This is v13 of my cleanup series to push a number of instances of people
defining their own io{read|write}64 functions into common headers seing
they don't exist in non-64bit systems. This series adds inline functions to the
io-64-nonatomic headers and then cleans up the drivers that defined their
own copies.
This cleanup was originally requested by Greg after he reviewed my
Switchtec NTB code. And I hope someone can pick it up or at least give
feedback on it soon as it's been around relatively unchanged for a few
cycles now and I'm getting a bit tired of resubmitting it with little to
no interest.
Thanks,
Logan
--
Changes since v12:
- Rebased onto v4.16-rc6
- Split patch 0001 into two and reworked the commit log as requested
by Luc Van Oostenryck
Changes since v11:
- Rebased onto v4.16-rc5
- Added a patch (0001) to fix some old and new sparse warnings
that the kbuild robot warned about this cycle. The latest version
of sparse was required to reproduce these.
- Added a patch (0002) to add io{read|write}64 to parisc which the kbuild
robot also found errors for this cycle
Changes since v10:
- Rebased onto v4.16-rc4, this droped the drm/tilcdc patch which was
picked up by that tree and is already in 4.16.
Changes since v9:
- Rebased onto v4.15-rc6
- Fixed a couple of issues in the new version of the CAAM patch as
pointed out by Horia
Changes since v8:
- Rebased onto v4.15-rc2, as a result rewrote patch 7 seeing someone did
some similar cleanup in that area.
- Added a patch to clean up the Switchtec NTB driver which landed in
v4.15-rc1
Changes since v7:
- Fix minor nits from Andy Shevchenko
- Rebased onto v4.14-rc1
Changes since v6:
** none **
Changes since v5:
- Added a fix to the tilcdc driver to ensure it doesn't use the
non-atomic operation. (This includes adding io{read|write}64[be]_is_nonatomic
defines).
Changes since v4:
- Add functions so the powerpc implementation of iomap.c compiles. (As
noticed by Horia)
Changes since v3:
- I noticed powerpc didn't use the appropriate functions seeing
readq/writeq were not defined when iomap.h was included. Thus I've
included a patch to adjust this
- Fixed some mistakes with a couple of the defines in io-64-nonatomic*
headers
- Fixed a typo noticed by Horia.
(earlier versions were drastically different)
--
Logan Gunthorpe (10):
iomap: Use correct endian conversion function in mmio_writeXXbe
iomap: Fix sparse endian check warnings
parisc: iomap: introduce io{read|write}64
powerpc: io.h: move iomap.h include so that it can use readq/writeq
defs
powerpc: iomap.c: introduce io{read|write}64_{lo_hi|hi_lo}
iomap: introduce io{read|write}64_{lo_hi|hi_lo}
io-64-nonatomic: add io{read|write}64[be]{_lo_hi|_hi_lo} macros
ntb: ntb_hw_intel: use io-64-nonatomic instead of in-driver hacks
crypto: caam: cleanup CONFIG_64BIT ifdefs when using io{read|write}64
ntb: ntb_hw_switchtec: Cleanup 64bit IO defines to use the common
header
arch/parisc/include/asm/io.h | 9 +++
arch/parisc/lib/iomap.c | 64 +++++++++++++++
arch/powerpc/include/asm/io.h | 6 +-
arch/powerpc/kernel/iomap.c | 40 ++++++++++
drivers/crypto/caam/regs.h | 30 +------
drivers/ntb/hw/intel/ntb_hw_intel.c | 30 +------
drivers/ntb/hw/mscc/ntb_hw_switchtec.c | 36 +--------
include/asm-generic/iomap.h | 26 ++++--
include/linux/io-64-nonatomic-hi-lo.h | 64 +++++++++++++++
include/linux/io-64-nonatomic-lo-hi.h | 64 +++++++++++++++
lib/iomap.c | 140 ++++++++++++++++++++++++++++++++-
11 files changed, 409 insertions(+), 100 deletions(-)
--
2.11.0
The semantics of the iowriteXXbe() functions are to write a
value in CPU endianess to an IO register that is known by the
caller to be in Big Endian. The mmio_writeXXbe() macro, which
is called by iowriteXXbe(), should therefore use cpu_to_beXX()
instead of beXX_to_cpu().
Seeing both beXX_to_cpu() and cpu_to_beXX() are both functionally
implemented as either null operations or swabXX operations there
was no noticable bug here. But it is confusing for both developers
and code analysis tools alike.
Signed-off-by: Logan Gunthorpe <[email protected]>
Cc: Philippe Ombredanne <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Kate Stewart <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: Luc Van Oostenryck <[email protected]>
---
lib/iomap.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/lib/iomap.c b/lib/iomap.c
index 541d926da95e..be120c13d6cc 100644
--- a/lib/iomap.c
+++ b/lib/iomap.c
@@ -106,8 +106,8 @@ EXPORT_SYMBOL(ioread32be);
#endif
#ifndef mmio_write16be
-#define mmio_write16be(val,port) __raw_writew(be16_to_cpu(val),port)
-#define mmio_write32be(val,port) __raw_writel(be32_to_cpu(val),port)
+#define mmio_write16be(val,port) __raw_writew(cpu_to_be16(val),port)
+#define mmio_write32be(val,port) __raw_writel(cpu_to_be32(val),port)
#endif
void iowrite8(u8 val, void __iomem *addr)
--
2.11.0
Clean up the extra ifdefs which defined the wr_reg64 and rd_reg64
functions in non-64bit cases in favour of the new common
io-64-nonatomic-lo-hi header.
To be consistent with CAAM engine HW spec: in case of 64-bit registers,
irrespective of device endianness, the lower address should be read from
/ written to first, followed by the upper address. Indeed the I/O
accessors in CAAM driver currently don't follow the spec, however this
is a good opportunity to fix the code.
