devres: provide and use devm_kstrdup_const()

PATCH RESEND In review Bartosz Golaszewski 2 participants 10 messages Oct 14, 2018 – Oct 19, 2018

Subsystems (6)

Memory management MM core Tegra architecture support Generic include/asm header files +2 more subsystems

lore

Share & tools

lore

Bartosz Golaszewski

author

2018-10-14 15:20 UTC #0 lore raw

v6 cover devres: provide and use devm_kstrdup_const()

Hi Greg,

this has been reviewed a lot and three first patches can be picked up. Please take it through your tree.

This series implements devm_kstrdup_const() together with some prerequisite changes and uses it in tegra-hsp driver.

v1 -> v2:
- fixed the changelog in the patch implementing devm_kstrdup_const()
- fixed the kernel doc
- moved is_kernel_rodata() to asm-generic/sections.h
- fixed constness

v2 -> v3:
- rebased on top of 4.19-rc5 as there were some conflicts in the pmc-atom driver
- collected Reviewed-by tags

v3 -> v4:
- Andy NAK'ed patch 4/4 so I added a different example
- collected more tags

v4 -> v5:
- instead of providing devm_kfree_const(), make devm_kfree() check if given pointer is not in .rodata and act accordingly

v5 -> v6:
- fixed the commit message in patch 2/4 (s/devm_kfree_const/devm_kfree/)
- collected even more tags

Bartosz Golaszewski (4):
devres: constify p in devm_kfree()
mm: move is_kernel_rodata() to asm-generic/sections.h
devres: provide devm_kstrdup_const()
mailbox: tegra-hsp: use devm_kstrdup_const()

 drivers/base/devres.c          | 36 +++++++++++++++++++++++++++--
 drivers/mailbox/tegra-hsp.c    | 41 ++++++++--------------------------
 include/asm-generic/sections.h | 14 ++++++++++++
 include/linux/device.h         |  4 +++-
 mm/util.c                      |  7 ------
 5 files changed, 60 insertions(+), 42 deletions(-)

Signature · 1 line

2.19.1

Bartosz Golaszewski

author

2018-10-14 15:20 UTC #1 lore raw

PATCH RESEND v6 1/4 devres: constify p in devm_kfree()

Make devm_kfree() signature uniform with that of kfree(). To avoid compiler warnings: cast p to (void *) when calling devres_destroy().

Signed-off-by Bartosz Golaszewski <[email protected]>

Reviewed-by Bjorn Andersson <[email protected]>

Reviewed-by Geert Uytterhoeven <[email protected]>

Acked-by Rasmus Villemoes <[email protected]>

Reviewed-by Andy Shevchenko <[email protected]>

---
 drivers/base/devres.c  | 5 +++--
 include/linux/device.h | 2 +-
 2 files changed, 4 insertions(+), 3 deletions(-)

drivers/base/devres.c +3 -2

index f98a097e73f2..438c91a43508 100644
@@ -885,11 +885,12 @@ EXPORT_SYMBOL_GPL(devm_kasprintf);
  *
  * Free memory allocated with devm_kmalloc().
  */
-void devm_kfree(struct device *dev, void *p)
+void devm_kfree(struct device *dev, const void *p)
 {
 	int rc;
 
-	rc = devres_destroy(dev, devm_kmalloc_release, devm_kmalloc_match, p);
+	rc = devres_destroy(dev, devm_kmalloc_release,
+			    devm_kmalloc_match, (void *)p);
 	WARN_ON(rc);
 }
 EXPORT_SYMBOL_GPL(devm_kfree);

include/linux/device.h +1 -1

index 8f882549edee..33f7cb271fbb 100644
@@ -692,7 +692,7 @@ static inline void *devm_kcalloc(struct device *dev,
 {
 	return devm_kmalloc_array(dev, n, size, flags | __GFP_ZERO);
 }
-extern void devm_kfree(struct device *dev, void *p);
+extern void devm_kfree(struct device *dev, const void *p);
 extern char *devm_kstrdup(struct device *dev, const char *s, gfp_t gfp) __malloc;
 extern void *devm_kmemdup(struct device *dev, const void *src, size_t len,
 			  gfp_t gfp);

Signature · 1 line

2.19.1

Bartosz Golaszewski

author

2018-10-14 15:20 UTC #2 lore raw

PATCH RESEND v6 2/4 mm: move is_kernel_rodata() to asm-generic/sections.h

In reply to Bartosz Golaszewski

Export this routine so that we can use it later in devm_kstrdup_const() and devm_kfree().

