Re: [RFC] MIPS: Install final length of TLB refill handler rather than 256 bytes

["Philippe Mathieu-Daudé" <f4bug@amsat.org>]

Hi Fredrik,

On 4/5/19 6:05 PM, Fredrik Noring wrote:
> The R5900 TLB refill handler is limited to 128 bytes, corresponding
> to 32 instructions.

There is a comment about the R4000 worst case:

 /* The worst case length of the handler is around 18 instructions for
  * R3000-style TLBs and up to 63 instructions for R4000-style TLBs.
  * Maximum space available is 32 instructions for R3000 and 64
  * instructions for R4000.

So you need to check the handler generated for your cpu doesn't exceed
your 32 instructions.

> Installing a 256 byte TLB refill handler for the R5900 at address
> 0x80000000 overwrites the performance counter handler at address
> 0x80000080, according to the TX79 manual[1]:
> 
>         Table 5-2. Exception Vectors for Level 1 exceptions
> 
>              Exceptions      |      Vector Address
>                              |   BEV = 0  |   BEV = 1
>         ---------------------+------------+-----------
>         TLB Refill (EXL = 0) | 0x80000000 | 0xBFC00200
>         TLB Refill (EXL = 1) | 0x80000180 | 0xBFC00380
>         Interrupt            | 0x80000200 | 0xBFC00400
>         Others               | 0x80000180 | 0xBFC00380
>         ---------------------+------------+-----------
> 
>         Table 5-3. Exception Vectors for Level 2 exceptions
> 
>              Exceptions      |      Vector Address
>                              |   DEV = 0  |   DEV = 1
>         ---------------------+------------+-----------
>         Reset, NMI           | 0xBFC00000 | 0xBFC00000
>         Performance Counter  | 0x80000080 | 0xBFC00280
>         Debug, SIO           | 0x80000100 | 0xBFC00300
>         ---------------------+------------+-----------
> 
> Reference:
> 
> [1] "TX System RISC TX79 Core Architecture" manual, revision 2.0,
>     Toshiba Corporation, p. 5-7, https://wiki.qemu.org/File:C790.pdf
> 
> Signed-off-by: Fredrik Noring <noring@nocrew.org>
> ---
> Hi MIPS maintainers,
> 
> Reading through the TX79 manual I noticed that the installation of
> the TLB refill handler seems to overwrite the performance counter
> handler. Is there any particular reason to not install the actual
> lengths of the handlers, such as memory boundaries or alignments?
> 
> I have a separate patch that checks the R5900 handler length limit,
> but it depends on R5900 support, which isn't merged (yet).
> 
> Fredrik
> ---
> 
> diff --git a/arch/mips/mm/tlbex.c b/arch/mips/mm/tlbex.c
> --- a/arch/mips/mm/tlbex.c
> +++ b/arch/mips/mm/tlbex.c
> @@ -1462,9 +1462,9 @@ static void build_r4000_tlb_refill_handler(void)
>  	pr_debug("Wrote TLB refill handler (%u instructions).\n",
>  		 final_len);
>  

> -	memcpy((void *)ebase, final_handler, 0x100);
> -	local_flush_icache_range(ebase, ebase + 0x100);
> -	dump_handler("r4000_tlb_refill", (u32 *)ebase, (u32 *)(ebase + 0x100));
> +	memcpy((void *)ebase, final_handler, 4 * final_len);
> +	local_flush_icache_range(ebase, ebase + 4 * final_len);
> +	dump_handler("r4000_tlb_refill", (u32 *)ebase, (u32 *)(ebase + 4 * final_len));

Maybe you could modify the logic few lines earlier that check and
panic("TLB refill handler space exceeded") and add a case for your cpu
type. There you could set a handler_max_size = 0x80, 0x100 else.

Take my comment as RFC too, I'm just wondering :)

Regards,

Phil.

>  }
>  
>  static void setup_pw(void)
> 
>