROUPS(turris_mox_rwtm);
+
 static int mox_get_status(enum mbox_cmd cmd, u32 retval)
 {
 	if (MBOX_STS_CMD(retval) != cmd)

@@ -175,16 +140,6 @@ static int mox_get_status(enum mbox_cmd cmd, u32 retval)
 		return MBOX_STS_VALUE(retval);
 }
 
-static const struct attribute *mox_rwtm_attrs[] = {
-	&mox_attr_serial_number.attr,
-	&mox_attr_board_version.attr,
-	&mox_attr_ram_size.attr,
-	&mox_attr_mac_address1.attr,
-	&mox_attr_mac_address2.attr,
-	&mox_attr_pubkey.attr,
-	NULL
-};
-
 static void mox_rwtm_rx_callback(struct mbox_client *cl, void *data)
 {
 	struct mox_rwtm *rwtm = dev_get_drvdata(cl->dev);

@@ -487,6 +442,11 @@ static inline void rwtm_unregister_debugfs(struct mox_rwtm *rwtm)
 }
 #endif
 
+static void rwtm_firmware_symlink_drop(void *)
+{
+	sysfs_remove_link(firmware_kobj, DRIVER_NAME);
+}
+
 static int turris_mox_rwtm_probe(struct platform_device *pdev)
 {
 	struct mox_rwtm *rwtm;

@@ -503,18 +463,6 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 	if (!rwtm->buf)
 		return -ENOMEM;
 
-	ret = mox_kobj_create(rwtm);
-	if (ret < 0) {
-		dev_err(dev, "Cannot create turris-mox-rwtm kobject!\n");
-		return ret;
-	}
-
-	ret = sysfs_create_files(rwtm_to_kobj(rwtm), mox_rwtm_attrs);
-	if (ret < 0) {
-		dev_err(dev, "Cannot create sysfs files!\n");
-		goto put_kobj;
-	}
-
 	platform_set_drvdata(pdev, rwtm);
 
 	mutex_init(&rwtm->busy);

@@ -528,7 +476,7 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 		if (ret != -EPROBE_DEFER)
 			dev_err(dev, "Cannot request mailbox channel: %i\n",
 				ret);
-		goto remove_files;
+		return ret;
 	}
 
 	init_completion(&rwtm->cmd_done);

@@ -562,14 +510,18 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 
 	dev_info(dev, "HWRNG successfully registered\n");
 
+	/*
+	 * For sysfs ABI compatibility, create symlink
+	 * /sys/firmware/turris-mox-rwtm to this device's sysfs directory.
+	 */
+	ret = sysfs_create_link(firmware_kobj, &dev->kobj, DRIVER_NAME);
+	if (!ret)
+		devm_add_action_or_reset(dev, rwtm_firmware_symlink_drop, NULL);
+
 	return 0;
 
 free_channel:
 	mbox_free_channel(rwtm->mbox);
-remove_files:
-	sysfs_remove_files(rwtm_to_kobj(rwtm), mox_rwtm_attrs);
-put_kobj:
-	kobject_put(rwtm_to_kobj(rwtm));
 	return ret;
 }

@@ -578,8 +530,6 @@ static void turris_mox_rwtm_remove(struct platform_device *pdev)
 	struct mox_rwtm *rwtm = platform_get_drvdata(pdev);
 
 	rwtm_unregister_debugfs(rwtm);
-	sysfs_remove_files(rwtm_to_kobj(rwtm), mox_rwtm_attrs);
-	kobject_put(rwtm_to_kobj(rwtm));
 	mbox_free_channel(rwtm->mbox);
 }

@@ -597,6 +547,7 @@ static struct platform_driver turris_mox_rwtm_driver = {
 	.driver	= {
 		.name		= DRIVER_NAME,
 		.of_match_table	= turris_mox_rwtm_match,
+		.dev_groups	= turris_mox_rwtm_groups,
 	},
 };
 module_platform_driver(turris_mox_rwtm_driver);

Signature · 1 line

2.44.2

Ilpo Järvinen

Reviewed-by on v2 8/17

2024-06-13 16:28 UTC #14 lore raw

Re: [PATCH v2 08/17] firmware: turris-mox-rwtm: Don't create own kobject type

Quoted 11 lines · Marek Behún

In order to create attribute files in /sys/firmware/turris-mox-rwtm, this driver creates it's own kobject type.
Simplify this by dropping this own kobject creation, and instead creating standard device attribute files.
For backwards compatibility with sysfs ABI, create a symlink /sys/firmware/turris-mox-rwtm, pointing to this device's sysfs directory.
Signed-off-by Marek Behún <[email protected]>

Much better, thanks.

Reviewed-by Ilpo Järvinen <[email protected]>

Signature · 1 line

i.

Quoted 206 lines

---
 drivers/firmware/turris-mox-rwtm.c | 107 ++++++++---------------------
 1 file changed, 29 insertions(+), 78 deletions(-)

drivers/firmware/turris-mox-rwtm.c +29 ~~-78~~

index 75a98386a2b0..d17dc0679439 100644
@@ -21,7 +21,6 @@
 #include <linux/module.h>
 #include <linux/mutex.h>
 #include <linux/platform_device.h>
-#include <linux/slab.h>
 #include <linux/sysfs.h>
 #include <linux/types.h>

@@ -60,13 +59,10 @@ enum mbox_cmd {
 	MBOX_CMD_OTP_WRITE	= 8,
 };
 
-struct mox_kobject;
-
 struct mox_rwtm {
 	struct device *dev;
 	struct mbox_client mbox_client;
 	struct mbox_chan *mbox;
-	struct mox_kobject *kobj;
 	struct hwrng hwrng;
 
 	struct armada_37xx_rwtm_rx_msg reply;

@@ -100,59 +96,17 @@ struct mox_rwtm {
 #endif
 };
 
-struct mox_kobject {
-	struct kobject kobj;
-	struct mox_rwtm *rwtm;
-};
-
-static inline struct kobject *rwtm_to_kobj(struct mox_rwtm *rwtm)
-{
-	return &rwtm->kobj->kobj;
-}
-
-static inline struct mox_rwtm *to_rwtm(struct kobject *kobj)
-{
-	return container_of(kobj, struct mox_kobject, kobj)->rwtm;
-}
-
-static void mox_kobj_release(struct kobject *kobj)
-{
-	kfree(to_rwtm(kobj)->kobj);
-}
-
-static const struct kobj_type mox_kobj_ktype = {
-	.release	= mox_kobj_release,
-	.sysfs_ops	= &kobj_sysfs_ops,
-};
-
-static int mox_kobj_create(struct mox_rwtm *rwtm)
-{
-	rwtm->kobj = kzalloc(sizeof(*rwtm->kobj), GFP_KERNEL);
-	if (!rwtm->kobj)
-		return -ENOMEM;
-
-	kobject_init(rwtm_to_kobj(rwtm), &mox_kobj_ktype);
-	if (kobject_add(rwtm_to_kobj(rwtm), firmware_kobj, "turris-mox-rwtm")) {
-		kobject_put(rwtm_to_kobj(rwtm));
-		return -ENXIO;
-	}
-
-	rwtm->kobj->rwtm = rwtm;
-
-	return 0;
-}
-
 #define MOX_ATTR_RO(name, format, cat)				\
 static ssize_t							\
-name##_show(struct kobject *kobj, struct kobj_attribute *a,	\
+name##_show(struct device *dev, struct device_attribute *a,	\
 	    char *buf)						\
 {								\
-	struct mox_rwtm *rwtm = to_rwtm(kobj);	\
+	struct mox_rwtm *rwtm = dev_get_drvdata(dev);		\
 	if (!rwtm->has_##cat)					\
 		return -ENODATA;				\
 	return sprintf(buf, format, rwtm->name);		\
 }								\
-static struct kobj_attribute mox_attr_##name = __ATTR_RO(name)
+static DEVICE_ATTR_RO(name)
 