Signed-off-by Bartosz Golaszewski <[email protected]>

Reviewed-by Bjorn Andersson <[email protected]>

Acked-by Mike Rapoport <[email protected]>

Acked-by Rasmus Villemoes <[email protected]>

Reviewed-by Geert Uytterhoeven <[email protected]>

Reviewed-by Andy Shevchenko <[email protected]>

---
 include/asm-generic/sections.h | 14 ++++++++++++++
 mm/util.c                      |  7 -------
 2 files changed, 14 insertions(+), 7 deletions(-)

include/asm-generic/sections.h +14 -0

index 849cd8eb5ca0..d79abca81a52 100644
@@ -141,4 +141,18 @@ static inline bool init_section_intersects(void *virt, size_t size)
 	return memory_intersects(__init_begin, __init_end, virt, size);
 }
 
+/**
+ * is_kernel_rodata - checks if the pointer address is located in the
+ *                    .rodata section
+ *
+ * @addr: address to check
+ *
+ * Returns: true if the address is located in .rodata, false otherwise.
+ */
+static inline bool is_kernel_rodata(unsigned long addr)
+{
+	return addr >= (unsigned long)__start_rodata &&
+	       addr < (unsigned long)__end_rodata;
+}
+
 #endif /* _ASM_GENERIC_SECTIONS_H_ */

mm/util.c +0 -7

index 9e3ebd2ef65f..470f5cd80b64 100644
@@ -15,17 +15,10 @@
 #include <linux/vmalloc.h>
 #include <linux/userfaultfd_k.h>
 
-#include <asm/sections.h>
 #include <linux/uaccess.h>
 
 #include "internal.h"
 
-static inline int is_kernel_rodata(unsigned long addr)
-{
-	return addr >= (unsigned long)__start_rodata &&
-		addr < (unsigned long)__end_rodata;
-}
-
 /**
  * kfree_const - conditionally free memory
  * @x: pointer to the memory

Signature · 1 line

2.19.1

Greg Kroah-Hartman

maintainer

2018-10-16 10:53 UTC #3 lore raw

Re: [RESEND PATCH v6 2/4] mm: move is_kernel_rodata() to asm-generic/sections.h

Quoted 24 lines · Bartosz Golaszewski

Export this routine so that we can use it later in devm_kstrdup_const() and devm_kfree().
Signed-off-by Bartosz Golaszewski <[email protected]>
Reviewed-by Bjorn Andersson <[email protected]>
Acked-by Mike Rapoport <[email protected]>
Acked-by Rasmus Villemoes <[email protected]>
Reviewed-by Geert Uytterhoeven <[email protected]>
Reviewed-by Andy Shevchenko <[email protected]>
---
 include/asm-generic/sections.h | 14 ++++++++++++++
 mm/util.c                      |  7 -------
 2 files changed, 14 insertions(+), 7 deletions(-)
include/asm-generic/sections.h +2 -0
index 849cd8eb5ca0..d79abca81a52 100644
@@ -141,4 +141,18 @@ static inline bool init_section_intersects(void *virt, size_t size)
 	return memory_intersects(__init_begin, __init_end, virt, size);
 }
 
+/**
+ * is_kernel_rodata - checks if the pointer address is located in the

include/asm-generic/sections.h static inline bool init_section_intersects(void *virt, size_t size) · line 155

Quoted 10 lines

include/asm-generic/sections.h +10 -0

+ *                    .rodata section
+ *
+ * @addr: address to check
+ *
+ * Returns: true if the address is located in .rodata, false otherwise.
+ */
+static inline bool is_kernel_rodata(unsigned long addr)
+{
+	return addr >= (unsigned long)__start_rodata &&
+	       addr < (unsigned long)__end_rodata;

Those symbols are not exported, are you sure this is going to work for a kernel module that ends up using this function?

