Skip to contentSkip to conversation

OpenRISC fixes for 5.9

PATCH Superseded Stafford Horne 3 participants 11 messages Sep 5, 2020 – Sep 6, 2020 Openrisc architecture

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Stafford Horne

author maintainer
#0 lore raw

v2 cover OpenRISC fixes for 5.9

Changes since v1:

- Now a series, v1 was only the "Reserve memblock for initrd" patch
- Added fixes for compiler issues pointed out by the kbuild robot

This is a few fixes found during testing 5.9.

Stafford Horne (3):
openrisc: Reserve memblock for initrd
openrisc: Fix cache API compile issue when not inlining
openrisc: Fix issue with get_user for 64-bit values

 arch/openrisc/include/asm/uaccess.h | 16 ++++++++--------
 arch/openrisc/kernel/setup.c        | 10 ++++++++++
 arch/openrisc/mm/cache.c            |  2 +-
 3 files changed, 19 insertions(+), 9 deletions(-)
Signature · 1 line
2.26.2
author maintainer
#1 lore raw

PATCH v2 1/3 openrisc: Reserve memblock for initrd

Recently OpenRISC added support for external initrd images, but I found some instability when using larger buildroot initrd images. It turned out that I forgot to reserve the memblock space for the initrd image.

This patch fixes the instability issue by reserving memblock space.

Fixes ff6c923dbec3 ("openrisc: Add support for external initrd images")
---
 arch/openrisc/kernel/setup.c | 10 ++++++++++
 1 file changed, 10 insertions(+)
arch/openrisc/kernel/setup.c +10 -0
index b18e775f8be3..13c87f1f872b 100644
@@ -80,6 +80,16 @@ static void __init setup_memory(void)
 	 */
 	memblock_reserve(__pa(_stext), _end - _stext);
 
+#ifdef CONFIG_BLK_DEV_INITRD
+	/* Then reserve the initrd, if any */
+	if (initrd_start && (initrd_end > initrd_start)) {
+		unsigned long aligned_start = ALIGN_DOWN(initrd_start, PAGE_SIZE);
+		unsigned long aligned_end = ALIGN(initrd_end, PAGE_SIZE);
+
+		memblock_reserve(__pa(aligned_start), aligned_end - aligned_start);
+	}
+#endif /* CONFIG_BLK_DEV_INITRD */
+
 	early_init_fdt_reserve_self();
 	early_init_fdt_scan_reserved_mem();
 
Signature · 1 line
2.26.2
author maintainer
#2 lore raw

Re: [PATCH v2 1/3] openrisc: Reserve memblock for initrd

Quoted 9 lines · Stafford Horne

Recently OpenRISC added support for external initrd images, but I found some instability when using larger buildroot initrd images. It turned out that I forgot to reserve the memblock space for the initrd image.

This patch fixes the instability issue by reserving memblock space.

Fixes ff6c923dbec3 ("openrisc: Add support for external initrd images")

---

Forgot to mention:

Changes since v1:
- Updated to use separate variables as suggested by Mike.

Quoted 27 lines
 arch/openrisc/kernel/setup.c | 10 ++++++++++
 1 file changed, 10 insertions(+)
arch/openrisc/kernel/setup.c +10 -0
index b18e775f8be3..13c87f1f872b 100644
@@ -80,6 +80,16 @@ static void __init setup_memory(void)
 	 */
 	memblock_reserve(__pa(_stext), _end - _stext);
 
+#ifdef CONFIG_BLK_DEV_INITRD
+	/* Then reserve the initrd, if any */
+	if (initrd_start && (initrd_end > initrd_start)) {
+		unsigned long aligned_start = ALIGN_DOWN(initrd_start, PAGE_SIZE);
+		unsigned long aligned_end = ALIGN(initrd_end, PAGE_SIZE);
+
+		memblock_reserve(__pa(aligned_start), aligned_end - aligned_start);
+	}
+#endif /* CONFIG_BLK_DEV_INITRD */
+
 	early_init_fdt_reserve_self();
 	early_init_fdt_scan_reserved_mem();
 
Signature · 1 line
2.26.2
Reviewed-by on v2 1/3
#3 lore raw

Re: [PATCH v2 1/3] openrisc: Reserve memblock for initrd

In reply to Stafford Horne
Quoted 8 lines · Stafford Horne

Recently OpenRISC added support for external initrd images, but I found some instability when using larger buildroot initrd images. It turned out that I forgot to reserve the memblock space for the initrd image.

