Skip to contentSkip to conversation

lib/iomem_copy: fix __iomem casts

PATCH In review Ben Dooks 3 participants 7 messages Jun 22 – Jun 29 Library code

lore
Share & tools
lore

Ben Dooks

author
#0 lore raw

PATCH 1/1 lib/iomem_copy: fix __iomem casts

The iomem_copy.c code discards __iomem address space when using the IS_ALIGNED() macro. It would make more sense to fix this in one place by aing a PTR_ALIGNED_LONG() macro and then doing the necessary casts there before invoking IS_ALIGNED().

As part of this, also force the pointer to an unsigned long as pointers are generally not signed, although there is no warning as yet on treating pointers as signed.

lib/iomem_copy.c:27:26: warning: cast removes address space '__iomem' of expression lib/iomem_copy.c:27:26: warning: cast removes address space '__iomem' of expression lib/iomem_copy.c:64:26: warning: cast removes address space '__iomem' of expression lib/iomem_copy.c:64:26: warning: cast removes address space '__iomem' of expression lib/iomem_copy.c:106:26: warning: cast removes address space '__iomem' of expression lib/iomem_copy.c:106:26: warning: cast removes address space '__iomem' of expression

---
 lib/iomem_copy.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
lib/iomem_copy.c +5 -3
index dec7eaea60e0..a8cd2642e3ee 100644
@@ -9,6 +9,8 @@
 #include <linux/types.h>
 #include <linux/unaligned.h>
 
+#define PTR_ALIGNED_LONG(__ptr) IS_ALIGNED((__force unsigned long)__ptr, sizeof(long))
+
 #ifndef memset_io
 /**
  * memset_io() - Set a range of I/O memory to a constant value
@@ -24,7 +26,7 @@ void memset_io(volatile void __iomem *addr, int val, size_t count)
 
 	qc *= ~0UL / 0xff;
 
-	while (count && !IS_ALIGNED((long)addr, sizeof(long))) {
+	while (count && !PTR_ALIGNED_LONG(addr)) {
 		__raw_writeb(val, addr);
 		addr++;
 		count--;
@@ -61,7 +63,7 @@ EXPORT_SYMBOL(memset_io);
  */
 void memcpy_fromio(void *dst, const volatile void __iomem *src, size_t count)
 {
-	while (count && !IS_ALIGNED((long)src, sizeof(long))) {
+	while (count && !PTR_ALIGNED_LONG(src)) {
 		*(u8 *)dst = __raw_readb(src);
 		src++;
 		dst++;
@@ -103,7 +105,7 @@ EXPORT_SYMBOL(memcpy_fromio);
  */
 void memcpy_toio(volatile void __iomem *dst, const void *src, size_t count)
 {
-	while (count && !IS_ALIGNED((long)dst, sizeof(long))) {
+	while (count && !PTR_ALIGNED_LONG(dst)) {
 		__raw_writeb(*(u8 *)src, dst);
 		src++;
 		dst++;
Signature · 1 line
2.37.2.352.g3c44437643
#1 lore raw
Quoted 10 lines · Ben Dooks

The iomem_copy.c code discards __iomem address space when using the IS_ALIGNED() macro. It would make more sense to fix this in one place by aing a PTR_ALIGNED_LONG() macro and then doing the necessary casts there before invoking IS_ALIGNED().

As part of this, also force the pointer to an unsigned long as pointers are generally not signed, although there is no warning as yet on treating pointers as signed.

+#define PTR_ALIGNED_LONG(__ptr) IS_ALIGNED((__force unsigned long)__ptr, sizeof(long))

Casting to unsigned long is fine (indeed, casting a pointer to long had been very odd in the first place), but... why __force? Casts to unsigned long (de facto uintptr_t) do *not* require __force - they are explicitly allowed, unless you pass -Wcast-from-as in sparse arguments. -Wall does not turn those on; -Wsparse-all would, but kbuild doesn't pass that.

