riscv: Fix debug instruction check and support trap-based WARN()
cover riscv: Fix debug instruction check and support trap-based WARN()
The handler for the debug exception will call is_valid_bugaddr(bugaddr) to check if the instruction in bugaddr is a real debug instruction. However, the expected instruction, ebreak, is possibly translated to c.ebreak by assmebler when C extension is supported. This patchset will add c.ebreak into the check mechanism. In addition, BUG() is currently unable to work in the kernel module due to an inappropriated condition in is_valid_bugaddr(). This issue will be fixed in this patchset. Finally, this patchset enables WARN() related functions to trap the code to help developers debug it.
Vincent Chen (3):
riscv: Add the support for c.ebreak check in is_valid_bugaddr()
riscv: Support BUG() in kernel module
riscv: Make WARN() related functions able to trigger a trap exception
_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
PATCH 1/3 riscv: Add the support for c.ebreak check in is_valid_bugaddr()
The is_valid_bugaddr() function compares the instruction pointed to by $sepc with macro __BUG_INSN to check whether the current trap exception is caused by an "ebreak" instruction. Hence the macro __BUG_INSN is defined as "ebreak". However, this check flow is possibly erroneous because the expected trap instruction possibly be translated to "c.ebreak" by assembler if C extension is supported. Therefore, it requires a mechanism to distinguish the length of the instruction in $spec and compare it to the correct __BUG_INSN.
By the way, I make the kernel go to die() like BUG() when the trap type is BUG_TRAP_TYPE_WARN because currently the WARN() does not trigger any trap exception.
--- arch/riscv/include/asm/bug.h | 7 ++++++- arch/riscv/kernel/traps.c | 8 +++++--- 2 files changed, 11 insertions(+), 4 deletions(-)
arch/riscv/include/asm/bug.h +6 -1
@@ -21,7 +21,12 @@#include <asm/asm.h> #ifdef CONFIG_GENERIC_BUG -#define __BUG_INSN _AC(0x00100073, UL)/* ebreak */ +#define __INSN_LENGTH_MASK _UL(0x3) +#define __INSN_LENGTH_32 _UL(0x3) +#define __COMPRESSED_INSN_MASK _UL(0xffff) + +#define __BUG_INSN_32 _UL(0x00100073)/* ebreak */ +#define __BUG_INSN_16 _UL(0x9002)/* c.ebreak */ #ifndef __ASSEMBLY__typedef u32 bug_insn_t ;
arch/riscv/kernel/traps.c +5 -3
@@ -129,8 +129,7 @@ asmlinkage void do_trap_break(struct pt_regs *regs)case BUG_TRAP_TYPE_NONE :break ;case BUG_TRAP_TYPE_WARN : - regs->sepc +=sizeof (bug_insn_t); -return ; +die (regs,"Kernel BUG. Kernel got an unexpected WARN trapped by ebreak" );case BUG_TRAP_TYPE_BUG :die (regs,"Kernel BUG" ); }
@@ -149,7 +148,10 @@ int is_valid_bugaddr(unsigned long pc)return 0 ;if (probe_kernel_address ((bug_insn_t *)pc, insn))return 0 ; - return (insn == __BUG_INSN); +if ((insn & __INSN_LENGTH_MASK) == __INSN_LENGTH_32) +return insn == __BUG_INSN_32; +else + return (insn & __COMPRESSED_INSN_MASK) == __BUG_INSN_16; }#endif /* CONFIG_GENERIC_BUG */
Signature · 7 lines
1.7.1 _______________________________________________ linux-riscv mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-riscv
PATCH 2/3 riscv: Support BUG() in kernel module
In reply to Vincent ChenThe kernel module is loaded into vmalloc region which is located below to the PAGE_OFFSET. Hence the condition, pc < PAGE_OFFSET, in the is_valid_bugaddr() will filter out all trap exceptions triggered by kernel module. To support BUG() in kernel module, the condition is changed to pc < VMALLOC_START.