Signed-off-by: Logan Gunthorpe <[email protected]>
Reviewed-by: Horia Geantă <[email protected]>
Cc: Andy Shevchenko <[email protected]>
Cc: Dan Douglass <[email protected]>
Cc: Herbert Xu <[email protected]>
Cc: "David S. Miller" <[email protected]>
---
drivers/crypto/caam/regs.h | 30 +++---------------------------
1 file changed, 3 insertions(+), 27 deletions(-)
diff --git a/drivers/crypto/caam/regs.h b/drivers/crypto/caam/regs.h
index fee363865d88..f887b371040f 100644
--- a/drivers/crypto/caam/regs.h
+++ b/drivers/crypto/caam/regs.h
@@ -10,7 +10,7 @@
#include <linux/types.h>
#include <linux/bitops.h>
-#include <linux/io.h>
+#include <linux/io-64-nonatomic-lo-hi.h>
/*
* Architecture-specific register access methods
@@ -136,10 +136,9 @@ static inline void clrsetbits_32(void __iomem *reg, u32 clear, u32 set)
* base + 0x0000 : least-significant 32 bits
* base + 0x0004 : most-significant 32 bits
*/
-#ifdef CONFIG_64BIT
static inline void wr_reg64(void __iomem *reg, u64 data)
{
- if (caam_little_end)
+ if (!caam_imx && caam_little_end)
iowrite64(data, reg);
else
iowrite64be(data, reg);
@@ -147,35 +146,12 @@ static inline void wr_reg64(void __iomem *reg, u64 data)
static inline u64 rd_reg64(void __iomem *reg)
{
- if (caam_little_end)
+ if (!caam_imx && caam_little_end)
return ioread64(reg);
else
return ioread64be(reg);
}
-#else /* CONFIG_64BIT */
-static inline void wr_reg64(void __iomem *reg, u64 data)
-{
- if (!caam_imx && caam_little_end) {
- wr_reg32((u32 __iomem *)(reg) + 1, data >> 32);
- wr_reg32((u32 __iomem *)(reg), data);
- } else {
- wr_reg32((u32 __iomem *)(reg), data >> 32);
- wr_reg32((u32 __iomem *)(reg) + 1, data);
- }
-}
-
-static inline u64 rd_reg64(void __iomem *reg)
-{
- if (!caam_imx && caam_little_end)
- return ((u64)rd_reg32((u32 __iomem *)(reg) + 1) << 32 |
- (u64)rd_reg32((u32 __iomem *)(reg)));
-
- return ((u64)rd_reg32((u32 __iomem *)(reg)) << 32 |
- (u64)rd_reg32((u32 __iomem *)(reg) + 1));
-}
-#endif /* CONFIG_64BIT */
-
static inline u64 cpu_to_caam_dma64(dma_addr_t value)
{
if (caam_imx)
--
2.11.0
In order to provide non-atomic functions for io{read|write}64 that will
use readq and writeq when appropriate. We define a number of variants
of these functions in the generic iomap that will do non-atomic
operations on pio but atomic operations on mmio.
These functions are only defined if readq and writeq are defined. If
they are not, then the wrappers that always use non-atomic operations
from include/linux/io-64-nonatomic*.h will be used.
Signed-off-by: Logan Gunthorpe <[email protected]>
Reviewed-by: Andy Shevchenko <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Michael Ellerman <[email protected]>
Cc: Arnd Bergmann <[email protected]>
Cc: Suresh Warrier <[email protected]>
Cc: Nicholas Piggin <[email protected]>
---
arch/powerpc/include/asm/io.h | 2 +
include/asm-generic/iomap.h | 26 +++++++--
lib/iomap.c | 132 ++++++++++++++++++++++++++++++++++++++++++
3 files changed, 154 insertions(+), 6 deletions(-)
diff --git a/arch/powerpc/include/asm/io.h b/arch/powerpc/include/asm/io.h
index af074923d598..4cc420cfaa78 100644
--- a/arch/powerpc/include/asm/io.h
+++ b/arch/powerpc/include/asm/io.h
@@ -788,8 +788,10 @@ extern void __iounmap_at(void *ea, unsigned long size);
#define mmio_read16be(addr) readw_be(addr)
#define mmio_read32be(addr) readl_be(addr)
+#define mmio_read64be(addr) readq_be(addr)
#define mmio_write16be(val, addr) writew_be(val, addr)
#define mmio_write32be(val, addr) writel_be(val, addr)
+#define mmio_write64be(val, addr) writeq_be(val, addr)
#define mmio_insb(addr, dst, count) readsb(addr, dst, count)
#define mmio_insw(addr, dst, count) readsw(addr, dst, count)
#define mmio_insl(addr, dst, count) readsl(addr, dst, count)
diff --git a/include/asm-generic/iomap.h b/include/asm-generic/iomap.h
index 5b63b94ef6b5..5a4af0199b32 100644
--- a/include/asm-generic/iomap.h
+++ b/include/asm-generic/iomap.h
@@ -31,9 +31,16 @@ extern unsigned int ioread16(void __iomem *);
extern unsigned int ioread16be(void __iomem *);
extern unsigned int ioread32(void __iomem *);
extern unsigned int ioread32be(void __iomem *);
-#ifdef CONFIG_64BIT
-extern u64 ioread64(void __iomem *);
-extern u64 ioread64be(void __iomem *);
+
+#ifdef readq
+#define ioread64_lo_hi ioread64_lo_hi
+#define ioread64_hi_lo ioread64_hi_lo
+#define ioread64be_lo_hi ioread64be_lo_hi
+#define ioread64be_hi_lo ioread64be_hi_lo
+extern u64 ioread64_lo_hi(void __iomem *addr);
+extern u64 ioread64_hi_lo(void __iomem *addr);
+extern u64 ioread64be_lo_hi(void __iomem *addr);
+extern u64 ioread64be_hi_lo(void __iomem *addr);
#endif
extern void iowrite8(u8, void __iomem *);
@@ -41,9 +48,16 @@ extern void iowrite16(u16, void __iomem *);
extern void iowrite16be(u16, void __iomem *);
extern void iowrite32(u32, void __iomem *);
extern void iowrite32be(u32, void __iomem *);
-#ifdef CONFIG_64BIT
-extern void iowrite64(u64, void __iomem *);
-extern void iowrite64be(u64, void __iomem *);
+
+#ifdef writeq
+#define iowrite64_lo_hi iowrite64_lo_hi
+#define iowrite64_hi_lo iowrite64_hi_lo
+#define iowrite64be_lo_hi iowrite64be_lo_hi
+#define iowrite64be_hi_lo iowrite64be_hi_lo
+extern void iowrite64_lo_hi(u64 val, void __iomem *addr);
+extern void iowrite64_hi_lo(u64 val, void __iomem *addr);
+extern void iowrite64be_lo_hi(u64 val, void __iomem *addr);
+extern void iowrite64be_hi_lo(u64 val, void __iomem *addr);
#endif
/*
diff --git a/lib/iomap.c b/lib/iomap.c
index 44645c4b516c..f3b7b2e4c625 100644
--- a/lib/iomap.c
+++ b/lib/iomap.c
@@ -67,6 +67,7 @@ static void bad_io_access(unsigned long port, const char *access)
#ifndef mmio_read16be
