Skip to contentSkip to conversation

net: emaclite: Fixes, enable driver for other archs

PATCH Superseded Samuel Obuch 5 participants 20 messages Jul 13, 2022 – Oct 5, 2022

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

These patches solve a few issues with the driver observed on a RISC-V platform.

#1 lore raw

PATCH 1/3 net: emaclite: fix broken build

Function ioremap_nocache seems to be defined only for mips and microblaze architectures. Therefore, the function call in the emaclite driver causes this driver to be unusable with other architectures, for example riscv.

---
 drivers/net/xilinx_emaclite.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/net/xilinx_emaclite.c +4 -0
index 6c9f1f7c27..5cd88e04fe 100644
@@ -615,8 +615,12 @@ static int emaclite_of_to_plat(struct udevice *dev)
 	int offset = 0;
 
 	pdata->iobase = dev_read_addr(dev);
+#if defined(CONFIG_MICROBLAZE) || defined(CONFIG_MIPS)
 	emaclite->regs = (struct emaclite_regs *)ioremap_nocache(pdata->iobase,
 								 0x10000);
+#else
+	emaclite->regs = (struct emaclite_regs *)pdata->iobase;
+#endif
 
 	emaclite->phyaddr = -1;
 
Signature · 1 line
2.31.1
#2 lore raw

Re: [PATCH 1/3] net: emaclite: fix broken build

Quoted 28 lines · Samuel Obuch

Function ioremap_nocache seems to be defined only for mips and microblaze architectures. Therefore, the function call in the emaclite driver causes this driver to be unusable with other architectures, for example riscv.

---
 drivers/net/xilinx_emaclite.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/net/xilinx_emaclite.c +4 -1
index 6c9f1f7c27..5cd88e04fe 100644
@@ -615,8 +615,12 @@ static int emaclite_of_to_plat(struct udevice *dev)
        int offset = 0;
 
        pdata->iobase = dev_read_addr(dev);
+#if defined(CONFIG_MICROBLAZE) || defined(CONFIG_MIPS)
        emaclite->regs = (struct emaclite_regs *)ioremap_nocache(pdata->iobase,
                                                                 0x10000);
+#else
+       emaclite->regs = (struct emaclite_regs *)pdata->iobase;
+#endif
 
        emaclite->phyaddr = -1;
 
--
Quoted 2 lines

2.31.1

Hm...
Well, this isn't right,The right solution is to replace ioremap_nocache() with ioremap().
This way it will work both for MIPS and other architectures. I can do it myself, you can fix your patch. let me know. Thanks,
Ramon.

maintainer
#3 lore raw

Re: [PATCH 1/3] net: emaclite: fix broken build

Quoted 28 lines · Ramon Fried
Quoted 27 lines · Samuel Obuch

Function ioremap_nocache seems to be defined only for mips and microblaze architectures. Therefore, the function call in the emaclite driver causes this driver to be unusable with other architectures, for example riscv.

---
  drivers/net/xilinx_emaclite.c | 4 ++++
  1 file changed, 4 insertions(+)
drivers/net/xilinx_emaclite.c +4 -0
index 6c9f1f7c27..5cd88e04fe 100644
@@ -615,8 +615,12 @@ static int emaclite_of_to_plat(struct udevice *dev)
         int offset = 0;
 
         pdata->iobase = dev_read_addr(dev);
+#if defined(CONFIG_MICROBLAZE) || defined(CONFIG_MIPS)
         emaclite->regs = (struct emaclite_regs *)ioremap_nocache(pdata->iobase,
                                                                  0x10000);
+#else
+       emaclite->regs = (struct emaclite_regs *)pdata->iobase;
+#endif
 
         emaclite->phyaddr = -1;
 
Quoted 8 lines
Quoted 3 lines

--
2.31.1

Hm...
Well, this isn't right,The right solution is to replace ioremap_nocache() with ioremap().
This way it will work both for MIPS and other architectures. I can do it myself, you can fix your patch. let me know.

Microblaze doesn't define it now. But I agree that using ioremap which has implicit nocache is the right way to go. It means please create the first patch which creates ioremap for microblaze,
Then second to replace ioremap_nocache() in emaclite driver to ioremap. And third to remove ioremap_nocache from microblaze io.h.

