Re: [PATCH 10/29] tcg_funcs: Add tlb_flush to TCGModuleOps

[Richard Henderson]

On 9/29/21 3:09 AM, Gerd Hoffmann wrote:
> TODO: Not sure whenever "check function pointer before call" or
> "function pointers to stubs" is better.  Right now the patch has both
> which clearly doesn't make sense.  Comments on that are welcome.

I guess I don't mind either way, but we do need to pick one.

> --- a/include/exec/exec-all.h
> +++ b/include/exec/exec-all.h
> @@ -25,6 +25,7 @@
>   #include "exec/cpu_ldst.h"
>   #endif
>   #include "sysemu/cpu-timers.h"
> +#include "tcg/tcg-module.h"

Surely not required here.

> +/**
> + * tlb_flush:
> + * @cpu: CPU whose TLB should be flushed
> + *
> + * Flush the entire TLB for the specified CPU. Most CPU architectures
> + * allow the implementation to drop entries from the TLB at any time
> + * so this is generally safe. If more selective flushing is required
> + * use one of the other functions for efficiency.
> + */
> +void tlb_flush(CPUState *cpu);
> +
>   #if !defined(CONFIG_USER_ONLY) && defined(CONFIG_TCG)

Going too far here, since CONFIG_USER_ONLY always uses TCG, never uses modules, and does 
not implement a softmmu tlb that requires flushing.  I think you can just change the ifdef 
to remove the CONFIG_TCG check.

Alternately, we could keep this slight inefficiency to fix...

> +
> +/* This is a wrapper for common code that can not use CONFIG_SOFTMMU */
> +void tcg_flush_softmmu_tlb(CPUState *cs)
> +{
> +#ifdef CONFIG_SOFTMMU
> +    tlb_flush(cs);
> +#endif
> +}

... this bit of silliness.


r~