#define mmio_read16be(addr) be16_to_cpu((__be16 __force)__raw_readw(addr))
#define mmio_read32be(addr) be32_to_cpu((__be32 __force)__raw_readl(addr))
+#define mmio_read64be(addr) be64_to_cpu((__be64 __force)__raw_readq(addr))
#endif
unsigned int ioread8(void __iomem *addr)
@@ -100,6 +101,80 @@ EXPORT_SYMBOL(ioread16be);
EXPORT_SYMBOL(ioread32);
EXPORT_SYMBOL(ioread32be);
+#ifdef readq
+static u64 pio_read64_lo_hi(unsigned long port)
+{
+ u64 lo, hi;
+
+ lo = inl(port);
+ hi = inl(port + sizeof(u32));
+
+ return lo | (hi << 32);
+}
+
+static u64 pio_read64_hi_lo(unsigned long port)
+{
+ u64 lo, hi;
+
+ hi = inl(port + sizeof(u32));
+ lo = inl(port);
+
+ return lo | (hi << 32);
+}
+
+static u64 pio_read64be_lo_hi(unsigned long port)
+{
+ u64 lo, hi;
+
+ lo = pio_read32be(port + sizeof(u32));
+ hi = pio_read32be(port);
+
+ return lo | (hi << 32);
+}
+
+static u64 pio_read64be_hi_lo(unsigned long port)
+{
+ u64 lo, hi;
+
+ hi = pio_read32be(port);
+ lo = pio_read32be(port + sizeof(u32));
+
+ return lo | (hi << 32);
+}
+
+u64 ioread64_lo_hi(void __iomem *addr)
+{
+ IO_COND(addr, return pio_read64_lo_hi(port), return readq(addr));
+ return 0xffffffffffffffffULL;
+}
+
+u64 ioread64_hi_lo(void __iomem *addr)
+{
+ IO_COND(addr, return pio_read64_hi_lo(port), return readq(addr));
+ return 0xffffffffffffffffULL;
+}
+
+u64 ioread64be_lo_hi(void __iomem *addr)
+{
+ IO_COND(addr, return pio_read64be_lo_hi(port),
+ return mmio_read64be(addr));
+ return 0xffffffffffffffffULL;
+}
+
+u64 ioread64be_hi_lo(void __iomem *addr)
+{
+ IO_COND(addr, return pio_read64be_hi_lo(port),
+ return mmio_read64be(addr));
+ return 0xffffffffffffffffULL;
+}
+
+EXPORT_SYMBOL(ioread64_lo_hi);
+EXPORT_SYMBOL(ioread64_hi_lo);
+EXPORT_SYMBOL(ioread64be_lo_hi);
+EXPORT_SYMBOL(ioread64be_hi_lo);
+
+#endif /* readq */
+
#ifndef pio_write16be
#define pio_write16be(val,port) outw(swab16(val),port)
#define pio_write32be(val,port) outl(swab32(val),port)
@@ -108,6 +183,7 @@ EXPORT_SYMBOL(ioread32be);
#ifndef mmio_write16be
#define mmio_write16be(val,port) __raw_writew((u16 __force)cpu_to_be16(val),port)
#define mmio_write32be(val,port) __raw_writel((u32 __force)cpu_to_be32(val),port)
+#define mmio_write64be(val,port) __raw_writeq((u64 __force)cpu_to_be64(val),port)
#endif
void iowrite8(u8 val, void __iomem *addr)
@@ -136,6 +212,62 @@ EXPORT_SYMBOL(iowrite16be);
EXPORT_SYMBOL(iowrite32);
EXPORT_SYMBOL(iowrite32be);
+#ifdef writeq
+static void pio_write64_lo_hi(u64 val, unsigned long port)
+{
+ outl(val, port);
+ outl(val >> 32, port + sizeof(u32));
+}
+
+static void pio_write64_hi_lo(u64 val, unsigned long port)
+{
+ outl(val >> 32, port + sizeof(u32));
+ outl(val, port);
+}
+
+static void pio_write64be_lo_hi(u64 val, unsigned long port)
+{
+ pio_write32be(val, port + sizeof(u32));
+ pio_write32be(val >> 32, port);
+}
+
+static void pio_write64be_hi_lo(u64 val, unsigned long port)
+{
+ pio_write32be(val >> 32, port);
+ pio_write32be(val, port + sizeof(u32));
+}
+
+void iowrite64_lo_hi(u64 val, void __iomem *addr)
+{
+ IO_COND(addr, pio_write64_lo_hi(val, port),
+ writeq(val, addr));
+}
+
+void iowrite64_hi_lo(u64 val, void __iomem *addr)
+{
+ IO_COND(addr, pio_write64_hi_lo(val, port),
+ writeq(val, addr));
+}
+
+void iowrite64be_lo_hi(u64 val, void __iomem *addr)
+{
+ IO_COND(addr, pio_write64be_lo_hi(val, port),
+ mmio_write64be(val, addr));
+}
+
+void iowrite64be_hi_lo(u64 val, void __iomem *addr)
+{
+ IO_COND(addr, pio_write64be_hi_lo(val, port),
+ mmio_write64be(val, addr));
+}
+
+EXPORT_SYMBOL(iowrite64_lo_hi);
+EXPORT_SYMBOL(iowrite64_hi_lo);
+EXPORT_SYMBOL(iowrite64be_lo_hi);
+EXPORT_SYMBOL(iowrite64be_hi_lo);
+
+#endif /* readq */
+
/*
* These are the "repeat MMIO read/write" functions.
* Note the "__raw" accesses, since we don't want to
--
2.11.0
Now that ioread64 and iowrite64 are available in io-64-nonatomic,
we can remove the hack at the top of ntb_hw_intel.c and replace it
with an include.
Signed-off-by: Logan Gunthorpe <[email protected]>
Reviewed-by: Andy Shevchenko <[email protected]>
Acked-by: Dave Jiang <[email protected]>
Acked-by: Allen Hubbe <[email protected]>
Acked-by: Jon Mason <[email protected]>
---
drivers/ntb/hw/intel/ntb_hw_intel.c | 30 +-----------------------------
1 file changed, 1 insertion(+), 29 deletions(-)
diff --git a/drivers/ntb/hw/intel/ntb_hw_intel.c b/drivers/ntb/hw/intel/ntb_hw_intel.c
index 156b45cd4a19..5cf40ab21366 100644
--- a/drivers/ntb/hw/intel/ntb_hw_intel.c
+++ b/drivers/ntb/hw/intel/ntb_hw_intel.c
@@ -59,6 +59,7 @@
#include <linux/random.h>
#include <linux/slab.h>
#include <linux/ntb.h>
+#include <linux/io-64-nonatomic-lo-hi.h>
#include "ntb_hw_intel.h"
@@ -149,35 +150,6 @@ MODULE_PARM_DESC(xeon_b2b_dsd_bar5_addr32,
static inline enum ntb_topo xeon_ppd_topo(struct intel_ntb_dev *ndev, u8 ppd);
static int xeon_init_isr(struct intel_ntb_dev *ndev);
-#ifndef ioread64
-#ifdef readq
-#define ioread64 readq
-#else
-#define ioread64 _ioread64
-static inline u64 _ioread64(void __iomem *mmio)
-{
- u64 low, high;
-
- low = ioread32(mmio);
- high = ioread32(mmio + sizeof(u32));
- return low | (high << 32);
-}
-#endif
-#endif
-
-#ifndef iowrite64
-#ifdef writeq
-#define iowrite64 writeq
-#else
-#define iowrite64 _iowrite64
-static inline void _iowrite64(u64 val, void __iomem *mmio)
-{
- iowrite32(val, mmio);
- iowrite32(val >> 32, mmio + sizeof(u32));
-}
-#endif
-#endif
-
static inline int pdev_is_xeon(struct pci_dev *pdev)
{
switch (pdev->device) {
--
2.11.0
Subsequent patches in this series makes use of the readq and writeq
defines in iomap.h. However, as is, they get missed on the powerpc
platform seeing the include comes before the define. This patch
moves the include down to fix this.