This patch fixes the instability issue by reserving memblock space.

Fixes ff6c923dbec3 ("openrisc: Add support for external initrd images")
Quoted 28 lines
---
 arch/openrisc/kernel/setup.c | 10 ++++++++++
 1 file changed, 10 insertions(+)
arch/openrisc/kernel/setup.c +10 -0
index b18e775f8be3..13c87f1f872b 100644
@@ -80,6 +80,16 @@ static void __init setup_memory(void)
 	 */
 	memblock_reserve(__pa(_stext), _end - _stext);
 
+#ifdef CONFIG_BLK_DEV_INITRD
+	/* Then reserve the initrd, if any */
+	if (initrd_start && (initrd_end > initrd_start)) {
+		unsigned long aligned_start = ALIGN_DOWN(initrd_start, PAGE_SIZE);
+		unsigned long aligned_end = ALIGN(initrd_end, PAGE_SIZE);
+
+		memblock_reserve(__pa(aligned_start), aligned_end - aligned_start);
+	}
+#endif /* CONFIG_BLK_DEV_INITRD */
+
 	early_init_fdt_reserve_self();
 	early_init_fdt_scan_reserved_mem();
 
Signature · 1 line
2.26.2
Signature · 2 lines
Sincerely yours,
Mike.
author maintainer
#4 lore raw

PATCH v2 2/3 openrisc: Fix cache API compile issue when not inlining

In reply to Stafford Horne

I found this when compiling a kbuild random config with GCC 11. The config enables CONFIG_DEBUG_SECTION_MISMATCH, which sets CFLAGS -fno-inline-functions-called-once. This causes the call to cache_loop in cache.c to not be inlined causing the below compile error.

    In file included from arch/openrisc/mm/cache.c:13:
    arch/openrisc/mm/cache.c: In function 'cache_loop':
    ./arch/openrisc/include/asm/spr.h:16:27: warning: 'asm' operand 0 probably does not match constraints
       16 | #define mtspr(_spr, _val) __asm__ __volatile__ (  \
	  |                           ^~~~~~~
    arch/openrisc/mm/cache.c:25:3: note: in expansion of macro 'mtspr'
       25 |   mtspr(reg, line);
	  |   ^~~~~
    ./arch/openrisc/include/asm/spr.h:16:27: error: impossible constraint in 'asm'
       16 | #define mtspr(_spr, _val) __asm__ __volatile__ (  \
	  |                           ^~~~~~~
    arch/openrisc/mm/cache.c:25:3: note: in expansion of macro 'mtspr'
       25 |   mtspr(reg, line);
	  |   ^~~~~
    make[1]: *** [scripts/Makefile.build:283: arch/openrisc/mm/cache.o] Error 1

The asm constraint "K" requires a immediate constant argument to mtspr, however because of no inlining a register argument is passed causing a failure. Fix this by using __always_inline.

---
 arch/openrisc/mm/cache.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/openrisc/mm/cache.c +1 -1
index 08f56af387ac..534a52ec5e66 100644
@@ -16,7 +16,7 @@
 #include <asm/cacheflush.h>
 #include <asm/tlbflush.h>
 
-static void cache_loop(struct page *page, const unsigned int reg)
+static __always_inline void cache_loop(struct page *page, const unsigned int reg)
 {
 	unsigned long paddr = page_to_pfn(page) << PAGE_SHIFT;
 	unsigned long line = paddr & ~(L1_CACHE_BYTES - 1);
Signature · 1 line
2.26.2
author maintainer
#5 lore raw

PATCH v2 3/3 openrisc: Fix issue with get_user for 64-bit values

In reply to Stafford Horne

A build failure was raised by kbuild with the following error.

drivers/android/binder.c: Assembler messages:
drivers/android/binder.c:3861: Error: unrecognized keyword/register name l.lwz ?ap,4(r24)' drivers/android/binder.c:3866: Error: unrecognized keyword/register name l.addi ?ap,r0,0'

The issue is with 64-bit get_user() calls on openrisc. I traced this to a problem where in the internally in the get_user macros there is a cast to long __gu_val this causes GCC to think the get_user call is 32-bit. This binder code is really long and GCC allocates register r30, which triggers the issue. The 64-bit get_user asm tries to get the 64-bit pair register, which for r30 overflows the general register names and returns the dummy register ?ap.

The fix is to just remove the assignment to the 32-bit temporary variable __gu_val.

---
 arch/openrisc/include/asm/uaccess.h | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)
arch/openrisc/include/asm/uaccess.h +8 -8
index f0390211236b..4a8976dda1a5 100644
@@ -165,19 +165,19 @@ struct __large_struct {
 
 #define __get_user_nocheck(x, ptr, size)			\
 ({								\
-	long __gu_err, __gu_val;				\
-	__get_user_size(__gu_val, (ptr), (size), __gu_err);	\
-	(x) = (__force __typeof__(*(ptr)))__gu_val;		\
+	long __gu_err;						\
+	__get_user_size((x), (ptr), (size), __gu_err);		\
 	__gu_err;						\
 })
 
 #define __get_user_check(x, ptr, size)					\
 ({									\
-	long __gu_err = -EFAULT, __gu_val = 0;				\
+	long __gu_err = -EFAULT;					\
 	const __typeof__(*(ptr)) __user *__gu_addr = (ptr);		\
-	if (access_ok(__gu_addr, size))			\
-		__get_user_size(__gu_val, __gu_addr, (size), __gu_err);	\
-	(x) = (__force __typeof__(*(ptr)))__gu_val;			\
+	if (access_ok(__gu_addr, size))					\
+		__get_user_size((x), __gu_addr, (size), __gu_err);	\
+	else								\
+		(x) = 0;						\
 	__gu_err;							\
 })
 
@@ -191,7 +191,7 @@ do {									\
 	case 2: __get_user_asm(x, ptr, retval, "l.lhz"); break;		\
 	case 4: __get_user_asm(x, ptr, retval, "l.lwz"); break;		\
 	case 8: __get_user_asm2(x, ptr, retval); break;			\
-	default: (x) = __get_user_bad();				\
+	default: (x) = (__typeof__(x)) __get_user_bad();		\
 	}								\
 } while (0)
 
Signature · 1 line
2.26.2
#6 lore raw

Re: [PATCH v2 3/3] openrisc: Fix issue with get_user for 64-bit values

On Sat, Sep 05, 2020 at 10:19:35PM +0900, Stafford Horne wrote:

Hi,

The change for 64-bit get_user() looks good to me.
But I wonder, given that openrisc is big-endian, what will happen you have the opposite situation:

	u32 *ptr;
	u64 val;
	...
	get_user(val, ptr);

Won't you end with the value in the most significant part of the register pair?

-- Luc

author maintainer
#7 lore raw

Re: [PATCH v2 3/3] openrisc: Fix issue with get_user for 64-bit values

Quoted 7 lines · Luc Van Oostenryck

On Sat, Sep 05, 2020 at 10:19:35PM +0900, Stafford Horne wrote:

Hi,

The change for 64-bit get_user() looks good to me.
But I wonder, given that openrisc is big-endian, what will happen you have the opposite situation:

Quoted 7 lines
	u32 *ptr;
	u64 val;
	...
	get_user(val, ptr);

Won't you end with the value in the most significant part of the register pair?

Hi Luc,

The get_user function uses the size of the ptr to determine how to do the load , so this case would not use the 64-bit pair register logic. I think it should be ok, the end result would be the same as c code:

  var = *ptr;

-Stafford

#8 lore raw

Re: [OpenRISC] [PATCH v2 3/3] openrisc: Fix issue with get_user for 64-bit values

Quoted 16 lines · Stafford Horne
Quoted 7 lines · Luc Van Oostenryck

On Sat, Sep 05, 2020 at 10:19:35PM +0900, Stafford Horne wrote:

Hi,

The change for 64-bit get_user() looks good to me.
But I wonder, given that openrisc is big-endian, what will happen you have the opposite situation:

Quoted 7 lines
	u32 *ptr;
	u64 val;
	...
	get_user(val, ptr);

Won't you end with the value in the most significant part of the register pair?

Quoted 7 lines

Hi Luc,

The get_user function uses the size of the ptr to determine how to do the load , so this case would not use the 64-bit pair register logic. I think it should be ok, the end result would be the same as c code:

  var = *ptr;

Hi,

Sorry to insist but both won't give the same result. The problem comes from the output part of the asm: "=r" (x).