--- arch/riscv/kernel/traps.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
arch/riscv/kernel/traps.c +1 -1
@@ -144,7 +144,7 @@ int is_valid_bugaddr(unsigned long pc) {bug_insn_t insn; -if (pc <PAGE_OFFSET ) +if (pc <VMALLOC_START )return 0 ;if (probe_kernel_address ((bug_insn_t *)pc, insn))return 0 ;
Signature · 7 lines
1.7.1 _______________________________________________ linux-riscv mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-riscv
PATCH 3/3 riscv: Make WARN() related functions able to trigger a trap exception
In reply to Vincent ChenThis can help developers to analyze the cause of WARN() because the control will be transferred to debugging environment if the debugger is connected.
--- arch/riscv/include/asm/bug.h | 27 ++++++++++++++++++--------- arch/riscv/kernel/traps.c | 19 ++++++++++++++++--- 2 files changed, 34 insertions(+), 12 deletions(-)
arch/riscv/include/asm/bug.h +18 -9
@@ -43,38 +43,47 @@#define __BUG_ENTRY \ __BUG_ENTRY_ADDR "\n\t" \ __BUG_ENTRY_FILE "\n\t" \ - RISCV_SHORT " %1" + RISCV_SHORT " %1 \n\t" \ + RISCV_SHORT " %2"#else #define __BUG_ENTRY \ - __BUG_ENTRY_ADDR + __BUG_ENTRY_ADDR "\n\t" \ + RISCV_SHORT " %2"#endif -#define BUG () \ +#define __BUG_FLAGS (flags) \ do { \ __asm__ __volatile__ ( \ "1:\n\t" \ "ebreak\n" \ - ".pushsection __bug_table,\"a\"\n\t" \ + ".pushsection __bug_table,\"aw\"\n\t" \ "2:\n\t" \ __BUG_ENTRY "\n\t" \ - ".org 2b + %2\n\t" \ + ".org 2b + %3\n\t" \ ".popsection" \ : \ : "i" (__FILE__), "i" (__LINE__), \ - "i" (sizeof(struct bug_entry))); \ - unreachable(); \ + "i" (flags), \ + "i" (sizeof(struct bug_entry))); \ } while (0) +#endif /* !__ASSEMBLY__ */ #else /* CONFIG_GENERIC_BUG */ #ifndef __ASSEMBLY__ -#define BUG () \ +#define __BUG_FLAGS (flags) \ do { \ __asm__ __volatile__ ("ebreak\n"); \ - unreachable(); \ } while (0)#endif /* !__ASSEMBLY__ */ #endif /* CONFIG_GENERIC_BUG */ +#define BUG () do { \ + __BUG_FLAGS(0); \ + unreachable(); \ +} while (0) + +#define __WARN_FLAGS (flags) __BUG_FLAGS(BUGFLAG_WARNING|(flags)) +#define HAVE_ARCH_BUG #include <asm-generic/bug.h>
arch/riscv/kernel/traps.c +16 -3
@@ -118,6 +118,17 @@ asmlinkage void name(struct pt_regs *regs) \DO_ERROR_INFO (do_trap_ecall_m,SIGILL ,ILL_ILLTRP ,"environment call from M-mode" ); +#ifdef CONFIG_GENERIC_BUG +static inline unsigned long get_break_insn_length (unsigned long pc) +{ +bug_insn_t insn; + +if (probe_kernel_address ((bug_insn_t *)pc, insn)) +return 0 ; +return ((insn & __INSN_LENGTH_MASK) == __INSN_LENGTH_32) ?4UL :2UL ; +} +#endif /* CONFIG_GENERIC_BUG */ + asmlinkage void do_trap_break(struct pt_regs *regs) {#ifdef CONFIG_GENERIC_BUG
@@ -129,7 +140,8 @@ asmlinkage void do_trap_break(struct pt_regs *regs)case BUG_TRAP_TYPE_NONE :break ;case BUG_TRAP_TYPE_WARN : -die (regs,"Kernel BUG. Kernel got an unexpected WARN trapped by ebreak" ); + regs->sepc +=get_break_insn_length (regs->sepc); +break ;case BUG_TRAP_TYPE_BUG :die (regs,"Kernel BUG" ); }
@@ -149,12 +161,13 @@ int is_valid_bugaddr(unsigned long pc)if (probe_kernel_address ((bug_insn_t *)pc, insn))return 0 ;if ((insn & __INSN_LENGTH_MASK) == __INSN_LENGTH_32) -return insn == __BUG_INSN_32; +return (insn == __BUG_INSN_32);else - return (insn & __COMPRESSED_INSN_MASK) == __BUG_INSN_16; + return ((insn & __COMPRESSED_INSN_MASK) == __BUG_INSN_16); }#endif /* CONFIG_GENERIC_BUG */ +void __init trap_init (void ) {/*
Signature · 7 lines
1.7.1 _______________________________________________ linux-riscv mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-riscv
Quoted 13 lines · On Thu, 28 Feb 2019 02:31:28 PST (-0800), [email protected] wrote:
The handler for the debug exception will call
is_valid_bugaddr(bugaddr)to check if the instruction in bugaddr is a real debug instruction. However, the expected instruction, ebreak, is possibly translated to c.ebreak by assmebler when C extension is supported. This patchset will add c.ebreak into the check mechanism. In addition,BUG()is currently unable to work in the kernel module due to an inappropriated condition inis_valid_bugaddr(). This issue will be fixed in this patchset. Finally, this patchset enablesWARN()related functions to trap the code to help developers debug it.