Signed-off-by: Logan Gunthorpe <[email protected]>
Acked-by: Michael Ellerman <[email protected]>
Reviewed-by: Andy Shevchenko <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Michael Ellerman <[email protected]>
Cc: Nicholas Piggin <[email protected]>
Cc: Suresh Warrier <[email protected]>
Cc: "Oliver O'Halloran" <[email protected]>
---
arch/powerpc/include/asm/io.h | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/arch/powerpc/include/asm/io.h b/arch/powerpc/include/asm/io.h
index 422f99cf9924..af074923d598 100644
--- a/arch/powerpc/include/asm/io.h
+++ b/arch/powerpc/include/asm/io.h
@@ -33,8 +33,6 @@ extern struct pci_dev *isa_bridge_pcidev;
#include <asm/mmu.h>
#include <asm/ppc_asm.h>
-#include <asm-generic/iomap.h>
-
#ifdef CONFIG_PPC64
#include <asm/paca.h>
#endif
@@ -663,6 +661,8 @@ static inline void name at \
#define writel_relaxed(v, addr) writel(v, addr)
#define writeq_relaxed(v, addr) writeq(v, addr)
+#include <asm-generic/iomap.h>
+
#ifdef CONFIG_PPC32
#define mmiowb()
#else
--
2.11.0
This patch adds generic io{read|write}64[be]{_lo_hi|_hi_lo} macros if
they are not already defined by the architecture. (As they are provided
by the generic iomap library).
The patch also points io{read|write}64[be] to the variant specified by the
header name.
This is because new drivers are encouraged to use ioreadXX, et al instead
of readX[1], et al -- and mixing ioreadXX with readq is pretty ugly.
[1] LDD3: section 9.4.2
Signed-off-by: Logan Gunthorpe <[email protected]>
Reviewed-by: Andy Shevchenko <[email protected]>
Cc: Christoph Hellwig <[email protected]>
Cc: Arnd Bergmann <[email protected]>
Cc: Alan Cox <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
---
include/linux/io-64-nonatomic-hi-lo.h | 64 +++++++++++++++++++++++++++++++++++
include/linux/io-64-nonatomic-lo-hi.h | 64 +++++++++++++++++++++++++++++++++++
2 files changed, 128 insertions(+)
diff --git a/include/linux/io-64-nonatomic-hi-lo.h b/include/linux/io-64-nonatomic-hi-lo.h
index 862d786a904f..ae21b72cce85 100644
--- a/include/linux/io-64-nonatomic-hi-lo.h
+++ b/include/linux/io-64-nonatomic-hi-lo.h
@@ -55,4 +55,68 @@ static inline void hi_lo_writeq_relaxed(__u64 val, volatile void __iomem *addr)
#define writeq_relaxed hi_lo_writeq_relaxed
#endif
+#ifndef ioread64_hi_lo
+#define ioread64_hi_lo ioread64_hi_lo
+static inline u64 ioread64_hi_lo(void __iomem *addr)
+{
+ u32 low, high;
+
+ high = ioread32(addr + sizeof(u32));
+ low = ioread32(addr);
+
+ return low + ((u64)high << 32);
+}
+#endif
+
+#ifndef iowrite64_hi_lo
+#define iowrite64_hi_lo iowrite64_hi_lo
+static inline void iowrite64_hi_lo(u64 val, void __iomem *addr)
+{
+ iowrite32(val >> 32, addr + sizeof(u32));
+ iowrite32(val, addr);
+}
+#endif
+
+#ifndef ioread64be_hi_lo
+#define ioread64be_hi_lo ioread64be_hi_lo
+static inline u64 ioread64be_hi_lo(void __iomem *addr)
+{
+ u32 low, high;
+
+ high = ioread32be(addr);
+ low = ioread32be(addr + sizeof(u32));
+
+ return low + ((u64)high << 32);
+}
+#endif
+
+#ifndef iowrite64be_hi_lo
+#define iowrite64be_hi_lo iowrite64be_hi_lo
+static inline void iowrite64be_hi_lo(u64 val, void __iomem *addr)
+{
+ iowrite32be(val >> 32, addr);
+ iowrite32be(val, addr + sizeof(u32));
+}
+#endif
+
+#ifndef ioread64
+#define ioread64_is_nonatomic
+#define ioread64 ioread64_hi_lo
+#endif
+
+#ifndef iowrite64
+#define iowrite64_is_nonatomic
+#define iowrite64 iowrite64_hi_lo
+#endif
+
+#ifndef ioread64be
+#define ioread64be_is_nonatomic
+#define ioread64be ioread64be_hi_lo
+#endif
+
+#ifndef iowrite64be
+#define iowrite64be_is_nonatomic
+#define iowrite64be iowrite64be_hi_lo
+#endif
+
#endif /* _LINUX_IO_64_NONATOMIC_HI_LO_H_ */
diff --git a/include/linux/io-64-nonatomic-lo-hi.h b/include/linux/io-64-nonatomic-lo-hi.h
index d042e7bb5adb..faaa842dbdb9 100644
--- a/include/linux/io-64-nonatomic-lo-hi.h
+++ b/include/linux/io-64-nonatomic-lo-hi.h
@@ -55,4 +55,68 @@ static inline void lo_hi_writeq_relaxed(__u64 val, volatile void __iomem *addr)
#define writeq_relaxed lo_hi_writeq_relaxed
#endif
+#ifndef ioread64_lo_hi
+#define ioread64_lo_hi ioread64_lo_hi
+static inline u64 ioread64_lo_hi(void __iomem *addr)
+{
+ u32 low, high;
+
+ low = ioread32(addr);
+ high = ioread32(addr + sizeof(u32));
+
+ return low + ((u64)high << 32);
+}
+#endif
+
+#ifndef iowrite64_lo_hi
+#define iowrite64_lo_hi iowrite64_lo_hi
+static inline void iowrite64_lo_hi(u64 val, void __iomem *addr)
+{
+ iowrite32(val, addr);
+ iowrite32(val >> 32, addr + sizeof(u32));
+}
+#endif
+
+#ifndef ioread64be_lo_hi
+#define ioread64be_lo_hi ioread64be_lo_hi
+static inline u64 ioread64be_lo_hi(void __iomem *addr)
+{
+ u32 low, high;
+
+ low = ioread32be(addr + sizeof(u32));
+ high = ioread32be(addr);
+
+ return low + ((u64)high << 32);
+}
+#endif
+
+#ifndef iowrite64be_lo_hi
+#define iowrite64be_lo_hi iowrite64be_lo_hi
+static inline void iowrite64be_lo_hi(u64 val, void __iomem *addr)
+{
+ iowrite32be(val, addr + sizeof(u32));
+ iowrite32be(val >> 32, addr);
+}
+#endif
+
+#ifndef ioread64
+#define ioread64_is_nonatomic
+#define ioread64 ioread64_lo_hi
+#endif
+
+#ifndef iowrite64
+#define iowrite64_is_nonatomic
+#define iowrite64 iowrite64_lo_hi
+#endif
+
+#ifndef ioread64be
+#define ioread64be_is_nonatomic
+#define ioread64be ioread64be_lo_hi
+#endif
+
+#ifndef iowrite64be
+#define iowrite64be_is_nonatomic
+#define iowrite64be iowrite64be_lo_hi
+#endif
+
#endif /* _LINUX_IO_64_NONATOMIC_LO_HI_H_ */
--
2.11.0
Clean up the ifdefs which conditionally defined the io{read|write}64
functions in favour of the new common io-64-nonatomic-lo-hi header.