; cat >/tmp/a.c <<'EOF'
#define __iomem __attribute__((noderef, address_space(__iomem)))

static int f(void __iomem *p)

{
	return (unsigned long)p;
}

static int g(void __iomem *p)

{
	return (long)p;
}

EOF
; sparse /tmp/a.c
/tmp/a.c:10:17: warning: cast removes address space '__iomem' of expression ; sparse -Wcast-from-as /tmp/a.c
/tmp/a.c:5:17: warning: cast removes address space '__iomem' of expression /tmp/a.c:10:17: warning: cast removes address space '__iomem' of expression ; sparse -Wall /tmp/a.c
/tmp/a.c:10:17: warning: cast removes address space '__iomem' of expression ; sparse -Wsparse-all /tmp/a.c
/tmp/a.c:5:17: warning: cast removes address space '__iomem' of expression /tmp/a.c:10:17: warning: cast removes address space '__iomem' of expression

#2 lore raw
Quoted 10 lines · Al Viro
Quoted 8 lines · Ben Dooks

The iomem_copy.c code discards __iomem address space when using the IS_ALIGNED() macro. It would make more sense to fix this in one place by aing a PTR_ALIGNED_LONG() macro and then doing the necessary casts there before invoking IS_ALIGNED().

As part of this, also force the pointer to an unsigned long as pointers are generally not signed, although there is no warning as yet on treating pointers as signed.

Quoted 7 lines
Quoted 1 line

+#define PTR_ALIGNED_LONG(__ptr) IS_ALIGNED((__force unsigned long)__ptr, sizeof(long))

Casting to unsigned long is fine (indeed, casting a pointer to long had been very odd in the first place), but... why __force? Casts to unsigned long (de facto uintptr_t) do *not* require __force - they are explicitly allowed, unless you pass -Wcast-from-as in sparse arguments. -Wall does not turn those on; -Wsparse-all would, but kbuild doesn't pass that.

FWIW, we have 170+ places where IS_ALIGNED((unsigned long)pointer, _) or IS_ALIGNED((uintptr_t)pointer, _) is used in the tree...

#3 lore raw
Quoted 13 lines · On Wed, 24 Jun 2026 01:04:36 +0100 Al Viro <[email protected]> wrote:
Quoted 12 lines · Al Viro
Quoted 10 lines · Ben Dooks

The iomem_copy.c code discards __iomem address space when using the IS_ALIGNED() macro. It would make more sense to fix this in one place by aing a PTR_ALIGNED_LONG() macro and then doing the necessary casts there before invoking IS_ALIGNED().

As part of this, also force the pointer to an unsigned long as pointers are generally not signed, although there is no warning as yet on treating pointers as signed.

+#define PTR_ALIGNED_LONG(__ptr) IS_ALIGNED((__force unsigned long)__ptr, sizeof(long))

Quoted 9 lines
Quoted 5 lines

Casting to unsigned long is fine (indeed, casting a pointer to long had been very odd in the first place), but... why __force? Casts to unsigned long (de facto uintptr_t) do *not* require __force - they are explicitly allowed, unless you pass -Wcast-from-as in sparse arguments. -Wall does not turn those on; -Wsparse-all would, but kbuild doesn't pass that.

FWIW, we have 170+ places where IS_ALIGNED((unsigned long)pointer, _) or IS_ALIGNED((uintptr_t)pointer, _) is used in the tree...

Would is be better to add an IS_PTR_UNALIGNED(ptr, type) that returns

	(unsigned long)(ptr) & (sizeof(type) - 1)

since I suspect quite a few of those places do fixups and need the offset?

David

author
#4 lore raw
In reply to Al Viro
Quoted 10 lines · Al Viro
Quoted 8 lines · Ben Dooks

The iomem_copy.c code discards __iomem address space when using the IS_ALIGNED() macro. It would make more sense to fix this in one place by aing a PTR_ALIGNED_LONG() macro and then doing the necessary casts there before invoking IS_ALIGNED().

As part of this, also force the pointer to an unsigned long as pointers are generally not signed, although there is no warning as yet on treating pointers as signed.

Quoted 7 lines
Quoted 1 line

+#define PTR_ALIGNED_LONG(__ptr) IS_ALIGNED((__force unsigned long)__ptr, sizeof(long))

Casting to unsigned long is fine (indeed, casting a pointer to long had been very odd in the first place), but... why __force? Casts to unsigned long (de facto uintptr_t) do *not* require __force - they are explicitly allowed, unless you pass -Wcast-from-as in sparse arguments. -Wall does not turn those on; -Wsparse-all would, but kbuild doesn't pass that.

Trying to get rid of the address space warning, which requires the __force part.