I'll take it, but be aware of the future complications that might happen here...

thanks,

greg k-h

Bartosz Golaszewski

author

2018-10-19 14:42 UTC #4 lore raw

Re: [RESEND PATCH v6 2/4] mm: move is_kernel_rodata() to asm-generic/sections.h

wt., 16 paź 2018 o 12:53 Greg Kroah-Hartman
<[email protected]> napisał(a):

Quoted 24 lines · Bartosz Golaszewski

Quoted 22 lines
Export this routine so that we can use it later in devm_kstrdup_const() and devm_kfree().
Signed-off-by Bartosz Golaszewski <[email protected]>
Reviewed-by Bjorn Andersson <[email protected]>
Acked-by Mike Rapoport <[email protected]>
Acked-by Rasmus Villemoes <[email protected]>
Reviewed-by Geert Uytterhoeven <[email protected]>
Reviewed-by Andy Shevchenko <[email protected]>
---
 include/asm-generic/sections.h | 14 ++++++++++++++
 mm/util.c                      |  7 -------
 2 files changed, 14 insertions(+), 7 deletions(-)
include/asm-generic/sections.h +0 -0
index 849cd8eb5ca0..d79abca81a52 100644
@@ -141,4 +141,18 @@ static inline bool init_section_intersects(void *virt, size_t size)
      return memory_intersects(__init_begin, __init_end, virt, size);
 }
 

Quoted 16 lines

include/asm-generic/sections.h static inline bool init_section_intersects(void *virt, size_t size) · line 155
Quoted 12 lines
include/asm-generic/sections.h +12 -0
+/**
+ * is_kernel_rodata - checks if the pointer address is located in the
+ *                    .rodata section
+ *
+ * @addr: address to check
+ *
+ * Returns: true if the address is located in .rodata, false otherwise.
+ */
+static inline bool is_kernel_rodata(unsigned long addr)
+{
+     return addr >= (unsigned long)__start_rodata &&
+            addr < (unsigned long)__end_rodata;
Those symbols are not exported, are you sure this is going to work for a kernel module that ends up using this function?

I tested it with a loaded module, yes.

Quoted 3 lines

I'll take it, but be aware of the future complications that might happen here...

These symbols are built-in for all architectures. Do they even need exporting? On the off-chance that something fails, we can move this function back to mm/util.c as it was in the first version of this series.

Best regards,
Bartosz Golaszewski

Bartosz Golaszewski

author

2018-10-14 15:20 UTC #5 lore raw

PATCH RESEND v6 3/4 devres: provide devm_kstrdup_const()

In reply to Bartosz Golaszewski

Provide a resource managed version of kstrdup_const(). This variant internally calls devm_kstrdup() on pointers that are outside of .rodata section and returns the string as is otherwise.

Make devm_kfree() check if the passed pointer doesn't point to .rodata and if so - don't actually destroy the resource.

Signed-off-by Bartosz Golaszewski <[email protected]>

Reviewed-by Bjorn Andersson <[email protected]>

Acked-by Mike Rapoport <[email protected]>

Acked-by Rasmus Villemoes <[email protected]>

Reviewed-by Geert Uytterhoeven <[email protected]>

Reviewed-by Andy Shevchenko <[email protected]>

---
 drivers/base/devres.c  | 31 +++++++++++++++++++++++++++++++
 include/linux/device.h |  2 ++
 2 files changed, 33 insertions(+)

drivers/base/devres.c +31 -0

index 438c91a43508..00c70f0fcdcd 100644
@@ -11,6 +11,8 @@
 #include <linux/slab.h>
 #include <linux/percpu.h>
 
+#include <asm/sections.h>
+
 #include "base.h"
 
 struct devres_node {

@@ -822,6 +824,28 @@ char *devm_kstrdup(struct device *dev, const char *s, gfp_t gfp)
 }
 EXPORT_SYMBOL_GPL(devm_kstrdup);
 