Per a nit from Andy Shevchenko, the include list is also made
alphabetical.
Signed-off-by: Logan Gunthorpe <[email protected]>
Cc: Jon Mason <[email protected]>
---
drivers/ntb/hw/mscc/ntb_hw_switchtec.c | 36 ++++------------------------------
1 file changed, 4 insertions(+), 32 deletions(-)
diff --git a/drivers/ntb/hw/mscc/ntb_hw_switchtec.c b/drivers/ntb/hw/mscc/ntb_hw_switchtec.c
index f624ae27eabe..f403da24b833 100644
--- a/drivers/ntb/hw/mscc/ntb_hw_switchtec.c
+++ b/drivers/ntb/hw/mscc/ntb_hw_switchtec.c
@@ -13,12 +13,13 @@
*
*/
-#include <linux/switchtec.h>
-#include <linux/module.h>
+#include <linux/interrupt.h>
+#include <linux/io-64-nonatomic-lo-hi.h>
#include <linux/delay.h>
#include <linux/kthread.h>
-#include <linux/interrupt.h>
+#include <linux/module.h>
#include <linux/ntb.h>
+#include <linux/switchtec.h>
MODULE_DESCRIPTION("Microsemi Switchtec(tm) NTB Driver");
MODULE_VERSION("0.1");
@@ -35,35 +36,6 @@ module_param(use_lut_mws, bool, 0644);
MODULE_PARM_DESC(use_lut_mws,
"Enable the use of the LUT based memory windows");
-#ifndef ioread64
-#ifdef readq
-#define ioread64 readq
-#else
-#define ioread64 _ioread64
-static inline u64 _ioread64(void __iomem *mmio)
-{
- u64 low, high;
-
- low = ioread32(mmio);
- high = ioread32(mmio + sizeof(u32));
- return low | (high << 32);
-}
-#endif
-#endif
-
-#ifndef iowrite64
-#ifdef writeq
-#define iowrite64 writeq
-#else
-#define iowrite64 _iowrite64
-static inline void _iowrite64(u64 val, void __iomem *mmio)
-{
- iowrite32(val, mmio);
- iowrite32(val >> 32, mmio + sizeof(u32));
-}
-#endif
-#endif
-
#define SWITCHTEC_NTB_MAGIC 0x45CC0001
#define MAX_MWS 128
--
2.11.0
Add support for io{read|write}64() functions in parisc architecture.
These are pretty straightforward copies of similar functions which
make use of readq and writeq.
Also, indicate that the lo_hi and hi_lo variants of these functions
are not provided by this architecture.
Signed-off-by: Logan Gunthorpe <[email protected]>
Cc: "James E.J. Bottomley" <[email protected]>
Cc: Helge Deller <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: Philippe Ombredanne <[email protected]>
Cc: Kate Stewart <[email protected]>
Cc: Thomas Gleixner <[email protected]>
---
arch/parisc/include/asm/io.h | 9 +++++++
arch/parisc/lib/iomap.c | 64 ++++++++++++++++++++++++++++++++++++++++++++
2 files changed, 73 insertions(+)
diff --git a/arch/parisc/include/asm/io.h b/arch/parisc/include/asm/io.h
index 27c62baa9c4e..34c133066d3c 100644
--- a/arch/parisc/include/asm/io.h
+++ b/arch/parisc/include/asm/io.h
@@ -311,6 +311,15 @@ extern void outsl (unsigned long port, const void *src, unsigned long count);
* value for either 32 or 64 bit mode */
#define F_EXTEND(x) ((unsigned long)((x) | (0xffffffff00000000ULL)))
+#define ioread64 ioread64
+#define ioread64be ioread64be
+#define iowrite64 iowrite64
+#define iowrite64be iowrite64be
+extern u64 ioread64(void __iomem *addr);
+extern u64 ioread64be(void __iomem *addr);
+extern void iowrite64(u64 val, void __iomem *addr);
+extern void iowrite64be(u64 val, void __iomem *addr);
+
#include <asm-generic/iomap.h>
/*
diff --git a/arch/parisc/lib/iomap.c b/arch/parisc/lib/iomap.c
index 4b19e6e64fb7..0195aec657e2 100644
--- a/arch/parisc/lib/iomap.c
+++ b/arch/parisc/lib/iomap.c
@@ -48,11 +48,15 @@ struct iomap_ops {
unsigned int (*read16be)(void __iomem *);
unsigned int (*read32)(void __iomem *);
unsigned int (*read32be)(void __iomem *);
+ u64 (*read64)(void __iomem *);
+ u64 (*read64be)(void __iomem *);
void (*write8)(u8, void __iomem *);
void (*write16)(u16, void __iomem *);
void (*write16be)(u16, void __iomem *);
void (*write32)(u32, void __iomem *);
void (*write32be)(u32, void __iomem *);
+ void (*write64)(u64, void __iomem *);
+ void (*write64be)(u64, void __iomem *);
void (*read8r)(void __iomem *, void *, unsigned long);
void (*read16r)(void __iomem *, void *, unsigned long);
void (*read32r)(void __iomem *, void *, unsigned long);
@@ -171,6 +175,16 @@ static unsigned int iomem_read32be(void __iomem *addr)
return __raw_readl(addr);
}
+static u64 iomem_read64(void __iomem *addr)
+{
+ return readq(addr);
+}
+
+static u64 iomem_read64be(void __iomem *addr)
+{
+ return __raw_readq(addr);
+}
+
static void iomem_write8(u8 datum, void __iomem *addr)
{
writeb(datum, addr);
@@ -196,6 +210,16 @@ static void iomem_write32be(u32 datum, void __iomem *addr)
__raw_writel(datum, addr);
}
+static void iomem_write64(u64 datum, void __iomem *addr)
+{
+ writel(datum, addr);
+}
+
+static void iomem_write64be(u64 datum, void __iomem *addr)
+{
+ __raw_writel(datum, addr);
+}
+
static void iomem_read8r(void __iomem *addr, void *dst, unsigned long count)
{
while (count--) {