 MOX_ATTR_RO(serial_number, "%016llX\n", board_info);
 MOX_ATTR_RO(board_version, "%i\n", board_info);

@@ -161,6 +115,17 @@ MOX_ATTR_RO(mac_address1, "%pM\n", board_info);
 MOX_ATTR_RO(mac_address2, "%pM\n", board_info);
 MOX_ATTR_RO(pubkey, "%s\n", pubkey);
 
+static struct attribute *turris_mox_rwtm_attrs[] = {
+	&dev_attr_serial_number.attr,
+	&dev_attr_board_version.attr,
+	&dev_attr_ram_size.attr,
+	&dev_attr_mac_address1.attr,
+	&dev_attr_mac_address2.attr,
+	&dev_attr_pubkey.attr,
+	NULL
+};
+ATTRIBUTE_GROUPS(turris_mox_rwtm);
+
 static int mox_get_status(enum mbox_cmd cmd, u32 retval)
 {
 	if (MBOX_STS_CMD(retval) != cmd)

@@ -175,16 +140,6 @@ static int mox_get_status(enum mbox_cmd cmd, u32 retval)
 		return MBOX_STS_VALUE(retval);
 }
 
-static const struct attribute *mox_rwtm_attrs[] = {
-	&mox_attr_serial_number.attr,
-	&mox_attr_board_version.attr,
-	&mox_attr_ram_size.attr,
-	&mox_attr_mac_address1.attr,
-	&mox_attr_mac_address2.attr,
-	&mox_attr_pubkey.attr,
-	NULL
-};
-
 static void mox_rwtm_rx_callback(struct mbox_client *cl, void *data)
 {
 	struct mox_rwtm *rwtm = dev_get_drvdata(cl->dev);

@@ -487,6 +442,11 @@ static inline void rwtm_unregister_debugfs(struct mox_rwtm *rwtm)
 }
 #endif
 
+static void rwtm_firmware_symlink_drop(void *)
+{
+	sysfs_remove_link(firmware_kobj, DRIVER_NAME);
+}
+
 static int turris_mox_rwtm_probe(struct platform_device *pdev)
 {
 	struct mox_rwtm *rwtm;

@@ -503,18 +463,6 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 	if (!rwtm->buf)
 		return -ENOMEM;
 
-	ret = mox_kobj_create(rwtm);
-	if (ret < 0) {
-		dev_err(dev, "Cannot create turris-mox-rwtm kobject!\n");
-		return ret;
-	}
-
-	ret = sysfs_create_files(rwtm_to_kobj(rwtm), mox_rwtm_attrs);
-	if (ret < 0) {
-		dev_err(dev, "Cannot create sysfs files!\n");
-		goto put_kobj;
-	}
-
 	platform_set_drvdata(pdev, rwtm);
 
 	mutex_init(&rwtm->busy);

@@ -528,7 +476,7 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 		if (ret != -EPROBE_DEFER)
 			dev_err(dev, "Cannot request mailbox channel: %i\n",
 				ret);
-		goto remove_files;
+		return ret;
 	}
 
 	init_completion(&rwtm->cmd_done);

@@ -562,14 +510,18 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 
 	dev_info(dev, "HWRNG successfully registered\n");
 
+	/*
+	 * For sysfs ABI compatibility, create symlink
+	 * /sys/firmware/turris-mox-rwtm to this device's sysfs directory.
+	 */
+	ret = sysfs_create_link(firmware_kobj, &dev->kobj, DRIVER_NAME);
+	if (!ret)
+		devm_add_action_or_reset(dev, rwtm_firmware_symlink_drop, NULL);
+
 	return 0;
 
 free_channel:
 	mbox_free_channel(rwtm->mbox);
-remove_files:
-	sysfs_remove_files(rwtm_to_kobj(rwtm), mox_rwtm_attrs);
-put_kobj:
-	kobject_put(rwtm_to_kobj(rwtm));
 	return ret;
 }

@@ -578,8 +530,6 @@ static void turris_mox_rwtm_remove(struct platform_device *pdev)
 	struct mox_rwtm *rwtm = platform_get_drvdata(pdev);
 
 	rwtm_unregister_debugfs(rwtm);
-	sysfs_remove_files(rwtm_to_kobj(rwtm), mox_rwtm_attrs);
-	kobject_put(rwtm_to_kobj(rwtm));
 	mbox_free_channel(rwtm->mbox);
 }

@@ -597,6 +547,7 @@ static struct platform_driver turris_mox_rwtm_driver = {
 	.driver	= {
 		.name		= DRIVER_NAME,
 		.of_match_table	= turris_mox_rwtm_match,
+		.dev_groups	= turris_mox_rwtm_groups,
 	},

Andy Shevchenko

2024-06-13 20:33 UTC #15 lore raw

Re: [PATCH v2 08/17] firmware: turris-mox-rwtm: Don't create own kobject type

In reply to Marek Behún

Quoted 3 lines · Marek Behún

In order to create attribute files in /sys/firmware/turris-mox-rwtm, this driver creates it's own kobject type.

its

Quoted 6 lines

Simplify this by dropping this own kobject creation, and instead creating standard device attribute files.
For backwards compatibility with sysfs ABI, create a symlink /sys/firmware/turris-mox-rwtm, pointing to this device's sysfs directory.

...

Quoted 1 line

+static void rwtm_firmware_symlink_drop(void *)

Interesting, can we actually name the parameter, like "unused"?

drivers/firmware/turris-mox-rwtm.c static inline void rwtm_unregister_debugfs(struct mox_rwtm *rwtm) · line 447

Quoted 2 lines

drivers/firmware/turris-mox-rwtm.c +2 -0
+{
+       sysfs_remove_link(firmware_kobj, DRIVER_NAME);

But why not provide a fimware_kobj pointer as a parameter?

Quoted 1 line

+}

...

drivers/firmware/turris-mox-rwtm.c static int turris_mox_rwtm_probe(struct platform_device *pdev) · line 519

Quoted 7 lines

drivers/firmware/turris-mox-rwtm.c +7 -0

+       /*
+        * For sysfs ABI compatibility, create symlink
+        * /sys/firmware/turris-mox-rwtm to this device's sysfs directory.
+        */
+       ret = sysfs_create_link(firmware_kobj, &dev->kobj, DRIVER_NAME);
+       if (!ret)
+               devm_add_action_or_reset(dev, rwtm_firmware_symlink_drop, NULL);

This means that it will remove the link in case devm_add_action() fails. Is it okay?