+/**
+ * devm_kstrdup_const - resource managed conditional string duplication
+ * @dev: device for which to duplicate the string
+ * @s: the string to duplicate
+ * @gfp: the GFP mask used in the kmalloc() call when allocating memory
+ *
+ * Strings allocated by devm_kstrdup_const will be automatically freed when
+ * the associated device is detached.
+ *
+ * RETURNS:
+ * Source string if it is in .rodata section otherwise it falls back to
+ * devm_kstrdup.
+ */
+const char *devm_kstrdup_const(struct device *dev, const char *s, gfp_t gfp)
+{
+	if (is_kernel_rodata((unsigned long)s))
+		return s;
+
+	return devm_kstrdup(dev, s, gfp);
+}
+EXPORT_SYMBOL(devm_kstrdup_const);
+
 /**
  * devm_kvasprintf - Allocate resource managed space and format a string
  *		     into that.

@@ -889,6 +913,13 @@ void devm_kfree(struct device *dev, const void *p)
 {
 	int rc;
 
+	/*
+	 * Special case: pointer to a string in .rodata returned by
+	 * devm_kstrdup_const().
+	 */
+	if (unlikely(is_kernel_rodata((unsigned long)p)))
+		return;
+
 	rc = devres_destroy(dev, devm_kmalloc_release,
 			    devm_kmalloc_match, (void *)p);
 	WARN_ON(rc);

include/linux/device.h +2 -0

index 33f7cb271fbb..e626acb93ef5 100644
@@ -694,6 +694,8 @@ static inline void *devm_kcalloc(struct device *dev,
 }
 extern void devm_kfree(struct device *dev, const void *p);
 extern char *devm_kstrdup(struct device *dev, const char *s, gfp_t gfp) __malloc;
+extern const char *devm_kstrdup_const(struct device *dev,
+				      const char *s, gfp_t gfp);
 extern void *devm_kmemdup(struct device *dev, const void *src, size_t len,
 			  gfp_t gfp);

Signature · 1 line

2.19.1

Greg Kroah-Hartman

maintainer

2018-10-16 10:50 UTC #6 lore raw

Re: [RESEND PATCH v6 3/4] devres: provide devm_kstrdup_const()

Quoted 46 lines · Bartosz Golaszewski

Provide a resource managed version of kstrdup_const(). This variant internally calls devm_kstrdup() on pointers that are outside of .rodata section and returns the string as is otherwise.
Make devm_kfree() check if the passed pointer doesn't point to .rodata and if so - don't actually destroy the resource.
Signed-off-by Bartosz Golaszewski <[email protected]>
Reviewed-by Bjorn Andersson <[email protected]>
Acked-by Mike Rapoport <[email protected]>
Acked-by Rasmus Villemoes <[email protected]>
Reviewed-by Geert Uytterhoeven <[email protected]>
Reviewed-by Andy Shevchenko <[email protected]>
---
 drivers/base/devres.c  | 31 +++++++++++++++++++++++++++++++
 include/linux/device.h |  2 ++
 2 files changed, 33 insertions(+)
drivers/base/devres.c +13 -0
index 438c91a43508..00c70f0fcdcd 100644
@@ -11,6 +11,8 @@
 #include <linux/slab.h>
 #include <linux/percpu.h>
 
+#include <asm/sections.h>
+
 #include "base.h"
 
 struct devres_node {
@@ -822,6 +824,28 @@ char *devm_kstrdup(struct device *dev, const char *s, gfp_t gfp)
 }
 EXPORT_SYMBOL_GPL(devm_kstrdup);
 
+/**
+ * devm_kstrdup_const - resource managed conditional string duplication
+ * @dev: device for which to duplicate the string
+ * @s: the string to duplicate
+ * @gfp: the GFP mask used in the kmalloc() call when allocating memory
+ *
+ * Strings allocated by devm_kstrdup_const will be automatically freed when
+ * the associated device is detached.
+ *
+ * RETURNS:
+ * Source string if it is in .rodata section otherwise it falls back to

drivers/base/devres.c char *devm_kstrdup(struct device *dev, const char *s, gfp_t gfp) · line 847