Thanks,
Michal

maintainer
#4 lore raw

Re: [PATCH 1/3] net: emaclite: fix broken build

Hi,

Quoted 40 lines · Michal Simek
Quoted 28 lines · Ramon Fried
Quoted 27 lines · Samuel Obuch

Function ioremap_nocache seems to be defined only for mips and microblaze architectures. Therefore, the function call in the emaclite driver causes this driver to be unusable with other architectures, for example riscv.

---
  drivers/net/xilinx_emaclite.c | 4 ++++

  1 file changed, 4 insertions(+)
drivers/net/xilinx_emaclite.c +4 -0
index 6c9f1f7c27..5cd88e04fe 100644
@@ -615,8 +615,12 @@ static int emaclite_of_to_plat(struct udevice *dev)
          int offset = 0;
 
          pdata->iobase = dev_read_addr(dev);
+#if defined(CONFIG_MICROBLAZE) || defined(CONFIG_MIPS)
          emaclite->regs = (struct emaclite_regs *)ioremap_nocache(pdata->iobase,
                                                                   0x10000);
+#else
+       emaclite->regs = (struct emaclite_regs *)pdata->iobase;
+#endif
 
          emaclite->phyaddr = -1;
 
Quoted 8 lines
Quoted 3 lines
Signature · 1 line
2.31.1

Hm...
Well, this isn't right,The right solution is to replace ioremap_nocache() with ioremap().
This way it will work both for MIPS and other architectures. I can do it myself, you can fix your patch. let me know.

Quoted 3 lines

Microblaze doesn't define it now. But I agree that using ioremap which has implicit nocache is the right way to go. It means please create the first patch which creates ioremap for microblaze,
Then second to replace ioremap_nocache() in emaclite driver to ioremap. And third to remove ioremap_nocache from microblaze io.h.

I did closer look when I looked at other patches. You should switch to linux/io.h which automatically create ioremap if not defined by architecture.

And ioremap_nocache for microblaze can be removed later.

drivers/net/xilinx_emaclite.c +1 -1
index 48aee77ab509..3299eefd999f 100644
@@ -21,7 +21,7 @@
 #include <linux/delay.h>
 #include <linux/errno.h>
 #include <linux/kernel.h>
-#include <asm/io.h>
+#include <linux/io.h>
 

M

#5 lore raw

Re: [PATCH 1/3] net: emaclite: fix broken build

Hi, using linux/io.h and ioremap works well for us, thanks. I will update the patch.

S

Quoted 78 lines · Michal Simek

Hi,

Quoted 4 lines · Michal Simek
Quoted 1 line · Ramon Fried

On Wed, Jul 13, 2022 at 5:02 PM Samuel Obuch <[email protected]>

Quoted 2 lines · wrote:
Quoted 2 lines
Quoted 2 lines

Function ioremap_nocache seems to be defined only for mips and

microblaze

Quoted 1 line
Quoted 1 line
Quoted 1 line

architectures. Therefore, the function call in the emaclite driver

causes

Quoted 8 lines
Quoted 8 lines

Review

Quoted 8 lines

this driver to be unusable with other architectures, for example riscv.

---
  drivers/net/xilinx_emaclite.c | 4 ++++
  1 file changed, 4 insertions(+)
(quoted diff) +0 -0

b/drivers/net/xilinx_emaclite.c

Quoted 4 lines
Quoted 4 lines

drivers/net/xilinx_emaclite.c static int emaclite_of_to_plat(struct udevice

Quoted 4 lines

index 6c9f1f7c27..5cd88e04fe 100644

drivers/net/xilinx_emaclite.c +0 -0
@@ -615,8 +615,12 @@ static int emaclite_of_to_plat(struct udevice

*dev)

Quoted 5 lines
Quoted 5 lines

drivers/net/xilinx_emaclite.c static int emaclite_of_to_plat(struct udevice · line 619

Quoted 5 lines
drivers/net/xilinx_emaclite.c +1 -0
         int offset = 0;
 
         pdata->iobase = dev_read_addr(dev);
+#if defined(CONFIG_MICROBLAZE) || defined(CONFIG_MIPS)
         emaclite->regs = (struct emaclite_regs

*)ioremap_nocache(pdata->iobase,

Quoted 1 line
Quoted 1 line
Quoted 1 line

0x10000);

Quoted 15 lines
Quoted 6 lines
Quoted 6 lines
drivers/net/xilinx_emaclite.c +3 -0
+#else
+       emaclite->regs = (struct emaclite_regs *)pdata->iobase;
+#endif
 
         emaclite->phyaddr = -1;
Quoted 8 lines
Quoted 3 lines

--
2.31.1

Hm...
Well, this isn't right,The right solution is to replace ioremap_nocache() with ioremap().
This way it will work both for MIPS and other architectures. I can do it myself, you can fix your patch. let me know.