Signature · 2 lines

With Best Regards,
Andy Shevchenko

Arnd Bergmann

2024-06-14 05:58 UTC #16 lore raw

Re: [PATCH v2 08/17] firmware: turris-mox-rwtm: Don't create own kobject type

Quoted 6 lines · Andy Shevchenko

Quoted 1 line · Marek Behún
+static void rwtm_firmware_symlink_drop(void *)
Interesting, can we actually name the parameter, like "unused"?

Agreed.

I was surprised that the compiler doesn't warn about it and
checked: apparently this is a c23 language feature, so newer versions of gcc now treats it as a gnu extension for --std=gnu11 as well. gcc-10 and older versions however fail:

error: parameter name omitted

and clang-19 still warns

warning: omitting the parameter name in a function definition is a C2x extension [-Wc2x-extensions]

Arnd

Marek Behún

author maintainer

2024-06-17 11:02 UTC #17 lore raw

Re: [PATCH v2 08/17] firmware: turris-mox-rwtm: Don't create own kobject type

In reply to Andy Shevchenko

Quoted 38 lines · On Thu, 13 Jun 2024 22:32:28 +0200 Andy Shevchenko <[email protected]>…

Quoted 3 lines · Marek Behún
In order to create attribute files in /sys/firmware/turris-mox-rwtm, this driver creates it's own kobject type.
its
Quoted 6 lines
Simplify this by dropping this own kobject creation, and instead creating standard device attribute files.
For backwards compatibility with sysfs ABI, create a symlink /sys/firmware/turris-mox-rwtm, pointing to this device's sysfs directory.
...
Quoted 1 line
+static void rwtm_firmware_symlink_drop(void *)
Interesting, can we actually name the parameter, like "unused"?
drivers/firmware/turris-mox-rwtm.c static inline void rwtm_unregister_debugfs(struct mox_rwtm *rwtm) · line 447
Quoted 2 lines
drivers/firmware/turris-mox-rwtm.c +2 -0
+{
+       sysfs_remove_link(firmware_kobj, DRIVER_NAME);  
But why not provide a fimware_kobj pointer as a parameter?
Quoted 1 line
+}
...
drivers/firmware/turris-mox-rwtm.c static int turris_mox_rwtm_probe(struct platform_device *pdev) · line 519
Quoted 7 lines
drivers/firmware/turris-mox-rwtm.c +7 -0
+       /*
+        * For sysfs ABI compatibility, create symlink
+        * /sys/firmware/turris-mox-rwtm to this device's sysfs directory.
+        */
+       ret = sysfs_create_link(firmware_kobj, &dev->kobj, DRIVER_NAME);
+       if (!ret)
+               devm_add_action_or_reset(dev, rwtm_firmware_symlink_drop, NULL);  

Quoted 2 lines

This means that it will remove the link in case devm_add_action() fails. Is it okay?

IMO rather to drop the symlink (which is created for backwards compatibility anyway) in the improbable case that devm_add_action() fails, then than to leave it dangling there on driver unbind.

Marek Behún

author maintainer

2024-06-17 11:04 UTC #18 lore raw

Re: [PATCH v2 08/17] firmware: turris-mox-rwtm: Don't create own kobject type

In reply to Andy Shevchenko

Quoted 38 lines · On Thu, 13 Jun 2024 22:32:28 +0200 Andy Shevchenko <[email protected]>…

Quoted 3 lines · Marek Behún
In order to create attribute files in /sys/firmware/turris-mox-rwtm, this driver creates it's own kobject type.
its
Quoted 6 lines
Simplify this by dropping this own kobject creation, and instead creating standard device attribute files.
For backwards compatibility with sysfs ABI, create a symlink /sys/firmware/turris-mox-rwtm, pointing to this device's sysfs directory.
...
Quoted 1 line
+static void rwtm_firmware_symlink_drop(void *)
Interesting, can we actually name the parameter, like "unused"?
drivers/firmware/turris-mox-rwtm.c static inline void rwtm_unregister_debugfs(struct mox_rwtm *rwtm) · line 447
Quoted 2 lines
drivers/firmware/turris-mox-rwtm.c +2 -0
+{
+       sysfs_remove_link(firmware_kobj, DRIVER_NAME);  
But why not provide a fimware_kobj pointer as a parameter?
Quoted 1 line
+}
...
drivers/firmware/turris-mox-rwtm.c static int turris_mox_rwtm_probe(struct platform_device *pdev) · line 519
Quoted 7 lines
drivers/firmware/turris-mox-rwtm.c +7 -0
+       /*
+        * For sysfs ABI compatibility, create symlink
+        * /sys/firmware/turris-mox-rwtm to this device's sysfs directory.
+        */
+       ret = sysfs_create_link(firmware_kobj, &dev->kobj, DRIVER_NAME);
+       if (!ret)
+               devm_add_action_or_reset(dev, rwtm_firmware_symlink_drop, NULL);  

Quoted 3 lines

This means that it will remove the link in case devm_add_action() fails. Is it okay?

Alternatively I could fail .probe() in case devm_add_action() fails. It is improbable anyway.

Marek Behún

author maintainer

2024-06-13 16:11 UTC #19 lore raw

PATCH v2 9/17 firmware: turris-mox-rwtm: Simplify debugfs code

In reply to Marek Behún

Simplify debugfs code: do not check for errors, as debugfs errors should be ignored, and use devm action for dropping the debugfs directory.

Signed-off-by Marek Behún <[email protected]>

---
 drivers/firmware/turris-mox-rwtm.c | 44 ++++++++----------------------
 1 file changed, 11 insertions(+), 33 deletions(-)

drivers/firmware/turris-mox-rwtm.c +11 ~~-33~~

index d17dc0679439..a536c9c461a7 100644
@@ -90,7 +90,6 @@ struct mox_rwtm {
 	 * It should be rewritten via crypto API once akcipher API is available
 	 * from userspace.
 	 */
-	struct dentry *debugfs_root;
 	u32 last_sig[MOX_ECC_SIGNATURE_WORDS];
 	bool last_sig_done;
 #endif

@@ -405,39 +404,23 @@ static const struct file_operations do_sign_fops = {
 	.llseek	= no_llseek,
 };
 
-static int rwtm_register_debugfs(struct mox_rwtm *rwtm)
+static void rwtm_debugfs_release(void *root)
 {
-	struct dentry *root, *entry;
-
-	root = debugfs_create_dir("turris-mox-rwtm", NULL);
-
-	if (IS_ERR(root))
-		return PTR_ERR(root);
-
-	entry = debugfs_create_file_unsafe("do_sign", 0600, root, rwtm,
-					   &do_sign_fops);
-	if (IS_ERR(entry))
-		goto err_remove;
-
-	rwtm->debugfs_root = root;
-
-	return 0;
-err_remove:
 	debugfs_remove_recursive(root);
-	return PTR_ERR(entry);
 }
 