Quoted 4 lines
Vincent Chen (3):
riscv: Add the support for c.ebreak check inis_valid_bugaddr()
riscv: SupportBUG()in kernel module
riscv: MakeWARN()related functions able to trigger a trap exception
I'm finding this patch set a bit hard to follow, and I think it has more diff than is necessary. For example, the first patch introduces a new die() only to have it removed by the third patch. There's also some unnecessary non-functional diff, like
@@ -149,12 +161,13 @@ int is_valid_bugaddr(unsigned long pc)
if (probe_kernel_address((bug_insn_t *)pc, insn))
return 0;
if ((insn & __INSN_LENGTH_MASK) == __INSN_LENGTH_32)
- return insn == __BUG_INSN_32;
+ return (insn == __BUG_INSN_32);
else
- return (insn & __COMPRESSED_INSN_MASK) == __BUG_INSN_16;
+ return ((insn & __COMPRESSED_INSN_MASK) == __BUG_INSN_16);
}
#endif /* CONFIG_GENERIC_BUG */ +
void __init trap_init(void)
{
/*I like the idea of the patch set, though. Do you have time to clean it up and submit a v2?
Quoted 42 lines · Palmer Dabbelt
Quoted 13 lines · On Thu, 28 Feb 2019 02:31:28 PST (-0800), [email protected] wrote:
The handler for the debug exception will call
is_valid_bugaddr(bugaddr)to check if the instruction in bugaddr is a real debug instruction. However, the expected instruction, ebreak, is possibly translated to c.ebreak by assmebler when C extension is supported. This patchset will add c.ebreak into the check mechanism. In addition,BUG()is currently unable to work in the kernel module due to an inappropriated condition inis_valid_bugaddr(). This issue will be fixed in this patchset. Finally, this patchset enablesWARN()related functions to trap the code to help developers debug it.Quoted 4 lines
Vincent Chen (3):
riscv: Add the support for c.ebreak check inis_valid_bugaddr()
riscv: SupportBUG()in kernel module
riscv: MakeWARN()related functions able to trigger a trap exceptionI'm finding this patch set a bit hard to follow, and I think it has more diff than is necessary. For example, the first patch introduces a new
die()only to have it removed by the third patch. There's also some unnecessary non-functional diff, like@@ -149,12 +161,13 @@ int is_valid_bugaddr(unsigned long pc) if (probe_kernel_address((bug_insn_t *)pc, insn)) return 0; if ((insn & __INSN_LENGTH_MASK) == __INSN_LENGTH_32) - return insn == __BUG_INSN_32; + return (insn == __BUG_INSN_32); else - return (insn & __COMPRESSED_INSN_MASK) == __BUG_INSN_16; + return ((insn & __COMPRESSED_INSN_MASK) == __BUG_INSN_16); } #endif /* CONFIG_GENERIC_BUG */ + void __init trap_init(void) { /*
Quoted 2 lines
I like the idea of the patch set, though. Do you have time to clean it up and submit a v2?
OK, I will improve it and submit a v2 patch.
_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv