Skip to contentSkip to conversation

semihosting: Improve ARMv7A support

PATCH New Andrey Smirnov 2 participants 3 messages Aug 8, 2015 – Sep 12, 2015 Arm (32-bit)

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 semihosting: Improve ARMv7A support

Previous implementation of semihosting didn't account for cases where doing an SVC call would clobber various data(most notable case would be 'lr' and 'spsr' when doing SVC call in Supervisor mode). This patch is an adaptation of the code from Newlib's Angel_SWI feature (can be found in libc/sys/arm/swi.h) which hopefuly fixes the problem.

Tested with modified OpenOCD and custom Vybrid VF610 based board

---
 arch/arm/lib/semihosting.c | 30 ++++++++++++++++++++++++++----
 1 file changed, 26 insertions(+), 4 deletions(-)
arch/arm/lib/semihosting.c +26 -4
index c3e964e..5f22c2d 100644
@@ -26,18 +26,40 @@
 /*
  * Call the handler
  */
+#if defined(CONFIG_ARM64)
 static noinline long smh_trap(unsigned int sysnum, void *addr)
 {
 	register long result asm("r0");
-#if defined(CONFIG_ARM64)
 	asm volatile ("hlt #0xf000" : "=r" (result) : "0"(sysnum), "r"(addr));
+	return result;
+}
 #else
-	/* Note - untested placeholder */
-	asm volatile ("svc #0x123456" : "=r" (result) : "0"(sysnum), "r"(addr));
+#if defined (CONFIG_SYS_THUMB_BUILD)
+#error "Support for semihosting in THUMB mode is not implemented"
 #endif
+/*
+ * Call the handler
+ */
+static noinline long smh_trap(unsigned int sysnum, void *addr)
+{
+	int result;
+
+	asm volatile ("mov r0, %1"	"\n\t"
+		      "mov r1, %2"	"\n\t"
+		      "svc #0x123456"	"\n\t"
+		      "mov %0, r0"	"\n\t"
+		      : "=r" (result)
+		      : "r" (sysnum), "r" (addr)
+		      : "r0", "r1", "r2", "r3", "ip", "lr", "memory", "cc");
+		      /* Clobbers r0 and r1, and lr if in supervisor
+			 mode Accordingly to page 13-77 of ARM DUI
+			 0040D other registers can also be clobbered.
+			 Some memory positions may also be changed by
+			 a system call, so they should not be kept in
+			 registers.  */
 	return result;
 }
-
+#endif
 /*
  * Open a file on the host. Mode is "r" or "rb" currently. Returns a file
  * descriptor or -1 on error.

--
2.1.4

#1 lore raw

PATCH 1/1 semihosting: Improve ARMv7A support

Hello Andrey,

Two nitpicks:

Quoted 18 lines · Andrey Smirnov

Previous implementation of semihosting didn't account for cases where doing an SVC call would clobber various data(most notable case would be 'lr' and 'spsr' when doing SVC call in Supervisor mode). This patch is an adaptation of the code from Newlib's Angel_SWI feature (can be found in libc/sys/arm/swi.h) which hopefuly fixes the problem.

Tested with modified OpenOCD and custom Vybrid VF610 based board

---
 arch/arm/lib/semihosting.c | 30 ++++++++++++++++++++++++++----
 1 file changed, 26 insertions(+), 4 deletions(-)
arch/arm/lib/semihosting.c +0 -0
index c3e964e..5f22c2d 100644

arch/arm/lib/semihosting.c line 32

Quoted 8 lines
arch/arm/lib/semihosting.c +1 -1
@@ -26,18 +26,40 @@
 /*
  * Call the handler
  */
+#if defined(CONFIG_ARM64)
 static noinline long smh_trap(unsigned int sysnum, void *addr)
 {
 	register long result asm("r0");
-#if defined(CONFIG_ARM64)

What's the point of moving the #ifdef up? It just forces you to duplicate the function start with no discernable benefit since the function name and arguments are the same in the 64 and 32 bit case.

arch/arm/lib/semihosting.c line 42

Quoted 12 lines
arch/arm/lib/semihosting.c +7 -2
 	asm volatile ("hlt #0xf000" : "=r" (result) : "0"(sysnum), "r"(addr));
+	return result;
+}
 #else
-	/* Note - untested placeholder */
-	asm volatile ("svc #0x123456" : "=r" (result) : "0"(sysnum), "r"(addr));
+#if defined (CONFIG_SYS_THUMB_BUILD)
+#error "Support for semihosting in THUMB mode is not implemented"
 #endif
+/*
+ * Call the handler
+ */

No point in duplicating this comment.

Quoted 25 lines
arch/arm/lib/semihosting.c +18 -2
+static noinline long smh_trap(unsigned int sysnum, void *addr)
+{
+	int result;
+
+	asm volatile ("mov r0, %1"	"\n\t"
+		      "mov r1, %2"	"\n\t"
+		      "svc #0x123456"	"\n\t"
+		      "mov %0, r0"	"\n\t"
+		      : "=r" (result)
+		      : "r" (sysnum), "r" (addr)
+		      : "r0", "r1", "r2", "r3", "ip", "lr", "memory", "cc");
+		      /* Clobbers r0 and r1, and lr if in supervisor
+			 mode Accordingly to page 13-77 of ARM DUI
+			 0040D other registers can also be clobbered.
+			 Some memory positions may also be changed by
+			 a system call, so they should not be kept in
+			 registers.  */
 	return result;
 }
-
+#endif
 /*
  * Open a file on the host. Mode is "r" or "rb" currently. Returns a file
  * descriptor or -1 on error.
--
Quoted 5 lines

2.1.4
_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
http://lists.denx.de/mailman/listinfo/u-boot

Amicalement,

Signature · 1 line
Albert.
#2 lore raw

PATCH 1/1 semihosting: Improve ARMv7A support

Quoted 4 lines · Albert ARIBAUD

Hello Andrey,

Two nitpicks:

Quoted 10 lines

arch/arm/lib/semihosting.c line 32

Quoted 5 lines
arch/arm/lib/semihosting.c +1 -1
+#if defined(CONFIG_ARM64)
 static noinline long smh_trap(unsigned int sysnum, void *addr)
 {
      register long result asm("r0");
-#if defined(CONFIG_ARM64)

What's the point of moving the #ifdef up? It just forces you to duplicate the function start with no discernable benefit since the function name and arguments are the same in the 64 and 32 bit case.

The point is to improve readability of the code. With #ifdef moved up you have to figure out the version of the code you care about first and then you can read the entirety of the function and reason about it. With #ifdefs inside of the body of the function, every time you read it you have to filter out all of the "irrelevant junk" that compiler would not even see on your platform.