Quoted 24 lines

; cat >/tmp/a.c <<'EOF'
#define __iomem __attribute__((noderef, address_space(__iomem)))

static int f(void __iomem *p)

{
	return (unsigned long)p;
}

static int g(void __iomem *p)

{
	return (long)p;
}

EOF
; sparse /tmp/a.c
/tmp/a.c:10:17: warning: cast removes address space '__iomem' of expression ; sparse -Wcast-from-as /tmp/a.c
/tmp/a.c:5:17: warning: cast removes address space '__iomem' of expression /tmp/a.c:10:17: warning: cast removes address space '__iomem' of expression ; sparse -Wall /tmp/a.c
/tmp/a.c:10:17: warning: cast removes address space '__iomem' of expression ; sparse -Wsparse-all /tmp/a.c
/tmp/a.c:5:17: warning: cast removes address space '__iomem' of expression /tmp/a.c:10:17: warning: cast removes address space '__iomem' of expression

Signature · 4 lines
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius

https://www.codethink.co.uk/privacy.html
#5 lore raw
Quoted 13 lines · Ben Dooks
Quoted 12 lines · Al Viro
Quoted 10 lines · Ben Dooks

The iomem_copy.c code discards __iomem address space when using the IS_ALIGNED() macro. It would make more sense to fix this in one place by aing a PTR_ALIGNED_LONG() macro and then doing the necessary casts there before invoking IS_ALIGNED().

As part of this, also force the pointer to an unsigned long as pointers are generally not signed, although there is no warning as yet on treating pointers as signed.

+#define PTR_ALIGNED_LONG(__ptr) IS_ALIGNED((__force unsigned long)__ptr, sizeof(long))

Quoted 8 lines
Quoted 5 lines

Casting to unsigned long is fine (indeed, casting a pointer to long had been very odd in the first place), but... why __force? Casts to unsigned long (de facto uintptr_t) do *not* require __force - they are explicitly allowed, unless you pass -Wcast-from-as in sparse arguments. -Wall does not turn those on; -Wsparse-all would, but kbuild doesn't pass that.

Trying to get rid of the address space warning, which requires the __force part.

It does not require __force - not with the type we are casting to being unsigned long. As the examples you've quoted demonstrate. See evaluate_cast() in sparse evaluate.c:

        if ((ttype == &ulong_ctype || ttype == uintptr_ctype) && !Wcast_from_as)
		tas = &bad_address_space;
with
        if (!tas && valid_as(sas))
		warning(expr->pos, "cast removes address space '%s' of expression", show_as(sas));

several lines below.

And yes, it is deliberate.

author
#6 lore raw
Quoted 26 lines · Al Viro
Quoted 13 lines · Ben Dooks
Quoted 12 lines · Al Viro
Quoted 10 lines · Ben Dooks

The iomem_copy.c code discards __iomem address space when using the IS_ALIGNED() macro. It would make more sense to fix this in one place by aing a PTR_ALIGNED_LONG() macro and then doing the necessary casts there before invoking IS_ALIGNED().

As part of this, also force the pointer to an unsigned long as pointers are generally not signed, although there is no warning as yet on treating pointers as signed.

+#define PTR_ALIGNED_LONG(__ptr) IS_ALIGNED((__force unsigned long)__ptr, sizeof(long))

Quoted 8 lines
Quoted 5 lines

Casting to unsigned long is fine (indeed, casting a pointer to long had been very odd in the first place), but... why __force? Casts to unsigned long (de facto uintptr_t) do *not* require __force - they are explicitly allowed, unless you pass -Wcast-from-as in sparse arguments. -Wall does not turn those on; -Wsparse-all would, but kbuild doesn't pass that.

Trying to get rid of the address space warning, which requires the __force part.

It does not require __force - not with the type we are casting to being unsigned long. As the examples you've quoted demonstrate. See evaluate_cast() in sparse evaluate.c:

Quoted 8 lines
         if ((ttype == &ulong_ctype || ttype == uintptr_ctype) && !Wcast_from_as)
		tas = &bad_address_space;
with
         if (!tas && valid_as(sas))
		warning(expr->pos, "cast removes address space '%s' of expression", show_as(sas));

several lines below.

And yes, it is deliberate.

Ah we should also have checked for intptr or long type there, the original code had long which causes sparse to complain here.

Signature · 4 lines
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius

https://www.codethink.co.uk/privacy.html