The following code:

	u32 getp(u32 *ptr)
	{
		u64 val;
		val = *ptr;
		return val;
	}
will compile to something like:
	getp:
		l.jr	r9
		l.lwz	r11, 0(r3)

The load is written to r11, which is what is returned. OK.

But the get_user() code with a u32 pointer *and* a u64 destination is equivalent to something like:

	u32 getl(u32 *ptr)
	{
		u64 val;

		asm("l.lwz %0,0(%1)" : "=r"(val) : "r"(ptr));
		return val;
	}
and this compiles to:
	getl:
		l.lwz	r17,0(r3)
		l.jr	r9
		l.or	r11, r19, r19

The load is written to r17 but what is returned is the content of r19. Not good.

I think that, in the get_user() code:
* if the pointer is a pointer to a 64-bit quantity, then variable used in as the output in the asm needs to be a 64-bit variable
* if the pointer is a pointer to a 32-bit quantity, then variable used in as the output in the asm needs to be a 64-bit variable
At least one way to guarantee this is to use a temporary variable that matches the size of the pointer.

-- Luc

author maintainer
#9 lore raw

Re: [OpenRISC] [PATCH v2 3/3] openrisc: Fix issue with get_user for 64-bit values

Quoted 62 lines · Luc Van Oostenryck
Quoted 16 lines · Stafford Horne
Quoted 7 lines · Luc Van Oostenryck

On Sat, Sep 05, 2020 at 10:19:35PM +0900, Stafford Horne wrote:

Hi,

The change for 64-bit get_user() looks good to me.
But I wonder, given that openrisc is big-endian, what will happen you have the opposite situation:

Quoted 7 lines
	u32 *ptr;
	u64 val;
	...
	get_user(val, ptr);

Won't you end with the value in the most significant part of the register pair?

Quoted 7 lines

Hi Luc,

The get_user function uses the size of the ptr to determine how to do the load , so this case would not use the 64-bit pair register logic. I think it should be ok, the end result would be the same as c code:

  var = *ptr;

Hi,

Sorry to insist but both won't give the same result. The problem comes from the output part of the asm: "=r" (x).

The following code:

	u32 getp(u32 *ptr)
	{
		u64 val;
		val = *ptr;
		return val;
	}
will compile to something like:
	getp:
		l.jr	r9
		l.lwz	r11, 0(r3)

The load is written to r11, which is what is returned. OK.

But the get_user() code with a u32 pointer *and* a u64 destination is equivalent to something like:

	u32 getl(u32 *ptr)
	{
		u64 val;

		asm("l.lwz %0,0(%1)" : "=r"(val) : "r"(ptr));
		return val;
	}
and this compiles to:
	getl:
		l.lwz	r17,0(r3)
		l.jr	r9
		l.or	r11, r19, r19

The load is written to r17 but what is returned is the content of r19. Not good.

Quoted 7 lines

I think that, in the get_user() code:
* if the pointer is a pointer to a 64-bit quantity, then variable used in as the output in the asm needs to be a 64-bit variable
* if the pointer is a pointer to a 32-bit quantity, then variable used in as the output in the asm needs to be a 64-bit variable
At least one way to guarantee this is to use a temporary variable that matches the size of the pointer.

Hello,

Thanks for taking the time to explain. I see your point, it makes sense I will fix this up.

-Stafford

author maintainer
#10 lore raw
In reply to Stafford Horne
Quoted 4 lines · Stafford Horne

Changes since v1:

- Now a series, v1 was only the "Reserve memblock for initrd" patch
- Added fixes for compiler issues pointed out by the kbuild robot

Forgot to mention:
- Updated "Reserve memblock for initrd", as per Mike's comments

Quoted 15 lines

This is a few fixes found during testing 5.9.

Stafford Horne (3):
openrisc: Reserve memblock for initrd
openrisc: Fix cache API compile issue when not inlining
openrisc: Fix issue with get_user for 64-bit values

 arch/openrisc/include/asm/uaccess.h | 16 ++++++++--------
 arch/openrisc/kernel/setup.c        | 10 ++++++++++
 arch/openrisc/mm/cache.c            |  2 +-
 3 files changed, 19 insertions(+), 9 deletions(-)
Signature · 1 line
2.26.2