Skip to contentSkip to conversation

csky: Fixup ioremap function losing

PATCH In review [email protected] 2 participants 5 messages Aug 15, 2019 – Aug 20, 2019

Subsystems (2)
lore
Share & tools
lore

[email protected]

author maintainer
#0 lore raw

PATCH 1/1 csky: Fixup ioremap function losing

From: Guo Ren <[email protected]>

Implement the following apis to meet usage in different scenarios.

 - ioremap          (NonCache + StrongOrder)
 - ioremap_nocache  (NonCache + StrongOrder)
 - ioremap_wc       (NonCache + WeakOrder  )
 - ioremap_cache    (   Cache + WeakOrder  )

Also change flag VM_ALLOC to VM_IOREMAP in get_vm_area_caller.

Cc Arnd Bergmann <[email protected]>
Cc Christoph Hellwig <[email protected]>
---
 arch/csky/include/asm/io.h | 23 ++++++++++++-----------
 arch/csky/mm/ioremap.c     | 23 +++++++++++++++++------
 2 files changed, 29 insertions(+), 17 deletions(-)
arch/csky/include/asm/io.h +12 -11
index c1dfa9c..80d071e 100644
@@ -4,17 +4,10 @@
 #ifndef __ASM_CSKY_IO_H
 #define __ASM_CSKY_IO_H
 
-#include <abi/pgtable-bits.h>
+#include <asm/pgtable.h>
 #include <linux/types.h>
 #include <linux/version.h>
 
-extern void __iomem *ioremap(phys_addr_t offset, size_t size);
-
-extern void iounmap(void *addr);
-
-extern int remap_area_pages(unsigned long address, phys_addr_t phys_addr,
-		size_t size, unsigned long flags);
-
 /*
  * I/O memory access primitives. Reads are ordered relative to any
  * following Normal memory access. Writes are ordered relative to any prior
@@ -40,9 +33,17 @@ extern int remap_area_pages(unsigned long address, phys_addr_t phys_addr,
 #define writel(v,c)		({ wmb(); writel_relaxed((v),(c)); mb(); })
 #endif
 
-#define ioremap_nocache(phy, sz)	ioremap(phy, sz)
-#define ioremap_wc ioremap_nocache
-#define ioremap_wt ioremap_nocache
+/*
+ * I/O memory mapping functions.
+ */
+extern void __iomem *ioremap_cache(phys_addr_t addr, size_t size);
+extern void __iomem *__ioremap(phys_addr_t addr, size_t size, pgprot_t prot);
+extern void iounmap(void *addr);
+
+#define ioremap(addr, size)		__ioremap((addr), (size), pgprot_noncached(PAGE_KERNEL))
+#define ioremap_wc(addr, size)		__ioremap((addr), (size), pgprot_writecombine(PAGE_KERNEL))
+#define ioremap_nocache(addr, size)	ioremap((addr), (size))
+#define ioremap_cache			ioremap_cache
 
 #include <asm-generic/io.h>
 
arch/csky/mm/ioremap.c +17 -6
index 4853111..e13cd34 100644
@@ -8,12 +8,12 @@
 
 #include <asm/pgtable.h>
 
-void __iomem *ioremap(phys_addr_t addr, size_t size)
+static void __iomem *__ioremap_caller(phys_addr_t addr, size_t size,
+				      pgprot_t prot, void *caller)
 {
 	phys_addr_t last_addr;
 	unsigned long offset, vaddr;
 	struct vm_struct *area;
-	pgprot_t prot;
 
 	last_addr = addr + size - 1;
 	if (!size || last_addr < addr)
@@ -23,14 +23,12 @@ void __iomem *ioremap(phys_addr_t addr, size_t size)
 	addr &= PAGE_MASK;
 	size = PAGE_ALIGN(size + offset);
 
-	area = get_vm_area_caller(size, VM_ALLOC, __builtin_return_address(0));
+	area = get_vm_area_caller(size, VM_IOREMAP, caller);
 	if (!area)
 		return NULL;
 
 	vaddr = (unsigned long)area->addr;
 
-	prot = pgprot_noncached(PAGE_KERNEL);
-
 	if (ioremap_page_range(vaddr, vaddr + size, addr, prot)) {
 		free_vm_area(area);
 		return NULL;
@@ -38,7 +36,20 @@ void __iomem *ioremap(phys_addr_t addr, size_t size)
 
 	return (void __iomem *)(vaddr + offset);
 }
-EXPORT_SYMBOL(ioremap);
+
+void __iomem *__ioremap(phys_addr_t phys_addr, size_t size, pgprot_t prot)
+{
+	return __ioremap_caller(phys_addr, size, prot,
+				__builtin_return_address(0));
+}
+EXPORT_SYMBOL(__ioremap);
+
+void __iomem *ioremap_cache(phys_addr_t phys_addr, size_t size)
+{
+	return __ioremap_caller(phys_addr, size, PAGE_KERNEL,
+				__builtin_return_address(0));
+}
+EXPORT_SYMBOL(ioremap_cache);
 
 void iounmap(void __iomem *addr)
 {
Signature · 1 line
2.7.4
#1 lore raw
Quoted 10 lines · On Thu, Aug 15, 2019 at 07:28:57PM +0800, [email protected] wrote:

From: Guo Ren <[email protected]>

Implement the following apis to meet usage in different scenarios.