Quoted 1 line

Microblaze doesn't define it now. But I agree that using ioremap which

has

Quoted 2 lines

implicit nocache is the right way to go.
It means please create the first patch which creates ioremap for

microblaze,

Quoted 1 line

Then second to replace ioremap_nocache() in emaclite driver to ioremap.

And

Quoted 1 line

third to remove ioremap_nocache from microblaze io.h.

I did closer look when I looked at other patches. You should switch to linux/io.h which automatically create ioremap if not defined by architecture.

And ioremap_nocache for microblaze can be removed later.

drivers/net/xilinx_emaclite.c +1 -1
index 48aee77ab509..3299eefd999f 100644
@@ -21,7 +21,7 @@
  #include <linux/delay.h>
  #include <linux/errno.h>
  #include <linux/kernel.h>
-#include <asm/io.h>
+#include <linux/io.h>
 

M

#6 lore raw

PATCH 2/3 net: emaclite: fix xemaclite_alignedread/write functions

In reply to Samuel Obuch

Use __raw_read* and __raw_write* functions to ensure read/write is passed to the memory-mapped regions, as non-volatile accesses may get optimised out.

---
 drivers/net/xilinx_emaclite.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)
drivers/net/xilinx_emaclite.c +4 -5
index 5cd88e04fe..de7a2dee0b 100644
@@ -113,12 +113,12 @@ static void xemaclite_alignedread(u32 *srcptr, void *destptr, u32 bytecount)
 	/* Word aligned buffer, no correction needed. */
 	to32ptr = (u32 *) destptr;
 	while (bytecount > 3) {
-		*to32ptr++ = *from32ptr++;
+		*to32ptr++ = __raw_readl(from32ptr++);
 		bytecount -= 4;
 	}
 	to8ptr = (u8 *) to32ptr;
 
-	alignbuffer = *from32ptr++;
+	alignbuffer = __raw_readl(from32ptr++);
 	from8ptr = (u8 *) &alignbuffer;
 
 	for (i = 0; i < bytecount; i++)
@@ -136,8 +136,7 @@ static void xemaclite_alignedwrite(void *srcptr, u32 *destptr, u32 bytecount)
 
 	from32ptr = (u32 *) srcptr;
 	while (bytecount > 3) {
-
-		*to32ptr++ = *from32ptr++;
+		__raw_writel(*from32ptr++, to32ptr++);
 		bytecount -= 4;
 	}
 
@@ -148,7 +147,7 @@ static void xemaclite_alignedwrite(void *srcptr, u32 *destptr, u32 bytecount)
 	for (i = 0; i < bytecount; i++)
 		*to8ptr++ = *from8ptr++;
 
-	*to32ptr++ = alignbuffer;
+	__raw_writel(alignbuffer, to32ptr++);
 }
 
 static int wait_for_bit(const char *func, u32 *reg, const u32 mask,
Signature · 1 line
2.31.1
#7 lore raw

Re: [PATCH 2/3] net: emaclite: fix xemaclite_alignedread/write functions

Quoted 49 lines · Samuel Obuch

Use __raw_read* and __raw_write* functions to ensure read/write is passed to the memory-mapped regions, as non-volatile accesses may get optimised out.

---
 drivers/net/xilinx_emaclite.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)
drivers/net/xilinx_emaclite.c +4 -6
index 5cd88e04fe..de7a2dee0b 100644
@@ -113,12 +113,12 @@ static void xemaclite_alignedread(u32 *srcptr, void *destptr, u32 bytecount)
        /* Word aligned buffer, no correction needed. */
        to32ptr = (u32 *) destptr;
        while (bytecount > 3) {
-               *to32ptr++ = *from32ptr++;
+               *to32ptr++ = __raw_readl(from32ptr++);
                bytecount -= 4;
        }
        to8ptr = (u8 *) to32ptr;
 
