Re: [PATCH v3 1/4] xen: introduce CONFIG_GENERIC_BUG_FRAME

[Jan Beulich <jbeulich@suse.com>]

On 28.02.2023 11:30, Oleksii wrote:
> On Mon, 2023-02-27 at 15:23 +0100, Jan Beulich wrote:
>> On 24.02.2023 12:31, Oleksii Kurochko wrote:
>>> --- /dev/null
>>> +++ b/xen/common/bug.c
>>> @@ -0,0 +1,109 @@
>>> +#include <xen/bug.h>
>>> +#include <xen/debugger.h>
>>> +#include <xen/errno.h>
>>> +#include <xen/kernel.h>
>>> +#include <xen/livepatch.h>
>>> +#include <xen/string.h>
>>> +#include <xen/types.h>
>>> +#include <xen/virtual_region.h>
>>> +
>>> +#include <asm/processor.h>
>>> +
>>> +/* Set default value for TRAP_invalid_op as it is defined only for
>>> X86 now */
>>> +#ifndef TRAP_invalid_op
>>> +#define TRAP_invalid_op 0
>>> +#endif
>>> +
>>> +int do_bug_frame(const struct cpu_user_regs *regs, unsigned long
>>> pc)
>>> +{
>>> +    const struct bug_frame *bug = NULL;
>>> +    const struct virtual_region *region;
>>> +    const char *prefix = "", *filename, *predicate;
>>> +    unsigned long fixup;
>>> +    unsigned int id = BUGFRAME_NR, lineno;
>>> +
>>> +    region = find_text_region(pc);
>>> +    if ( region )
>>> +    {
>>> +        for ( id = 0; id < BUGFRAME_NR; id++ )
>>> +        {
>>> +            const struct bug_frame *b;
>>> +            unsigned int i;
>>> +
>>> +            for ( i = 0, b = region->frame[id].bugs;
>>> +                  i < region->frame[id].n_bugs; b++, i++ )
>>> +            {
>>> +                if ( bug_loc(b) == pc )
>>> +                {
>>> +                    bug = b;
>>> +                    goto found;
>>> +                }
>>> +            }
>>> +        }
>>> +    }
>>> +
>>> + found:
>>> +    if ( !bug )
>>> +        return -EINVAL;
>>> +
>>> +    if ( id == BUGFRAME_run_fn )
>>> +    {
>>> +#ifdef BUG_FN_REG
>>> +        void (*fn)(const struct cpu_user_regs *) = (void *)regs-
>>>> BUG_FN_REG;
>>> +#else
>>> +        void (*fn)(const struct cpu_user_regs *) = bug_ptr(bug);
>>> +#endif
>>> +
>>> +        fn(regs);
>>> +
>>> +        return id;
>>> +    }
>>> +
>>> +    /* WARN, BUG or ASSERT: decode the filename pointer and line
>>> number. */
>>> +    filename = bug_ptr(bug);
>>> +    if ( !is_kernel(filename) && !is_patch(filename) )
>>> +        return -EINVAL;
>>> +    fixup = strlen(filename);
>>> +    if ( fixup > 50 )
>>> +    {
>>> +        filename += fixup - 47;
>>> +        prefix = "...";
>>> +    }
>>> +    lineno = bug_line(bug);
>>> +
>>> +    switch ( id )
>>> +    {
>>> +    case BUGFRAME_warn:
>>> +        printk("Xen WARN at %s%s:%d\n", prefix, filename, lineno);
>>> +        show_execution_state(regs);
>>> +
>>> +        return id;
>>> +
>>> +    case BUGFRAME_bug:
>>> +        printk("Xen BUG at %s%s:%d\n", prefix, filename, lineno);
>>> +
>>> +        if ( debugger_trap_fatal(TRAP_invalid_op, regs) )
>>
>> TRAP_invalid_op is, as said, about to disappear on x86 as well. I
>> think
>> this construct wants abstracting by another asm/bug.h provided macro
>> (taking just regs).
>>
> Thanks for the link.
> 
> Nice idea to abstract 'debugger_trap_fatal(TRAP_invalid_op, regs)'.
> Actually we have to options here:
> 1. As you proposed abstract in <asm/bug.h>:
>    x86:  #define DEBUG_TRAP_FATAL(regs) debugger_trap_fatal(X86_EXC_GP,
> regs)
>    ARM: #define DEBUG_TRAP_FATAL(regs) 0
>    RISC-V: #define DEBUG_TRAP_FATAL(regs) 0
>   For ARM and RISC-V it doesn't use so we can skip the check if (
> DEBUG_TRAP_FATAL ).
> 
> 2. Abstract only TRAP_invalid_op in <asm/bug.h>
>   x86: #define TRAP_invalud_op X86_EXC_GP
>   RISC-V: #define TRAP_invalid_op 0
>   ARN: #define TRAP_invalid_op 0
>   
>   I am not sure if we have to provide real invalid opcodes for RISC-V
> and ARM as it looks like debug_trap_fatal() isn't used in ARM&RISC-V
> now.
> 
> Could you please suggest which one option is better?

I don't view 2 as a viable option. How an arch deals with invalid opcodes
is entirely arch-specific (including the naming). As to 1 - since we want
this solely for bug.c, I'd prefer if the wrapper macro's name would start
with BUG_, e.g. BUG_DEBUGGER_TRAP_FATAL() or BUG_TRAP_FATAL() or just
BUG_FATAL(). Further adding ARCH_ may also be wanted by other maintainers
(I'm neither pro nor con there).

Jan