 - ioremap          (NonCache + StrongOrder)
 - ioremap_nocache  (NonCache + StrongOrder)
 - ioremap_wc       (NonCache + WeakOrder  )
 - ioremap_cache    (   Cache + WeakOrder  )

Also change flag VM_ALLOC to VM_IOREMAP in get_vm_area_caller.

Looks generally fine, but two comments:

- do you have a need for ioremap_cache? We are generally try to phase it out in favour of memremap, and it is generally only used by arch specific code.
- I have a big series pending to clean up the mess with our ioremap_* functions, including adding a generic implementation that csky should be able to use. Unless this patch is urgent it might make sense to rebase it on top. Here is my current tree, I plan to post it soon:

http://git.infradead.org/users/hch/misc.git/shortlog/refs/heads/generic-ioremap

author maintainer
#2 lore raw

Thx Christoph,

Quoted 11 lines · Christoph Hellwig
Quoted 9 lines · On Thu, Aug 15, 2019 at 07:28:57PM +0800, [email protected] wrote:

From: Guo Ren <[email protected]>

Implement the following apis to meet usage in different scenarios.

 - ioremap          (NonCache + StrongOrder)
 - ioremap_nocache  (NonCache + StrongOrder)
 - ioremap_wc       (NonCache + WeakOrder  )
 - ioremap_cache    (   Cache + WeakOrder  )
Quoted 7 lines
Quoted 1 line

Also change flag VM_ALLOC to VM_IOREMAP in get_vm_area_caller.

Looks generally fine, but two comments:

- do you have a need for ioremap_cache? We are generally try to phase it out in favour of memremap, and it is generally only used by arch specific code.

Yes, some drivers of our customers use ioremap_cache to map phy_addr which isn't belong to system memory.

Quoted 7 lines

- I have a big series pending to clean up the mess with our ioremap_* functions, including adding a generic implementation that csky should be able to use. Unless this patch is urgent it might make sense to rebase it on top. Here is my current tree, I plan to post it soon:

http://git.infradead.org/users/hch/misc.git/shortlog/refs/heads/generic-ioremap

I agree to use GENERIC_IOREMAP, but I want to add csky support GENERIC_IOREMAP patch by myself.
You could remove "csky: use generic ioremap" in your patchset first and I'll add support GENERIC_IORMAP patch later.
Then we won't get confilct :)

--
Best Regards
Guo Ren

ML: https://lore.kernel.org/linux-csky/

#3 lore raw
Quoted 9 lines · Guo Ren
Quoted 7 lines
Quoted 1 line

Also change flag VM_ALLOC to VM_IOREMAP in get_vm_area_caller.

Looks generally fine, but two comments:

- do you have a need for ioremap_cache? We are generally try to phase it out in favour of memremap, and it is generally only used by arch specific code.

Yes, some drivers of our customers use ioremap_cache to map phy_addr which isn't belong to system memory.

Which driver? We should move it over to memremap instead of adding a new ioremap_cache.

Quoted 5 lines

I agree to use GENERIC_IOREMAP, but I want to add csky support GENERIC_IOREMAP patch by myself.
You could remove "csky: use generic ioremap" in your patchset first and I'll add support GENERIC_IORMAP patch later.
Then we won't get confilct :)

Ok.

author maintainer
#4 lore raw
Quoted 6 lines · Christoph Hellwig
Quoted 4 lines · Guo Ren
Quoted 4 lines
Quoted 1 line

Also change flag VM_ALLOC to VM_IOREMAP in get_vm_area_caller.

Looks generally fine, but two comments:

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

- do you have a need for ioremap_cache? We are generally try to phase it out in favour of memremap, and it is generally only used by arch specific code.

Yes, some drivers of our customers use ioremap_cache to map phy_addr which isn't belong to system memory.

Which driver? We should move it over to memremap instead of adding a new ioremap_cache.

The driver hasn't been upstreamed. OK, just remove ioremap_cache and seems it's not a big problem.

Signature · 4 lines
Best Regards
 Guo Ren

ML: https://lore.kernel.org/linux-csky/