-       alignbuffer = *from32ptr++;
+       alignbuffer = __raw_readl(from32ptr++);
        from8ptr = (u8 *) &alignbuffer;
 
        for (i = 0; i < bytecount; i++)
@@ -136,8 +136,7 @@ static void xemaclite_alignedwrite(void *srcptr, u32 *destptr, u32 bytecount)
 
        from32ptr = (u32 *) srcptr;
        while (bytecount > 3) {
-
-               *to32ptr++ = *from32ptr++;
+               __raw_writel(*from32ptr++, to32ptr++);
                bytecount -= 4;
        }
 
@@ -148,7 +147,7 @@ static void xemaclite_alignedwrite(void *srcptr, u32 *destptr, u32 bytecount)
        for (i = 0; i < bytecount; i++)
                *to8ptr++ = *from8ptr++;
 
-       *to32ptr++ = alignbuffer;
+       __raw_writel(alignbuffer, to32ptr++);
 }
 
 static int wait_for_bit(const char *func, u32 *reg, const u32 mask,
--
Quoted 2 lines

2.31.1

I think that using readl/writel is fine, no need for raw_...

maintainer
#8 lore raw

Re: [PATCH 2/3] net: emaclite: fix xemaclite_alignedread/write functions

Quoted 48 lines · Ramon Fried
Quoted 47 lines · Samuel Obuch

Use __raw_read* and __raw_write* functions to ensure read/write is passed to the memory-mapped regions, as non-volatile accesses may get optimised out.

---
  drivers/net/xilinx_emaclite.c | 9 ++++-----
  1 file changed, 4 insertions(+), 5 deletions(-)
drivers/net/xilinx_emaclite.c +4 -5
index 5cd88e04fe..de7a2dee0b 100644
@@ -113,12 +113,12 @@ static void xemaclite_alignedread(u32 *srcptr, void *destptr, u32 bytecount)
         /* Word aligned buffer, no correction needed. */
         to32ptr = (u32 *) destptr;
         while (bytecount > 3) {
-               *to32ptr++ = *from32ptr++;
+               *to32ptr++ = __raw_readl(from32ptr++);
                 bytecount -= 4;
         }
         to8ptr = (u8 *) to32ptr;
 
-       alignbuffer = *from32ptr++;
+       alignbuffer = __raw_readl(from32ptr++);
         from8ptr = (u8 *) &alignbuffer;
 
         for (i = 0; i < bytecount; i++)
@@ -136,8 +136,7 @@ static void xemaclite_alignedwrite(void *srcptr, u32 *destptr, u32 bytecount)
 
         from32ptr = (u32 *) srcptr;
         while (bytecount > 3) {
-
-               *to32ptr++ = *from32ptr++;
+               __raw_writel(*from32ptr++, to32ptr++);
                 bytecount -= 4;
         }
 
@@ -148,7 +147,7 @@ static void xemaclite_alignedwrite(void *srcptr, u32 *destptr, u32 bytecount)
         for (i = 0; i < bytecount; i++)
                 *to8ptr++ = *from8ptr++;
 
-       *to32ptr++ = alignbuffer;
+       __raw_writel(alignbuffer, to32ptr++);
  }
 
Quoted 5 lines
Quoted 4 lines
drivers/net/xilinx_emaclite.c +0 -1
  static int wait_for_bit(const char *func, u32 *reg, const u32 mask,
--

2.31.1

I think that using readl/writel is fine, no need for raw_...

For microblaze that should be fine but let me ask my team to rest it on ARM.
I think that __raw version are safer because this IP can also run on big endian systems and I think that was the reason why readl/writel wasn't used in past.

Thanks,
Michal

#9 lore raw

Re: [PATCH 2/3] net: emaclite: fix xemaclite_alignedread/write functions

Quoted 58 lines · Michal Simek
Quoted 48 lines · Ramon Fried
Quoted 47 lines · Samuel Obuch

Use __raw_read* and __raw_write* functions to ensure read/write is passed to the memory-mapped regions, as non-volatile accesses may get optimised out.