@@ -250,11 +274,15 @@ static const struct iomap_ops iomem_ops = {
.read16be = iomem_read16be,
.read32 = iomem_read32,
.read32be = iomem_read32be,
+ .read64 = iomem_read64,
+ .read64be = iomem_read64be,
.write8 = iomem_write8,
.write16 = iomem_write16,
.write16be = iomem_write16be,
.write32 = iomem_write32,
.write32be = iomem_write32be,
+ .write64 = iomem_write64,
+ .write64be = iomem_write64be,
.read8r = iomem_read8r,
.read16r = iomem_read16r,
.read32r = iomem_read32r,
@@ -304,6 +332,20 @@ unsigned int ioread32be(void __iomem *addr)
return *((u32 *)addr);
}
+u64 ioread64(void __iomem *addr)
+{
+ if (unlikely(INDIRECT_ADDR(addr)))
+ return iomap_ops[ADDR_TO_REGION(addr)]->read64(addr);
+ return le64_to_cpup((u64 *)addr);
+}
+
+u64 ioread64be(void __iomem *addr)
+{
+ if (unlikely(INDIRECT_ADDR(addr)))
+ return iomap_ops[ADDR_TO_REGION(addr)]->read64be(addr);
+ return *((u64 *)addr);
+}
+
void iowrite8(u8 datum, void __iomem *addr)
{
if (unlikely(INDIRECT_ADDR(addr))) {
@@ -349,6 +391,24 @@ void iowrite32be(u32 datum, void __iomem *addr)
}
}
+void iowrite64(u64 datum, void __iomem *addr)
+{
+ if (unlikely(INDIRECT_ADDR(addr))) {
+ iomap_ops[ADDR_TO_REGION(addr)]->write64(datum, addr);
+ } else {
+ *((u64 *)addr) = cpu_to_le64(datum);
+ }
+}
+
+void iowrite64be(u64 datum, void __iomem *addr)
+{
+ if (unlikely(INDIRECT_ADDR(addr))) {
+ iomap_ops[ADDR_TO_REGION(addr)]->write64be(datum, addr);
+ } else {
+ *((u64 *)addr) = datum;
+ }
+}
+
/* Repeating interfaces */
void ioread8_rep(void __iomem *addr, void *dst, unsigned long count)
@@ -449,11 +509,15 @@ EXPORT_SYMBOL(ioread16);
EXPORT_SYMBOL(ioread16be);
EXPORT_SYMBOL(ioread32);
EXPORT_SYMBOL(ioread32be);
+EXPORT_SYMBOL(ioread64);
+EXPORT_SYMBOL(ioread64be);
EXPORT_SYMBOL(iowrite8);
EXPORT_SYMBOL(iowrite16);
EXPORT_SYMBOL(iowrite16be);
EXPORT_SYMBOL(iowrite32);
EXPORT_SYMBOL(iowrite32be);
+EXPORT_SYMBOL(iowrite64);
+EXPORT_SYMBOL(iowrite64be);
EXPORT_SYMBOL(ioread8_rep);
EXPORT_SYMBOL(ioread16_rep);
EXPORT_SYMBOL(ioread32_rep);
--
2.11.0
These functions will be introduced into the generic iomap.c so
they can deal with PIO accesses in hi-lo/lo-hi variants. Thus,
the powerpc version of iomap.c will need to provide the same
functions even though, in this arch, they are identical to the
regular io{read|write}64 functions.
Signed-off-by: Logan Gunthorpe <[email protected]>
Tested-by: Horia Geantă <[email protected]>
Reviewed-by: Andy Shevchenko <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Michael Ellerman <[email protected]>
---
arch/powerpc/kernel/iomap.c | 40 ++++++++++++++++++++++++++++++++++++++++
1 file changed, 40 insertions(+)
diff --git a/arch/powerpc/kernel/iomap.c b/arch/powerpc/kernel/iomap.c
index aab456ed2a00..5ac84efc6ede 100644
--- a/arch/powerpc/kernel/iomap.c
+++ b/arch/powerpc/kernel/iomap.c
@@ -45,12 +45,32 @@ u64 ioread64(void __iomem *addr)
{
return readq(addr);
}
+u64 ioread64_lo_hi(void __iomem *addr)
+{
+ return readq(addr);
+}
+u64 ioread64_hi_lo(void __iomem *addr)
+{
+ return readq(addr);
+}
u64 ioread64be(void __iomem *addr)
{
return readq_be(addr);
}
+u64 ioread64be_lo_hi(void __iomem *addr)
+{
+ return readq_be(addr);
+}
+u64 ioread64be_hi_lo(void __iomem *addr)
+{
+ return readq_be(addr);
+}
EXPORT_SYMBOL(ioread64);
+EXPORT_SYMBOL(ioread64_lo_hi);
+EXPORT_SYMBOL(ioread64_hi_lo);
EXPORT_SYMBOL(ioread64be);
+EXPORT_SYMBOL(ioread64be_lo_hi);
+EXPORT_SYMBOL(ioread64be_hi_lo);
#endif /* __powerpc64__ */
void iowrite8(u8 val, void __iomem *addr)
@@ -83,12 +103,32 @@ void iowrite64(u64 val, void __iomem *addr)
{
writeq(val, addr);
}
+void iowrite64_lo_hi(u64 val, void __iomem *addr)
+{
+ writeq(val, addr);
+}
+void iowrite64_hi_lo(u64 val, void __iomem *addr)
+{
+ writeq(val, addr);
+}
void iowrite64be(u64 val, void __iomem *addr)
{
writeq_be(val, addr);
}
+void iowrite64be_lo_hi(u64 val, void __iomem *addr)
+{
+ writeq_be(val, addr);
+}
+void iowrite64be_hi_lo(u64 val, void __iomem *addr)
+{
+ writeq_be(val, addr);
+}
EXPORT_SYMBOL(iowrite64);
+EXPORT_SYMBOL(iowrite64_lo_hi);
+EXPORT_SYMBOL(iowrite64_hi_lo);
EXPORT_SYMBOL(iowrite64be);
+EXPORT_SYMBOL(iowrite64be_lo_hi);
+EXPORT_SYMBOL(iowrite64be_hi_lo);
#endif /* __powerpc64__ */
/*
--
2.11.0
Newer version of sparse produce a few warnings of the form:
lib/iomap.c:84:9: warning: cast to restricted __be16
when run with:
make C=2 CF=-D__CHECK_ENDIAN__
(The kbuild robot has recently started running such checks)
The warning is not valid because the __raw_readX() and __raw_writeX()
functions have an endianess determined by the semantics of whichever
register they are operating on (which is intern dependant on the address
passed to the function).
