[PATCH 0/5] metag: second set of misc patches for v3.9

[PATCH 0/5] metag: second set of misc patches for v3.9

[James Hogan]

Here is the second set of miscellaneous metag patches that I'll be
adding to the tip of the metag tree for v3.9.

James Hogan (5):
  metag: export metag_code_cache_flush_all
  metag: export clear_page and copy_page
  metag: hugetlb: convert to vm_unmapped_area()
  genksyms: fix metag symbol prefix on crc symbols
  metag: move irq enable out of irqflags.h on SMP

 arch/metag/include/asm/irqflags.h |   11 +++----
 arch/metag/kernel/metag_ksyms.c   |    2 +
 arch/metag/kernel/traps.c         |   15 ++++++++++-
 arch/metag/mm/cache.c             |    1 +
 arch/metag/mm/hugetlbpage.c       |   52 +++++++-----------------------------
 scripts/genksyms/genksyms.c       |    3 +-
 6 files changed, 34 insertions(+), 50 deletions(-)

-- 
1.7.7.6

[PATCH 1/5] metag: export metag_code_cache_flush_all

[James Hogan]

Various file systems indirectly use metag_code_cache_flush_all(), so
when they're built as modules we get build errors like the following:

ERROR: "metag_code_cache_flush_all" [fs/xfs/xfs.ko] undefined!

Therefore export this function to modules to fix the errors. This was
hit by a randconfig build.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
---
 arch/metag/mm/cache.c |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/arch/metag/mm/cache.c b/arch/metag/mm/cache.c
index 4dd96e4..b5d3b2e 100644
--- a/arch/metag/mm/cache.c
+++ b/arch/metag/mm/cache.c
@@ -460,6 +460,7 @@ void metag_code_cache_flush_all(const void *start)
 
 	metag_phys_code_cache_flush(start, 4096);
 }
+EXPORT_SYMBOL(metag_code_cache_flush_all);
 
 void metag_code_cache_flush(const void *start, int bytes)
 {
-- 
1.7.7.6

[PATCH 2/5] metag: export clear_page and copy_page

[James Hogan]

Various file systems use clear_page() and copy_page(), so when they're
built as modules we get build errors like the following:

ERROR: "clear_page" [fs/ntfs/ntfs.ko] undefined!
ERROR: "copy_page" [fs/nilfs2/nilfs2.ko] undefined!

Therefore export these functions to modules from metag_ksyms.c to fix
the errors. This was hit by a randconfig build.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
---
 arch/metag/kernel/metag_ksyms.c |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)

diff --git a/arch/metag/kernel/metag_ksyms.c b/arch/metag/kernel/metag_ksyms.c
index 5381514..1b27bea 100644
--- a/arch/metag/kernel/metag_ksyms.c
+++ b/arch/metag/kernel/metag_ksyms.c
@@ -37,6 +37,8 @@ EXPORT_SYMBOL(get_trigger_mask);
 EXPORT_SYMBOL(global_trigger_mask);
 #endif
 
+EXPORT_SYMBOL(clear_page);
+EXPORT_SYMBOL(copy_page);
 EXPORT_SYMBOL(empty_zero_page);
 
 EXPORT_SYMBOL(pfn_base);
-- 
1.7.7.6

[PATCH 3/5] metag: hugetlb: convert to vm_unmapped_area()

[James Hogan]

Convert hugetlb_get_unmapped_area_new_pmd() to use vm_unmapped_area()
rather than searching the virtual address space itself. This fixes the
following errors in linux-next due to the specified members being
removed after other architectures have already been converted:

arch/metag/mm/hugetlbpage.c: In function 'hugetlb_get_unmapped_area_new_pmd':
arch/metag/mm/hugetlbpage.c:199: error: 'struct mm_struct' has no member named 'cached_hole_size'
arch/metag/mm/hugetlbpage.c:200: error: 'struct mm_struct' has no member named 'free_area_cache'
arch/metag/mm/hugetlbpage.c:215: error: 'struct mm_struct' has no member named 'cached_hole_size'