---
  drivers/net/xilinx_emaclite.c | 9 ++++-----
  1 file changed, 4 insertions(+), 5 deletions(-)
drivers/net/xilinx_emaclite.c +4 -5
index 5cd88e04fe..de7a2dee0b 100644
@@ -113,12 +113,12 @@ static void xemaclite_alignedread(u32 *srcptr, void *destptr, u32 bytecount)
         /* Word aligned buffer, no correction needed. */
         to32ptr = (u32 *) destptr;
         while (bytecount > 3) {
-               *to32ptr++ = *from32ptr++;
+               *to32ptr++ = __raw_readl(from32ptr++);
                 bytecount -= 4;
         }
         to8ptr = (u8 *) to32ptr;
 
-       alignbuffer = *from32ptr++;
+       alignbuffer = __raw_readl(from32ptr++);
         from8ptr = (u8 *) &alignbuffer;
 
         for (i = 0; i < bytecount; i++)
@@ -136,8 +136,7 @@ static void xemaclite_alignedwrite(void *srcptr, u32 *destptr, u32 bytecount)
 
         from32ptr = (u32 *) srcptr;
         while (bytecount > 3) {
-
-               *to32ptr++ = *from32ptr++;
+               __raw_writel(*from32ptr++, to32ptr++);
                 bytecount -= 4;
         }
 
@@ -148,7 +147,7 @@ static void xemaclite_alignedwrite(void *srcptr, u32 *destptr, u32 bytecount)
         for (i = 0; i < bytecount; i++)
                 *to8ptr++ = *from8ptr++;
 
-       *to32ptr++ = alignbuffer;
+       __raw_writel(alignbuffer, to32ptr++);
  }
 
Quoted 5 lines
Quoted 4 lines
drivers/net/xilinx_emaclite.c +0 -1
  static int wait_for_bit(const char *func, u32 *reg, const u32 mask,
--

2.31.1

I think that using readl/writel is fine, no need for raw_...

Quoted 6 lines

For microblaze that should be fine but let me ask my team to rest it on ARM. I think that __raw version are safer because this IP can also run on big endian systems and I think that was the reason why readl/writel wasn't used in past.

Thanks,
Michal

Hi Michal,

Do you have any new information on this? For the v2, it would be good to have this resolved.

Thanks,
Jan

maintainer
#10 lore raw

Re: [PATCH 2/3] net: emaclite: fix xemaclite_alignedread/write functions

Quoted 63 lines · Jan Remes
Quoted 62 lines · Michal Simek
Quoted 48 lines · Ramon Fried
Quoted 47 lines · Samuel Obuch

Use __raw_read* and __raw_write* functions to ensure read/write is passed to the memory-mapped regions, as non-volatile accesses may get optimised out.

---
   drivers/net/xilinx_emaclite.c | 9 ++++-----
   1 file changed, 4 insertions(+), 5 deletions(-)
drivers/net/xilinx_emaclite.c +4 -5
index 5cd88e04fe..de7a2dee0b 100644
@@ -113,12 +113,12 @@ static void xemaclite_alignedread(u32 *srcptr, void *destptr, u32 bytecount)
          /* Word aligned buffer, no correction needed. */
          to32ptr = (u32 *) destptr;
          while (bytecount > 3) {
-               *to32ptr++ = *from32ptr++;
+               *to32ptr++ = __raw_readl(from32ptr++);
                  bytecount -= 4;
          }
          to8ptr = (u8 *) to32ptr;
 
-       alignbuffer = *from32ptr++;
+       alignbuffer = __raw_readl(from32ptr++);
          from8ptr = (u8 *) &alignbuffer;
 
          for (i = 0; i < bytecount; i++)
@@ -136,8 +136,7 @@ static void xemaclite_alignedwrite(void *srcptr, u32 *destptr, u32 bytecount)
 
          from32ptr = (u32 *) srcptr;
          while (bytecount > 3) {
-
-               *to32ptr++ = *from32ptr++;
+               __raw_writel(*from32ptr++, to32ptr++);
                  bytecount -= 4;
          }
 
@@ -148,7 +147,7 @@ static void xemaclite_alignedwrite(void *srcptr, u32 *destptr, u32 bytecount)
          for (i = 0; i < bytecount; i++)
                  *to8ptr++ = *from8ptr++;
 
-       *to32ptr++ = alignbuffer;
+       __raw_writel(alignbuffer, to32ptr++);
   }
 
Quoted 5 lines
Quoted 4 lines
drivers/net/xilinx_emaclite.c +0 -1
   static int wait_for_bit(const char *func, u32 *reg, const u32 mask,
--

2.31.1

I think that using readl/writel is fine, no need for raw_...