-static void rwtm_unregister_debugfs(struct mox_rwtm *rwtm)
+static void rwtm_register_debugfs(struct mox_rwtm *rwtm)
 {
-	debugfs_remove_recursive(rwtm->debugfs_root);
+	struct dentry *root;
+
+	root = debugfs_create_dir("turris-mox-rwtm", NULL);
+
+	devm_add_action_or_reset(rwtm->dev, rwtm_debugfs_release, root);
+
+	debugfs_create_file_unsafe("do_sign", 0600, root, rwtm, &do_sign_fops);
 }
 #else
-static inline int rwtm_register_debugfs(struct mox_rwtm *rwtm)
-{
-	return 0;
-}
-
-static inline void rwtm_unregister_debugfs(struct mox_rwtm *rwtm)
+static inline void rwtm_register_debugfs(struct mox_rwtm *rwtm)
 {
 }
 #endif

@@ -502,11 +485,7 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 		goto free_channel;
 	}
 
-	ret = rwtm_register_debugfs(rwtm);
-	if (ret < 0) {
-		dev_err(dev, "Failed creating debugfs entries: %i\n", ret);
-		goto free_channel;
-	}
+	rwtm_register_debugfs(rwtm);
 
 	dev_info(dev, "HWRNG successfully registered\n");

@@ -529,7 +508,6 @@ static void turris_mox_rwtm_remove(struct platform_device *pdev)
 {
 	struct mox_rwtm *rwtm = platform_get_drvdata(pdev);
 
-	rwtm_unregister_debugfs(rwtm);
 	mbox_free_channel(rwtm->mbox);
 }

Signature · 1 line

2.44.2

Andy Shevchenko

2024-06-13 20:38 UTC #20 lore raw

Re: [PATCH v2 09/17] firmware: turris-mox-rwtm: Simplify debugfs code

Quoted 3 lines · Marek Behún

Simplify debugfs code: do not check for errors, as debugfs errors should be ignored, and use devm action for dropping the debugfs directory.

...

drivers/firmware/turris-mox-rwtm.c static const struct file_operations do_sign_fops = { · line 418

Quoted 9 lines

drivers/firmware/turris-mox-rwtm.c +6 -2

-static void rwtm_unregister_debugfs(struct mox_rwtm *rwtm)
+static void rwtm_register_debugfs(struct mox_rwtm *rwtm)
 {
-       debugfs_remove_recursive(rwtm->debugfs_root);
+       struct dentry *root;
+
+       root = debugfs_create_dir("turris-mox-rwtm", NULL);
+
+       devm_add_action_or_reset(rwtm->dev, rwtm_debugfs_release, root);

This is incorrect. If devm_add_action() fails, the root will be removed...

Quoted 1 line

+ debugfs_create_file_unsafe("do_sign", 0600, root, rwtm, &do_sign_fops);

...and this most likely will use the dangling pointer.

Quoted 1 line

}

Signature · 2 lines

With Best Regards,
Andy Shevchenko

Marek Behún

author maintainer

2024-06-17 11:10 UTC #21 lore raw

Re: [PATCH v2 09/17] firmware: turris-mox-rwtm: Simplify debugfs code

Quoted 19 lines · On Thu, 13 Jun 2024 22:37:39 +0200 Andy Shevchenko <[email protected]>…

Quoted 3 lines · Marek Behún
Simplify debugfs code: do not check for errors, as debugfs errors should be ignored, and use devm action for dropping the debugfs directory.
...
drivers/firmware/turris-mox-rwtm.c static const struct file_operations do_sign_fops = { · line 418
Quoted 9 lines
drivers/firmware/turris-mox-rwtm.c +6 -2
-static void rwtm_unregister_debugfs(struct mox_rwtm *rwtm)
+static void rwtm_register_debugfs(struct mox_rwtm *rwtm)
 {
-       debugfs_remove_recursive(rwtm->debugfs_root);
+       struct dentry *root;
+
+       root = debugfs_create_dir("turris-mox-rwtm", NULL);
+
+       devm_add_action_or_reset(rwtm->dev, rwtm_debugfs_release, root);  
This is incorrect. If devm_add_action() fails, the root will be removed...

Quoted 3 lines

Quoted 1 line
+ debugfs_create_file_unsafe("do_sign", 0600, root, rwtm, &do_sign_fops);
...and this most likely will use the dangling pointer.

OK, I will first creae the file and then add the action.

Marek Behún

author maintainer

2024-06-13 16:11 UTC #22 lore raw

PATCH v2 10/17 firmware: turris-mox-rwtm: Convert rest to devm_* and get rid of driver .remove()

In reply to Marek Behún

Use devm resource management for driver's mailbox. This allows us to get rid of driver's .remove() method and gotos in .probe().

Signed-off-by Marek Behún <[email protected]>

---
 drivers/firmware/turris-mox-rwtm.c | 25 +++++++++++--------------
 1 file changed, 11 insertions(+), 14 deletions(-)

drivers/firmware/turris-mox-rwtm.c +11 ~~-14~~

index a536c9c461a7..3017a8c19005 100644
@@ -425,6 +425,11 @@ static inline void rwtm_register_debugfs(struct mox_rwtm *rwtm)
 }
 #endif
 
+static void rwtm_devm_mbox_release(void *mbox)
+{
+	mbox_free_channel(mbox);
+}
+
 static void rwtm_firmware_symlink_drop(void *)
 {
 	sysfs_remove_link(firmware_kobj, DRIVER_NAME);

@@ -462,6 +467,10 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 		return ret;
 	}
 
+	ret = devm_add_action_or_reset(dev, rwtm_devm_mbox_release, rwtm->mbox);
+	if (ret < 0)
+		return ret;
+
 	init_completion(&rwtm->cmd_done);
 
 	ret = mox_get_board_info(rwtm);

@@ -472,7 +481,7 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 	if (ret < 0) {
 		dev_notice(dev,
 			   "Firmware does not support the GET_RANDOM command\n");
-		goto free_channel;
+		return ret;
 	}
 
 	rwtm->hwrng.name = DRIVER_NAME "_hwrng";

@@ -482,7 +491,7 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 	ret = devm_hwrng_register(dev, &rwtm->hwrng);
 	if (ret < 0) {
 		dev_err(dev, "Cannot register HWRNG: %i\n", ret);
-		goto free_channel;
+		return ret;
 	}
 
 	rwtm_register_debugfs(rwtm);

@@ -498,17 +507,6 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 		devm_add_action_or_reset(dev, rwtm_firmware_symlink_drop, NULL);
 
 	return 0;
-
-free_channel:
-	mbox_free_channel(rwtm->mbox);
-	return ret;
-}
-
-static void turris_mox_rwtm_remove(struct platform_device *pdev)
-{
-	struct mox_rwtm *rwtm = platform_get_drvdata(pdev);
-
-	mbox_free_channel(rwtm->mbox);
 }
 
 static const struct of_device_id turris_mox_rwtm_match[] = {

@@ -521,7 +519,6 @@ MODULE_DEVICE_TABLE(of, turris_mox_rwtm_match);
 
 static struct platform_driver turris_mox_rwtm_driver = {
 	.probe	= turris_mox_rwtm_probe,
-	.remove_new = turris_mox_rwtm_remove,
 	.driver	= {
 		.name		= DRIVER_NAME,
 		.of_match_table	= turris_mox_rwtm_match,

Signature · 1 line

2.44.2

Andy Shevchenko

2024-06-13 20:40 UTC #23 lore raw

Re: [PATCH v2 10/17] firmware: turris-mox-rwtm: Convert rest to devm_* and get rid of driver .remove()

Quoted 3 lines · Marek Behún

Use devm resource management for driver's mailbox. This allows us to get rid of driver's .remove() method and gotos in .probe().

of the driver's

...

drivers/firmware/turris-mox-rwtm.c static int turris_mox_rwtm_probe(struct platform_device *pdev) · line 471

Quoted 2 lines

drivers/firmware/turris-mox-rwtm.c +2 -0

+       ret = devm_add_action_or_reset(dev, rwtm_devm_mbox_release, rwtm->mbox);
+       if (ret < 0)

Can it return positive codes? If not, why ' < 0'?

