Re: [PATCH] Fix for TLB errata on early Xilinx Virtex-II Pro silicon

[Andrei Konovalov <akonovalov@ru.mvista.com>]

Grant Likely wrote:
> Early versions of the Xilinx Virtex-II Pro have a TLB errata where
> only even numbered TLB entries work correctly.  Occurs on chips where
> PVR == 0x20010820 || 0x20010860
> 
> See Record #14052, solution #12 in the Xilinx answers database
> http://www.xilinx.com/xlnx/xil_ans_display.jsp?getPagePath=14052
> 
> This patch adds a config option to use only even TLB entries on the V2Pro
> It also makes a trivial change to the Kconfig so that Xilinx options depend
> on VIRTEX_II_PRO instead of XILINX_ML300
> 
> Signed-off-by: Grant Likely <grant.likely@gdcanada.com>
> ---
> 
>  arch/ppc/kernel/head_4xx.S     |   11 +++++++++++
>  arch/ppc/platforms/4xx/Kconfig |   28 ++++++++++++++++++++++------
>  2 files changed, 33 insertions(+), 6 deletions(-)
> 
> 322a82cd190a777e4ebe728cad2a2a3759039260
> diff --git a/arch/ppc/kernel/head_4xx.S b/arch/ppc/kernel/head_4xx.S
> --- a/arch/ppc/kernel/head_4xx.S
> +++ b/arch/ppc/kernel/head_4xx.S
> @@ -769,7 +769,11 @@ finish_tlb_load:
>         /* load the next available TLB index.
>         */
>         lwz     r9, tlb_4xx_index@l(0)
> +#if defined(CONFIG_VIRTEX_II_PRO_TLB_FIX)
> +       addi    r9, r9, 2
> +#else
>         addi    r9, r9, 1
> +#endif
>         andi.   r9, r9, (PPC4XX_TLB_SIZE-1)
>         stw     r9, tlb_4xx_index@l(0)

I would also fix the comment in this file
(it has nothing to do with the TLB fix but...):

@@ -915,10 +919,10 @@ initial_mmu:
	mtspr	SPRN_PID,r0
	sync

-	/* Configure and load two entries into TLB slots 62 and 63.
-	 * In case we are pinning TLBs, these are reserved in by the
+	/* Configure and load an entry into TLB slot 63.
+	 * In case we are pinning TLBs, it is reserved in by the
	 * other TLB functions.  If not reserving, then it doesn't
-	 * matter where they are loaded.
+	 * matter where it is loaded.
	 */
	clrrwi	r4,r4,10		/* Mask off the real page number */
	ori	r4,r4,(TLB_WR | TLB_EX)	/* Set the write and execute bits */

> @@ -926,7 +930,14 @@ initial_mmu:
>         clrrwi  r3,r3,10                /* Mask off the effective page number */        ori     r3,r3,(TLB_VALID | TLB_PAGESZ(PAGESZ_16M))
> 
> +#if defined(CONFIG_VIRTEX_II_PRO_TLB_FIX)
> +       /* Odd numbered TLB slots are broken on Xilinx V2Pro processors
> +        * where PVR = 20010820 | 20010860
> +        */
> +        li      r0,62                    /* TLB slot 62 */
> +#else
>          li      r0,63                    /* TLB slot 63 */
> +#endif
> 
>         tlbwe   r4,r0,TLB_DATA          /* Load the data portion of the entry */        tlbwe   r3,r0,TLB_TAG           /* Load the tag portion of the entry */
> diff --git a/arch/ppc/platforms/4xx/Kconfig b/arch/ppc/platforms/4xx/Kconfig
> --- a/arch/ppc/platforms/4xx/Kconfig
> +++ b/arch/ppc/platforms/4xx/Kconfig
> @@ -161,11 +161,6 @@ config IBM_OCP
>         depends on ASH || BAMBOO || BUBINGA || CPCI405 || EBONY || EP405 || LUAN || OCOTEA || REDWOOD_5 || REDWOOD_6 || SYCAMORE || WALNUT
>         default y
> 
> -config XILINX_OCP
> -       bool
> -       depends on XILINX_ML300
> -       default y
> -
>  config IBM_EMAC4
>         bool
>         depends on 440GX || 440SP
> @@ -201,6 +196,27 @@ config VIRTEX_II_PRO
>         depends on XILINX_ML300
>         default y
> 
> +config VIRTEX_II_PRO_TLB_FIX
> +       bool "Virtex-II Pro TLB bugfix"
> +       depends on VIRTEX_II_PRO
> +       default n
> +       help
> +         Early versions of the Xilinx Virtex-II Pro have a TLB errata where
> +         only even numbered TLB entries work correctly.  Say Y here if
> +         PVR == 0x20010820 || 0x20010860, or if your board crashes early
> +         after enabling the MMU
> +
> +         See Record #14052, solution #12 in the Xilinx answers database
> +         http://www.xilinx.com/xlnx/xil_ans_display.jsp?getPagePath=14052
> +
> +         It is safe to say Y here, but there is a performance impact.
> +         Say N if unsure.
> +
> +config XILINX_OCP
> +       bool
> +       depends on VIRTEX_II_PRO
> +       default y
> +
>  config STB03xxx
>         bool
>         depends on REDWOOD_5 || REDWOOD_6
> @@ -208,7 +224,7 @@ config STB03xxx
> 
>  config EMBEDDEDBOOT
>         bool
> -       depends on EP405 || XILINX_ML300
> +       depends on EP405 || VIRTEX_II_PRO
>         default y
> 
>  config IBM_OPENBIOS

I would drop the last chunk.
I agree that VIRTEX_II_PRO_TLB_FIX and XILINX_OCP should depend on the
chip (VIRTEX_II_PRO), but using EMBEDDEDBOOT or, say, U-Boot is at least
board specific. Well, even the same board can use different bootloaders,
but tying together EMBEDDEDBOOT and VIRTEX_II_PRO is even worse IMO.

Otherwise the patch looks OK.

These changes to head_4xx.S were tested OK on ML300:
in both cases (odd numbered TLB slots used or not) there were no problems.
(For some unknown reason my ML300 doesn't need this workaround though
'cat /proc/cpuinfo' gets "revision        : 8.32 (pvr 2001 0820)")

Thanks,
Andrei

> _______________________________________________
> Linuxppc-embedded mailing list
> Linuxppc-embedded@ozlabs.org
> https://ozlabs.org/mailman/listinfo/linuxppc-embedded