For microblaze that should be fine but let me ask my team to rest it on ARM. I think that __raw version are safer because this IP can also run on big endian systems and I think that was the reason why readl/writel wasn't used in past.

Quoted 7 lines
Quoted 2 lines

Thanks,
Michal

Hi Michal,

Do you have any new information on this? For the v2, it would be good to have this resolved.

we are not testing emaclite on any ARM design. But in Linux you can find in this driver.

#ifdef __BIG_ENDIAN
#define xemaclite_readl ioread32be
#define xemaclite_writel iowrite32be
#else
#define xemaclite_readl ioread32
#define xemaclite_writel iowrite32
#endif

If you keep __raw variants it will ensure that native endian access is doing to be used.
On ARM IIRC readl/writel also have barriers. Origin patch is also using raw variant that's why I expect it is working on your system.

Thanks,
Michal

#11 lore raw

Re: [PATCH 2/3] net: emaclite: fix xemaclite_alignedread/write functions

Hi, I tested both versions to be sure, but the results are as can be expected:

1. both __raw_readl/__raw_writel and readl/writel functions work ok on riscv - only the original nonvolatile accesses were problematic
2. the additional barrier in readl/writel functions can introduce noticeable slowdown - e.g. with our setup, download speed over tftp for the same image decreased from 460 KiB/s to 118 KiB/s

Can we conclude to keep the raw versions in the patch?

Thanks,
Samuel

Quoted 107 lines · Michal Simek
Quoted 1 line · Jan Remes

On Mon, Aug 8, 2022 at 10:05 AM Michal Simek <[email protected]>

Quoted 5 lines · wrote:
Quoted 5 lines · Ramon Fried
Quoted 1 line

On Wed, Jul 13, 2022 at 5:02 PM Samuel Obuch <[email protected]>

Quoted 11 lines · wrote:
Quoted 11 lines
Quoted 11 lines

Review

Quoted 11 lines

Use __raw_read* and __raw_write* functions to ensure read/write is passed to the memory-mapped regions, as non-volatile accesses may get optimised out.

---
   drivers/net/xilinx_emaclite.c | 9 ++++-----
   1 file changed, 4 insertions(+), 5 deletions(-)
(quoted diff) +0 -0

b/drivers/net/xilinx_emaclite.c

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

drivers/net/xilinx_emaclite.c static void xemaclite_alignedread(u32 *srcptr,

Quoted 4 lines

index 5cd88e04fe..de7a2dee0b 100644

drivers/net/xilinx_emaclite.c +0 -0
@@ -113,12 +113,12 @@ static void xemaclite_alignedread(u32 *srcptr,

void *destptr, u32 bytecount)

Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
drivers/net/xilinx_emaclite.c +1 -1
          /* Word aligned buffer, no correction needed. */
          to32ptr = (u32 *) destptr;
          while (bytecount > 3) {
-               *to32ptr++ = *from32ptr++;
+               *to32ptr++ = __raw_readl(from32ptr++);
                  bytecount -= 4;
          }
          to8ptr = (u8 *) to32ptr;
 
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

drivers/net/xilinx_emaclite.c static void xemaclite_alignedwrite(void *srcptr,

Quoted 6 lines
drivers/net/xilinx_emaclite.c +1 -1
-       alignbuffer = *from32ptr++;
+       alignbuffer = __raw_readl(from32ptr++);
          from8ptr = (u8 *) &alignbuffer;
 
          for (i = 0; i < bytecount; i++)
@@ -136,8 +136,7 @@ static void xemaclite_alignedwrite(void *srcptr,

u32 *destptr, u32 bytecount)

Quoted 10 lines
Quoted 10 lines
Quoted 10 lines

drivers/net/xilinx_emaclite.c static void xemaclite_alignedwrite(void *srcptr,

Quoted 10 lines
drivers/net/xilinx_emaclite.c +1 -2
          from32ptr = (u32 *) srcptr;
          while (bytecount > 3) {
-
-               *to32ptr++ = *from32ptr++;
+               __raw_writel(*from32ptr++, to32ptr++);
                  bytecount -= 4;
          }
 
@@ -148,7 +147,7 @@ static void xemaclite_alignedwrite(void *srcptr,

u32 *destptr, u32 bytecount)

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
drivers/net/xilinx_emaclite.c +1 -1
          for (i = 0; i < bytecount; i++)
                  *to8ptr++ = *from8ptr++;
 
-       *to32ptr++ = alignbuffer;
+       __raw_writel(alignbuffer, to32ptr++);
   }
 
Quoted 7 lines
Quoted 7 lines
Quoted 5 lines
Quoted 4 lines
drivers/net/xilinx_emaclite.c +0 -1
   static int wait_for_bit(const char *func, u32 *reg, const u32 mask,
--

2.31.1

I think that using readl/writel is fine, no need for raw_...