Quoted 10 lines

drivers/base/devres.c +10 -0

+ * devm_kstrdup.
+ */
+const char *devm_kstrdup_const(struct device *dev, const char *s, gfp_t gfp)
+{
+	if (is_kernel_rodata((unsigned long)s))
+		return s;
+
+	return devm_kstrdup(dev, s, gfp);
+}
+EXPORT_SYMBOL(devm_kstrdup_const);

This should be EXPORT_SYMBOL_GPL() to match the rest of the file. I'll go fix that up when I apply it.

thanks,

greg k-h

Bartosz Golaszewski

author

2018-10-14 15:20 UTC #7 lore raw

PATCH RESEND v6 4/4 mailbox: tegra-hsp: use devm_kstrdup_const()

In reply to Bartosz Golaszewski

Use devm_kstrdup_const() in the tegra-hsp driver. This mostly serves as an example of how to use this new routine to shrink driver code.

Also use devm_kzalloc() instead of regular kzalloc() to get shrink the driver even more.

Doorbell objects are only removed in the driver's remove callback so it's safe to convert all memory allocations to devres.

Signed-off-by Bartosz Golaszewski <[email protected]>

---
 drivers/mailbox/tegra-hsp.c | 41 ++++++++-----------------------------
 1 file changed, 9 insertions(+), 32 deletions(-)

drivers/mailbox/tegra-hsp.c +9 ~~-32~~

index 0cde356c11ab..106c94dedbf1 100644
@@ -183,14 +183,15 @@ static irqreturn_t tegra_hsp_doorbell_irq(int irq, void *data)
 }
 
 static struct tegra_hsp_channel *
-tegra_hsp_doorbell_create(struct tegra_hsp *hsp, const char *name,
-			  unsigned int master, unsigned int index)
+tegra_hsp_doorbell_create(struct device *dev, struct tegra_hsp *hsp,
+			  const char *name, unsigned int master,
+			  unsigned int index)
 {
 	struct tegra_hsp_doorbell *db;
 	unsigned int offset;
 	unsigned long flags;
 
-	db = kzalloc(sizeof(*db), GFP_KERNEL);
+	db = devm_kzalloc(dev, sizeof(*db), GFP_KERNEL);
 	if (!db)
 		return ERR_PTR(-ENOMEM);

@@ -200,7 +201,7 @@ tegra_hsp_doorbell_create(struct tegra_hsp *hsp, const char *name,
 	db->channel.regs = hsp->regs + offset;
 	db->channel.hsp = hsp;
 
-	db->name = kstrdup_const(name, GFP_KERNEL);
+	db->name = devm_kstrdup_const(dev, name, GFP_KERNEL);
 	db->master = master;
 	db->index = index;

@@ -211,13 +212,6 @@ tegra_hsp_doorbell_create(struct tegra_hsp *hsp, const char *name,
 	return &db->channel;
 }
 
-static void __tegra_hsp_doorbell_destroy(struct tegra_hsp_doorbell *db)
-{
-	list_del(&db->list);
-	kfree_const(db->name);
-	kfree(db);
-}
-
 static int tegra_hsp_doorbell_send_data(struct mbox_chan *chan, void *data)
 {
 	struct tegra_hsp_doorbell *db = chan->con_priv;

@@ -332,31 +326,16 @@ static struct mbox_chan *of_tegra_hsp_xlate(struct mbox_controller *mbox,
 	return chan ?: ERR_PTR(-EBUSY);
 }
 
-static void tegra_hsp_remove_doorbells(struct tegra_hsp *hsp)
-{
-	struct tegra_hsp_doorbell *db, *tmp;
-	unsigned long flags;
-
-	spin_lock_irqsave(&hsp->lock, flags);
-
-	list_for_each_entry_safe(db, tmp, &hsp->doorbells, list)
-		__tegra_hsp_doorbell_destroy(db);
-
-	spin_unlock_irqrestore(&hsp->lock, flags);
-}
-
-static int tegra_hsp_add_doorbells(struct tegra_hsp *hsp)
+static int tegra_hsp_add_doorbells(struct device *dev, struct tegra_hsp *hsp)
 {
 	const struct tegra_hsp_db_map *map = hsp->soc->map;
 	struct tegra_hsp_channel *channel;
 