Quoted 1 line

+ return ret;

Signature · 2 lines

With Best Regards,
Andy Shevchenko

Marek Behún

author maintainer

2024-06-17 11:13 UTC #24 lore raw

Re: [PATCH v2 10/17] firmware: turris-mox-rwtm: Convert rest to devm_* and get rid of driver .remove()

Quoted 5 lines · On Thu, 13 Jun 2024 22:39:33 +0200 Andy Shevchenko <[email protected]>…

Quoted 3 lines · Marek Behún
Use devm resource management for driver's mailbox. This allows us to get rid of driver's .remove() method and gotos in .probe().

Quoted 9 lines

of the driver's
...
drivers/firmware/turris-mox-rwtm.c static int turris_mox_rwtm_probe(struct platform_device *pdev) · line 471
Quoted 2 lines
drivers/firmware/turris-mox-rwtm.c +2 -0
+       ret = devm_add_action_or_reset(dev, rwtm_devm_mbox_release, rwtm->mbox);
+       if (ret < 0)  
Can it return positive codes? If not, why ' < 0'?

Right...

Marek Behún

author maintainer

2024-06-13 16:11 UTC #25 lore raw

PATCH v2 11/17 firmware: turris-mox-rwtm: Use dev_err_probe() where possible

In reply to Marek Behún

Use dev_err_probe() where possible in the driver's .probe() method.

Signed-off-by Marek Behún <[email protected]>

---
 drivers/firmware/turris-mox-rwtm.c | 15 +++++++--------
 1 file changed, 7 insertions(+), 8 deletions(-)

drivers/firmware/turris-mox-rwtm.c +7 -8

index 3017a8c19005..81a82b1ef515 100644
@@ -461,10 +461,11 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 	rwtm->mbox = mbox_request_channel(&rwtm->mbox_client, 0);
 	if (IS_ERR(rwtm->mbox)) {
 		ret = PTR_ERR(rwtm->mbox);
-		if (ret != -EPROBE_DEFER)
-			dev_err(dev, "Cannot request mailbox channel: %i\n",
-				ret);
-		return ret;
+		if (ret == -EPROBE_DEFER)
+			return ret;
+
+		return dev_err_probe(dev, ret,
+				     "Cannot request mailbox channel!\n");
 	}
 
 	ret = devm_add_action_or_reset(dev, rwtm_devm_mbox_release, rwtm->mbox);

@@ -489,10 +490,8 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 	rwtm->hwrng.priv = (unsigned long) rwtm;
 
 	ret = devm_hwrng_register(dev, &rwtm->hwrng);
-	if (ret < 0) {
-		dev_err(dev, "Cannot register HWRNG: %i\n", ret);
-		return ret;
-	}
+	if (ret < 0)
+		return dev_err_probe(dev, ret, "Cannot register HWRNG!\n");
 
 	rwtm_register_debugfs(rwtm);

Signature · 1 line

2.44.2

Andy Shevchenko

2024-06-13 20:49 UTC #26 lore raw

Re: [PATCH v2 11/17] firmware: turris-mox-rwtm: Use dev_err_probe() where possible

Quoted 2 lines · Marek Behún

Use dev_err_probe() where possible in the driver's .probe() method.

...

drivers/firmware/turris-mox-rwtm.c static int turris_mox_rwtm_probe(struct platform_device *pdev) · line 465

Quoted 10 lines

        rwtm->mbox = mbox_request_channel(&rwtm->mbox_client, 0);
        if (IS_ERR(rwtm->mbox)) {
                ret = PTR_ERR(rwtm->mbox);

drivers/firmware/turris-mox-rwtm.c +2 -4

-               if (ret != -EPROBE_DEFER)
-                       dev_err(dev, "Cannot request mailbox channel: %i\n",
-                               ret);
-               return ret;
 
+               if (ret == -EPROBE_DEFER)
+                       return ret;

What;s the point of this check, please?

Quoted 3 lines

drivers/firmware/turris-mox-rwtm.c +2 -0

+               return dev_err_probe(dev, ret,
+                                    "Cannot request mailbox channel!\n");
        }

Signature · 2 lines

With Best Regards,
Andy Shevchenko

Marek Behún

author maintainer

2024-06-17 11:14 UTC #27 lore raw

Re: [PATCH v2 11/17] firmware: turris-mox-rwtm: Use dev_err_probe() where possible

Quoted 14 lines · On Thu, 13 Jun 2024 22:49:18 +0200 Andy Shevchenko <[email protected]>…

Quoted 2 lines · Marek Behún

Use dev_err_probe() where possible in the driver's .probe() method.

...

Quoted 7 lines

        rwtm->mbox = mbox_request_channel(&rwtm->mbox_client, 0);
        if (IS_ERR(rwtm->mbox)) {
                ret = PTR_ERR(rwtm->mbox);

drivers/firmware/turris-mox-rwtm.c +0 -4

-               if (ret != -EPROBE_DEFER)
-                       dev_err(dev, "Cannot request mailbox channel: %i\n",
-                               ret);
-               return ret;

Quoted 10 lines

drivers/firmware/turris-mox-rwtm.c static int turris_mox_rwtm_probe(struct platform_device *pdev) · line 465
Quoted 2 lines
drivers/firmware/turris-mox-rwtm.c +2 -0
+               if (ret == -EPROBE_DEFER)
+                       return ret;  
What;s the point of this check, please?
Quoted 3 lines
drivers/firmware/turris-mox-rwtm.c +2 -0
+               return dev_err_probe(dev, ret,
+                                    "Cannot request mailbox channel!\n");
        }  

The point is to not print the error message if we just need to wait for the mailbox driver.

Andy Shevchenko

2024-06-17 12:26 UTC #28 lore raw

Re: [PATCH v2 11/17] firmware: turris-mox-rwtm: Use dev_err_probe() where possible

Quoted 3 lines · Marek Behún

Quoted 1 line · On Thu, 13 Jun 2024 22:49:18 +0200 Andy Shevchenko <[email protected]>…
On Thu, Jun 13, 2024 at 6:11 PM Marek Behún <[email protected]> wrote:

...