For microblaze that should be fine but let me ask my team to rest it on

ARM.

Quoted 1 line
Quoted 1 line

I think that __raw version are safer because this IP can also run on

big endian

Quoted 1 line
Quoted 1 line

systems and I think that was the reason why readl/writel wasn't used in

past.

Quoted 8 lines
Quoted 3 lines

Thanks,
Michal

Hi Michal,

Do you have any new information on this? For the v2, it would be good to have this resolved.

we are not testing emaclite on any ARM design. But in Linux you can find in this
driver.

#ifdef __BIG_ENDIAN
#define xemaclite_readl ioread32be
#define xemaclite_writel iowrite32be
#else
#define xemaclite_readl ioread32
#define xemaclite_writel iowrite32
#endif

If you keep __raw variants it will ensure that native endian access is doing to
be used.
On ARM IIRC readl/writel also have barriers. Origin patch is also using raw
variant that's why I expect it is working on your system.

Thanks,
Michal

maintainer
#12 lore raw

Re: [PATCH 2/3] net: emaclite: fix xemaclite_alignedread/write functions

Hi,

Quoted 6 lines · Samuel Obuch

Hi, I tested both versions to be sure, but the results are as can be expected:

1. both __raw_readl/__raw_writel and readl/writel functions work ok on riscv - only the original nonvolatile accesses were problematic
2. the additional barrier in readl/writel functions can introduce noticeable slowdown - e.g. with our setup, download speed over tftp for the same image decreased from 460 KiB/s to 118 KiB/s

Can we conclude to keep the raw versions in the patch?

I prefer raw versions too.

M

#13 lore raw

Re: [PATCH 2/3] net: emaclite: fix xemaclite_alignedread/write functions

Quoted 12 lines · Michal Simek

Hi,

Quoted 8 lines · Samuel Obuch

Hi, I tested both versions to be sure, but the results are as can be expected:

1. both __raw_readl/__raw_writel and readl/writel functions work ok on riscv - only the original nonvolatile accesses were problematic
2. the additional barrier in readl/writel functions can introduce noticeable slowdown - e.g. with our setup, download speed over tftp for the same image decreased from 460 KiB/s to 118 KiB/s

Quoted 5 lines
Quoted 1 line

Can we conclude to keep the raw versions in the patch?

I prefer raw versions too.

Let's keep it raw.

#14 lore raw

PATCH 3/3 net: emaclite: fix handling for IP packets with specific lengths

In reply to Samuel Obuch

The maximum length is capped similarly to the emaclite_send function. Avoid integer underflow for values of ip->ip_len < 30, the minimum length of an IP packet is 21 bytes.

---
 drivers/net/xilinx_emaclite.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/xilinx_emaclite.c +3 -1
index de7a2dee0b..21c450ec46 100644
@@ -518,6 +518,8 @@ try_again:
 		length = ntohs(ip->ip_len);
 		length += ETHER_HDR_SIZE + ETH_FCS_LEN;
 		debug("IP Packet %x\n", length);
+		if (length > PKTSIZE)
+			length = PKTSIZE;
 		break;
 	default:
 		debug("Other Packet\n");
@@ -526,7 +528,7 @@ try_again:
 	}
 
 	/* Read the rest of the packet which is longer then first read */
-	if (length != first_read)
+	if (length > first_read)
 		xemaclite_alignedread(addr + first_read,
 				      etherrxbuff + first_read,
 				      length - first_read);
Signature · 1 line
2.31.1
#15 lore raw

Re: [PATCH 3/3] net: emaclite: fix handling for IP packets with specific lengths

Quoted 33 lines · Samuel Obuch

The maximum length is capped similarly to the emaclite_send function. Avoid integer underflow for values of ip->ip_len < 30, the minimum length of an IP packet is 21 bytes.