In the iomap case, the register being operated on is known,
semantically, to be Big Endian when an ioXXbe() function is used
by the caller. These functions wrap a raw read or write function
in an endianness conversion function.
Sparse enforces that all values that aren't in CPU endianness
be marked with types like __be16 and similar and said types cannot
be mixed with other types. However, per above, the raw functions
cannot be marked as such seeing the endianness is indeterminate.
Thus, the output of __raw_readX() and the input of __raw_writeX()
must be cast with the __force keyword to supress the invalid warning.
Signed-off-by: Logan Gunthorpe <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: Philippe Ombredanne <[email protected]>
Cc: Kate Stewart <[email protected]>
Cc: Luc Van Oostenryck <[email protected]>
---
lib/iomap.c | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/lib/iomap.c b/lib/iomap.c
index be120c13d6cc..44645c4b516c 100644
--- a/lib/iomap.c
+++ b/lib/iomap.c
@@ -65,8 +65,8 @@ static void bad_io_access(unsigned long port, const char *access)
#endif
#ifndef mmio_read16be
-#define mmio_read16be(addr) be16_to_cpu(__raw_readw(addr))
-#define mmio_read32be(addr) be32_to_cpu(__raw_readl(addr))
+#define mmio_read16be(addr) be16_to_cpu((__be16 __force)__raw_readw(addr))
+#define mmio_read32be(addr) be32_to_cpu((__be32 __force)__raw_readl(addr))
#endif
unsigned int ioread8(void __iomem *addr)
@@ -106,8 +106,8 @@ EXPORT_SYMBOL(ioread32be);
#endif
#ifndef mmio_write16be
-#define mmio_write16be(val,port) __raw_writew(cpu_to_be16(val),port)
-#define mmio_write32be(val,port) __raw_writel(cpu_to_be32(val),port)
+#define mmio_write16be(val,port) __raw_writew((u16 __force)cpu_to_be16(val),port)
+#define mmio_write32be(val,port) __raw_writel((u32 __force)cpu_to_be32(val),port)
#endif
void iowrite8(u8 val, void __iomem *addr)
--
2.11.0
On Wed, Mar 21, 2018 at 10:37:36AM -0600, Logan Gunthorpe wrote:
> The semantics of the iowriteXXbe() functions are to write a
> value in CPU endianess to an IO register that is known by the
> caller to be in Big Endian. The mmio_writeXXbe() macro, which
> is called by iowriteXXbe(), should therefore use cpu_to_beXX()
> instead of beXX_to_cpu().
>
> Seeing both beXX_to_cpu() and cpu_to_beXX() are both functionally
> implemented as either null operations or swabXX operations there
> was no noticable bug here. But it is confusing for both developers
> and code analysis tools alike.
>
> Signed-off-by: Logan Gunthorpe <[email protected]>
> Cc: Philippe Ombredanne <[email protected]>
> Cc: Thomas Gleixner <[email protected]>
> Cc: Kate Stewart <[email protected]>
> Cc: Greg Kroah-Hartman <[email protected]>
> Cc: Luc Van Oostenryck <[email protected]>
> ---
> lib/iomap.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/lib/iomap.c b/lib/iomap.c
> index 541d926da95e..be120c13d6cc 100644
> --- a/lib/iomap.c
> +++ b/lib/iomap.c
> @@ -106,8 +106,8 @@ EXPORT_SYMBOL(ioread32be);
> #endif
>
> #ifndef mmio_write16be
> -#define mmio_write16be(val,port) __raw_writew(be16_to_cpu(val),port)
> -#define mmio_write32be(val,port) __raw_writel(be32_to_cpu(val),port)
> +#define mmio_write16be(val,port) __raw_writew(cpu_to_be16(val),port)
> +#define mmio_write32be(val,port) __raw_writel(cpu_to_be32(val),port)
> #endif
>
> void iowrite8(u8 val, void __iomem *addr)
> --
> 2.11.0
>
LGTM, feel free to add my
Reviewed-by: Luc Van Oostenryck <[email protected]>
-- Luc
On Wed, Mar 21, 2018 at 6:37 PM, Logan Gunthorpe <[email protected]> wrote:
>
> This is v13 of my cleanup series to push a number of instances of people
> defining their own io{read|write}64 functions into common headers seing
> they don't exist in non-64bit systems. This series adds inline functions to the
> io-64-nonatomic headers and then cleans up the drivers that defined their
> own copies.
>
> This cleanup was originally requested by Greg after he reviewed my
> Switchtec NTB code. And I hope someone can pick it up or at least give
> feedback on it soon as it's been around relatively unchanged for a few
> cycles now and I'm getting a bit tired of resubmitting it with little to
> no interest.
>
The first patches that I didn't reviewed before makes sense to me, the
rest (at the end) FWIW,
Reviewed-by: Andy Shevchenko <[email protected]>
> Thanks,
>
> Logan
>
> --
>
> Changes since v12:
> - Rebased onto v4.16-rc6
> - Split patch 0001 into two and reworked the commit log as requested
> by Luc Van Oostenryck
>
> Changes since v11:
> - Rebased onto v4.16-rc5
> - Added a patch (0001) to fix some old and new sparse warnings
> that the kbuild robot warned about this cycle. The latest version
> of sparse was required to reproduce these.
> - Added a patch (0002) to add io{read|write}64 to parisc which the kbuild
> robot also found errors for this cycle
>
> Changes since v10:
> - Rebased onto v4.16-rc4, this droped the drm/tilcdc patch which was
> picked up by that tree and is already in 4.16.
>
> Changes since v9:
> - Rebased onto v4.15-rc6
> - Fixed a couple of issues in the new version of the CAAM patch as
> pointed out by Horia
>
> Changes since v8:
> - Rebased onto v4.15-rc2, as a result rewrote patch 7 seeing someone did
> some similar cleanup in that area.
> - Added a patch to clean up the Switchtec NTB driver which landed in
> v4.15-rc1
>
> Changes since v7:
> - Fix minor nits from Andy Shevchenko
> - Rebased onto v4.14-rc1
>
> Changes since v6:
> ** none **
>
> Changes since v5:
> - Added a fix to the tilcdc driver to ensure it doesn't use the
> non-atomic operation. (This includes adding io{read|write}64[be]_is_nonatomic
> defines).
>
> Changes since v4:
> - Add functions so the powerpc implementation of iomap.c compiles. (As
> noticed by Horia)
>
> Changes since v3:
>
> - I noticed powerpc didn't use the appropriate functions seeing
> readq/writeq were not defined when iomap.h was included. Thus I've
> included a patch to adjust this
> - Fixed some mistakes with a couple of the defines in io-64-nonatomic*
> headers
> - Fixed a typo noticed by Horia.