Quoted 11 lines

Quoted 10 lines

        rwtm->mbox = mbox_request_channel(&rwtm->mbox_client, 0);
        if (IS_ERR(rwtm->mbox)) {
                ret = PTR_ERR(rwtm->mbox);

drivers/firmware/turris-mox-rwtm.c +2 -4

-               if (ret != -EPROBE_DEFER)
-                       dev_err(dev, "Cannot request mailbox channel: %i\n",
-                               ret);
-               return ret;
 
+               if (ret == -EPROBE_DEFER)
+                       return ret;

Quoted 8 lines

Quoted 5 lines
What;s the point of this check, please?
Quoted 3 lines
drivers/firmware/turris-mox-rwtm.c +2 -0
+               return dev_err_probe(dev, ret,
+                                    "Cannot request mailbox channel!\n");
        }
The point is to not print the error message if we just need to wait for the mailbox driver.

Right, but that was the initial idea behind dev_err_probe(). Have you checked its implementation (and I hope it's documented as well)?

Signature · 2 lines

With Best Regards,
Andy Shevchenko

Marek Behún

author maintainer

2024-06-17 13:04 UTC #29 lore raw

Re: [PATCH v2 11/17] firmware: turris-mox-rwtm: Use dev_err_probe() where possible

Quoted 24 lines · On Mon, 17 Jun 2024 14:26:00 +0200 Andy Shevchenko <[email protected]>…

Quoted 3 lines · Marek Behún
Quoted 1 line · On Thu, 13 Jun 2024 22:49:18 +0200 Andy Shevchenko <[email protected]>…
On Thu, Jun 13, 2024 at 6:11 PM Marek Behún <[email protected]> wrote:
...
Quoted 17 lines
Quoted 16 lines
drivers/firmware/turris-mox-rwtm.c static int turris_mox_rwtm_probe(struct platform_device *pdev) · line 465
Quoted 10 lines
        rwtm->mbox = mbox_request_channel(&rwtm->mbox_client, 0);
        if (IS_ERR(rwtm->mbox)) {
                ret = PTR_ERR(rwtm->mbox);
drivers/firmware/turris-mox-rwtm.c +2 -4
-               if (ret != -EPROBE_DEFER)
-                       dev_err(dev, "Cannot request mailbox channel: %i\n",
-                               ret);
-               return ret;  
 
+               if (ret == -EPROBE_DEFER)
+                       return ret;  
What;s the point of this check, please?
Quoted 3 lines
drivers/firmware/turris-mox-rwtm.c +2 -0
+               return dev_err_probe(dev, ret,
+                                    "Cannot request mailbox channel!\n");
        }  

Quoted 6 lines

Quoted 2 lines
The point is to not print the error message if we just need to wait for the mailbox driver.
Right, but that was the initial idea behind dev_err_probe(). Have you checked its implementation (and I hope it's documented as well)?

OK I am blushing now :-) Thanks, I'll drop the check

Marek Behún

author maintainer

2024-06-13 16:11 UTC #30 lore raw

PATCH v2 12/17 firmware: turris-mox-rwtm: Initialize completion before mailbox

In reply to Marek Behún

Initialize the completion before the mailbox channel is requested.

Signed-off-by Marek Behún <[email protected]>

---
 drivers/firmware/turris-mox-rwtm.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

drivers/firmware/turris-mox-rwtm.c +1 -2

index 81a82b1ef515..35c2a899caab 100644
@@ -454,6 +454,7 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 	platform_set_drvdata(pdev, rwtm);
 
 	mutex_init(&rwtm->busy);
+	init_completion(&rwtm->cmd_done);
 
 	rwtm->mbox_client.dev = dev;
 	rwtm->mbox_client.rx_callback = mox_rwtm_rx_callback;

@@ -472,8 +473,6 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 	if (ret < 0)
 		return ret;
 
-	init_completion(&rwtm->cmd_done);
-
 	ret = mox_get_board_info(rwtm);
 	if (ret < 0)
 		dev_warn(dev, "Cannot read board information: %i\n", ret);

Signature · 1 line

2.44.2

Andy Shevchenko

2024-06-13 20:56 UTC #31 lore raw

Re: [PATCH v2 12/17] firmware: turris-mox-rwtm: Initialize completion before mailbox

Quoted 2 lines · Marek Behún

Initialize the completion before the mailbox channel is requested.

Sounds like a fix, does it need the Fixes tag?

Signature · 2 lines

With Best Regards,
Andy Shevchenko

Marek Behún

author maintainer

2024-06-13 16:11 UTC #32 lore raw

PATCH v2 13/17 firmware: turris-mox-rwtm: Drop redundant device pointer

In reply to Marek Behún

Drop redundant device pointer from driver's private structure.

Signed-off-by Marek Behún <[email protected]>

---
 drivers/firmware/turris-mox-rwtm.c | 18 +++++++++++-------
 1 file changed, 11 insertions(+), 7 deletions(-)

drivers/firmware/turris-mox-rwtm.c +11 -7

index 35c2a899caab..e1142083186c 100644
@@ -60,7 +60,6 @@ enum mbox_cmd {
 };
 
 struct mox_rwtm {
-	struct device *dev;
 	struct mbox_client mbox_client;
 	struct mbox_chan *mbox;
 	struct hwrng hwrng;

@@ -95,6 +94,11 @@ struct mox_rwtm {
 #endif
 };
 
+static inline struct device *rwtm_dev(struct mox_rwtm *rwtm)
+{
+	return rwtm->mbox_client.dev;
+}
+
 #define MOX_ATTR_RO(name, format, cat)				\
 static ssize_t							\
 name##_show(struct device *dev, struct device_attribute *a,	\

@@ -163,6 +167,7 @@ static void reply_to_mac_addr(u8 *mac, u32 t1, u32 t2)
 
 static int mox_get_board_info(struct mox_rwtm *rwtm)
 {
+	struct device *dev = rwtm_dev(rwtm);
 	struct armada_37xx_rwtm_tx_msg msg;
 	struct armada_37xx_rwtm_rx_msg *reply = &rwtm->reply;
 	int ret;

@@ -177,10 +182,10 @@ static int mox_get_board_info(struct mox_rwtm *rwtm)
 
 	ret = mox_get_status(MBOX_CMD_BOARD_INFO, reply->retval);
 	if (ret == -ENODATA) {
-		dev_warn(rwtm->dev,
+		dev_warn(dev,
 			 "Board does not have manufacturing information burned!\n");
 	} else if (ret == -ENOSYS) {
-		dev_notice(rwtm->dev,
+		dev_notice(dev,
 			   "Firmware does not support the BOARD_INFO command\n");
 	} else if (ret < 0) {
 		return ret;

@@ -212,9 +217,9 @@ static int mox_get_board_info(struct mox_rwtm *rwtm)
 
 	ret = mox_get_status(MBOX_CMD_ECDSA_PUB_KEY, reply->retval);
 	if (ret == -ENODATA) {
-		dev_warn(rwtm->dev, "Board has no public key burned!\n");
+		dev_warn(dev, "Board has no public key burned!\n");
 	} else if (ret == -ENOSYS) {
-		dev_notice(rwtm->dev,
+		dev_notice(dev,
 			   "Firmware does not support the ECDSA_PUB_KEY command\n");
 	} else if (ret < 0) {
 		return ret;

@@ -415,7 +420,7 @@ static void rwtm_register_debugfs(struct mox_rwtm *rwtm)
 
 	root = debugfs_create_dir("turris-mox-rwtm", NULL);
 
-	devm_add_action_or_reset(rwtm->dev, rwtm_debugfs_release, root);
+	devm_add_action_or_reset(rwtm_dev(rwtm), rwtm_debugfs_release, root);
 
 	debugfs_create_file_unsafe("do_sign", 0600, root, rwtm, &do_sign_fops);
 }

@@ -445,7 +450,6 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 	if (!rwtm)
 		return -ENOMEM;
 
-	rwtm->dev = dev;
 	rwtm->buf = dmam_alloc_coherent(dev, RWTM_DMA_BUFFER_SIZE,
 					&rwtm->buf_phys, GFP_KERNEL);
 	if (!rwtm->buf)

Signature · 1 line

2.44.2

Marek Behún

author maintainer

2024-06-13 16:11 UTC #33 lore raw

PATCH v2 14/17 firmware: turris-mox-rwtm: Use devm_mutex_init() instead of mutex_init()

In reply to Marek Behún

Use devm_mutex_init() instead of mutex_init(), to properly call mutex_destroy() on probe failure / driver unbind.

Signed-off-by Marek Behún <[email protected]>

---
 drivers/firmware/turris-mox-rwtm.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

drivers/firmware/turris-mox-rwtm.c +4 -1

index e1142083186c..a9a3a189ea47 100644
@@ -457,7 +457,10 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 
 	platform_set_drvdata(pdev, rwtm);
 
-	mutex_init(&rwtm->busy);
+	ret = devm_mutex_init(dev, &rwtm->busy);
+	if (ret < 0)
+		return ret;
+
 	init_completion(&rwtm->cmd_done);
 
 	rwtm->mbox_client.dev = dev;

Signature · 1 line

2.44.2

Marek Behún

author maintainer

2024-06-13 16:11 UTC #34 lore raw

PATCH v2 15/17 firmware: turris-mox-rwtm: Use container_of() instead of hwrng .priv member

In reply to Marek Behún

Use container_of() to get driver private structure from hwnrg structure, instead of the hwrng's .priv member, as suggested by Herbert for another driver [1].

[1] https://lore.kernel.org/soc/[email protected]/

Signed-off-by Marek Behún <[email protected]>

---
 drivers/firmware/turris-mox-rwtm.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

drivers/firmware/turris-mox-rwtm.c +1 -2

index a9a3a189ea47..4c445f6b8140 100644
@@ -258,7 +258,7 @@ static int check_get_random_support(struct mox_rwtm *rwtm)
 
 static int mox_hwrng_read(struct hwrng *rng, void *data, size_t max, bool wait)
 {
-	struct mox_rwtm *rwtm = (struct mox_rwtm *) rng->priv;
+	struct mox_rwtm *rwtm = container_of(rng, struct mox_rwtm, hwrng);
 	struct armada_37xx_rwtm_tx_msg msg;
 	int ret;

@@ -493,7 +493,6 @@ static int turris_mox_rwtm_probe(struct platform_device *pdev)
 
 	rwtm->hwrng.name = DRIVER_NAME "_hwrng";
 	rwtm->hwrng.read = mox_hwrng_read;
-	rwtm->hwrng.priv = (unsigned long) rwtm;
 
 	ret = devm_hwrng_register(dev, &rwtm->hwrng);
 	if (ret < 0)

Signature · 1 line

2.44.2

Marek Behún

author maintainer

2024-06-13 16:11 UTC #35 lore raw

PATCH v2 16/17 firmware: turris-mox-rwtm: Use EOPNOTSUPP instead of ENOSYS

In reply to Marek Behún

Checkpatch warns agains -ENOSYS:
WARNING: ENOSYS means 'invalid syscall nr' and nothing else
Use EOPNOTSUPP instead.

Signed-off-by Marek Behún <[email protected]>

---
 drivers/firmware/turris-mox-rwtm.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

drivers/firmware/turris-mox-rwtm.c +3 -3

index 4c445f6b8140..2ea378fec414 100644
@@ -136,7 +136,7 @@ static int mox_get_status(enum mbox_cmd cmd, u32 retval)
 	else if (MBOX_STS_ERROR(retval) == MBOX_STS_FAIL)
 		return -(int)MBOX_STS_VALUE(retval);
 	else if (MBOX_STS_ERROR(retval) == MBOX_STS_BADCMD)
-		return -ENOSYS;
+		return -EOPNOTSUPP;
 	else if (MBOX_STS_ERROR(retval) != MBOX_STS_SUCCESS)
 		return -EIO;
 	else

@@ -184,7 +184,7 @@ static int mox_get_board_info(struct mox_rwtm *rwtm)
 	if (ret == -ENODATA) {
 		dev_warn(dev,
 			 "Board does not have manufacturing information burned!\n");
-	} else if (ret == -ENOSYS) {
+	} else if (ret == -EOPNOTSUPP) {
 		dev_notice(dev,
 			   "Firmware does not support the BOARD_INFO command\n");
 	} else if (ret < 0) {

@@ -218,7 +218,7 @@ static int mox_get_board_info(struct mox_rwtm *rwtm)
 	ret = mox_get_status(MBOX_CMD_ECDSA_PUB_KEY, reply->retval);
 	if (ret == -ENODATA) {
 		dev_warn(dev, "Board has no public key burned!\n");
-	} else if (ret == -ENOSYS) {
+	} else if (ret == -EOPNOTSUPP) {
 		dev_notice(dev,
 			   "Firmware does not support the ECDSA_PUB_KEY command\n");
 	} else if (ret < 0) {

Signature · 1 line

2.44.2

Marek Behún

author maintainer

2024-06-13 16:11 UTC #36 lore raw

PATCH v2 17/17 firmware: turris-mox-rwtm: Deduplicate command execution code

In reply to Marek Behún

Deduplicate rWTM command execution calls
mbox_send_message()
wait_for_completion()
mox_get_status()
to one function
mox_rwtm_exec()

Signed-off-by Marek Behún <[email protected]>

---
 drivers/firmware/turris-mox-rwtm.c | 98 ++++++++++++------------------
 1 file changed, 39 insertions(+), 59 deletions(-)

drivers/firmware/turris-mox-rwtm.c +39 ~~-59~~

index 2ea378fec414..f4b56269f507 100644
@@ -155,6 +155,34 @@ static void mox_rwtm_rx_callback(struct mbox_client *cl, void *data)
 	complete(&rwtm->cmd_done);
 }
 