---
 drivers/net/xilinx_emaclite.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/xilinx_emaclite.c +3 -2
index de7a2dee0b..21c450ec46 100644
@@ -518,6 +518,8 @@ try_again:
                length = ntohs(ip->ip_len);
                length += ETHER_HDR_SIZE + ETH_FCS_LEN;
                debug("IP Packet %x\n", length);
+               if (length > PKTSIZE)
+                       length = PKTSIZE;
                break;
        default:
                debug("Other Packet\n");
@@ -526,7 +528,7 @@ try_again:
        }
 
        /* Read the rest of the packet which is longer then first read */
-       if (length != first_read)
+       if (length > first_read)
                xemaclite_alignedread(addr + first_read,
                                      etherrxbuff + first_read,
                                      length - first_read);
--
Quoted 2 lines

2.31.1

+ Michal

maintainer Reviewed-by
#16 lore raw

Re: [PATCH 3/3] net: emaclite: fix handling for IP packets with specific lengths

Quoted 27 lines · Ramon Fried
Quoted 26 lines · Samuel Obuch

The maximum length is capped similarly to the emaclite_send function. Avoid integer underflow for values of ip->ip_len < 30, the minimum length of an IP packet is 21 bytes.

---
  drivers/net/xilinx_emaclite.c | 4 +++-
  1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/xilinx_emaclite.c +2 -0
index de7a2dee0b..21c450ec46 100644
@@ -518,6 +518,8 @@ try_again:
                 length = ntohs(ip->ip_len);
                 length += ETHER_HDR_SIZE + ETH_FCS_LEN;
                 debug("IP Packet %x\n", length);
+               if (length > PKTSIZE)
+                       length = PKTSIZE;
                 break;
         default:
                 debug("Other Packet\n");
@@ -526,7 +528,7 @@ try_again:
         }
 
Quoted 10 lines
Quoted 9 lines
drivers/net/xilinx_emaclite.c +1 -2
         /* Read the rest of the packet which is longer then first read */
-       if (length != first_read)
+       if (length > first_read)
                 xemaclite_alignedread(addr + first_read,
                                       etherrxbuff + first_read,
                                       length - first_read);
--

2.31.1

+ Michal

This looks good.

Thanks,
Michal

maintainer
#17 lore raw
In reply to Samuel Obuch

Hi,

Quoted 1 line · Samuel Obuch

These patches solve a few issues with the driver observed on a RISC-V platform.

How does that system look like that you are using standard Xilinx IPs? Do you use any other standard Xilinx IP like console?

Is it available somewhere that I can run it too?

Thanks,
Michal

#18 lore raw

Hi,

its a soft-core in an FPGA. Other than the ethernet, I think we only use Xilinx bridges.

Unfortunately theres no publicly available version.

S.

Quoted 14 lines · Michal Simek

Hi,

Quoted 1 line · Samuel Obuch

These patches solve a few issues with the driver observed on a RISC-V

platform.

How does that system look like that you are using standard Xilinx IPs? Do you use any other standard Xilinx IP like console?

Is it available somewhere that I can run it too?

Thanks,
Michal

#19 lore raw

Hi Samuel,

po 8. 8. 2022 v 10:34 odesílatel Samuel Obuch <[email protected]> napsal:

Quoted 22 lines

Hi,

its a soft-core in an FPGA. Other than the ethernet, I think we only use Xilinx bridges.

Unfortunately theres no publicly available version.

S.

Quoted 10 lines · Michal Simek

Hi,

Quoted 1 line · Samuel Obuch

These patches solve a few issues with the driver observed on a RISC-V

platform.

How does that system look like that you are using standard Xilinx IPs? Do you use any other standard Xilinx IP like console?

Is it available somewhere that I can run it too?

Are you going to send a new version of this patchset?

Thanks,
Michal

Signature · 5 lines
Michal Simek, Ing. (M.Eng), OpenPGP -> KeyID: FE3D1F91
w: www.monstr.eu p: +42-0-721842854
Maintainer of Linux kernel - Xilinx Microblaze
Maintainer of Linux kernel - Xilinx Zynq ARM and ZynqMP ARM64 SoCs
U-Boot custodian - Xilinx Microblaze/Zynq/ZynqMP/Versal SoCs