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Michel Lespinasse <walken@google.com>
---
 arch/metag/mm/hugetlbpage.c |   52 ++++++++----------------------------------
 1 files changed, 10 insertions(+), 42 deletions(-)

diff --git a/arch/metag/mm/hugetlbpage.c b/arch/metag/mm/hugetlbpage.c
index 24ceed4..3c52fa6 100644
--- a/arch/metag/mm/hugetlbpage.c
+++ b/arch/metag/mm/hugetlbpage.c
@@ -192,43 +192,15 @@ new_search:
 static unsigned long
 hugetlb_get_unmapped_area_new_pmd(unsigned long len)
 {
-	struct mm_struct *mm = current->mm;
-	struct vm_area_struct *vma;
-	unsigned long start_addr, addr;
-
-	if (ALIGN_HUGEPT(len) > mm->cached_hole_size)
-		start_addr = mm->free_area_cache;
-	else
-		start_addr = TASK_UNMAPPED_BASE;
-
-new_search:
-	addr = ALIGN_HUGEPT(start_addr);
-
-	for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
-		if (TASK_SIZE - len < addr) {
-			/*
-			 * Start a new search - just in case we missed
-			 * some holes.
-			 */
-			if (start_addr != TASK_UNMAPPED_BASE) {
-				start_addr = TASK_UNMAPPED_BASE;
-				mm->cached_hole_size = 0;
-				goto new_search;
-			}
-			return 0;
-		}
-		/* skip ahead if we've aligned right over some vmas */
-		if (vma && vma->vm_end <= addr)
-			continue;
-		if (!vma || ALIGN_HUGEPT(addr + len) <= vma->vm_start) {
-#if HPAGE_SHIFT < HUGEPT_SHIFT
-			if (len & HUGEPT_MASK)
-				mm->context.part_huge = addr + len;
-#endif
-			return addr;
-		}
-		addr = ALIGN_HUGEPT(vma->vm_end);
-	}
+	struct vm_unmapped_area_info info;
+
+	info.flags = 0;
+	info.length = len;
+	info.low_limit = TASK_UNMAPPED_BASE;
+	info.high_limit = TASK_SIZE;
+	info.align_mask = PAGE_MASK & HUGEPT_MASK;
+	info.align_offset = 0;
+	return vm_unmapped_area(&info);
 }
 
 unsigned long
@@ -266,11 +238,7 @@ hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 	 * Find an unmapped naturally aligned set of 4MB blocks that we can use
 	 * for huge pages.
 	 */
-	addr = hugetlb_get_unmapped_area_new_pmd(len);
-	if (likely(addr))
-		return addr;
-
-	return -EINVAL;
+	return hugetlb_get_unmapped_area_new_pmd(len);
 }
 
 #endif /*HAVE_ARCH_HUGETLB_UNMAPPED_AREA*/
-- 
1.7.7.6

[PATCH 4/5] genksyms: fix metag symbol prefix on crc symbols

[James Hogan]

Meta uses symbol prefixes, so add "metag" to the list of architectures
to set the mod_prefix to "_" for. This fixes __crc_* symbols to add the
extra underscore to match _CRC_SYMBOL macro in <linux/export.h> and so
that modpost finds them.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
---
 scripts/genksyms/genksyms.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)

diff --git a/scripts/genksyms/genksyms.c b/scripts/genksyms/genksyms.c
index 8a10649..d25e4a1 100644
--- a/scripts/genksyms/genksyms.c
+++ b/scripts/genksyms/genksyms.c
@@ -826,7 +826,8 @@ int main(int argc, char **argv)
 			genksyms_usage();
 			return 1;
 		}
-	if ((strcmp(arch, "h8300") == 0) || (strcmp(arch, "blackfin") == 0))
+	if ((strcmp(arch, "h8300") == 0) || (strcmp(arch, "blackfin") == 0) ||
+	    (strcmp(arch, "metag") == 0))
 		mod_prefix = "_";
 	{
 		extern int yydebug;
-- 
1.7.7.6