>
> (earlier versions were drastically different)
>
> --
>
> Logan Gunthorpe (10):
> iomap: Use correct endian conversion function in mmio_writeXXbe
> iomap: Fix sparse endian check warnings
> parisc: iomap: introduce io{read|write}64
> powerpc: io.h: move iomap.h include so that it can use readq/writeq
> defs
> powerpc: iomap.c: introduce io{read|write}64_{lo_hi|hi_lo}
> iomap: introduce io{read|write}64_{lo_hi|hi_lo}
> io-64-nonatomic: add io{read|write}64[be]{_lo_hi|_hi_lo} macros
> ntb: ntb_hw_intel: use io-64-nonatomic instead of in-driver hacks
> crypto: caam: cleanup CONFIG_64BIT ifdefs when using io{read|write}64
> ntb: ntb_hw_switchtec: Cleanup 64bit IO defines to use the common
> header
>
> arch/parisc/include/asm/io.h | 9 +++
> arch/parisc/lib/iomap.c | 64 +++++++++++++++
> arch/powerpc/include/asm/io.h | 6 +-
> arch/powerpc/kernel/iomap.c | 40 ++++++++++
> drivers/crypto/caam/regs.h | 30 +------
> drivers/ntb/hw/intel/ntb_hw_intel.c | 30 +------
> drivers/ntb/hw/mscc/ntb_hw_switchtec.c | 36 +--------
> include/asm-generic/iomap.h | 26 ++++--
> include/linux/io-64-nonatomic-hi-lo.h | 64 +++++++++++++++
> include/linux/io-64-nonatomic-lo-hi.h | 64 +++++++++++++++
> lib/iomap.c | 140 ++++++++++++++++++++++++++++++++-
> 11 files changed, 409 insertions(+), 100 deletions(-)
>
> --
> 2.11.0
--
With Best Regards,
Andy Shevchenko
On 21/03/18 11:40 AM, Andy Shevchenko wrote:
> The first patches that I didn't reviewed before makes sense to me, the
> rest (at the end) FWIW,
>
> Reviewed-by: Andy Shevchenko <[email protected]>
Great, thanks!
Logan
On Wed, Mar 21, 2018 at 5:37 PM, Logan Gunthorpe <[email protected]> wrote:
> The semantics of the iowriteXXbe() functions are to write a
> value in CPU endianess to an IO register that is known by the
> caller to be in Big Endian. The mmio_writeXXbe() macro, which
> is called by iowriteXXbe(), should therefore use cpu_to_beXX()
> instead of beXX_to_cpu().
>
> Seeing both beXX_to_cpu() and cpu_to_beXX() are both functionally
> implemented as either null operations or swabXX operations there
> was no noticable bug here. But it is confusing for both developers
> and code analysis tools alike.
>
> Signed-off-by: Logan Gunthorpe <[email protected]>
Your patch is a clear improvement of what we had before, but I notice
that we have a weird asymmetry between big-endian and little-endian
accessors before and after this patch:
void iowrite32(u32 val, void __iomem *addr)
{
IO_COND(addr, outl(val,port), writel(val, addr));
}
void iowrite32be(u32 val, void __iomem *addr)
{
IO_COND(addr, pio_write32be(val,port), mmio_write32be(val, addr));
}
The little-endian iowrite32() when applied to mmio registers uses
a 32-bit wide atomic store to a little-endian register with barriers to
order against both spinlocks and DMA.
The big-endian iowrite32be() on the same pointer uses a nonatomic
store with no barriers whatsoever and the opposite endianess.
On most architectures, this is not important:
- For x86, the stores are aways atomic and no additional barriers
are needed, so the two are the same
- For ARM (both 32 and 64-bit), powerpc and many others, we don't
use the generic iowrite() and just fall back to writel() or
writel(swab32()).
However, shouldn't we just use the writel(swab32()) logic here as well
for the common case rather than risking missing barriers?
Arnd
On 26/03/18 04:53 AM, Arnd Bergmann wrote:
> On most architectures, this is not important:
> - For x86, the stores are aways atomic and no additional barriers
> are needed, so the two are the same
> - For ARM (both 32 and 64-bit), powerpc and many others, we don't
> use the generic iowrite() and just fall back to writel() or
> writel(swab32()).
>
> However, shouldn't we just use the writel(swab32()) logic here as well
> for the common case rather than risking missing barriers?
Hmm, I don't know... it's complicated?
Doing a bit of digging shows that the existing code was written during a
time when writel() did not include extra barriers over __raw_writel() in
any of the common arches.
The commit logs don't seem to provide any guidance as to why this it was
done this way, but I'd assume it was done to avoid a double swab() call
on BE arches. Seeing writel() is typically implemented as:
__raw_writel(__cpu_to_le32(value), addr);
Then on BE arches, writel(swab32()) would become:
__raw_writel(swab32(swab32(value)), addr)
Which seems undesirable.
Logan
On Mon, Mar 26, 2018 at 6:21 PM, Logan Gunthorpe <[email protected]> wrote:
>
>
> On 26/03/18 04:53 AM, Arnd Bergmann wrote:
>> On most architectures, this is not important:
>> - For x86, the stores are aways atomic and no additional barriers
>> are needed, so the two are the same
>> - For ARM (both 32 and 64-bit), powerpc and many others, we don't
>> use the generic iowrite() and just fall back to writel() or
>> writel(swab32()).
>>
>> However, shouldn't we just use the writel(swab32()) logic here as well
>> for the common case rather than risking missing barriers?
>
> Hmm, I don't know... it's complicated?
>
> Doing a bit of digging shows that the existing code was written during a
> time when writel() did not include extra barriers over __raw_writel() in
> any of the common arches.
>
> The commit logs don't seem to provide any guidance as to why this it was
> done this way, but I'd assume it was done to avoid a double swab() call
> on BE arches. Seeing writel() is typically implemented as:
>
> __raw_writel(__cpu_to_le32(value), addr);
>
> Then on BE arches, writel(swab32()) would become:
>
> __raw_writel(swab32(swab32(value)), addr)
>
> Which seems undesirable.
I wouldn't expect it to matter: the byte swap is almost always much
cheaper compared to the actual bus access for the MMIO, and I
would also guess that modern compilers can eliminate the double
byte swap on architectures where writel() is an inline function. Most of
the important architectures use ARCH_USE_BUILTIN_BSWAP, which
guarantees that.
Arnd
On 26/03/18 01:50 PM, Arnd Bergmann wrote:
> I wouldn't expect it to matter: the byte swap is almost always much
> cheaper compared to the actual bus access for the MMIO, and I
> would also guess that modern compilers can eliminate the double
> byte swap on architectures where writel() is an inline function. Most of
> the important architectures use ARCH_USE_BUILTIN_BSWAP, which
> guarantees that.
Fair enough. Sometime this week I'll update my patch set to change that.
Thanks,
Logan