 	while (map->name) {
-		channel = tegra_hsp_doorbell_create(hsp, map->name,
+		channel = tegra_hsp_doorbell_create(dev, hsp, map->name,
 						    map->master, map->index);
-		if (IS_ERR(channel)) {
-			tegra_hsp_remove_doorbells(hsp);
+		if (IS_ERR(channel))
 			return PTR_ERR(channel);
-		}
 
 		map++;
 	}

@@ -412,7 +391,7 @@ static int tegra_hsp_probe(struct platform_device *pdev)
 	if (!hsp->mbox.chans)
 		return -ENOMEM;
 
-	err = tegra_hsp_add_doorbells(hsp);
+	err = tegra_hsp_add_doorbells(&pdev->dev, hsp);
 	if (err < 0) {
 		dev_err(&pdev->dev, "failed to add doorbells: %d\n", err);
 		return err;

@@ -423,7 +402,6 @@ static int tegra_hsp_probe(struct platform_device *pdev)
 	err = mbox_controller_register(&hsp->mbox);
 	if (err) {
 		dev_err(&pdev->dev, "failed to register mailbox: %d\n", err);
-		tegra_hsp_remove_doorbells(hsp);
 		return err;
 	}

@@ -443,7 +421,6 @@ static int tegra_hsp_remove(struct platform_device *pdev)
 	struct tegra_hsp *hsp = platform_get_drvdata(pdev);
 
 	mbox_controller_unregister(&hsp->mbox);
-	tegra_hsp_remove_doorbells(hsp);
 
 	return 0;
 }

Signature · 1 line

2.19.1

Bartosz Golaszewski

author

2018-10-14 15:21 UTC #8 lore raw

In reply to Bartosz Golaszewski

niedz., 14 paź 2018 o 17:20 Bartosz Golaszewski <[email protected]> napisał(a):

Quoted 6 lines

Hi Greg,
this has been reviewed a lot and three first patches can be picked up. Please take it through your tree.

Except for patch 4/4 of course - it needs Thierry's Ack, but that can wait until the first three are merged.

Bart

Quoted 42 lines

This series implements devm_kstrdup_const() together with some prerequisite changes and uses it in tegra-hsp driver.
v1 -> v2:
- fixed the changelog in the patch implementing devm_kstrdup_const()
- fixed the kernel doc
- moved is_kernel_rodata() to asm-generic/sections.h
- fixed constness
v2 -> v3:
- rebased on top of 4.19-rc5 as there were some conflicts in the pmc-atom driver
- collected Reviewed-by tags
v3 -> v4:
- Andy NAK'ed patch 4/4 so I added a different example
- collected more tags
v4 -> v5:
- instead of providing devm_kfree_const(), make devm_kfree() check if given pointer is not in .rodata and act accordingly
v5 -> v6:
- fixed the commit message in patch 2/4 (s/devm_kfree_const/devm_kfree/)
- collected even more tags
Bartosz Golaszewski (4):
devres: constify p in devm_kfree()
mm: move is_kernel_rodata() to asm-generic/sections.h
devres: provide devm_kstrdup_const()
mailbox: tegra-hsp: use devm_kstrdup_const()
 drivers/base/devres.c          | 36 +++++++++++++++++++++++++++--
 drivers/mailbox/tegra-hsp.c    | 41 ++++++++--------------------------
 include/asm-generic/sections.h | 14 ++++++++++++
 include/linux/device.h         |  4 +++-
 mm/util.c                      |  7 ------
 5 files changed, 60 insertions(+), 42 deletions(-)
--
2.19.1

Greg Kroah-Hartman

maintainer

2018-10-16 11:33 UTC #9 lore raw

Quoted 10 lines · Bartosz Golaszewski

niedz., 14 paź 2018 o 17:20 Bartosz Golaszewski <[email protected]> napisał(a):
Quoted 6 lines
Hi Greg,
this has been reviewed a lot and three first patches can be picked up. Please take it through your tree.
Except for patch 4/4 of course - it needs Thierry's Ack, but that can wait until the first three are merged.

first 3 are merged now.

greg k-h