+static int mox_rwtm_exec(struct mox_rwtm *rwtm, enum mbox_cmd cmd,
+			 struct armada_37xx_rwtm_tx_msg *msg,
+			 bool interruptible)
+{
+	struct armada_37xx_rwtm_tx_msg _msg = {};
+	int ret;
+
+	if (!msg)
+		msg = &_msg;
+
+	msg->command = cmd;
+
+	ret = mbox_send_message(rwtm->mbox, msg);
+	if (ret < 0)
+		return ret;
+
+	if (interruptible) {
+		ret = wait_for_completion_interruptible(&rwtm->cmd_done);
+		if (ret < 0)
+			return ret;
+	} else {
+		if (!wait_for_completion_timeout(&rwtm->cmd_done, HZ / 2))
+			return -ETIMEDOUT;
+	}
+
+	return mox_get_status(cmd, rwtm->reply.retval);
+}
+
 static void reply_to_mac_addr(u8 *mac, u32 t1, u32 t2)
 {
 	mac[0] = t1 >> 8;

@@ -168,19 +196,10 @@ static void reply_to_mac_addr(u8 *mac, u32 t1, u32 t2)
 static int mox_get_board_info(struct mox_rwtm *rwtm)
 {
 	struct device *dev = rwtm_dev(rwtm);
-	struct armada_37xx_rwtm_tx_msg msg;
 	struct armada_37xx_rwtm_rx_msg *reply = &rwtm->reply;
 	int ret;
 
-	msg.command = MBOX_CMD_BOARD_INFO;
-	ret = mbox_send_message(rwtm->mbox, &msg);
-	if (ret < 0)
-		return ret;
-
-	if (!wait_for_completion_timeout(&rwtm->cmd_done, HZ / 2))
-		return -ETIMEDOUT;
-
-	ret = mox_get_status(MBOX_CMD_BOARD_INFO, reply->retval);
+	ret = mox_rwtm_exec(rwtm, MBOX_CMD_BOARD_INFO, NULL, false);
 	if (ret == -ENODATA) {
 		dev_warn(dev,
 			 "Board does not have manufacturing information burned!\n");

@@ -207,15 +226,7 @@ static int mox_get_board_info(struct mox_rwtm *rwtm)
 		pr_info("           burned RAM size %i MiB\n", rwtm->ram_size);
 	}
 
-	msg.command = MBOX_CMD_ECDSA_PUB_KEY;
-	ret = mbox_send_message(rwtm->mbox, &msg);
-	if (ret < 0)
-		return ret;
-
-	if (!wait_for_completion_timeout(&rwtm->cmd_done, HZ / 2))
-		return -ETIMEDOUT;
-
-	ret = mox_get_status(MBOX_CMD_ECDSA_PUB_KEY, reply->retval);
+	ret = mox_rwtm_exec(rwtm, MBOX_CMD_ECDSA_PUB_KEY, NULL, false);
 	if (ret == -ENODATA) {
 		dev_warn(dev, "Board has no public key burned!\n");
 	} else if (ret == -EOPNOTSUPP) {

@@ -238,38 +249,24 @@ static int mox_get_board_info(struct mox_rwtm *rwtm)
 
 static int check_get_random_support(struct mox_rwtm *rwtm)
 {
-	struct armada_37xx_rwtm_tx_msg msg;
-	int ret;
-
-	msg.command = MBOX_CMD_GET_RANDOM;
-	msg.args[0] = 1;
-	msg.args[1] = rwtm->buf_phys;
-	msg.args[2] = 4;
-
-	ret = mbox_send_message(rwtm->mbox, &msg);
-	if (ret < 0)
-		return ret;
-
-	if (!wait_for_completion_timeout(&rwtm->cmd_done, HZ / 2))
-		return -ETIMEDOUT;
+	struct armada_37xx_rwtm_tx_msg msg = {
+		.args = { 1, rwtm->buf_phys, 4 },
+	};
 
-	return mox_get_status(MBOX_CMD_GET_RANDOM, rwtm->reply.retval);
+	return mox_rwtm_exec(rwtm, MBOX_CMD_GET_RANDOM, &msg, false);
 }
 
 static int mox_hwrng_read(struct hwrng *rng, void *data, size_t max, bool wait)
 {
 	struct mox_rwtm *rwtm = container_of(rng, struct mox_rwtm, hwrng);
-	struct armada_37xx_rwtm_tx_msg msg;
+	struct armada_37xx_rwtm_tx_msg msg = {
+		.args = { 1, rwtm->buf_phys, (max + 3) & ~3 },
+	};
 	int ret;
 
 	if (max > RWTM_DMA_BUFFER_SIZE)
 		max = RWTM_DMA_BUFFER_SIZE;
 
-	msg.command = MBOX_CMD_GET_RANDOM;
-	msg.args[0] = 1;
-	msg.args[1] = rwtm->buf_phys;
-	msg.args[2] = (max + 3) & ~3;
-
 	if (!wait) {
 		if (!mutex_trylock(&rwtm->busy))
 			return -EBUSY;

@@ -277,15 +274,7 @@ static int mox_hwrng_read(struct hwrng *rng, void *data, size_t max, bool wait)
 		mutex_lock(&rwtm->busy);
 	}
 
-	ret = mbox_send_message(rwtm->mbox, &msg);
-	if (ret < 0)
-		goto unlock_mutex;
-
-	ret = wait_for_completion_interruptible(&rwtm->cmd_done);
-	if (ret < 0)
-		goto unlock_mutex;
-
-	ret = mox_get_status(MBOX_CMD_GET_RANDOM, rwtm->reply.retval);
+	ret = mox_rwtm_exec(rwtm, MBOX_CMD_GET_RANDOM, &msg, true);
 	if (ret < 0)
 		goto unlock_mutex;

@@ -333,7 +322,6 @@ static ssize_t do_sign_write(struct file *file, const char __user *buf,
 			     size_t len, loff_t *ppos)
 {
 	struct mox_rwtm *rwtm = file->private_data;
-	struct armada_37xx_rwtm_rx_msg *reply = &rwtm->reply;
 	struct armada_37xx_rwtm_tx_msg msg;
 	loff_t dummy = 0;
 	ssize_t ret;

@@ -366,23 +354,15 @@ static ssize_t do_sign_write(struct file *file, const char __user *buf,
 		goto unlock_mutex;
 	be32_to_cpu_array(rwtm->buf, rwtm->buf, MOX_ECC_NUMBER_WORDS);
 
-	msg.command = MBOX_CMD_SIGN;
 	msg.args[0] = 1;
 	msg.args[1] = rwtm->buf_phys;
 	msg.args[2] = rwtm->buf_phys + MOX_ECC_NUMBER_LEN;
 	msg.args[3] = rwtm->buf_phys + 2 * MOX_ECC_NUMBER_LEN;
-	ret = mbox_send_message(rwtm->mbox, &msg);
-	if (ret < 0)
-		goto unlock_mutex;
 
-	ret = wait_for_completion_interruptible(&rwtm->cmd_done);
+	ret = mox_rwtm_exec(rwtm, MBOX_CMD_SIGN, &msg, true);
 	if (ret < 0)
 		goto unlock_mutex;
 
-	ret = MBOX_STS_VALUE(reply->retval);
-	if (MBOX_STS_ERROR(reply->retval) != MBOX_STS_SUCCESS)
-		goto unlock_mutex;
-
 	/*
 	 * Here we read the R and S values of the ECDSA signature
 	 * computed by the rWTM firmware and convert their words from

Signature · 1 line

2.44.2