[PATCH 5/5] metag: move irq enable out of irqflags.h on SMP

[James Hogan]

The SMP version of arch_local_irq_enable() uses preempt_disable(), but
<asm/irqflags.h> doesn't include <linux/preempt.h> causing the following
errors on SMP when pstore/ftrace is enabled (caught by buildbot smp
allyesconfig):

In file included from include/linux/irqflags.h:15,
                 from fs/pstore/ftrace.c:16:
arch/metag/include/asm/irqflags.h: In function 'arch_local_irq_enable':
arch/metag/include/asm/irqflags.h:84: error: implicit declaration of function 'preempt_disable'
arch/metag/include/asm/irqflags.h:86: error: implicit declaration of function 'preempt_enable_no_resched'

However <linux/preempt.h> cannot be easily included from
<asm/irqflags.h> as it can cause circular include dependencies in the
!SMP case, and potentially in the SMP case in the future. Therefore move
the SMP implementation of arch_local_irq_enable() into traps.c and use
an inline version of get_trigger_mask() which is also defined in traps.c
for SMP.

This adds an extra layer of function call / stack push when
preempt_disable needs to call other functions, however in the
non-preemptive SMP case it should be about as fast, as it was already
calling the get_trigger_mask() function which is now used inline.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
---
 arch/metag/include/asm/irqflags.h |   11 +++++------
 arch/metag/kernel/traps.c         |   15 ++++++++++++++-
 2 files changed, 19 insertions(+), 7 deletions(-)

diff --git a/arch/metag/include/asm/irqflags.h b/arch/metag/include/asm/irqflags.h
index cba5e13..339b16f 100644
--- a/arch/metag/include/asm/irqflags.h
+++ b/arch/metag/include/asm/irqflags.h
@@ -78,16 +78,15 @@ static inline void arch_local_irq_disable(void)
 	asm volatile("MOV TXMASKI,%0\n" : : "r" (flags) : "memory");
 }
 
-static inline void arch_local_irq_enable(void)
-{
 #ifdef CONFIG_SMP
-	preempt_disable();
-	arch_local_irq_restore(get_trigger_mask());
-	preempt_enable_no_resched();
+/* Avoid circular include dependencies through <linux/preempt.h> */
+void arch_local_irq_enable(void);
 #else
+static inline void arch_local_irq_enable(void)
+{
 	arch_local_irq_restore(get_trigger_mask());
-#endif
 }
+#endif
 
 #endif /* (__ASSEMBLY__) */
 
diff --git a/arch/metag/kernel/traps.c b/arch/metag/kernel/traps.c
index 2808042..3f7764b 100644
--- a/arch/metag/kernel/traps.c
+++ b/arch/metag/kernel/traps.c
@@ -15,6 +15,7 @@
 #include <linux/types.h>
 #include <linux/init.h>
 #include <linux/interrupt.h>
+#include <linux/preempt.h>
 #include <linux/ptrace.h>
 #include <linux/module.h>
 #include <linux/kallsyms.h>
@@ -776,17 +777,29 @@ int traps_restore_context(void)
 #endif
 
 #ifdef CONFIG_SMP
-unsigned int get_trigger_mask(void)
+static inline unsigned int _get_trigger_mask(void)
 {
 	unsigned long cpu = smp_processor_id();
 	return per_cpu(trigger_mask, cpu);
 }
 
+unsigned int get_trigger_mask(void)
+{
+	return _get_trigger_mask();
+}
+
 static void set_trigger_mask(unsigned int mask)
 {
 	unsigned long cpu = smp_processor_id();
 	per_cpu(trigger_mask, cpu) = mask;
 }
+
+void arch_local_irq_enable(void)
+{
+	preempt_disable();
+	arch_local_irq_restore(_get_trigger_mask());
+	preempt_enable_no_resched();
+}
 #else
 static void set_trigger_mask(unsigned int mask)
 {
-- 
1.7.7.6

Re: [PATCH 3/5] metag: hugetlb: convert to vm_unmapped_area()

[Michel Lespinasse]

On Wed, Feb 13, 2013 at 4:03 AM, James Hogan <james.hogan@imgtec.com> wrote:
> Convert hugetlb_get_unmapped_area_new_pmd() to use vm_unmapped_area()
> rather than searching the virtual address space itself. This fixes the
> following errors in linux-next due to the specified members being
> removed after other architectures have already been converted:
>
> arch/metag/mm/hugetlbpage.c: In function 'hugetlb_get_unmapped_area_new_pmd':
> arch/metag/mm/hugetlbpage.c:199: error: 'struct mm_struct' has no member named 'cached_hole_size'
> arch/metag/mm/hugetlbpage.c:200: error: 'struct mm_struct' has no member named 'free_area_cache'
> arch/metag/mm/hugetlbpage.c:215: error: 'struct mm_struct' has no member named 'cached_hole_size'
>
> Signed-off-by: James Hogan <james.hogan@imgtec.com>
> Cc: Michel Lespinasse <walken@google.com>

Looks good. Just one thing, which I don't have full context to evaluate:

> +       info.align_mask = PAGE_MASK & HUGEPT_MASK;

This will work only if HUGEPT_MASK == HUGEPT_SIZE - 1, as opposed to
the PAGE_MASK definition which is ~(PAGE_SIZE - 1).
Not sure how HUGEPT_MASK is defined, so please double check that.

Acked-by: Michel Lespinasse <walken@google.com>

-- 
Michel "Walken" Lespinasse
A program is never fully debugged until the last user dies.

Re: [PATCH 3/5] metag: hugetlb: convert to vm_unmapped_area()

[James Hogan]

On 15/02/13 08:04, Michel Lespinasse wrote:
> On Wed, Feb 13, 2013 at 4:03 AM, James Hogan <james.hogan@imgtec.com> wrote:
>> Convert hugetlb_get_unmapped_area_new_pmd() to use vm_unmapped_area()
>> rather than searching the virtual address space itself. This fixes the
>> following errors in linux-next due to the specified members being
>> removed after other architectures have already been converted:
>>
>> arch/metag/mm/hugetlbpage.c: In function 'hugetlb_get_unmapped_area_new_pmd':
>> arch/metag/mm/hugetlbpage.c:199: error: 'struct mm_struct' has no member named 'cached_hole_size'
>> arch/metag/mm/hugetlbpage.c:200: error: 'struct mm_struct' has no member named 'free_area_cache'
>> arch/metag/mm/hugetlbpage.c:215: error: 'struct mm_struct' has no member named 'cached_hole_size'
>>
>> Signed-off-by: James Hogan <james.hogan@imgtec.com>
>> Cc: Michel Lespinasse <walken@google.com>
> 
> Looks good. Just one thing, which I don't have full context to evaluate:
> 
>> +       info.align_mask = PAGE_MASK & HUGEPT_MASK;
> 
> This will work only if HUGEPT_MASK == HUGEPT_SIZE - 1, as opposed to
> the PAGE_MASK definition which is ~(PAGE_SIZE - 1).
> Not sure how HUGEPT_MASK is defined, so please double check that.
> 
> Acked-by: Michel Lespinasse <walken@google.com>
> 

Thanks Michel. HUGEPT_MASK is indeed equal to HUGEPT_SIZE - 1. I
originally did info.align_mask = PAGE_MASK & ~HUGEPT_MASK accidentally
and as you say it just didn't work :)

Cheers
James